An Optimized Hybrid Deep Learning Framework for Early PCOS Prediction Using Clinical and Ultrasound data

Abstract

Poly Cystic Ovaries Syndrome (PCOS) is one of the most common endocrine disorders that should be diagnosed as early as possible to minimise potential long-term metabolic and reproductive consequences. However, due to the disparate nature of clinical and ultrasound data, traditional computational diagnostic methodologies are quite challenging. This paper presents an intelligent system called Progressive Multi-Channel Diagnostic Calibration Network (PMDC-Net) which consists of four stages: medical information processing, systematical processing, progressive classification, and imaginary information synthesis. In this paper, an intelligent system Progressive Multi-Channel Diagnostic Calibration Network (PMDC-Net) to inductively process the heterogeneous medical information is proposed, which includes four stages: processing of medical information, systematical processing, progressive classification and processing of imagination information. The model well integrates clinical observations, ultrasound characteristics and medical information to form a consistent and unified representation of the image, so as to improve the consistency and reliability of medical diagnosis prediction. To strengthen data quality and further improve data before diagnostic analysis, extensive and comprehensive pre-processing and feature engineering has been implemented. According to experimental evaluation PMDC-Net provides better classification performance than other existing deep learning methods with an accuracy of 98%, precision of 97%, recall of 97%, F1-Score of 97% and Specificity of 98%. The suggested platform offers a comprehensive and expandable decision support system for precise prediction of PCOS by leveraging complementary clinical and ultrasound information.