Enhanced DenseNet121 Framework for Accurate PCOS Detection and Recurrent Miscarriage Prediction Using Statistical Features and Advanced Convolutional Techniques

Polycystic Ovary Syndrome (PCOS) is a prevalent endocrine disorder affecting women of reproductive age and is closely associated with infertility and recurrent miscarriages. Early and accurate detection is essential for timely intervention. This study proposes an enhanced computational framework for PCOS detection and recurrent miscarriage prediction using a multi-hospital clinical dataset consisting of 10 hospitals in Kerala, India, comprising 541 patient records (both clinical and physical parameters). The preprocessing pipeline integrates Z-score normalisation, SMOTE-based augmentation, and median imputation to address scale variation, class imbalance, and missing data. Statistical features, Pearson Correlation Coefficient (PCC), and an Improved IQR-PCA approach are employed for robust feature extraction. These features are further refined using a Self-Attention GRU (SA-GRU) network. For classification, DenseNet121 is enhanced with dilation rates and residual dense blocks to improve contextual learning and feature reuse. Hyperparameters are optimised using an Improved Secretary Bird Optimisation Algorithm (I-SBOA). The proposed model achieves a superior accuracy of 0.9415, outperforming existing architectures. Results demonstrate that integrating statistical analysis with advanced deep learning mechanisms significantly improves diagnostic reliability for both PCOS detection and miscarriage risk prediction.