Sie sind auf Seite 1von 14

ESWA 9208

No. of Pages 14, Model 5G

20 March 2014
Expert Systems with Applications xxx (2014) xxxxxx
1

Contents lists available at ScienceDirect

Expert Systems with Applications


journal homepage: www.elsevier.com/locate/eswa
5
6

Q1

A data mining approach to identify cognitive NeuroRehabilitation Range


in Traumatic Brain Injury patients

Q2

Alejandro Garca-Rudolph a,1, Karina Gibert b,

8
9
10
11
1
2 3
0
14
15
16
17
18
19

a
b

Institut Guttmann, Hospital de Neurorehabilitaci, Cami Can Ruti s/n, 08916 Badalona, Barcelona, Spain
Departament dEstadstica i Investigaci Operativa, Universitat Politcnica de Catalunya BarcelonaTech, Jordi Girona 1-3, 08034 Barcelona, Spain

a r t i c l e

i n f o

Keywords:
Machine learning
Algorithms
Statistics

a b s t r a c t
Cognitive rehabilitation (CR) treatment consists of hierarchically organized tasks that require repetitive
use of impaired cognitive functions in a progressively more demanding sequence. Active monitoring of
the progress of the subjects is therefore required, and the difculty of the tasks must be progressively
increased, always pushing the subjects to reach a goal just beyond what they can attain. There is an
important lack of well-established criteria by which to identify the right tasks to propose to the patient.
In this paper, the NeuroRehabilitation Range (NRR) is introduced as a means of identifying formal
operational models. These are to provide the therapist with dynamic decision support information for
assigning the most appropriate CR plan to each patient. Data mining techniques are used to build
data-driven models for NRR. The Sectorized and Annotated Plane (SAP) is proposed as a visual tool by
which to identify NRR, and two data-driven methods to build the SAP are introduced and compared.
Application to a specic representative cognitive task is presented. The results obtained suggest that
the current clinical hypothesis about NRR might be reconsidered. Prior knowledge in the area is taken
into account to introduce the number of task executions and task performance into NRR models and a
new model is proposed which outperforms the current clinical hypothesis. The NRR is introduced as a
key concept to provide an operational model identifying when a patient is experiencing activities in
his or her Zone of Proximal Development and, consequently, experiencing maximum improvement.
For the rst time, data collected through a CR platform has been used to nd a model for the NRR.
2014 Elsevier Ltd. All rights reserved.

21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40

41
42

1. Introduction

43

Acquired Brain Injury (ABI) of either vascular or traumatic nature is one of the most important causes of neurological disabilities.
According to the World Health Organization, Traumatic Brain Injury (TBI) is the leading cause of death and disability in children
and young adults around the world and is a factor in nearly half
of all trauma deaths (Walsh, Donal, Stephen, & Muldoon, 2012).
In Europe, brain injuries from trauma are responsible for more
years of disability than any other cause (Maas, Stocchetti, &
Bullock, 2008).
Despite new techniques for early intervention and intensive
ABI, both of which increase the survival rate, there is still no surgical or pharmacological treatment for the re-establishment of lost
functions following brain injury. Cognitive rehabilitation (CR) is

44
45
46
47
48
49
50
51
52
53
54
55

Corresponding author. Tel.: +34 93 401 73 23; fax: +34 93 401 58 55.
E-mail addresses: agarciar@guttmann.com (A. Garca-Rudolph), karina.gibert@
upc.edu (K. Gibert).
1
Tel.: +34 93 497 77 00; fax: +34 93 497 77 07.

currently considered the therapeutic process for re-establishing


functioning in everyday life (Pascual-Leone & et al., 2005). A typical
CR program mainly provides exercises which require repetitive use
of the impaired cognitive system in a progressively more demanding (Sohlberg, 2001) sequence of tasks. The rehabilitating impact of
a task or exercise depends on the ratio between the skills of the
treated patient and the challenges involved in the execution of
the task itself. Thus, determining the correct training schedule requires a quite precise trade-off between sufcient stimulation and
sufciently achievable tasks, which is far from intuition, and is still
an open issue, both empirically and theoretically (Green & Bavelier,
2005). It is difcult to identify this maximum effective level of
stimulation and therapists use their expertise in daily practice,
without precise guidelines on these issues.
In this work, the NeuroRehabilitation Range (NRR) is introduced
as the conceptual framework to describe the degree of performance of a CR task that produces maximum rehabilitation effects.
A data mining approach is used to induce an operational model for
the NRR of CR tasks. The aim is to help create useful guidelines for
CR therapists that can help them select the most appropriate tasks

http://dx.doi.org/10.1016/j.eswa.2014.03.001
0957-4174/ 2014 Elsevier Ltd. All rights reserved.

Please cite this article in press as: Garca-Rudolph, A., & Gibert, K. A data mining approach to identify cognitive NeuroRehabilitation Range in Traumatic

Q1 Brain Injury patients. Expert Systems with Applications (2014), http://dx.doi.org/10.1016/j.eswa.2014.03.001

56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75

ESWA 9208

No. of Pages 14, Model 5G

20 March 2014
2
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104

A. Garca-Rudolph, K. Gibert / Expert Systems with Applications xxx (2014) xxxxxx

for each single patient at a given moment in their rehabilitation


plan, as well as correctly to determine the most appropriate level
of difculty for the proposed task.
The Sectorized and Annotated Plane (SAP) is proposed here as a
visual tool to nd both the NRR and operational denitions to be
used in real clinical practice.
Two data-driven methods to build the SAP are introduced and
compared. One of them (DT-SAP) is based on a decision tree model,
the other (Vis-SAP) on a visualization of available data that promotes model induction from a graphical representation. A quality
criterion to assess NRR models is also introduced, based on the correct prediction ratio provided by the tool.
The performance of NRR model obtained with both DT-SAP and
Vis-SAP approaches is evaluated and the advantages and drawbacks are analyzed over a real application.
Data comes from the PREVIRNEC platform (Tormos, GarciaMolina, Garcia Rudolph, & Roig, 2009) which contains rich data
monitoring the CR process on real neurorehabilitation patients.
The real performance of a representative cognitive task is analyzed
under both approaches and discussed for a sample of patients following a CR treatment at Institut Guttmann (IG) hospital de Neurorehabilitaci, Barcelona, Spain.
The structure of the paper is: Section 2 briey presents the state
of the art. Section 3 presents the IG conceptual framework for the
research of NRR. Section 4 introduces the analysis methodology
and Section 5 its application to a typical cognitive rehabilitation
task in the proposed framework. Section 6 presents a discussion
of the results obtained and Section 7 the conclusions and future
lines of research.

105

2. State of the art

106

CR, as part of neuropsychological rehabilitation, tries to improve the decits caused by ABI in daily living activities (Bernabeu
& Roig, 1999) by retraining attention, memory, reasoning/problem
solving, and executive functions. The plasticity of the central nervous system plays a central role (Pascual-Leone et al., 2005) in
CR, based on therapeutic plans to stimulate that non-damaged
neurons can modify their structure by learning from experience
the damaged functions, through repetition (Luria, 1978). Plasticity
may represent a surrogate marker of functional recovery, indicating behavioral change that is resistant to decay. In Kleim and Jones
(2008) is suggested that a sufcient level of rehabilitation is likely
to be required in order to get the subject over the hump i.e. repetition may be needed to obtain a sufcient level of improvement and
brain reorganization for the patient to continue using the affected
function outside of therapy and to achieve and maintain further
functional gains. A great deal of research indicates that behavioral
experience can enhance behavioral performance and optimize
restorative brain plasticity after brain damage. Simply engaging a
neural circuit in task performance is not sufcient to drive plasticity. Repetition of a newly learned (or relearned) behavior may be
required to induce lasting neural changes. In fact, from the experts
point of view, there is a clear perception that the effectiveness of
the task also depends on the replication, as Luria also asserts.
A typical CR program mainly provides exercises that require
repetitive use of the impaired cognitive system in a progressively
more demanding (Sohlberg, 2001) sequence of tasks. Each task targets a principal cognitive function and can be performed at different levels of difculty, according to the response of the patient. The
design of a CR program has become an essential issue for patient
recovery.
As said before, the rehabilitating effect of a task or exercise depends on the ratio between the skills of the treated patient and
the challenges involved in the execution of the task itself. The dif-

