Unified Techniques using Statistical and Clinical Data Analysis for Enhanced Accuracy in Medical Diagnosis

Abstract

Epilepsy affects one to two percent of the global population, marked by seizures due to abnormal brain activity. Electroencephalogram signals aid in diagnosis and seizure prediction. EEG signals are classified as focal that is the affected brain areas and non-focal means unaffected regions. Accurate detection and localization of affected areas are crucial for effective treatment. newlineThis research introduces two innovative methods for seizure detection. The first uses entropy measures namely Sample, Permutation, Shannon, and Spectral entropy to capture EEG signal randomness. These features are classified using LSTM networks, offering improved accuracy. The second method transforms EEG signals into recurrence plots for feature extraction via RESNET 50 using transfer learning. Dimensionality reduction and validation are performed using PCA, followed by classification with machine learning models like Decision Tree, SVM, and Random Forest. newlineThe comparative analysis highlights that feature selection enhances classifier performance, proving the effectiveness of deep learning-based feature extraction. These methods improve seizure detection by integrating deep learning, transfer learning, and statistical techniques, ensuring better sensitivity, accuracy, and robustness. newline newline