A Hybrid Machine Learning Approach with SMOTE for Improved Endometriosis Diagnosis

Endometriosis is a widespread condition characterized by pelvic discomfort and subfertility. Endometriosis arises when tissue similar to the uterine lining spreads outside the female reproductive organs. While laparoscopic excision remains an effective treatment, delayed diagnosis remains a significant barrier, allowing symptoms to progress unnecessarily. Due to its visual diversity, non-specialized physicians often struggle to recognize it, and even image-based classification proves difficult. Advances in deep learning have revolutionized medical imaging analysis, with algorithms now matching clinician-level accuracy in identifying pathological features. This research aims to help doctors recognize endometriosis by analyzing images from laparoscopic surgery. Our methodology combines Convolutional Neural Networks (CNNs) with Vision Transformers (ViTs), harnessing their combined power for accurate endometriosis lesion detection. CNNs extract localized features, while the ViTs attention mechanism captures global context, together enabling precise lesion identification. The study employed the publicly accessible Gynecologic Laparoscopy Endometriosis Dataset (GLENDA) for the model training, validation, and testing. The proposed models adopted the synthetic minority oversampling technique (SMOTE) to address class imbalance issues. The ViTs and CNNs models were assessed using four key metrics: accuracy, precision, recall, and F1 score. Our hybrid framework achieved state-of-the-art performance, with an accuracy of 97.61%, precision of 97.74%, recall of 96.67%, and an F1 score of 97.18%, demonstrating its effectiveness in endometriosis lesion detection. We additionally evaluate our model's performance using Receiver Operating Characteristic (ROC) curves and their corresponding Area Under the Curve (AUC) values. Finally, a confusion matrix is employed to evaluate the model's effectiveness comprehensively. The integration of CNNs and ViTs represents a transformative advancement in medical and therapeutic research, showing significant potential to accelerate innovation in diagnostic and treatment methodologies.