107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138

culty is related to the level of stimulation of cognitively involved


functions; maximum activation occurs when the task is just barely
too difcult (Green & Bavelier, 2005). If the task is too easy for the
patient, or too hard, it appears to be less effective. Active monitoring of the subjects progress is therefore required to adapt the difculty of the tasks to the potential capacities and progress of the
subject, always pushing them to reach a goal just beyond what they
can attain, but not too far. Thus, determining the correct training
schedule requires a very precise trade-off between sufciently stimulating and sufciently achievable tasks, which is far from intuitive,
and is still an open problem, both empirically and theoretically.
In the early 1930s, Vygotsky introduced the concept of Zone of
Proximal Development (ZPD) (Vygotsky, 1934) in the eld of child
learning, being the distance between the actual capacities of the child
by himself and their potential capacities when being guided
(Vygotsky, 1978). In 1986, Cicerone and Tupper (1986) transferred
ZPD ideas to the neurorehabilitation eld by introducing the zone
of rehabilitation potential (ZRP), i.e. the zone in which maximum
recovery of cognitive functions might occur, provided that the
proper help is given to the subject. They propose the use of ZPD
as a guiding principle in CR. This zone is supposed to reect the patients region of potential restoration thanks to cognitive plasticity
(Calero & Navarro, 2007). Current neurorehabilitation practice tries
to design therapeutic plans that keep the subject working in this
area during treatment. However, determining when the patient
works in ZPD or not is still an open issue. Thus in most cases CR
therapists design CR plans from scratch, determining clinical settings for specic patients based mainly on their own expertise.
Each specic plan evolves according to each therapists own criteria and evaluation of the patients follow-up. There is as yet not enough in-eld knowledge regarding which specic intervention
(task or exercise assignation) is more appropriate to help CR therapists design their clinical therapeutic plans.
There is a common belief that CR is effective for TBI patients,
based on a large number of studies and extensive clinical experience. Different statistical methodologies and predictive data mining methods have been applied to predict clinical outcomes of
TBI rehabilitation (Rughani et al., 2010; Ji, Smith, Huynh,& Najarian, 2009; Pang et al., 2007; Segal et al., 2006; Brown et al., 2005;
Rovlias & Kotsou, 2004; Andrews et al., 2002). Most of these stud- Q3
ies focus on determining survival, predicting disability or the
recovery of patients, and looking for the factors that better predict
the patients condition after an ABI.
However, current knowledge about the factors that determine a
favorable outcome is mainly empirical and the benet of such
interventions is still controversial (Ecri, 2011; Rohling, Faust,
et al., 2009). The development of new tools to evaluate scientic
evidence of such effectiveness will contribute to a better understanding of CR.
Several meta-analyses (Cicerone, Langenbahn, Braden, Malec, &
Kalmar, 2011) identify structural limitations to nd scientic evidence under classical approaches, related mainly to the existence
of uncontrolled factors and the intrinsic difculty of guaranteeing
the sample heterogeneity. Classical approaches tend to generate
evidence about effectiveness by comparing two or more interventions in selected and comparable groups. Determining the comparable groups relies on identifying the factors that inuence
recovery or chronicity, which should be controlled during the
study, and these factors are unknown in neurorehabilitation. It
seems that patient improvement might depend inter alia on the
location of the injuries, cognitive prole, duration, and intensity
of proposed treatments and their level of completion (Cicerone
et al., 2011; Norea et al., 2010; Whyte & Hart, 2003).
However, these seem to be only some of the determining factors
and they cannot by themselves explain the overall phenomenon.
Although these factors are considered in the design of rehabilitation

Please cite this article in press as: Garca-Rudolph, A., & Gibert, K. A data mining approach to identify cognitive NeuroRehabilitation Range in Traumatic

Q1 Brain Injury patients. Expert Systems with Applications (2014), http://dx.doi.org/10.1016/j.eswa.2014.03.001

139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204

ESWA 9208

No. of Pages 14, Model 5G

20 March 2014
A. Garca-Rudolph, K. Gibert / Expert Systems with Applications xxx (2014) xxxxxx

265

treatments, other relevant factors exist that are much more difcult
to control, and which are related to the high variability of the lesions, the complexity of cognitive functions, and the lack of proper
instrumentation by which to systematize interventions. This produces intrinsic group heterogeneity and the classical comparative
studies do not perform well (Gibert & Garca-Rudolph, 2007), which
makes it difcult to advances knowledge on the pathophysiology of
cognitive neurorehabilitation.
In Serra et al. (2013) basic machine learning algorithms were
used to predict the probability of improvement of a patient given
their initial neuropsychological assessment. This approach,
although it is able to identify subpopulations of patients more suitable for improvement using CR treatments, did not provide any
information to help CR therapists adapt CR programs to increase
the improvement itself or to enlarge the subpopulations that might
activate improvement to CR treatments. Going a little bit further, in
Marcano, Chausa, Garcia-Rudolph, Cceres, and Tormos (2013) the
performance obtained by the patient in a certain task has been included in the model together with the initial assessment. Machine
learning methods signicantly improved predictive capacity. This
work provided evidence that task performance is involved in patient improvement; However, it did not provide information on
successful patterns of tasks to be proposed to the patients so that
they improved as much as possible.
For these reasons, other approaches have to be found to better
understand the CR process, with the aim of obtaining scientic evidence about its effectiveness and providing relevant information
for the establishment of general guidelines for CR program design
that can assist CR therapists in clinical practice.
Analyzing data from new perspectives can contribute to this
eld (Jagaroo, 2009). Our proposal is trying to approach the problem from a data-driven perspective, by developing new tools that
can reduce uncertainty in the eld. This paper introduces elements
to assess when a patient is performing a task under a NeuroRehabilitation Range, as an indicator that maximum improvement of
the patient might be expected on the targeted cognitive function.
This contributes to a better understanding of the role over clinical
improvement of a particular degree of performance of a CR task. In
fact, the NRR helps provide an operational denition for the zone of
maximum rehabilitation potential and represents an operationalization of the ZPD.
The work is based on the experience of Institut Guttmann Neurorehabilitation Hospital (IG) regarding the introduction of Information and Communication Technologies (ICTs) in CR. Data used
in this work comes from a CR computerized platform conceived
by IG and developed in collaboration with clinical and technological
partners, namely PREVIRNEC a serious game platform for CR (Tormos et al., 2009). PREVIRNEC is specically designed to manage
the CR plans assigned to subjects, as well as obtaining precise follow-up information about the process. At the time of submission
(December 2013) PREVIRNEC has already been integrated into
the clinical practices of more than 24 clinical centers. As the whole
behavior of the patients working in PREVIRNEC is registered in a
central server, this provides a unique database of on-eld interventions, with detailed information that is potentially valuable for a
better understanding of the circumstances under which CR therapy
would be most benecial to individual patients. The PREVIRNEC
database permits knowledge extraction from data, and provides a
framework in which new knowledge can be introduced into the
system to be veried and continuously rened, also contributing
to elaborate data-driven personalized treatments.

266

2.1. Serious games in cognitive rehabilitation

267

According to Lurias theory outlined above, repeated taxing of the


same neurological system facilitates and guides the reorganization

205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264

268

of the targeted cognitive function. This approach requires implementation of repetitive exercises within the planned program which
require patients to use their impaired cognitive skills at a productive
level. In this work therefore, given the initial assessment of the
patient impairments, the therapists assign repetitive tasks to be executed in PREVIRNEC, a serious game platform for cognitive
rehabilitation.
The emergence of serious games broadens the discipline of
entertainment-education in numerous dimensions. Serious games
have recently been applied in diverse areas e.g. military training,
health, higher education, city planning (Rego, Moreira, & Reis,
2010). Prior research demonstrates that videogame attributes, such
as task difculty, realism, and interactivity, affect learning outcomes in game-based learning environments (Orvis, Horn, & Belanich, 2008). These prior works suggest that in order to be most
effective, instructional games should present an optimal level of
difculty to learners. This optimal range of difculty is aligned
with the Vygotskys concept of ZPD, where training should be difcult to the learner, but not beyond his or her capabilities.
Videogames involving the sensory-motor system and problemsolving skills are serious candidates for neuro-rehabilitation and
motor or cognitive training. In Green and Bavelier (2007) several
improvements in gaming activity were identied, from reaction
times to spatial skills. The opportunities for using this kind of media to improve cognitive functions in individuals with particular
needs (as reviewed for surgeons and soldiers) or for training and
retraining of individuals with special health-related problems
(such as young disabled or elder people) involving the nervous system were also highlighted. An improvement in the spatial resolution of attention in videogame players has been observed (Green
& Bavelier, 2007).
A persistent difculty is that training can be more or less efcient depending on how it is administered and this is directly related with tasks difculty management (Linkenhoker & Knudsen,
2002).

269

2.2. Flow in computer-mediated environments

304

Learning is enhanced when the match between the skills of the


learner and the challenges of the subject matter are optimized
(Whalen, 1998). Csikszentmihalyis Flow Theory (Csikszentmihalyi,
1991) provides a framework and vocabulary for understanding the
experiential nexus between the active person and the facilitative
environment. The experience of Flow creates information that
melds actor and activity into one transactive system. In this sense,
Flow may be seen as the experiential dimension of the ZPD (Whalen, 1998).
Flow or optimal experiences, also referred to as the zone
(Csikszentmihalyi, 1991) represents a state of consciousness where
a person is so absorbed in an activity that he or she excels in performance without consciously being aware of his or her every
movement.
Within a Computer Mediated Environment (CME), the
experience of ow in the past 20 years has demonstrated an increase in communication, ofce productivity software on desktop
computers, learning, general web activity, online consumer settings, and online search experiences among others (Finneran &
Zhang, 2005).
The practical implications of the consequences of ow experiences are clear, important, and promising. It is expected that a
good understanding of the ow phenomenon would guide ICTs
designers to build products that lead users to ow experiences. Little research is available concerning the application of data mining
techniques in Flow. In Mathwick and RigdonPlay (2004) cluster
analysis is used to identify a ow cluster comprised of individu-

305

Please cite this article in press as: Garca-Rudolph, A., & Gibert, K. A data mining approach to identify cognitive NeuroRehabilitation Range in Traumatic

Q1 Brain Injury patients. Expert Systems with Applications (2014), http://dx.doi.org/10.1016/j.eswa.2014.03.001

270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303

306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331

ESWA 9208

No. of Pages 14, Model 5G

20 March 2014
4
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346

A. Garca-Rudolph, K. Gibert / Expert Systems with Applications xxx (2014) xxxxxx

als with high Internet search skills and a search task that presents a
high navigational challenge.
From a research perspective however, ow is poorly dened in
CME because of the numerous ways it is conceptualized, operationalized, and measured. Flow experience is associated with a person doing an activity. In traditional ow studies, the activities tend
to be very clear: playing music, climbing a cliff, playing chess or
reading a book. Most existing ow studies in CME do not clearly
differentiate between factors that are related to the task and those
that are related to the artifact.
Thus, there is a need to re-conceptualize ow in CME to consider the uniqueness of the artifacts and the complexity they add
to the ow phenomenon. Indeed one of the aims of this work is
to use PREVIRNEC to produce ow experiences in the subject,
thus incrementing the benets of the neurorehabilitation process.

