Enhanced automatic diagnosis of cecum colorectal cancer using novel artificial neural network on abdominal CT radiological scans

Abstract

One of the most common causes of death is colorectal cancer (CRC). The spread of cancer cells to other organs increases dramatically because of delayed detection. Presently, the only ways to increase survival rates and reduce cancer-related mortality are via prompt diagnosis and customized therapies. Artificial intelligence (AI) may significantly aid professionals in identifying CRC cases with less effort, time, and cost. This paper presents a novel convolutional neural network (CNN) for detection known as COCDNet and two sets of modifications to CNN models for identifying cecum CRC in computed tomography (CT) radiological scans. Before images are included in the architecture, they are preprocessed to reduce the noise. The data is then sent into a COCDNet model that holds 22 layers. On other hand, two types of transfer learning (TL) are used to four popular CNN models: DarkNet19, VGG16, VGG19, and AlexNet. The dataset comprises 1695 images of abdomen CT scans, categorized into two main classes as cecum cancer and normal images. COCDNet achieves the highest performance, proving an accuracy of 97.04%, an F1-score of 95.80%, and recall approaching 100%. These measures demonstrate that COCDNet is a dependable tool for early CRC diagnosis because it can both reliably detect cancer and reduce false positives. The suggested model success in detecting cecum CRC demonstrates the value of this work that improves AI models for bettering healthcare systems and saving lives.