AI-Based Deep Learning Model for Classification and Segmentation of Colorectal Cancer

Abstract

Colorectal cancer is the most common malignancy in the world, accounting for almost 35% of all cancer-related deaths. Colorectal cancer is third in terms of cancer diagnoses, behind only lung and breast cancers. In particular, AI-guided clinical treatment can aid in the reduction of health inequities. In modern times, digital pathology is crucial in the evaluation of tumors. Despite the abundance of extensively annotated datasets, current approaches struggle to handle the large size and high resolution of Whole Slide Images (WSIs). When it comes to histopathology image segmentation and tissue classification, these models appear like a promising option, especially considering the scalability of Deep Learning (DL) methods. Using DL architectures, this research focuses on colon cancer location classification and highlighting in a sparsely annotated histopathology data setting. To begin, we will examine and contrast many cutting-edge Convolutional Neural Networks (CNN). Due to the scarcity of high-quality WSI datasets, we have turned to transfer learning approaches. The technique's defining characteristic is the extensive collection of learnt features that is produced by training the network on a massive computer vision dataset. With VGG, we were able to achieve an accurate patch-level classification rate of up to 94.56% during testing and evaluation on our colon cancer dataset. This paper's proposed method outperformed the state-of-the-art algorithms for histopathology image classification, which had the lowest error rate. A simple, effective, and efficient method for histopathology image categorization. Through effective utilization of the dataset, we achieved state-of-the-art results.