347

2.3. PAT model

348

As introduced in the previous section, there is a need to re-conceptualize ow in CME to consider the uniqueness of the artifacts
and the complexity they add to the ow phenomenon. As an attempt to re-conceptualization, in Finneran and Zhang (2005) a conceptual model for ow antecedents is proposed: the PersonArtifact-Task (PAT) model.
PAT removes the ambiguities among the ow antecedents by
considering the task and the artifact as separate entities when
looking at the factors that lead to a ow state.
The PAT model considers each of the three main components of
person, artifact, and task independently and their interactions, to
understand the holistic picture of ow antecedents.
The intention is therefore to conceptualize the major components of a person working on a computer-related activity that
can inuence the ow experience the person may have. Individual
differences, which are shown to be important in early non-computer-mediated ow studies, are probably even more important
in CME. According to Finneran and Zhang (2005) much empirical
research is needed to validate or clarify which individual factors
inuence the ow experience and where they occur in the process.

349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367

368

3. Conceptual framework

369

3.1. NeuroRehabilitation Range

370

In Clinical Pharmacokinetics, therapeutic range is dened as a


range of drug concentrations within which the probability of the
desired clinical response is relatively high and the probability of
unacceptable toxicity is relatively low. Within this therapeutic
range the desired effects of the drug are observed. Below it there
is a greater probability that the therapeutic benets are not realized (non-response or treatment-resistance); above it, toxic effects
may occur (DiPiro & Spruill, 2010).
In this work, the concept of NeuroRehabilitation Range (NRR) is
introduced as a translation of the classical therapeutic range from
pharmacology to the eld of neurorehabilitation. The role of pharmacs in disease treatment is assumed in neurology by the role of
neurorehabilitation tasks. The effect of treatment corresponds here
to the restoration of cognitive functions.
Using this analogy, we will consider that a cognitive rehabilitation treatment task behaves in NRR if the desired clinical response
is obtained i.e. if an observable improvement in the targeted cognitive function is registered for the patient. As nding therapeutic
range in pharmacokinetics consists of determining the proper drug
concentration to be administered to a patient, nding NRR of a cognitive rehabilitation task is dened as determining the proper level

371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390

of task difculty to be proposed to the patient to obtain an optimal


cognitive improvement of the targeted cognitive function.
We presume that being able to determine the NRR will provide
a model that can help us to know better the relevant factors determining the ZRP proposed by Cicerone and Tupper (1986).
In PREVIRNEC, following the execution of a given task T the
subject gets a result RT ranging from 0 to 100. Section 3.3.3 details Q4
how this result is obtained, in this section we merely remark that a
0 result denotes the lowest level of task completion and a 100 the
highest. Being the NRR of task T dened as NRR(T) = [r, r+], and
being r, r+ in [0, 100], using a simple test it is easy to determine
whether or not the patient performed the task in NRR:

in NRR RT iff RT 2 NRRT  r 6 RT 6 r


 Tasks that are too easy will produce results higher than r+ and
are probably out of ZRP because they only involve undamaged
brain areas and do not demand impaired cognitive functions
to be activated. In this case, we say the task has been executed
in SupraNeuroRehabilitation Range (SNRR).
 Tasks that are too difcult will produce results lower than
r and are also likely to be out of ZRP. This is because they
intensively required the implication of the impaired brain areas
that cannot react to the excessively difcult cognitive stimulus.
In this case we talk about InfraNeuroRehabilitationRange
(INRR).

391
392
393
394
395
396
397
398
399
400
401
402

403
405
407
406
408
409
410
411
412
413
414
415
416
417
418

Currently, some hypotheses are being tested for the values of


r, r+. The aim of this work is to use data-driven models and PREVIRNEC database to extract useful knowledge to determine a
proper model for NRR(T).

419

3.2. PAT model

423

In this section we identify the elements of the PAT model used


in our application.

424

3.2.1. Person
One hundred and twenty-three Traumatic Brain Injury (TBI) patients with moderate to severe cognitive affectation (according to
Glasgow Coma Scale) and who underwent rehabilitation at IG were
included in this analysis. All subjects gave informed consent to the
neuropsychological procedure, which was approved by IGs Ethical
Committee. The patient (mean age 36.56 6.5, range 1868 years;
91 male and 32 female) diagnosis was made according to the
clinical protocols of the IG Neuropsychological department. All
patients met criteria to initiate IG neuropsychological rehabilitation
treatment. It includes a Neuropsychological Assessment Battery
(NAB), 28 items covering the major cognitive domains (language,
attention, memory and learning, and executive functions)
measured using standardized cognitive tests.
After NAB initial evaluation all patients started a CR program
lasting four to six months based on personalized interventions,
where patients worked in each one of the specic cognitive
domains, considering the degree of the decit and the residual
functional capacity. All patients were administered the same NAB
neuropsychological assessment at the end of the rehabilitation
program. All NAB items are normalized to a 04 scale (where
0 = no affectation, 1 = mild affectation, 2 = moderate affectation,
3 = severe affectation and 4 = acute affectation). Differences
between pre- and post-treatment NAB test scores were used to
measure particular patient improvement in the elds of attention,
memory, and executive functions. Improvement criteria in the
respective cognitive functions are dened in IG cognitive rehabilitation protocols.

426

Please cite this article in press as: Garca-Rudolph, A., & Gibert, K. A data mining approach to identify cognitive NeuroRehabilitation Range in Traumatic

Q1 Brain Injury patients. Expert Systems with Applications (2014), http://dx.doi.org/10.1016/j.eswa.2014.03.001

420
421
422

425

427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453

ESWA 9208

No. of Pages 14, Model 5G

20 March 2014
A. Garca-Rudolph, K. Gibert / Expert Systems with Applications xxx (2014) xxxxxx
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515

3.2.2. Artifact
PREVIRNEC (Tormos et al., 2009) comprises a series of rehabilitation tasks for training different cognitive functions: attention,
memory, executive functions, and language. For each specic task
that the patient executes, the result of the task (0100 real number) is assigned into one of the following three ranges: NRR, INRR,
and SNRR. As a rst hypothesis, PREVIRNEC is currently assuming
that NRR(T) = [65, 85] These limits have been dened according to
the expertise of the CR therapists of making the task difcult enough not to be innocuous, but not so difcult as to be blocking.
To maintain the interest of the patient throughout the execution of tasks, PREVIRNEC includes an algorithm that proposes
tasks to patients while trying to keep them inside NRR limits. PREVIRNEC is a cognitive tele-rehabilitation platform, developed
over an architecture based on web 2.0 technologies. It is conceived
as a tool for the enhancement of cognitive rehabilitation, the
strengthening of the relationship between the neuropsychologist
and the patient, the personalization of treatment, the monitoring
of results, and the performance of tasks. The platform architecture
consists of four main modules that group related functionalities
vertically, sharing the user interface that is personalized depending
on the users role. This interface is also multi-language, with Catalan, Spanish and English already implemented, but being open to
support any other language. The system also has a help module,
which guides the user in order to complete each action. Security
aspects are transversal and have to be taken into account in every
module to keep information and all connections safe, due to the
condentiality concerns of medical applications. The security module is responsible for controlling every access, including the ones
related to the patients Electronic Health Record (EHR). The four
modules are briey described below:
Q5 3.2.2.1. Information management. This module groups functionalities related to the generation and edition of information that depends on the patients EHR, as well as the tests used to
determine the grade of affection of each cognitive function. These
tests are used to dene the affection prole of the patient. In addition, this module controls the assignation of therapies to the patients, determining which computerized tasks a patient has to do
on a certain day. The results of the execution of these tasks are registered in the system, and can then be used by the clinicians to see
the evolution of the therapy, as well as showing graphics and reports related to the completion of the sessions, tasks that have
been used, both global and individual results, and much more.
3.2.2.2. Monitoring. To comply with data protection laws, every action carried out by a user is stored in both the database and also in
a log le, so that the administrator can track every action related to
a patient and their data. The system also offers a module for monitoring the execution of the tasks, so the therapist can then reproduce a task as it was done by the patient. This allows the therapist
to see exactly what a patient did in the monitored task. This is very
useful because sometimes merely seeing the numeric results is not
enough.
3.2.2.3. Administration. This module, although it is the one with
fewer functionalities and users, includes very important functionalities such as the users management and their proles, as well
as system monitoring (using logs).
3.2.2.4. Communication. The main element of this module is the video conference that allows users to communicate using video,
audio, and chat. Using the videoconference therapists can hold
tele-appointments with patients or other therapists, removing
the distance barriers between users, and helping the patients to
feel closer to the clinical team. In addition to the videoconference,

this module has a mailing service for the exchange of internal


asynchronous messages and an alert service that lets users know
what tasks they have to accomplish.
The platform is based on open source web 2.0 technologies. The
main architecture of the platform is based on a clientserver communication using HTTP and XML-RPC, as it is shown in Fig. 1.
A Model-View-Controller pattern was followed during the
development phase. As a result, the view and the logic to access
and process data are separated.
The new web application requires Java (jdk 1.6, jre 6.x) and it
runs over Apache Tomcat 6.X, as it is based on Servlet/JSP. The
database used is MySQL Server 5.X and MySQL Java Connector
5.X (JDBC).
With regard to the programming languages used, all the environment is Java 2 Platform (J2EE, Enterprise Edition), using JavaScript and AJAX (SACK library) to dynamically change the data
showed on the HTML pages, thereby avoiding the need to reload
the page every time the user wants to show or edit content.
For the videoconference module, OpenMeetings has been used.
This implements the Real Time Multimedia Protocol (RTMP) using
a red5 server for audio and video streaming.
For each person or entity using the system in a determined context and for a specic goal, four different user proles have been
dened:

516

3.2.2.5. Patients. Man or woman of any age with one or some cognitive functions affected, as a consequence of suffering ABI. The
caregiver role appears here, considered a secondary actor that will
help the patient use the system when necessary.

