3D CT Scans for Colorectal Cancer Classification using VGG16 and Data Augmentation

Abstract

Computed Tomography (CT) scans play a vital role in diagnosing and monitoring various medical conditions. Applying deep learning models for accurate and automated CT scan classification has shown promising results. Among these models, VGG16, a deep convolutional neural network (CNN) architecture, has gained considerable popularity due to its exceptional performance in image recognition tasks. This study adapts VGG16 architecture to classify 3D volumetric CT scans, aiming to detect the presence or absence of polyps in the colon and prevent Colorectal Cancer CRC. The three-dimensional nature of the data poses unique challenges in terms of model complexity and overfitting on limited resources. Accordingly, we employed data augmentation techniques to augment the dataset and improve model generalization. Results suggest that data augmentation did increase the accuracy of VGG16 by +3.3%.