Advanced Machine Learning Framework for Early Detection and Classification of Polycystic Ovary Syndrome (PCOS)

Abstract

Polycystic Ovary Syndrome (PCOS) is a common endocrine and metabolic condition in women of reproductive age that causes infertility, endocrine disturbances, obesity, insulin resistance and cardiovascular issues. The diagnosis of PCOS can be difficult, as clinical presentations are not the same and conventional diagnosis is limited. This study introduces a novel machine learning approach to early detection and classification of PCOS from multi-modal clinical and imaging data. The proposed methodology is a combination of Adaptive preprocessing technique, Adaptive Ant-Lion Optimization (AALO) based feature selection technique, Hybrid Convolutional Neural Network (CNN) and Bidirectional Long Short-Term Memory (BiLSTM) architectures, Transformer based attention mechanism, Explainable Artificial Intelligence (XAI) techniques such as SHAP and Grad-CAM. The framework can successfully obtain the discriminative spatial and temporal features from ultrasound images and clinical records to accurately predict the disease. The experimental results showed better accuracy (99.12%), precision (98.96%), recall (98.84%), F1-score (98.90%), AUC (99.28%)), which were better than the existing-state-of-the-art approaches. The study suggests that the proposed framework for an intelligent, interpretable, and scalable approach for early diagnosis of PCOS and managing reproductive healthcare in an intelligent manner.