540

3.2.2.6. Therapist. A neuropsychologist who specializes in cognitive


rehabilitation in patients with ABI, who will have a number of assigned patients and be responsible for their treatment, scheduling
and the monitoring of personalized and individualized therapies.

544

3.2.2.7. Supervisor. Person in charge of the user management for


each center, both patients and therapists and their assignments,
apart from other management and control functions applied to
the supervising center/s.

548

3.2.2.8. Administrator. Apart from all the typical administration


tasks for every informatics system, the administrator will be the
person responsible for managing the categories, functions and
tasks dened in the system, which is the content the therapist will
use to schedule therapy sessions for their patients.
The content used in the tele-rehabilitation sessions consists of
computerized tasks, grouped by cognitive functions and categories.
The neuropsychologist creates a tele-rehabilitation session by
assigning a set of tasks to a certain day. He or she is able to congure the difculty of each task because they all have a set of input
parameters.
PREVIRNEC comprises a series of rehabilitation tasks for training different cognitive functions: attention, memory, executive
functions and language. For each specic task that the patient executes, the result of the task (0100 real number) is assigned to one
of the following three ranges: the NRR, INRR, and SNRR. In PREVIRNEC the initial hypothesis is that the patient has completed a task
within NRR if they obtained result is higher than 65 and lower than
85, in INRR if it is less than 65 and in SNRR if it is higher than 85.
PREVIRNEC dynamically adjusts each task level of difculty
according to the above dened ranges, aiming to scaffold them in
NRR. Therefore if the actual task is performed in INRR its difculty
level is automatically lowered, and if it is performed in SNRR it is
automatically raised.

552

Please cite this article in press as: Garca-Rudolph, A., & Gibert, K. A data mining approach to identify cognitive NeuroRehabilitation Range in Traumatic

Q1 Brain Injury patients. Expert Systems with Applications (2014), http://dx.doi.org/10.1016/j.eswa.2014.03.001

517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539

541
542
543

545
546
547

549
550
551

553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575

ESWA 9208

No. of Pages 14, Model 5G

20 March 2014
6

A. Garca-Rudolph, K. Gibert / Expert Systems with Applications xxx (2014) xxxxxx

Fig. 1. PREVIRNEC client/server communication schema.

576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628

3.2.3. Tasks
As introduced in Section 3.2.1, following NAB initial evaluation
all analyzed subjects initiated a CR program lasting four to six
months based on personalized interventions in PREVIRNEC
platform.
The therapeutic content used in PREVIRNEC tele-rehabilitation sessions consists of computerized tasks, grouped by cognitive
functions. The neuropsychologist creates a tele-rehabilitation session by assigning a set of tasks to a certain session day. He or
she is able to congure the difculty of each task because they
all have a set of input parameters.
At the time of this analysis, PREVIRNEC includes one hundred
and fteen rehabilitations tasks. Each task is dened by a series of
parameters that determine its level of difculty. The therapist selects for each task the parameter to be used for the automatic
adjustment of the difculty level described above. This dynamic
adjustment of the difculty level is performed twice for each task
as necessary. This means that if the patient does not obtain a task
result in NRR in the rst execution, PREVIRNEC automatically
generates the task with the adjusted difculty level once; if again
the obtained result is not in TR, PREVIRNEC likewise generates
a second version of the task.
3.2.3.1. Visual memory task description. For illustrative purposes
one such task designed for visual memory treatment is described
below in more detail. This task (identied as idTask = 151) has
been one of the most extensively administrated by neuropsychologists and executed by participants during the analyzed period
(described in Section 3.2.1.) and will be used throughout the different sections of this paper.
The objective of the task is to recall the position of pairs of identical images in a grid. A grid of xed size (e.g. 5  5 dark colored
cells) is presented to the participant at the start. When the participant left-clicks on a cell in the grid, an image of an object on a
white background appears in the cell. This image remains until a
second cell is clicked, then both images are shown for a period of
time (e.g. 1500 ms) for the participant to remember them; afterwards both images are covered. Only two cells can be simultaneously discovered in one go. When two identical images are
discovered, both of them remain visible in their cells. The aim of
the task is to discover all the images in the grid with the minimum
number of clicks. The parameters that determine the different difculty levels are shown in Table 1.
The quantied result parameters for the evaluation of task completion are: the total execution time, the total number of discovering clicks, the total number of wrong clicks (this number increases
if the participant clicks on an image already discovered before,
meaning that errors are computed after an initial exploration
phase), the total number of correct clicks (in this case, although
it is computed for homogeneity with other tasks, the number of
clicks for all participants is constant because the task is considered
unnished until all the images are discovered; this also means that
task151 does not produce omissions, and they are presumed to be
zero). The task result is computed as:

Task result correct=correct wrong omissions  100

629
631

4. Methods

632

As a rst attempt to nd the NeuroRehabilitation Range of a


cognitive task executed by means of PREVIRNEC, the result obtained by the patient in the execution of the task is used. The number of executions of the task performed by the patient is
considered, as it is known that repetition is highly related to activation of brain plasticity, which is in the core of cognitive functions
re-establishment, as discussed in the State of The Art Section.
The proposed methodology presents two strategies for the analytical and graphical identication/visualization of neurorehabilitation and non-NeuroRehabilitation Ranges based on the notion
of Sectorized and Annotated Plane introduced below. The two
models for NRR obtained are compared and discussed for the specic case of task151, related to visual memory cognitive function.

633

4.1. Sectorized and Annotated Plane (SAP)

646

Given three variables Y, X1, X2, where Y is a qualitative response


variable, with values {y1, y2, . . .}, and X1, X2 numerical explanatory
variables, the SAP is a 2-dimensional plot with X1 in the x axis, X2
in the y axis and rectangular regions with constant Y displayed and
labeled with Y values as outlined in Fig. 2. An SAP is therefore a
graphical support tool aimed at visualization, where the response
variable is constant in certain regions of the X1  X2 space. Eventually, allowing a relaxation of strict constant Y in the marked regions, the SAP might include an indicator of region purity, adding
the probability of occurrence of the labeling value.
Given a particular CR task, and assuming Y as a binary variable
reporting improvement of the patient in the cognitive function targeted by the task (yes, no), the SAP leads to response zones where
participants show similar response to treatment. The SAP shows a
plane sectorization directly related to treatment response. This allows identication of logical restrictions (rules) determining the
different outcomes of treatment.
The SAP is built following two methodologies that implement
two different strategies, the rst one based on graphical visualization and the second based on the plane partitions induced by decision trees, as introduced below.

647

4.1.1. Visualization-based SAP (Vis-SAP)


Data is plotted regarding X1 and X2, and each point is marked
with different colors according to the values of Y. This categorized
scatterplot (sometimes known as letterplot) is an exploratory technique for investigating relationships between X1 and X2 within the
sub-groups determined by Y. For the particular application presented here, X1 is the number of executions of the task performed
by the subject (Execs151), X2 is the result obtained at every single
execution (Results), while Y is the effect of the neurorehabilitation
process (improvement/non-improvement).

668

Please cite this article in press as: Garca-Rudolph, A., & Gibert, K. A data mining approach to identify cognitive NeuroRehabilitation Range in Traumatic

Q1 Brain Injury patients. Expert Systems with Applications (2014), http://dx.doi.org/10.1016/j.eswa.2014.03.001

634
635
636
637
638
639
640
641
642
643
644
645

648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667

669
670
671
672
673
674
675
676
677

ESWA 9208

No. of Pages 14, Model 5G

20 March 2014
7

A. Garca-Rudolph, K. Gibert / Expert Systems with Applications xxx (2014) xxxxxx


Table 1
Parameters that determine idTask = 151 level of difculty.

678
679
680
681
682
683

Number of
cells

Stimulus type

44
55
66
88

Abstract objects
Numbers
Animals
Colors

Proximity of
the
second image

Presentation time
(ms)

2 Cells
3 Cells
4 Cells
Random

1500
3000
4000

This exploratory analysis is used to identify systematic relationships between variables when there is no previous knowledge
about the nature of those relationships. The constant-Y regions detected in the plot can be expressed in the form of logical rules
involving the implied variables. The SAP is built on the basis of
these rules.

699

4.1.2. Decision tree-based SAP


The tuple (X1, X2, Y) = (Execs151, Results, Improvement) can be
treated as a classical classication problem, where Improvement
has to be recognized on the basis of Execs151 and Results. The
training dataset is used to induce a decision tree classier which
is later evaluated with the test set in the usual way. For the testing,
the class label is ignored and predicted by the classier. Performance of classier is evaluated by comparing both predicted and
real class. The confusion matrix and taxes of misclassication can
be provided.
Here the Weka (Hall, Frank, Holmes, Pfahringer Reutemann, &
Witten, 2009) software has been used to apply the J48 (Witten &
Frank, 2005) decision tree algorithm which implements Quinlans
C4.5 algorithm (Quinlan, 1993) building an unpruned tree.
Once the decision tree has been constructed, it is converted into
an equivalent set of rules in the usual way.

700

4.2. Quality indicators

684
685
686
687
688
689
690
691
692
693
694
695
696
697
698

701
702
703
704
705
706
707
708
709

4.2.1. Sector condence


Given a sector S from the SAP graph, labeled with Y = y, the sector condence corresponds to the empirical probability of occurrence of event y inside the sector. P(Y = y|S) is computed as the
ratio between the number of positive cases and the sector size. This
is in one sense a measurement of the purity of the sector and
provides the quality of the assignment of class y to all elements
in the sector. The higher the condences of the SAP sectors, the
better the model is considered.

