Deep learning framework for colon and lung cancer detection using CNNs

Abstract

Cancer remains one of the leading causes of mortality worldwide. The World Health Organization (2025) reports approximately 10 million cancer deaths in 2020, with lung and colorectal cancers among the largest contributors, accounting for an estimated 20% and 9% of cancer deaths, respectively. This study presents a convolutional neural network (CNN)-based framework for classifying histopathological images into five classes: lung adenocarcinoma (lungaca), lung squamous cell carcinoma (lungscc), benign lung tissue (lungn), colon adenocarcinoma (colonaca), and benign colon tissue (colonn), with the goal of supporting diagnostic workflows across clinical contexts. The model was trained on the publicly available LC25000 dataset and evaluated using accuracy, precision, recall, F1-score, confusion matrices, receiver operating characteristic (ROC) curves, and area under the curve (AUC). The architecture consists of four convolutional blocks, global average pooling, dropout, and a 512-unit dense layer, achieving 98.53% accuracy, 98.57% precision, and 98.53% F1-score. These results suggest that the proposed framework could serve as a decision-support tool to assist pathologists in the earlier and more consistent detection of lung and colon cancer from histopathological images.