Colon cancer remains a leading cause of cancer-related mortality, primarily due to delayed diagnosis and rapid metastatic progression. Early-stage detection is therefore critical for improving survival outcomes. From a complex systems perspective, medical imaging data exhibit high-dimensional, nonlinear, and hierarchical structures that are well suited to deep learning-based analysis. In this study, we propose a convolutional neural network (CNN) framework for the automated detection of transverse colon cancer from computed tomography (CT) colonography images. A dataset of 1,860 abdominal CT scans, categorized into transverse colon cancer and normal classes, is utilized to model the underlying structural patterns of the pathology. A novel deep learning architecture, COCDNet, is designed to capture multiscale spatial features and complex morphological variations characteristic of cancerous colon tissue. In addition, transfer learning strategies are systematically applied to four established CNN models (AlexNet, VGG16, VGG19, and DarkNet19) to enhance feature extraction efficiency and convergence behavior. Comparative performance analysis demonstrates that COCDNet achieves superior classification performance, with an accuracy of 93.53%, an F1-score of 94.69%, and near-perfect specificity. These results indicate that the proposed model effectively encodes complex image structures and nonlinear spatial dependencies inherent in medical imaging data, enabling reliable early-stage cancer detection while minimizing false positives. The study highlights the relevance of deep learning (DL) as a computational framework for modeling complex biological systems, reinforcing its role in AI-driven healthcare diagnostics and intelligent decision-support systems.
Alharbi et al. (Fri,) studied this question.