Polycystic ovary syndrome (PCOS) is a prevalent endocrine disorder affecting women of reproductive age. Early and accurate diagnosis is crucial for managing its associated health complications. This research proposes a novel probabilistic approach for PCOS detection using ovarian ultrasound images, integrating deep learning models with Bayesian inference and automatic report generation. Our approach leverages the strengths of two convolutional neural networks (CNNs): PCONet and MobileNet. PCONet, specifically designed for PCOS detection, captures subtle ultrasound features indicative of polycystic ovaries. MobileNet, known for its lightweight architecture, facilitates efficient deployment on resource-constrained platforms. We employ Bayesian CNN ensembles to learn from the combined knowledge of these models, generating probabilistic predictions for each image. Further, we incorporate XGBoost, a gradient boosting algorithm, to refine these probabilities and improve diagnostic accuracy. Through our study and work, we are successfully able to make a high-performing ensemble model, exhibiting 98.6% accuracy on validation data. To streamline the diagnostic workflow, we develop an automatic report generation module. This module extracts key findings from the ultrasound images, analyzes predictions from the probabilistic model, and generates a comprehensive report summarizing the PCOS diagnosis and potential clinical implications. This automation expedites decision-making and enhances healthcare provider efficiency. Our proposed framework, combining deep learning, Bayesian inference, and automatic report generation, offers a promising avenue for accurate and early PCOS detection, ultimately improving patient care and management of this prevalent condition.