When Y is a binary variable P(Y = yes | S) it provides the sensitivity of S, while P(Y = No | S) provides the specicity. As usual,
the higher the sensibility and specicity, the higher the quality of S.
We dene the global quality of the SAP as the pooled condence
of all sectors. Additionally for the SAP of binary variables a pooled
specicity and a pooled sensitivity can be used as quality
indicators.

710

4.2.2. Hypothesis testing


Thus, the range of results determining those sectors where S is
labeled as Y = yes (Improvement) determine the NRR of the task.
The sensitivity of the NRR is related to the fact that patients in
NRR improve, or in a more relaxed formulation, that there is a high
proportion of patient improvement within NRR.
The specicity is related to the fact that patients out of NRR do
not improve. This can be measured by the high proportion of nonimproving patients in INRR or SNRR or equivalently by the low proportion of improving patients in INRR or SNRR.
A classic 2-sample probability test is used to see whether the
response to the CR therapy is signicantly different for those executing tasks in NRR than those obtaining results out of NRR. It is
expected that the probability of improvement is signicantly higher for those in NRR. SAP models that provide sectors without significant differences should be disregarded, as they provide NRR with
poor identication of the improving population. Thus, being

717

711
712
713
714
715
716

718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733

pMR = Probability of improving being in NRR

734

pMR = Probability of improving being out of NRR

735
736

the hypothesis tested is

737

H0 : pMR pMR

738
740

H1 : pMR > pMR

741
743

The statistics used are

744

pMR  pMR
e r

 H0 z
1
1
p0 1  p0 nMR nMNoR

747

where p0 is the weighted common estimator of


the Ho,

745

pMR and pMR under

nMR pMR nMR pMR


p0
nMR nMR

748
749

750
752
753

^ MR
pMR p

number of patients in NRR that improve

nMR

^ MR
pMR p

number of patients out of NRR that improve

nMR

755
756

Fig. 2. General Sectorized Annotated Plane (SAP) description.

758

The test is solved under the z-distribution, with alpha = 0.05. The
greater the difference between pMR and pMR (pMR > pMR ) the more
sensitive and specic is the NRR criterion tested, the lower the p-value of the test, and better performs over real patients.

759

5. Application and results

763

For the 123 patients followed, we have 3366 executions of the


idTask = 151 relative to visual memory. Each task execution is described by an instance. Table 2 shows an instance representing an
execution of a task by a patient, the number of previous executions
of the task by the participant is 23, the result obtained in that specic task execution is 83.4, and an improvement in the addressed
cognitive sub-function was obtained according to NAB pre and post
differences.

764

Please cite this article in press as: Garca-Rudolph, A., & Gibert, K. A data mining approach to identify cognitive NeuroRehabilitation Range in Traumatic

Q1 Brain Injury patients. Expert Systems with Applications (2014), http://dx.doi.org/10.1016/j.eswa.2014.03.001

760
761
762

765
766
767
768
769
770
771

ESWA 9208

No. of Pages 14, Model 5G

20 March 2014
8

A. Garca-Rudolph, K. Gibert / Expert Systems with Applications xxx (2014) xxxxxx

Table 2
Attributes of classied instances.
Execs151

Result

Improvement

23

83.4

YES

784

Execs151. Number of previous executions of the task already


performed by the participant at the time of current execution.
Result. Is the result obtained by the participant in the execution
of the task as described above; it is a [0, 100] real value.
Improvement. Improvement is a binary variable with values
(YES, NO) and indicates whether or not the patient improved
with the neurorehabilitation treatment. It is dened as the class
label Y described in Section 4.1.2. The improvement of a patient
is determined according to the criteria described in IG clinical
protocols. Those protocols consider the comparison of the
NAB neuropsychological assessment both before and after
treatment. For each cognitive function the protocol is dened
with the relevant NAB scales.

785

5.1. Analysis of PREVIRNEC visual memory task using basic criterion

786

As introduced in Section 3.2.2, nowadays in PREVIRNEC a rst


hypothesis is being tested which considers that any participant patient has completed any task within NRR if they obtained result is
in the 6585 range, in INRR if it is less than 65, and in SNRR if it is
higher than 85. As a reference, the current NRR used (Result e [65, 85]) is visualized in a manually built SAP shown in
Fig. 3 with an overall sensitivity = 0.5660, overall specicity = 0.5012 and overall quality = 0.5022. The percentages represented in the SAP provide the empirical proportion of patients
that improved after the treatment in every area. About 60% of patients performing idTask = 151 in NRR really improved. This is far
from random improvement. However, to evaluate the quality of
the basic NRR used as a reference, a 2-sample probability test is
used as described in Section 4.2.2
This enables verication of whether being in NRR really implies
a signicantly higher probability of improvement.
Table 3 contains the relevant information to compute the test
by crossing the classication of the patients regarding two factors.
Improving/not improving and performing the idTask = 151 within
NRR or not (according to the basic criterion currently used).

772
773
774
775
776
777
778
779
780
781
782
783

787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805

The test provided a result that is not statistically signicant


(z = 0.7316, p = 0.2323). This appears to be evidence that using
the single result of the task is not enough to detect either the
NRR or the ZRP zone. In the next sections, a model including the
number of executions per task is tested.

806

5.2. Analysis of PREVIRNEC visual memory task using visualizationbased SAP (Vis-SAP)

811

The relationship between Results and Number of executions of


patients is shown in Fig. 4. Improving patients improving are
shown in green and non-improving in red. Areas with a single category of patients (improving or not improving) are visually
identied.
Looking at Fig. 4, an area with no improvement is obtained for
participants obtaining results higher than 70 and number of executions lower than 40 at the top left of the scatter plot. Furthermore,
if the number of executions is higher than 60 and results higher
than 20, a large region can be easily identied in which every participant is labeled in the improvement group, leading to the identication of a therapeutic range of results depending on the
number of executions.
Two neat regions emerge from the above rules which can be expressed in the form of logical restriction rules, and visualized in an
SAP diagram (shown in Fig. 5 with an overall sensitivity = 0.939,
overall specicity = 0.5523 and overall quality = 0.994).
Identied rules are:

813

(Execs151 < = 40) AND (Res > 70) ? Not NRR


(Execs151 > 60) AND (Res > 20) ? NRR

807
808
809
810

812

814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833

The quality of the induced denition for NRR is assessed by


means of the 2-sample proportion test (described in Section
4.2.2). Table 4 contains the relevant information to compute the
test by crossing the classication of the patients regarding two factors. Improving/not improving and performing the idTask = 151
within NRR or not (according to the rules directly induced over
the SAP). The results are z = 12.42, p<<< 0.00001 is statistically
signicant.
In this case we can ensure that repeating idTask = 151 more
than 60 times and getting results higher than 20 in all executions
is associated with a group of patients with a sensibly higher prob-

Fig. 3. Vis-SAP for [68, 85] NRR for idTask 151.

Please cite this article in press as: Garca-Rudolph, A., & Gibert, K. A data mining approach to identify cognitive NeuroRehabilitation Range in Traumatic

Q1 Brain Injury patients. Expert Systems with Applications (2014), http://dx.doi.org/10.1016/j.eswa.2014.03.001

834
835
836
837
838
839
840
841
842
843
844

ESWA 9208

No. of Pages 14, Model 5G

20 March 2014
A. Garca-Rudolph, K. Gibert / Expert Systems with Applications xxx (2014) xxxxxx
Table 3
Contingency table for NRR with Result e [65, 85] for idTask = 151.
Improvement

Yes
No
Total
^p(YES)

In NRR
Yes

No

Total

30
23
53
0.5660

1652
1661
3313
0.4986

1682
1684
3366
0.4997

846

ability of improving visual memory function. Thus, upon the VisSAP criterion, NRR(task151) = Execs151 > 60 and Results > 20.

847

5.3. Analysis of PREVIRNEC visual memory task using DT-SAP

848

Although the visual-based SAP method seems to produce good


results, the NRR has been dened on the basis of the visual expertise of the data miner, and this approach is totally dependent of the
ability of the data miner itself. This is why an attempt to nd the
NRR automatically from data is presented. As introduced in Section
4.1.2 a supervised classication method, the J48 decision tree is applied to the target dataset (3366 instances) for SAP generation.
Experiments were conducted in Weka with J48 decision tree for
default conguration input parameters (condence factor = 0.25),
since default values produced the best accuracy (57.45%), precision
and recall. As usual [32], 10-fold cross validation was used to evaluate the goodness of results. As shown in Fig. 6 three leaves of the
resulting tree are labeled as an improvement.
Following the path from the root to those leaves, a condition for
a patients improvement can be induced, and the NRR dened as:

845

849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866

(Execs151<=58) AND (Execs151 > 44) AND (Res > 4) AND


(Res<=32) OR
(Execs151<=58) AND (Execs151 > 45) AND (Res > 57) OR
(Execs151 > 58) AND (Res > 8) ? NRR

867
868
869
870
871

The quality of this induced criterion is assessed through the two


samples proportion test presented in Section 4.2.2. The relevant
information for assessing the goodness of the NRR found for
Task151 is presented in Table 5.

With this information, the test is statistically signicant


(z = 13.45 p<<<<0.00001).
In fact, J48 identies linear partitions of the space by means of
linear separators that can be represented in an SAP diagram and
compared with the zones identied in Fig. 7 with an overall sensitivity = 0.87, overall specicity = 0.689 and overall quality = 0.716.
Fig. 8 shows the ROC curve of the three models, and it can be
seen that both Vis-Sap and DT-SAP perform signicantly better
than the experts-based current criterion. The Vis-SAP method provides a slightly better, also giving the higher global quality.

