سمانه السادات سعيدي نيا

Epilepsy ranks among the most preva‎lent chronic neurological disorders globally. Regrettably, over one-third of patients undergoing treatment do not achieve freedom from seizure, often due to inappropriate treatment strategies stemming from human error, or they may endure treatment for over five years. Consequently, predicting treatment efficacy could enhance overall treatment outcomes. In this context, machine learning tools can significantly aid in selecting the appropriate treatment strategy, given their strong capacity for processing large datasets and simulating various scenarios. This necessitates the development of a robust intelligent system that integrates clinical information and individual patient characteristics to forecast the effectiveness of initial treatments as recommended by medical guidelines. This thesis presents a three-dimensional model that combines EEG and MRI data from the patient's brain, along with the design of a novel spiking neural network. This model successfully predicts EEG data in unsupervised channels and estimates one step ahead for supervised channels, utilizing 70% of the data for training and 30% for testing. The proposed model achieved a mean square error of 0.357 and a standard deviation of 0.057, improving one-step-ahead prediction error by 50% compared to the neuron cube model. To stabilize the system—specifically, to achieve seizure freedom—a stability analysis of the new neural network was conducted using the reduced state space model based on mean-field theory, bifurcation stability analysis, and eigenvalue assessment. We eva‎luated the permissible stimulation input based on the pre-trained model tailored to patient characteristics. Furthermore, to incorporate statistical data alongside analytical calculations, we developed a new Markov decision-making model addressing three primary interventions: drug treatment, electrical stimulation, and surgery. The eva‎luation of the intelligent treatment decision-making system on two patients—one recommended for electrical stimulation and the other for surgery—demonstrated that the proposed method not only effectively identifies the optimal treatment options to support the medical team's decision-making but also assesses the potential efficacy and scope of suggested treatments based on the patient's clinical information and records in managing adult focal epilepsy.

صرع , سيستم هوشمند تصميم‌يار , پايدارسازي , پيش‌بيني ميزان اثربخشي , مدل‌سازي مغز , شبكه‌هاي عصبي ضربه‌اي

Epilepsy , Intelligent decision support system , Stabilization , Efficacy predictions , Brain modeling , Spiking neural network

Samaneh Alsadat Saeedinia

Prof. Mohammad Reza Jahed-Motlagh