Development and Performance Evaluation of Various Approaches to Predict the Occurrence of Colorectal Cancer Using Data Analysis for Enhanced Diagnosis and Prognosis

Abstract

Colorectal Cancer (CRC) remains a critical global health challenge, ranking third in both incidence and mortality rates among cancers. Accurate prediction and early detection are crucial for improving CRC prognosis and treatment. This study emphasizes the critical role of integrating advanced Machine Learning (ML) techniques for improving the prediction and management of colorectal cancer. The proposed study presents a hybrid model integrating Convolutional Neural Networks (CNN) and Vision Transformers (ViT) for CRC prediction using the NCT-CRC-HE-100K dataset. CNNs are utilized for capturing spatial hierarchies in images through convolutional layers, making them highly effective for analyzing histological patterns in tissue samples. In contrast, ViTs are employed for capturing global contextual information by processing images as sequences of patches, allowing for a more comprehensive understanding of the tissue architecture. The proposed hybrid framework demonstrates remarkable performance on various metrics, such as with precision, recall, and F1-scores reaching near-perfect values across various tissue classes, highlighting its robustness in differentiating between histological components. Comparative analysis revealed that the proposed hybrid CNN+ViT model outperformed other models, achieving an accuracy of 99.89%, compared to 99.77% for ResNet-50 with Adam optimizer and 99.76% for Deep Adaptive Regularized Clustering (DARC). These results underscore the model's potential in clinical applications for early diagnosis and personalized treatment strategies, ultimately contributing to better patient outcomes.