IJSTE - International Journal of Science Technology & Engineering | Volume 3 | Issue 11 | May 2017

ISSN (online): 2349-784X

A Mixture of Experts Model for Extubation

Vinod Kulkarni
PG Student
Department of Computer Network Engineering
Visvesvaraya Technological University, Jnana Sangama,
Machhe, Belagavi, Karnataka 590018
Anand Unakal
PG Student
Department of Computer Network Engineering
Visvesvaraya Technological University, Jnana Sangama,
Machhe, Belagavi, Karnataka 590018

Rohit More Dr. R.H. Goudar

PG Student Assistant Professor
Department of Computer Network Engineering Department of Computer Network Engineering
Visvesvaraya Technological University, Jnana Sangama, Visvesvaraya Technological University, Jnana Sangama,
Machhe, Belagavi, Karnataka 590018 Machhe, Belagavi, Karnataka 590018

Abstract
Extubation refers to removal of endotracheal tube (ETT).Intubation refers to the process of putting a tube into a hollow organ or
passageway. Mixture of Experts (MoE) model is an architecture consisting of n expert networks. These experts are connected via
gating network. The expert is a person who has effective knowledge and experience in a particular field. The expert is taken as a
doctor/clinician who diagnoses patients. The patient data is collected and processed and a suitable time is decided to extubate.
Currently, the data collection is limited to a known number of patients. For each patient, about 100 features are extracted. In this
paper, a system is proposed which contains a model that takes the data from the patients irrespective of their count. The model
takes both clinical and physiological data. Gating function is used which chooses weight for each expert for a given input. The
model is predictive in which probability of success of extubation is more.
Keywords: Extubation, Mixture of Experts, Patient, Gating Function, Probability
________________________________________________________________________________________________________

I. INTRODUCTION

Now days, medical science is touching heights initiating from birth of a child to laser surgeries. This has led to increase in
survival of individuals, irrespective of individual’s age. One of these technologies has been discussed in this paper. This
technology is named as Extubation. In this technique, patient is discontinued the supply of artificial airway. This is done under
the care of adequate physician, nursing and therapist staffs. Extubation is performed when an artificial airway is no longer
needed. This technique will extend the survival of the patient as intubation is done in initial phase in terms of artificial airway.
When the patient is healthy enough, airway is removed which is called extubation. Proper planning should be done by
communication between physician, therapist and nursing staff before extubation. The equipments should be arranged required
for the treatment. After extubation is performed, the patient should be continuously assessed to verify his/her comfort and ease of
breathing.
Initiating from infants to older ones, there is some or other health problems such as fever, common cold and several breathing
problems. Due to these situations, several people lose their lives due to improper treatment. Taking above factors into
consideration, the solution is proposed in this paper.
In earlier days, the dataset taken for the patients was small. So, it was difficult to make classification. So, if dataset is large, it
will be easier for classification. Also, larger dataset will provide more features. This technique requires both physiological and
clinical data. So, taking these two together, the performance of the model can be increased.

II. THE PROPOSED SYSTEM

The proposed system for extubation is shown below. The system contains a model that takes the data from the patients
irrespective of their count. The model takes both clinical and physiological data.

All rights reserved by www.ijste.org 309

 A Mixture of Experts Model for Extubation
(IJSTE/ Volume 3 / Issue 11 / 057)

Fig. 1: Proposed System

As shown in Fig. 1, there are n patients considered which are shown on the top of the system. Next, there are experts which are
also taken ‘n’ in number. The lower part of the system shows the procedure of extubation.
The model used here is a tree-like structure which is shown in Fig. 2. It is based on “Divide and Conquer” method. It has three
important components: Probabilistic Model, Gate and Expert. The Expert is considered as either regression function or classifier.
These experts are more in number. The Gate acts as a partition of input space. It takes only those regions in which individual
expert opinions are of more trust. The Probabilistic Model combines gate and experts. It is a sum-weight of experts where the
weight is defined as an input-dependent gate. The MoE has three important properties: (i) Allows an expert to specialize on sub
problems of a large problem (ii) Makes data to be allocated to applications within same operating system (iii) Form partitions
along one dimension less than the surrounding space at random alignments in the input space. In MoE, a gate and a group of
experts collaborate by separating an input space into a clustered group of sections.

Fig. 2: The Probabilistic Model

Suppose there is “an input vector ‘a’ and target vector ‘b’, the total probability of declaring b” can be given in respect of experts:

Where,
k-Number of Experts
P (k|a, ϴg)-Gate’s Rating
P (b|k, a, ϴe)-Probability of kth expert generating b given a
Based on above calculations, the system is proposed. A number of patients visit a particular hospital on a daily basis. They
consult their concerned doctors/physicians depending on their availability. The doctors/physicians are considered as experts here.

III. ALGORITHMS FOR EXTUBATION

Firstly, there is “Patient Arrival” algorithm. It deals with patient’s description and his/her clinical health.

All rights reserved by www.ijste.org 310

 A Mixture of Experts Model for Extubation
(IJSTE/ Volume 3 / Issue 11 / 057)

Patient Arrival
1) Patient arrives at a particular hospital.
2) Required documents will be submitted.
3) Documents will be processed by hospital officials.
Second one, doctor diagnoses patients. Here, doctors treat the patients the disease they are suffering from and diagnose
according to their knowledge.

