Early Detection of Ovarian Cancer From Histopathological Images Using Deep Learning Models and Explainable AI

Ovarian cancer remains a significant public health challenge due to difficulties in early detection. It is often tricky to catch early on, making treatment more difficult. Histopathological examination of tissue samples is the gold standard for diagnosis, but traditional methods can be subjective and time-consuming. This study explores the potential of deep learning models for accurate and early detection of ovarian cancer from histopathological images. Using a binary classification approach, we compare the performance of various pre-trained convolutional neural networks (CNNs), including VGG16, ResNet50, EfficientNet B0, and EfficientNet B1. Our findings demonstrate that EfficientNet B0 achieves the highest accuracy in distinguishing cancerous from non-cancerous tissues achieving the highest test accuracy of 99.09%. Furthermore, we leverage Score-CAM, an XAI technique, to highlight the image regions most crucial in the model’s classification decisions. This approach improves model interpretability and fosters trust in its predictions. Our research suggests that deep learning models, coupled with XAI techniques, hold significant promise for improving the accuracy and efficiency of ovarian cancer detection, potentially leading to enhanced patient outcomes.