Explainable Deep Learning Framework for Early Cancer Detection from Histopathological Images Using LIME and SHAP
Year: 2026
Type: article
Abstract: Early and accurate classification of malignant and non-malignant tissues is essential for improving diagnostic accuracy and patient outcomes. With the rapid advancement of artificial intelligence, machine learning , deep learning methods have demonstrated significant potential in automating histopathological image analysis and assisting clinical diagnosis. This paper proposes an explainable deep learning framework for early cancer detection using histopathological images. A total of six deep learning architectures, which are Convolutional Neural Network CNN, VGG16, ResNet50, MobileNetV2, EfficientNetB0, and InceptionV3 were tested with LC25000. It has 25,000 histopathological images which are classified into five groups such as lung squamous cell carcinoma, lung adenocarcinoma, lung benign tissue, and colon adenocarcinoma. The data was separated into 15% test, 15% validation and 70 % training of the models considered, ResNet50 had the best classification accuracy of 99% with excellent ability to generalize to all tissue classes. CNN and VGG16 also achieved high accuracy of 98%., whereas EfficientNetB0 was 97%. Conversely, MobileNetV2 (86% accuracy) and InceptionV3 (69% accuracy) demonstrated relatively worse performance especially when it comes to differentiating lung cancer subtypes. Explainable artificial intelligence methods such as Grad CAM, LIME, and SHAP were implemented to the most successful model to increase the level of transparency in the model. These visual explanations identified clinically significant tissue locations that influenced model predictions, hence improving the diagnostic system's interpretability and reliability. The findings establish that high performance deep learning architectures can be used in conjunction with explainable artificial intelligence techniques to offer a strong and reliable system of computer aided diagnosis of cancer.
DOI: 10.13005/bpj/3498
Institution: Mepco Schlenk Engineering College
Language: English
Topic: AI in cancer detection
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: diamond
OpenAlex ID: W7220658255