ISOLATED FEATURE EXTRACTION AND QUANTITATIVE METRICS FOR HEART DISEASE PREDICTION

Abstract

Accurate and discriminative feature extraction is a critical prerequisite for effective medical image analysis, particularly in detecting and characterizing heart disease. This study proposes a structured feature extraction pipeline that integrates texture, frequency, and shape-based descriptors to comprehensively represent cardiac imaging data. Texture features are derived using Gray-Level Co-occurrence Matrix (GLCM) analysis in four directional orientations (0°, 45°, 90°, and 135°) to capture spatial intensity relationships. Frequency-domain features are obtained through multi-level Discrete Wavelet Transform (DWT) decomposition, enabling the identification of both coarse and fine details across multiple scales. Shape descriptors are calculated to preserve structural information relevant to anatomical interpretation. The extracted features are quantitatively evaluated to ensure stability, distinctiveness, and interpretability, providing a robust dataset for downstream classification or diagnostic modelling. By emphasizing precision in calculation and complementarity among feature types, the proposed methodology offers a flexible foundation adaptable to diverse medical imaging scenarios. This work does not address classification directly but establishes the essential feature-level groundwork upon which accurate, reliable, and clinically relevant automated diagnostic systems can be built.