Interpretable Lightweight CNN for Colon and Lung Cancer Classification with LIME Based Explainability

Colon and lung cancers are one of the major reasons reported for cancer related death across the globe, which underscores the vital need for early and accurate method for its diagnosis. Our study presents a lightweight CNN model for classification of colon and lung cancer disease using histopathological images. The proposed model fuses Convolutional Block Attention Module (CBAM) within adaptive residual blocks where skip connections are created by adding input tensor to the output of residual block, so as to enhance feature extraction and refine the feature map through inclusion of attention mechanism. Comprehensive evaluations on LC25000 dataset for both lung and colon cancers demonstrate the model's effectiveness in classification, with overall accuracy of 99.8%. In spite of its compact architecture with only 0.9M trainable parameters our model achieved high classification accuracy. Further to facilitate transparency and interpretability Local Interpretable Model-agnostic Explanation (LIME) is being generated which provides insights into the region which contributes to model's decision. The incorporation of LIME facilitates a deeper comprehension of the model's behavior, supporting its potential for clinical adoption and improving diagnostic reliability.