1

A Soft Computing Approach for Osteoporosis Risk Factor Estimation

Dimitrios Mantzaris, George Anastassopoulos, Lazaros Iliadis, Konstantinos Kazakos and Harris Papadopoulos Artificial Intelligence Applications and Innovations IFIP Advances in Information and Communication Technology, 2010, Volume 339/2010, 120127, DOI: 10.1007/978-3-642-16239-8_18

Abstract
This research effort deals with the application of Artificial Neural Networks (ANNs) in order to help the diagnosis of cases with an orthopaedic disease, namely osteoporosis. Probabilistic Neural Networks (PNNs) and Learning Vector Quantization (LVQ) ANNs, were developed for the estimation of osteoporosis risk. PNNs and LVQ ANNs are both feed-forward networks; however they are diversified in terms of their architecture, structure and optimization approach. The obtained results of successful prognosis over pathological cases lead to the conclusion that in this case the PNNs (96.58%) outperform LVQ (96.03%) networks, thus they provide an effective potential soft computing technique for the evaluation of osteoporosis risk. The ANN with the best performance was used for the contribution assessment of each risk feature towards the prediction of this medical disease. Moreover, the available data underwent statistical processing using the Receiver Operating Characteristic (ROC) analysis in order to determine the most significant factors for the estimation of osteoporosis risk. The results of the PNN model are in accordance with the ROC analysis and identify age as the most significant factor. Keywords Artificial Neural Networks - Probabilistic Neural Networks - Learning Vector Quantization Neural Networks - ROC analysis - Osteoporosis

Diagnosis Classification Research for Disease of Osteoporosis Based on BP Neural Network

CHEN Ruo-zhu1, YANG Zi-juan1,2, WEI Zhe2 (1.College of Electrical Engineering and Information Engineering, Lanzhou University of Technology, Lanzhou 730050, China; 2.General Hospital of Lanzhou, Lanzhou Military Area Command, Lanzhou 730050, China) Objective To reduce the misdiagnosis rate of osteoporosis. A diagnostic classification facilitates rapid and accurate diagnosis. Methods We utilized the theory of BP neural network model to construct osteoporosis diagnosis classification model for analyzing osteoporosis assessment indicator system. BP neural network system is designed according to the indicator system. The artificial neural network was used to compute actual data of a large number of cases in Orthopaedics Department of General Hospital of Lanzhou Military Area Command. Results The error of the model was controlled within predetermined error range. It is demonstrated to be feasibile and accurate. Conclusion The diagnosis of osteoporosis by using BP neural network is feasible and effective.

Hybrid Artificial Intelligence Ensemble for Early Detection of Osteoporosis Dr W Wang , Dr S Rae

Summary The osteoporosis disease has been identified by WHO as a global "silent epidemic" which affects 1 in 3 women. There is currently no cure but it could be effectively prevented. Early detection of the disease is key to prevention, and forms the basis of our research. The principal goal of the proposed research is to investigate a methodology of constructing hybrid ensemble of data mining techniques for identifying salient risk factors and then to develop an artificial intelligence system to assist early detection. The proposed project is viewed as an extension of our current and past work on both the chosen technologies and their application for the osteoporosis problem. The proposed investigations will explore reasoning techniques such as, automatically induced decision trees and neural networks, as well as some conventional techniques commonly applied in medical fields, such as correlation, logistic regression and probabilistic analysis. A hybrid ensemble pursues an effective combination of various approaches in order to utilise the strengths of each individual technique and compensate for their weaknesses and therefore produce 'optimal' decisions. The focus of this proposal is to investigate and develop such a system Final Report This project has achieved the objectives that were defined in the original proposal. Specially, based on our evaluations(objective 4) of some commonly used related techniques, we developed the techniques for (objective 1) identifying the risk factors from patients' data and their impact value (salience) on the development of the osteoporosis, and for (objective 3) building hybrid ensembles of artificial neural networks and decision trees for detecting patients who have an increased risk of developing the osteoporosis disease. The techniques developed were tested on 12 benchmark data sets and then applied to the osteoporosis data collected in this project. The data of over 3000 osteoporosis patients have been collected in this project. All the patients were referred by their GPs to hospitals for further screening examinations by using either the Dual-energy X-ray Absorptiometry (DEXA or DXA) or PIXI (Peripheral DXA) or Quantitative Ultrasound Scanner (QUS). Thus three data sets were generated according to their diagnosis devices i.e. DXA, PIXI and QUS.

4 Each set consists of around 1000 records and each record has 30 to 45 factors. Based on the data, (for objective 2) we analysed the performance of these screening instruments in terms of ability in picking up risk factors and high risk patients firstly with the traditional statistic methods before being put through the developed techniques for estimating the salience of each factor and selecting the relevant risk factors for inducing the models for building various types of ensembles for classifying patients into normal or abnormal in terms of their likelihood of osteoporosis risk. Our feature salience estimation techniques have estimated the relative impact value for each factor and then identified the most relevant risk factors accordingly. The effectiveness of the selected factors was tested by comparing the classification accuracy of the ensemble of the models that were used with only selected factors and the ensemble of the models that were trained with all the factors in the data sets. The comparison results indicated that the hybrid ensembles of neural net and decision trees trained with 15 selected factors produced the highest classification accuracy. The sensitivity (accuracy for classifying true positive cases) and specificity (accuracy for classifying true negative cases) were used to further assess the performance of the models and ensembles. On average, our ensembles produced 80% and 90% respectively. If sacrificing specificity to as low as 50%, then the ensembles are able to achieve a higher sensitivity of about 90% on average. Compared with other existing risk-factor based osteoporosis assessment tools such as OSIRIS, our models and ensembles are more accurate and reliable but more complex. It should be noted that the techniques and ensemble systems developed are not intended to replace the existing screening devices but to aid identifying high risk patients in primary care. On the methodology, this project has produced 6 papers published in Journals or major international conferences on both computing and health care fields. Built upon our technical achievements, a joint research grant was award by the ESRC to apply our techniques to a social problem. In addition, this research has attracted a high profile of attentions from various communities from its beginning to the end. The research has been reported in regional newspapers and the EPSRCs magazine Newsline. These activities have disseminated our research and findings to all levels of society and directly helped to increase the public awareness of the osteoporosis, a benefit brought to society. Further research will focus on verifying the findings (identified risk factors and the robustness of the ensembles) by using new independent data and monitoring the changes of the recorded patients bone mineral density over time, and these followup work will be essential to exploit the achievements of this project and help to transfer the knowledge into practice.

