Deep Learning Driven Colon Cancer Diagnosis: Performance Assessment of Five Convolutional Neural Network Architectures on Histopathological Image Classification

Abstract

The integration of artificial intelligence (AI), particularly deep learning, has significantly advanced disease diagnosis and clinical decision making. This study evaluates the performance of five prominent deep learning architectures MobileNetV1, ResNet50, AlexNet, DenseNet201, and Inception v2for classifying colon adenocarcinoma versus benign colon tissue, a task essential for effective colon cancer management. Using a dataset of 3,000 histopathological images, each model was trained and tested to assess classification accuracy. Among the evaluated models, MobileNetV1 and AlexNet demonstrated the highest performance, achieving test accuracies of 96.33% and 95.67%, respectively. In contrast, ResNet50 and DenseNet201 showed comparatively lower accuracies of 85.80% and 87.40%, while Inception v2 reached 92.87%. These findings underscore the strong potential of lightweight architectures particularly MobileNetV1 and AlexNet in improving colon cancer detection and supporting clinical workflows. Future work will explore additional model architectures, evaluation metrics, and optimization techniques to further enhance diagnostic reliability. This study contributes to the expanding body of research on AI driven oncology, highlighting deep learning’s role in advancing early and accurate cancer classification.