From Pixels to Diagnosis: An Innovative EfficientNetB3 Approach for Interpreting Lung and Colon Cancer Histopathology

Globally, lung and colon cancers are predominant contributors to cancer mortality. Early and accurate diagnosis is crucial for effective treatment. While deep learning models have shown promise in automating cancer classification from histopathological images, challenges remain in achieving high accuracy while maintaining interpretability. This research focuses on creating a high-efficiency and interpretable deep-learning model for identifying lung and colon cancer from histopathological images. In this paper, a novel approach utilizing a modified EfficientNetB3 architecture is introduced. The model was trained on a 25,000 high-resolution histopathological images dataset across five classes. Explainable AI (XAI) techniques, specifically LIME and SHAP, were integrated to enhance model interpretability. The proposed model achieved an exceptional accuracy of 99.66%, outperforming other state-of-the-art architectures. The model demonstrated high precision (99.78%), recall (98.58%), and F1-score (99.18%). Incorporating LIME and SHAP offered a significant understanding of how the model makes decisions, emphasizing relevant histological characteristics that affect classification.