Walid MOUDANI, Ahmad SHAHIN, Fadi CHAKIK and Dima RAJAB. Article:Intelligent Predictive Osteoporosis System. International Journal of Computer Applications 32(5):28-37, October 2011. Published by Foundation of Computer Science, New York, USA. BibTeX
@article{key:article, author = {Walid MOUDANI and Ahmad SHAHIN and Fadi CHAKIK and Dima RAJAB}, title = {Article:Intelligent Predictive Osteoporosis System}, journal = {International Journal of Computer Applications}, year = {2011}, volume = {32}, number = {5}, pages = {28-37}, month = {October}, note = {Published by Foundation of Computer Science, New York, USA} }

Abstract

The healthcare environment is generally perceived as being information rich yet knowledge poor. The healthcare industry collects huge amounts of healthcare data which, unfortunately, are not mined to discover hidden information. However, there is a lack of effective analysis tools to discover hidden relationships and trends in data. The information technology may provide alternative approaches to Osteoporosis disease diagnosis. In this study, we examine the potential use of classification techniques on a massive volume of healthcare data, particularly in prediction of patients that may have Osteoporosis Disease (OD) through its risk factors. For this purpose, we propose to develop a new solution approach based on Random Forest (RF) decision tree to identify the osteoporosis cases. There has been no research in using the afore-mentioned algorithm for Osteoporosis patients prediction. The reduction of the attributes consists to enumerate dynamically the optimal subsets of the reduced attributes of high interest by reducing the degree of complexity. A computer-aided system is developed for this purpose. The study population consisted of 2845 adults. The performance of the proposed model is analyzed and evaluated based on set of benchmark techniques applied in this classification problem.

Medical disease prediction using Artificial Neural Networks

Mantzaris, D.H.; Anastassopoulos, G.C.; Lymberopoulos, D.K.;
+Mantzaris, D.H.+

Med. Inf. Lab., Univ. of Thrace, Alexandroupolis This paper appears in: BioInformatics and BioEngineering, 2008. BIBE 2008. 8th IEEE International Conference on Issue Date: 8-10 Oct. 2008 On page(s): 1 - 6 Location: Athens Print ISBN: 978-1-4244-2844-1 References Cited: 29 INSPEC Accession Number: 10412565 Digital Object Identifier: 10.1109/BIBE.2008.4696782 Date of Current Version: 08 December 2008

Abstract
This study examines a variety of artificial neural network (ANN) models in terms of their classification efficiency in an orthopedic disease, namely osteoporosis. Osteoporosis risk prediction may be viewed as a pattern classification problem, based on a set of clinical parameters. Multilayer perceptrons (MLPs) and probabilistic neural networks (PNNs) were used in order to face the osteoporosis risk factor prediction. This approach is the first computational intelligence technique based on ANNs for osteoporosis risk study on Greek population. MLPs and PNNs are both feed-forward networks; however, their modus operandi is different. Various MPL architectures were examined after modifying the number of nodes in the hidden layer, the transfer functions and the learning algorithms. Moreover, PNNs were implemented with spread values ranging from 0.1 to 50, and 4 or 2 neurons in output layer, according to coding of osteoporosis desired outcome. The obtained results lead to the conclusion that the PNNs outperform to MLPs, thus they are proved as appropriate computation intelligence technique for osteoporosis risk factor prediction. Furthermore, the overfitting problem was more frequent to MLPs, contrary to PNNs as their spread value increased. The aim of proposed PNN is to assist specialists in osteoporosis prediction, avoiding unnecessary further testing with bone densitometry.

Osteoporosis classification using fuzzy rule based and neural networks

Badawi, A.M.;
+Badaw i, A.M.

Fac. of Eng., Cairo Univ. This paper appears in: Circuits and Systems, 2003 IEEE 46th Midwest Symposium on Issue Date: 30-30 Dec. 2003 On page(s): 405 - 407 Vol. 1 Location: Cairo ISSN: 1548-3746 Print ISBN: 0-7803-8294-3 INSPEC Accession Number: 8823944 Digital Object Identifier: 10.1109/MWSCAS.2003.1562304 Date of Current Version: 27 December 2005

Abstract
Most bone densitometry ultrasound devices measure only single predefined peripheral skeletal site. We propose a classification system to study the ability of combining speed of sound (SOS) measured at multiple bone sites to differentiate subjects with osteoporosis fractures from normal subjects based on fuzzy logic and neural networks systems. Classification rates were found to be 100% for training set and 97% for testing set for a dataset of 66 subjects