872

5.4. Clinical validation

882

Following the NRR identication phase, 327 patients not included so far in this study were considered for participation in order to validate the results. The clinical staff of the hospital
randomly selected 10 of them after participants provided consent
in the usual way for these interventions to test the validity of the
clinical hypothesis about NRR of task 151 arising from Section 5.2
of this study. Patients were evaluated before treatment according
to the standard clinical protocol (Neuropsychological Assessment
Battery) introduced in Section 3.2.1 of this paper. The neuropsychologists in charge of the NR program of each patient included
in the program the execution of task 151 a minimum of 60 times
in such a difculty conguration as to guarantee that the patient
obtained a result higher than 20. In this validation phase, the conguration of the tasks were manually tuned for each patient by the
specialist, according to the performance shown in previous executions and the specic clinical condition of each participating
patient.
All patients were evaluated after treatment following the same
standard protocol and the improvement of the patient was assessed in the usual way by comparing scores before the treatment
with scores at the end of it.
Of the 317 patients following the classical NR program, 189
showed improvement and 128 did not. Meanwhile, we could verify
that all of the participating patients under the NRR recommendations improved in the targeted cognitive function. A twofold
impact was observed: SAP recommendations can support cognitive
therapies with new (previously unknown and specic) congurations of tasks and those recommendations show a higher probability

883

Fig. 4. Letterplot of TaskExecs vs. Result vs. Improving/not for idTask 151.

Please cite this article in press as: Garca-Rudolph, A., & Gibert, K. A data mining approach to identify cognitive NeuroRehabilitation Range in Traumatic

Q1 Brain Injury patients. Expert Systems with Applications (2014), http://dx.doi.org/10.1016/j.eswa.2014.03.001

873
874
875
876
877
878
879
880
881

884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910

ESWA 9208

No. of Pages 14, Model 5G

20 March 2014
10

A. Garca-Rudolph, K. Gibert / Expert Systems with Applications xxx (2014) xxxxxx

Fig. 5. Vis-SAP for idTask 151.

Table 4
Contingency table for VIS-SAP TR for idTask = 151.
Improvement

Yes
No
Total
^p(YES)

In NRR
Yes

No

Total

199
13
212
0.9386

1483
1671
3154
0.4701

1682
1684
3366
0.4997

of obtaining measurable improvements in the participating


patients.
The authors are aware that the sample size is small enough to
guarantee improvement, but it can be claimed a guarantee of
increasing the probabilities of improvement of the participating
patients. For the standard treatment group, an improvement tax
of 59% was found with a 95% CI: [0.536, 0.644]. This means that
the tax of improvement would rarely be higher than the 64% of patients. The whole set of 10 patients submitted to the NRR recommendations improved. The 95% CI with a 100% improvement tax

911
912
913
914
915
916
917
918
919
920

Q12

is [1, 1], since the length of the CI is computed as a function of


p(1  p). However, even in the hypothetical case of having one patient without improvement in this small size set, this would lead to
a 90% of improvement with a 95% CI of [0.714, 1], meaning that
rarely the tax of improvement goes down to 71.4%. This lower
bound is much higher than the upper bound of the general group,
thus proving the efcacy of the recommendation. We can also conrm this issue by the standard test of comparing two proportions
as shown in Table 6.

921

6. Discussion

930

This work aims to identify the conditions in which performing a


certain cognitive rehabilitation task maximizes the potential to
activate brain plasticity and to establish the damaged cognitive
function as much as possible. The NeuroRehabilitation Range has
been introduced as a formal concept to express these conditions
and it has been induced on the basis of the SAP tool. As far as we

931

Fig. 6. DT-SAP for idTask 151.

Please cite this article in press as: Garca-Rudolph, A., & Gibert, K. A data mining approach to identify cognitive NeuroRehabilitation Range in Traumatic

Q1 Brain Injury patients. Expert Systems with Applications (2014), http://dx.doi.org/10.1016/j.eswa.2014.03.001

922
923
924
925
926
927
928
929

932
933
934
935
936

ESWA 9208

No. of Pages 14, Model 5G

20 March 2014
A. Garca-Rudolph, K. Gibert / Expert Systems with Applications xxx (2014) xxxxxx
Table 5
Contingency table for DT-SAT TR for idTask = 151.
Improvement

Yes
NO
Total
^p(YES)

In NRR
YES

NO

Total

375
71
446
0.8408

1307
1613
2920
0.4476

1682
1684
3366
0.4997

11

Two methods to induce the NRR are proposed, one based in


visualization of raw data, the other based on DT. Both provide an
easy data-driven tool to determine the NRR of a task.
The Vis-SAP method induced the NRR of task15: (NRR(151) = Execs151 > 60 and Res > 20) The DT-SAP provides a more complex
model:

969
970
971
972
973
974

975

NRR151 Execs151 < 58 AND Execs151


> 44 AND Res > 4ANDRes <

937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968

know, this is the rst time that the number of repetitions of the
task is considered for modeling the NRR of the task.
This goes one step ahead on the current state of the art. Most of
the work done on maximizing the rehabilitative effect of a task has
been oriented to a proper management of the level of difculty of
the tasks presented to the patient. In the context of rehabilitation
through a serious game strategy, the determination of tasks and
game difculty is made statically by the therapists in most therapeutic games proposed in the literature. For example (Heuser
Q6 et al., 2006) suggest ve therapeutic games exercises, each one
involving a set of difculty levels following patients recovery;
the difculty level is statically designed in an increasing way and
the simulation stops the exercise when the patient fails. In Ma
et al. (2007) the therapeutic game trains visual discrimination
and selective attention using three difculty levels: Beginner,
Intermediate and Expert; the system includes a matrix assigning
suitable difculty levels for a set of patient proles, that is used
to suggest a difculty level to each patient.
PREVIRNEC is assuming an expert-based constant NRR for the
whole set of available tasks: [65, 85]; the system automatically increases difculty if the patient performs over the NRR and decreases it if he/she performs below NRR. All these systems try to
deal with the degree of difculty of the task to be proposed to
the patient, but none of them include any kind of guide about
the number of executions required to empower the rehabilitation
effect.
According to Lurias theory (Luria, 1978), present research
shows that repetition is important in CR. The main result emerged
from this analysis is that the number of executions of the CR tasks
is clearly relevant to determine NRR and that including them in the
model provides signicant improvement with regards to current
practice.

32 OR Execs151 < 58 AND Execs151


> 45ANDRes > 57 OR Execs151
> 58 AND Res > 8
But both methods determine a region for Results and Number of
Executions where improvements are concentrated. Bounds for both
Results and Number of Executions dene the conditions of executing the task that maximizes probability of improvement.
Statistical test assessed that indeed, patients executing the
tasks in these conditions, are expected to improve much
easily (with signicant higher probability) than those executing
the task out of NRR (pVis-SAP(improvement|NRR) = 0.9386, pDTSAP(improvement|NRR) = 0.8408).
Even if both models involve both Results and Number of executions, the conditions provided by the DT-SAP are more complex,
less intuitive for the end-user, and from the numerical point of
view, provide lower quality results. Both specicity and sensitivity
as well as the global quality of the SAP are lower than those provided by Vis-SAP. For this reason, the Vis-SAP criterion seems to
be more appropriated for real use.
Despite of that, it is interesting to remark that the denitions
provided by Vis-SAP and DT-SAP are consistent. Indeed DT-SAP
correctly recognizes the NRR zone identied by Vis-SAP for
task151. However, the decision tree procedure, cannot detect the
non-improvement zone. In fact, partitions induced by decision
trees on the plane are complete; thus, the procedure tries to assign
improvement or not to every area in the SAP diagram, thus forcing
assignments in areas where the behavior of patients is not so
homogeneous. This, generates a decreasing of the quality indicators as predictive power decreases in those zones where heterogeneity of treatment outcome is higher, leading to a lower global rate.

Fig. 7. DT-SAP for idTask 151.

Please cite this article in press as: Garca-Rudolph, A., & Gibert, K. A data mining approach to identify cognitive NeuroRehabilitation Range in Traumatic

Q1 Brain Injury patients. Expert Systems with Applications (2014), http://dx.doi.org/10.1016/j.eswa.2014.03.001

977

978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
1001
1002
1003
1004

ESWA 9208

No. of Pages 14, Model 5G

20 March 2014
12

A. Garca-Rudolph, K. Gibert / Expert Systems with Applications xxx (2014) xxxxxx

Fig. 8. ROC curves comparison for VIS-SAP current hypothesis, VIS-SAP proposed NRR and DT -SAP.

1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
1016
1017
1018
1019
1020
1021
1022
1023
1024
1025
1026
1027
1028
1029
1030
1031
1032
1033
1034
1035
1036
1037

In fact, VIS-SAP strategy identied smaller zones showing higher


homogeneity, and this impacts into higher predictive power.
Same kind of behavior has been observed when analyzing other
neurorehabilitation tasks, addressed to different cognitive functions (e.g. idTask = 146 for sustained attention, idTask = 153 for
verbal memory).
For this particular application the Vis-SAP induced a NRR for
Task 151 that suggests more than 60 executions of the task with
results scoring over 20 for a maximum benecial effect. From the
practical point of view, the proposed tools are able to give clinicians clear criteria to decide how many times the task must be repeated by the patient and which is the range of performance that
can produce better improvement. In fact, the NRR conditions
emerged from the SAP were rarely envisaged by therapists along
typical CR treatments. Also it was observed that assuming these
guidelines to design the CR program signicantly increased the
possibilities of improving the targeted cognitive functions of the
participating patients.
The main drawbacks of the Vis-SAP proposal are two. On the
one hand, the lack of completeness of the Vis-SAP criterion proposed. Indeed, looking at the SAP diagram, VIS-SAP it is not assigning improvement or non improvement to the whole surface, but
only to small parts of the diagram corresponding to concrete and
reduced areas where improvement can be ensured, or non response can be ensured. This means that, even though the probability of improving is very high inside the NRR, this is not necessarily
implying that people out of this area do not improve. In fact, the
Vis-SAP is determining a small area where the non improvement
is dominant, but, out of the NRR area almost half of the patients
also improve, being this close to random response to the treatment. So, one could say that Vis-SAP provides a semi-deterministic
procedure where a particular conguration for both results and
repetitions ensure improvement, a second conguration where

Table 6
Contingency table for validation of idTask = 151.
Improvement

Yes
NO
Total
^p(YES)

In NRR
YES

NO

Total

10
0
10
1.000

189
128
317
0.5962

199
128
327
0.6100

the task do not produce patients improvement, and out of these


regions the outcome is undetermined.
On the other hand, the proposed analysis considers each task
individually, being NRR dened for every single task. However, it
is already known that the CR is based on sequences of tasks that
interact among them, and taking in consideration the whole sequence of tasks involved in the treatment is likely to improve the
quality of the model.

