Enhanced Feature Extraction-Based Hybrid Deep Learning Model for Polycystic Ovary Syndrome Detection

Abstract

Polycystic ovary Syndrome (PCOS) has been recognized as one of the greatest health complications affecting majority of women around the world. Early identification and management of PCOS are crucial to minimize the risk of long-term morbidity. Proper and early diagnosis can assist the healthcare systems in reducing the problems and complications of the disease. The conventional methods of diagnosis, including blood examination and ultrasounds, are time-consuming and expensive. Although there have been previous methods with efforts to counter these shortcomings, they are still limited by poor feature learning ability and extreme overfitting. To overcome these challenges, this study introduces an Enhanced Feature Extraction Module Enclosed Hybrid Deep Learning (EFEM-HDL) model. The proposed methodology addresses class imbalance and overfitting through Generative Adversarial Networks (GAN) for synthetic data generation, employs a Feature Interaction Channel Attention-enclosed Variational Autoencoder (FICA-VAE) for feature extraction, and utilizes a Cross Attention-based Convolutional Neural Network (CA-CNN) for classification. Experimental findings show that the EFEM-HDL model greatly improves diagnostic performance with an accuracy rate of 99.1%, hence presenting a strong solution to early PCOS detection.