Reliable and High-Precision PCOS Detection Using C2T-Transformer with Uncertainty-Calibrated Learning on Clinical Text Data

Objectives: To propose a reliable and high-precision deep learning-based PCOS detection model using clinical text data in order to maximize diagnostic accuracy, minimize false predictions and clinician dependency. The model improves robustness under class imbalance and enables scalable, real-time deployment as a support system across diverse healthcare environments. Methods: A Causal-Contrastive Tabular Transformer with uncertainty-calibrated learning (C2T-UCL) is employed to learn clinically valid feature interactions and latent symptom dependencies among heterogeneous text features for accurate PCOS detection and to ensure reliable, low-error and confidence-aware predictions suitable for real-time decision support. The text attributes are encoded as tabular tokens and optimized using supervised contrastive and focal uncertainty losses. Three quantitative PCOS text datasets are extracted from the Kaggle and Bio-GPS repositories, which comprise 285, 250 and 541 patient records with PCOS and non-PCOS labels, and are utilized for training, testing and validation. The proposed model is assessed using Python, Scikit-learn, PyTorch, and Google Colab, and the results are compared against the baseline models such as Re-PSO, SVM-ICA, and DBMM-KPCA. Findings: The proposed C2T-UCL model demonstrates consistent performance with promising results across all three benchmarked datasets with 99.3% prediction accuracy, 99.2% validation accuracy, 99.6% precision, 98.8% F1-score, and a reduced RMSE of 0.6%, which indicates superior generalization and a stable PCOS diagnostic support system. Novelty: This research integrates a unique combination of casual-guided attention approach with supervised contrastive separation and uncertainty within a tabular transformer for efficient PCOS detection. The proposed C2T-UCL model directly learns complex feature interactions from structured text input data, which enables accurate and reliable screening process and helps practical clinical workflows for timely, cost-effective, resource-constrained and high-volume diagnostic settings, making it a promising solution for scalable PCOS diagnosis. Keywords: PCOS Detection, Deep Learning, Tabular Transformer, Decision Support System, High-Precision System, Diagnostic Model