1038

7. Conclusions and future work

1046

This work is a contribution towards the personalized, predictable, and data driven CR design from both a theoretical and practical point of view.
From the theoretical point of view, the paper introduces a new
concept, the NeuroRehabilitation Range (NRR) as the framework to
describe the degree of performance of a CR task which produces
maximum rehabilitation effects. The NRR contributes to provide
an operational denition for the zone of maximum rehabilitation
potential and represents an operationalization of the Zone of Proximal Development referred in Vygotsky (1934).
Analytical and visual tools are also proposed in this paper, dened and validated, to nd an operational denition of a NRR from
a data driven approach. On the one hand, the SAP has been introduced as a general visualization tool to nd areas with high probability of occurrence of a target event. A particular case of SAP for
detecting cognitive improvement in relation with results and repetitions of a certain cognitive rehabilitation task is presented in the
paper. For this particular application, the SAP identies areas with
high probability of cognitive improvement. Although SAP is not a
complex concept, it has shown a great potential to nd the NRR region of a cognitive rehabilitation task in a quick, simple and very
intuitive way, which has shown to be highly useful at clinical practice level. Also, for the rst time, the NRR is dened as a bivariate
structure involving conditions in both results and repetitions of the
tasks.
Another contribution of the paper is to propose two different
methodologies to build the SAP in a given real problem: Direct construction of SAP by visualization of raw data (Vis-SAP method); and
DT-SAP, which is based on decision-tree induction and could be
automated. Decision trees have been considered because their
inherent structure is directly providing the NRR model, which is
built as the OR of all branches bringing to a leaf labeled as improvement. Both methods effectively determine the areas where proba-

1047

Please cite this article in press as: Garca-Rudolph, A., & Gibert, K. A data mining approach to identify cognitive NeuroRehabilitation Range in Traumatic

Q1 Brain Injury patients. Expert Systems with Applications (2014), http://dx.doi.org/10.1016/j.eswa.2014.03.001

1039
1040
1041
1042
1043
1044
1045

1048
1049
1050
1051
1052
1053
1054
1055
1056
1057
1058
1059
1060
1061
1062
1063
1064
1065
1066
1067
1068
1069
1070
1071
1072
1073
1074
1075
1076
1077
1078
1079

ESWA 9208

No. of Pages 14, Model 5G

20 March 2014
A. Garca-Rudolph, K. Gibert / Expert Systems with Applications xxx (2014) xxxxxx
1080
1081
1082
1083
1084
1085
1086
1087
1088
1089
1090
1091
1092
1093
1094
1095
1096
1097
1098
1099
1100
1101
1102
1103
1104
1105
1106
1107
1108
1109
1110
1111
1112
1113
1114
1115
1116
1117
1118
1119
1120
1121
1122
1123
1124
1125
1126
1127
1128
1129
1130
1131
1132
1133
1134
1135
1136
1137
1138
1139
1140
1141
1142
1143
1144
1145

bility of improvement are higher; a statistical two-proportion test


has been used to assess the goodness of the NRR models, by checking whereas the probability of improvement is signicantly higher
when tasks are performed according to NRR than out of it. Whereas
DT-SAP is a deterministic method that can be automated, the VisSAP is a semi-deterministic method that requires visual inspection
in its last step. However, it seems to produce better results in practical applications, as the incomplete sectorization of the plane in
very homogeneous areas provided by Vis-SAP outperforms the results induced from a DT where the leaves are often contaminated,
containing both improving and non-improving patients.
The last theoretical contribution of this work is the denition of
a quality criterion to assess NRR models, based on pooled condence and pooled specicity. The dened criterion is based on
the capacity of a NRR model to detect the patients improving with
the execution of a task. This is somehow giving a global performance indicator, although ROC curves have also been used to test
the quality of obtained models, and it conrms that both proposed
methods outperform the univariate and static NRR [65,85] currently used by the experts, as well as that Vis-SAP performs slightly
better than DT-SAP.
Finally, all those elements have been applied to a real case
study. The application shows how the proposed methodology
could identify NRR for a given cognitive rehabilitation task, and
how the NRR obtained provides clear guidelines to the therapists
about the number of repetitions of the task to be proposed to the
patient together with the acceptable range of performance desirable to maximize the effect of the rehabilitation. From the clinical
point of view, main contributions of this paper are to show that
repetition of tasks is really relevant for rehabilitation (as stated
by Luria in 1978), to provide an intuitive tool that permits the
therapists to obtain guidelines about how much repetitions of
the task must be proposed to the patient, and to show that the
desirable difculty level of a task is specic for each task. This
is an excellent complement to the previous state of the art in
which some advances were done regarding how to manage the
difculty level of the tasks, but no works assessing repetitions
were addressed.
Clinicians established an initial hypothesis about the NRR,
assuming it is xed and task independent (NRR(T) = [65, 85]); these
bounds have been dened according to CR therapists expertise.
PREVIRNEC allows a systematic pre and post evaluation of participants covering the major cognitive domains. This provides empirical data useful to validate or clarify clinical hypothesis. For the
rst time, data collected through PREVIRNEC platform has been
used to learn more about the NRR. Although the ratio of improvement of patients in that initial NRR was not low, this work provided evidence that a simple formulation for NRR regarding only
the Results obtained is insufcient to identify the group of patients
with better response to CR treatment. According to our results NRR
cannot be dened by means of univariant analysis (considering
only the Result of performing a task). A predictive model considering other implied covariables needs to be developed. The present
analysis is a rst attempt into that direction. It has been shown
that the number of repetitions that a patient performs of a certain
task is also relevant for the patients outcome, according to literature. Bidimensional NRR, depending not only on performances, but
also on repetition, signicantly improve the CR treatment design.
On the other hand, the range of therapeutic performances might
change from task to task. This work points to target a specic performance-range for each task, instead of the current [65, 85] range
used for the whole set of cognitive tasks available.
An old wondering of the Institute Guttmann was to better
understand how NRR could be found, and since 2007, the institute
has been leading research in this line. This work is providing objective criteria for NRR that can be integrated in daily clinical practice

13

of the institution, as well as operationalized for PREVIRNEC platform, that provides the support to verify clinical hypothesis.
As discussed in previous section, the Vis-SAP provides a semideterministic criterion that outperforms DT-SAP, but the later is
automatable. On the other hand, the proposed analysis considers
each task individually
Till now, CR plans are mainly built from scratch for every patient, on the basis of the expertise of therapist and the follow-up
of the patient, as no standard guidelines are available in this domain yet. The ndings from the present study led to new actionable knowledge in the eld of rehabilitation practice, opening the
door towards more precise, predictable and powerful CR treatments, customized for the individual patient. Some clinical
hypothesis are being formulated by specialists on the basis of these
results and currently under validation, as a previous step to the
establishment of a methodology for personalized therapeutic
interventions based on clinical evidence.
As future research lines, the automatic construction of SAP still
requires more work since decision trees imply, by construction,
some intrinsic error taxes in every branch that will be always propagated to the NRR performance and automation from Vis-SAP has
to be faced from scratch.
This work is currently being enriched by analyzing how patients
walk through the SAP areas (or sectors) during their rehabilitation
process. This can be analyzed by connecting the points corresponding to a same patient in the SAP and nding prototypical patterns
according to the form of the paths designed on the SAP. This dynamic analysis can be later generalized to nd dynamic patterns
on the global treatment of the patient involving the whole sequence of tasks performed during the treatment, and providing
information about the possible positive interactions among tasks
that empower the improvement capacity.
Although the NRR models using number of executions and results seem to provide quite high sensitiveness and specicity, there
are other factors supposed to be highly determinant of cognitive
improvement, like task difculty. Extension of the current proposals to include such other factors is currently being explored.
Finally, obtained results are expected to be more interpretable
by clinicians when other demographic and clinical variables are included in the model, e.g. participants educational level, age, time
since injury, obtained results in pre-treatment evaluation.

1146

Acknowledgments

1187

This research was supported by: Ministry of Industry, Tourism


and Trade (Spain) AVANZA PLAN-Digital Citizen Subprogram (PT:
NEUROLEARNING Grant Nr: TSI-020501-2008-0154). Institute of
Health Carlos III (Spain) Strategic Action Healths Call (PT: Clinical
implantation of PREVIRNEC platform in TBI and stroke patients/
Grant Nr: PI08/900525). Ministry of Science and Innovation (Spain)
INNPACTO Program (PT NEUROCONTENT - Grant Nr 300000-201030). Ministry of Education Social Policy and Social Services (Spain)
IMSERSO Program (PT COGNIDAC Grant Nr 41/2008). MARAT
TV3 Foundation (PT: Improving Social Cognition and meta-cognition in schizophrenia: A tele-rehabilitation project Nr 091330)
and EU CIP-ICT-PSP-2007-1 (PT: CLEAR Grant Nr.: 224985)

1188

References

1200

Bernabeu, M., & Roig. T. (1999). La rehabilitacin del traumatismo craneoenceflico:


un enfoque interdisciplinar. Barcelona: Fundaci Institut Guttmann.
Calero, M. D., & Navarro, E. (2007). Cognitive plasticity as a modulating variable on
the effects of memory training in elderly persons. Archives of Clinical
Neuropsychology, 22, 6372.
Cicerone, K. D., & Tupper, D. E. (1986). Cognitive assessment in the
neuropsychological rehabilitation of head-injured adults. In B. P. Uzzell & Y.
Gross (Eds.) (pp. 5984). Boston: Martinus Nijhoff.

