Ovarian Tumor Diagnosis and Characterization of CT Scan Images Using Ensemble Deep Learning and Explainable AI

The application of ensemble deep neural network strategies for classifying ovarian tumors using CT-scanned images has become increasingly important in recent years. Accurate classification of ovarian tumors from these images is pivotal for early diagnosis and effective treatment planning. Our research proposes an integrated approach that combines explainable Artificial Intelligence (XAI) techniques with Ensemble Deep Neural Networks to enhance ovarian tumor classification. We utilize the interpretability aspects of XAI to provide insights into the decision-making process of the ensemble deep neural networks. By visualizing regions of interest and highlighting contributing features in CT-scanned images, we aim to bridge the gap between deep learning models and clinical interpretability, offering an accurate yet explainable solution. The proposed ensemble framework incorporates a diverse set of deep neural networks, each with distinct architectures and initializations, to improve tumor detection performance in ovarian CT images. This is achieved using SHAP, SmoothGrad, and GradCAM for enhanced explainability. Besides performance improvements, our methodology is robust, generalizable, and scalable, contributing new insights to radiological diagnostics. It empowers clinicians and radiologists to validate and understand the model's predictions, fostering trust and acceptance in real-world medical settings. We validated our methodology on a comprehensive in-house dataset of CT-scanned images, demonstrating quantitative performance improvements with a mean accuracy of 98.96 ± 1.269, and Dice and Jaccard scores of 0.98 and 0.97, respectively. The methodology uses Transformers for both quantitative and qualitative interpretability, paving the way for standardizing diagnostics in ovarian tumor characterization.