PAMBA: A Novel Hybrid Architecture Combining Selective State Space Models (Mamba) and Ensemble Stacking for High-Precision PCOS Detection
PAMBA: A Novel Hybrid Architecture Combining Selective State Space Models (Mamba) and Ensemble Stacking for High-Precision PCOS Detection
Authors:
1st K. Geetha Uma Maheswari
Department of Computer Science Engineering
Rajiv Gandhi University of Knowledge Technologies, Basar
Basar, India geethaumamaheswari07@gmail.com
2nd N. Rachana
Department of Computer Science Engineering
Rajiv Gandhi University of Knowledge Technologies, Basar
Basar, India rachananellutla@gmail.com
3rd M. Geetha Maheshwari
Department of Computer Science Engineering
Rajiv Gandhi University of Knowledge Technologies, Basar
Basar, India geethamaheshwari888@gmail.com
4th P. Sarika
Department of Computer Science Engineering
Rajiv Gandhi University of Knowledge Technologies, Basar
Basar, India sarikavittalrao@gmail.com
Abstract—The high prevalence of Polycystic Ovary Syndrome (PCOS) among women of reproductive age has created an urgent need for accurate automated systems for clinical detection. Although manual analysis of ovarian ultrasound images is tedious and prone to human error, existing machine learning techniques face difficulties in processing high-dimensional medical images. This paper proposes a novel hybrid model named PAMBA (PodBoost-Augmented Mamba), which combines Deep Learning and Ensemble Learning for automated PCOS detection. The proposed model uses ResNet-50 as a visual backbone to ex-tract 2,048-dimensional feature vectors from ovarian ultrasound images. These features are then processed by a Selective State Space Model (Mamba-SSM) to capture long-range contextual dependencies. The outputs are fused in a Phase III Meta-Fusion layer that stacks the Mamba-SSM probability scores with those of a Random Forest classifier, with a Logistic Regression meta-learner producing the final diagnostic decision. The proposed model was evaluated on a dataset of 11,784 ovarian ultrasound images using an 80/20 stratified train-test split. PAMBA achieved an overall accuracy of 0.9876, with an F1-score of 0.9888 for PCOS-Infected cases and a high recall of 0.9923, demonstrating its strong sensitivity in detecting pathological cases. The frame-work shows strong potential for use in intelligent decision-support systems for the early detection of PCOS.