1201
1202
1203
1204
1205
1206
1207
1208

Please cite this article in press as: Garca-Rudolph, A., & Gibert, K. A data mining approach to identify cognitive NeuroRehabilitation Range in Traumatic

Q1 Brain Injury patients. Expert Systems with Applications (2014), http://dx.doi.org/10.1016/j.eswa.2014.03.001

1147
1148
1149
1150
1151
1152
1153
1154
1155
1156
1157
1158
1159
1160
1161
1162
1163
1164
1165
1166
1167
1168
1169
1170
1171
1172
1173
1174
1175
1176
1177
1178
1179
1180
1181
1182
1183
1184
1185
1186

1189
1190
1191
1192
1193
1194
1195
1196
1197
1198
1199

Q7

ESWA 9208

No. of Pages 14, Model 5G

20 March 2014
14
1209
1210
1211
1212
1213
1214
1215
1216
1217
1218
1219
1220
1221
1222
1223
1224
1225
1226
1227
1228
1229
1230
1231
1232
1233
1234
1235
1236
1237
1238
1239
1240
1241
1242
1243
1244
1245
1246
1247
1248
1249
1250
1251
1252
1253
1254
1255
1256
1257
1258
1259

Q8

A. Garca-Rudolph, K. Gibert / Expert Systems with Applications xxx (2014) xxxxxx

Cicerone, K. D., Langenbahn, D. M., Braden, C., Malec, J. F., & Kalmar, K. (2011).
Evidence-based cognitive rehabilitation: Updated review of the literature from
2003 through 2008. Archives of Physical Medicine and Rehabilitation, 92(5),
1930.
Csikszentmihalyi, M. (1991). Flow: The psychology of optimal experience. Harper
perennial
DiPiro, J. T., & Spruill, W. J. (2010). Concepts in clinical pharmacokinetics (5th ed.)
American Society of Health-System Pharmacists (January 1) ISBN-13: 9781585282418.
Ecri. Cognitive rehabilitation therapy for traumatic brain injury: What we know and
dont know about its efcacy. EDITORIAL NOTE 10/11/11: IOMs New Report on
Brain Injury Treatments Draws Conclusions Similar to ECRI Institutes Earlier
Findings.
Finneran, C. M., & Zhang, P. (2005). Flow in computer-mediated environments:
Promises and challenges. Communications of the Association for Information
Systems, 15, 82101.
Gibert, K., & Garca-Rudolph, A. (2007). Desarrollo de herramientas para evaluar el
resultado de las tecnologas aplicadas al proceso rehabilitador Estudio a partir
de dos modelos concretos: Lesin Medular y Dao Cerebral Adquirido. Cap 6:
Posibilidades de aplicacin de minera de datos para el descubrimiento de
conocimiento a partir de la prctica clnica. Madrid: Plan Nacional para el
Sistema Nacional de Salud del Ministerio de Sanidad y Consumo. Barcelona:
Agncia dAvaluaci de Tecnologia I Recerca Mdiques. Informes de Evaluacin
de Tecnologias Sanitarias. AATRM nm. 2006/11. Barcelona.
Green, C., & Bavelier, D. (2005). In Messaris & Humphreys (Eds.), Digital media:
Transformations in human communication.
Green, C. S., & Bavelier, D. (2007). Action-video-game experience alters the spatial
resolution of vision. Psychological Science, 18(1), 8894.
Hall, M., Frank, E., Holmes, G., Pfahringer Reutemann, P., & Witten, I. H. (2009). The
WEKA data mining software: An update. SIGKDD Explorations, 11(1).
Heuser, H. A., Kourtev, H., Winter, S., Fensterheim, D., Burdea, G., Hentz, V., et al.
(2006). Tele-rehabilitation using the Rutgers master II glove following carpal
tunnel release surgery. In 2006 International workshop on virtual rehabilitation
(pp. 8893). IEEE publisher.
Jagaroo, V. (2009). Neuroinformatics for neuropsychologists (1st ed). Springer (August
21).
Kleim, J. A., & Jones, T. A. (2008). Principles of experience-dependent neural
plasticity: Implications for rehabilitation after brain damage. Journal of Speech,
Language, and Hearing Research, 51, S225S239. February D American SpeechLanguage-Hearing Association 10924388/08/5101-S225.
Linkenhoker, B. A., & Knudsen, E. I. (2002). Incremental training increases the
plasticity of the auditory space map in adult barn owls. Nature, 419(6904),
293296.
Luria, A. (1978). Neuropsychology. In Great soviet encyclopedia: A translation of the
third edition (pp. 514515). New York, London: MacMillan, Collier Macmillan.
Vol. 17.
Ma, M., Charles, D., McDonough S., Crosbie, Oliver, L., & McGoldrick C., 2007.
Adaptive virtual reality games for rehabilitation of motor disorders. Universal
Access in HumanComputer Interaction, Ambient, Interaction.
Maas, A. I., Stocchetti, N., & Bullock, R. (2008). Moderate and severe traumatic brain
injury in adults. Lancet Neurology, 7(8), 728741.

Marcano, A., Chausa, P., Garcia-Rudolph, A., Cceres, C., & Tormos, J. M. (2013). Data
mining applied to the cognitive rehabilitation of patients with acquired brain
injury. Data mining applied to the cognitive rehabilitation of patients with
acquired brain injury. Expert Systems with Applications, 40(4), 10541060. ISSN
0957-4174.
Mathwick, C., & RigdonPlay, E. (2004). Flow, and the online search experience.
Journal of Consumer Research, 31(2), 324332.
Norea, D., Rios-Lago, M., Bombin-Gonzalez, I., Sanchez-Cubillo, I., Garcia-Molina,
A., & Tirapu-Ustarroz, J. (2010). Efectividad de la rehabilitacin neuropsicolgica
en el dao cerebral adquirido (I): atencin, velocidad de procesamiento,
memoria y lenguaje. Revista de Neurologia, 51, 687698.
Orvis, K., Horn, D., & Belanich, J. (2008). The roles of task difculty and prior
videogame experience on performance and motivation in instructional
videogames. Computers in Human Behavior, 24, 24152433.
Pascual-Leone, A. et al. (2005). The plastic human brain cortex. Annual Review of
Neuroscience, 28, 377401.
Quinlan, R. (1993). C4.5: Programs for machine learning. San Mateo, CA: Morgan
Kaufmann Publishers.
Rego, P., Moreira, P. M. & Reis, L. P. (2010). Serious games for rehabilitation a survey
and a classication towards a taxonomy. In Information systems and technologies
(CISTI) (pp. 16).
Rohling, M. L., Faust, M. E., et al. (2009). Effectiveness of cognitive rehabilitation
following acquired brain injury: A meta-analytic re-examination of Cicerone
et al.s (2000, 2005) systematic reviews. Neuropsychology, 23(1), 2039.
Serra, J., Arcos, J. Ll., Garcia-Rudolph, A., Garca-Molina, A., Roig, T., & Tormos, J. M.
(2013). Cognitive prognosis of acquired brain injury patients using machine
learning techniques. In International conference on advanced cognitive
technologies and applications (COGNITIVE), IARIA (pp. 108113). Valencia, Spain.
Sohlberg, M. M. (2001). In A. Mateer (Ed.), Cognitive rehabilitation. An interactive
neuropsychological approach. Catherine. ISBN: 9781572306134.
Tormos, J. M., Garcia-Molina, A., Garcia Rudolph, A., & Roig, T. (2009). Information
and communications technology in learning development and rehabilitation.
International Journal of Integrated Care, 9. 22 June, ISSN 1568-4156.
Vygotsky, L. S. (1934). Dinamika umstvennogo razvitiia shkolnika v sviazi s
obucheniem. In Umstvennoe razvitie detei v protsesse obucheniia (pp. 3352).
Gosuchpedgiz: Moscow-Leningrad.
Vygotsky, L. S. (1978). Mind and society: The development of higher mental processes.
Cambridge, MA: Harvard University Press.
Walsh, R. S., Donal, G., Stephen, G., & Muldoon, T. (2012). Acquired brain injury:
Combining social psychological and neuropsychological perspectives. Health
Psychology Review.
Whalen, S. (1998). Revisiting the problem of match. In N. Colangelo, S. Assouline
(Eds.), Proceedings from the Henry B. and Jocelyn Wallace national research
symposium on talent development.
Whyte, J., & Hart, T. (2003). Its more than a black box; its a Russian doll: Dening
rehabilitation treatments. American Journal of Physical Medicine and
Rehabilitation, 82, 639652.
Witten, I. H., & Frank, E. (2005). Data mining: Practical machine learning tools and
techniques (3rd ed.). The Morgan Kaufmann series in data management systems.

Please cite this article in press as: Garca-Rudolph, A., & Gibert, K. A data mining approach to identify cognitive NeuroRehabilitation Range in Traumatic

Q1 Brain Injury patients. Expert Systems with Applications (2014), http://dx.doi.org/10.1016/j.eswa.2014.03.001

1260
1261
1262
1263
1264
1265
1266
1267
1268
1269
1270
1271
1272
1273
1274
1275
1276
1277
1278
1279
1280
1281
1282
1283
1284
1285
1286
1287
1288
1289
1290
Q9 1291
Q101292
1293
1294
1295
1296
1297
1298
1299
1300
1301
1302
1303
1304
1305
1306
1307
Q111308
1309