Leveraging Deep Learning for Automated Analysis of Colorectal Cancer Histology Images to Elevate Diagnosis Precision

Abstract

Histopathology plays a vital role in the microscopic examination of colorectal cancer tissues, with a historical focus on the tumor-stroma ratio using texture analysis. However, due to the time-consuming and labor-intensive nature of this approach, there's a need for innovative solutions. This study introduces a groundbreaking shift by employing deep transfer learning to automate tissue classification within colorectal cancer histology samples. Through a comprehensive evaluation of various pre-trained models, including ResNet50V2, VGG19, Xception, InceptionV3, and MobileNet, we have achieved remarkable results. Notably, the ResNet50V2 model stands out with an impressive accuracy of 95%. Beyond its potential to significantly enhance operational responses, this research underscores the effectiveness and consistency of transfer learning as a rapid and efficient tool for colorectal cancer detection and classification.