Loading...

Computer Aided Prognosis of Epileptic Patients Using Multi-Modality Data and Artificial Intelligence Techniques

Latifi-Navid, Masoud | 2009

517 Viewed
  1. Type of Document: M.Sc. Thesis
  2. Language: English
  3. Document No: 39912 (58)
  4. University: Sharif University of Technology, International Campus, Kish Island
  5. Department: Science and Engineering
  6. Advisor(s): Soltanian-Zadeh, Hamid
  7. Abstract:
  8. Abnormality detection and prognosis of epileptic patients with artificial intelligence and machine learning techniques is still in its early experimental stages. Surgical candidacy determination for epilepsy depends on the clinical actions which involve an intracranial electrode implantation followed by prolonged electrographic monitoring (EEG phase II) .This invasive test is very costly, painful and time consuming. Here the goal is integration of the two following paradigms: 1-Non invasive multimodality data of epilepsy. 2- Artificial intelligence and machine learning techniques. We have used human brain multi-modality database system that includes patient’s demographics, clinical and EEG phase I data as well as localized anatomical landmarks and identified/initialized brain structures in volumetric MRI images. Such database is very useful for meaningful data mining and knowledge discovery. Morphological and visual features of hippocampus, as a key anatomical structure involved in epilepsy disease, are the focus of attention in this thesis. In the first part of the present study we focused on non-invasive and non-visual multimodality data, like EEG phase I, medication information, medical history, neuropsychology and WADA tests in order to design an algorithm for determination of surgery candidates. Our algorithms were composed of some feature selection, clustering and classification approaches. We used “Correlation based feature selection CFS” and “Classifier subset evaluator CSE” with “Genetic algorithm GA” search tool and “ReliefF attribute evaluation” as the methods of feature selection, “Expectation maximization class EM” and “Incremental conceptual clustering COBWEB ” as the clustering methods and “Multilayer Perceptron classifier MLP” as classification tool at all stages of the study. In addition, the output of classifiers was based on Engel classification of the epileptic patients, which is a suitable standard for determination of surgery success. Using these inexpensive and available data, up to 56.11% of all Engel class and 31.65% of both Engel class and subclass were predicted correctly. Furthermore using this kind of database we could correctly predict 70.73% of class 1 instances in database that were more appropriate candidates for surgery. In the second part of this study we focused on image modality data. Using mentioned methods and combination of them we yielded up to 64.93% in correct prediction of all Engel classes and 46.75% of both Engel class and subclass were correctly predicted. Also regarding the prediction class1 instances for determination of surgical candidates, up to 88.23% of class 1 instances in dataset were predicted correctly with acceptable precision rate. The results obtained in this study can be a starting point for detecting more valuable features on the epilepsy and define an effective algorithm in order to predict the surgery outcome with low-cost and non-invasive data. It seems that the extended population size of the samples can lead to accuracy increase in the classification process.
  9. Keywords:
  10. Epilepsy ; Data Mining ; Artificial Intelligence ; Machine Learning ; Engel Classification

 Digital Object List

 Bookmark

No TOC