Diagnose by Doctor
1) Patients are classified according to their health issues.
2) Clinical features are collected from individual patients.
3) The patients are diagnosed by the respective doctors.
Lastly, extubation is continued. Prior to this, patient is treated for his/her comfort.

Extubation Procedure
1) Heart rate and respiratory measurements are recorded immediately.
2) Clinical database is prepared for each patient.
3) Finally, extubation probability is predicted.

IV. RELATED WORK

In this section, we will go through the work done by different authors on Mixture of Experts (MoE) and Extubation. Combining
these two concepts, the survival analysis of patients can be increased.
J.Kaczmarek, et.al [2] discussed a new predictor to help doctors to decide the suitable time to extubate. It uses modern
machine learning approach. Pascale Gourdeau, et.al [3] explained an approach for analysis of clinical data from extremely
preterm infants. It includes about 100 clinical features. Also, they have used SMOTE algorithm. Seniha E. Yuksel, et.al [4]
discussed about the fundamentals of Mixture of Experts (ME).Also they have discussed about the fundamental models for
regression and classification. They have also listed statistical properties of ME and its applications. B.R.Matam, et.al [5]
evaluated that automated prediction systems have ability to predict unplanned extubations.Also investigated whether system
which is capable of determining prediction could be utilized to intercept unplanned further extubations.
Lara J. Kanbar, et.al [1] described the automated validation and quality control procedure without any user supervision and
takes care of data acquired from different locations. It is mainly for APEX project for preterm infants. Martina Mueller, et.al [6]
discussed a decision-support tool to assist clinicians to decide the best time to extubate for premature infants. Allou Same [7] has
discussed a new approach for dynamic modeling and dimensionality reduction from time series data.Rahul Gupta, et.al [8]
discussed MoE approach for intelligible classification of pathological speech.
In the above discussion made, it shows the work done in both Mixture of Experts (MoE) and Extubation. Some authors have
explained the applications of MoE whereas; some have discussed extubation using different techniques and algorithms. Some
authors have described Mixture of Experts (MoE) only whereas, some authors have explained only extubation.

V. CONCLUSION

In this paper, extubation is discussed using Mixture of Experts (MoE).The clinicians/doctors are considered as experts. Patients
are taken as input variables and doctors are taken as target variables. The probability of extubation depends on choosing the
target variable. The system proposed here makes distribution of patients which eases classification. Prior to extubation, patient is
diagnosed regularly which makes the technique strong enough to extend the survival of patient and best time can be chosen to
extubate.

REFERENCES
[1] Lara J. Kanbar, Wissam Shalish, Doina Precup, Karen Brown, G. M. Sant’Anna, Robert E. Kearney. “Automated ongoing data validation and quality
control of multi-institutional studies”, 38th Annual International Conference of the IEEE Engineering in Medicine and Biology Society (EMBC), pp 2504-
2507, 2016.
[2] Doina Precup, Carlos A. Robles-Rubio, Karen A. Brown, L.Kanbar, J.Kaczmarek, S.Chawla, G.M. Sant’ Anna, Robert E. Kearney. “Prediction of
extubation readiness in extreme preterm infants based on measures of cardio respiratory variability”, Annual International Conference of the IEEE
Engineering in Medicine and Biology Society, pp 5630-5633, 2012.
[3] Pascale Gourdeau, Lara Kanbar, Wissam Shalish, G.Sant’Anna, Robert Kearney, Doina Precup. “Feature selection and oversampling in analysis of
clinical data for extubation readiness in extreme preterm infants”, 37th Annual International Conference of the IEEE Engineering in Medicine and Biology
Society (EMBC), pp 4427-4430, 2015.
[4] Seniha E. Yuksel, Joseph N. Wilson, Paul D. Gader. “Twenty Years of Mixture of Experts”, IEEE Transactions on Neural Networks and Learning Systems,
pp 1177-1193, 2012.
[5] B.R.Matam, B.K.Fule, H.P.Duncan, D.Lowe. “Predictability of unplanned extubations”, IEEE EMBS International Conference on Biomedical and Health
Informatics (BHI), pp 488-491, 2014.

All rights reserved by www.ijste.org 311

 A Mixture of Experts Model for Extubation
(IJSTE/ Volume 3 / Issue 11 / 057)

[6] Martina Mueller, Carol C. Wagner, Romesh Stanislaus, Jonas S. Almeida. “Machine learning to predict extubation outcome in premature infants”, The
International Joint Conference on Neural Networks (IJCNN), pp 1-6, 2013.
[7] Allou Same “Dynamic Factor Mixture of Experts for Functional Time Series Modeling”, 15th IEEE International Conference on Machine Learning and
Applications (ICMLA), pp 19-25, 2016.
[8] Rahul Gupta, Kartik Audhkhasi, Shrikanth Narayanan. “A mixture of experts approach towards intelligibility classification of pathological speech”, IEEE
International Conference on Acoustics, Speech and Signal Processing (ICASSP), pp 1986-1990, 2015.
[9] Joseph Santarcangelo, Xiao-Ping Zhang. “Kernel-based mixture of experts models for linear regression”, IEEE International Symposium on Circuits and
Systems (ISCAS), pp 1526-1529, 2015.
[10] Everson Verissimo, Diogo da Silva Severo, George D. C. Cavalcanti, Tsang I. Ren. “Diversity in task decomposition: A strategy for combining mixtures
of experts”, The International Joint Conference on Neural Networks (IJCNN), pp 1-5, 2013.

All rights reserved by www.ijste.org 312