A Visionary Approach for Enhanced CRC Diagnosis by Leveraging Visual Geometry Groups Classic Deep Network

Colorectal cancer remains a prevalent and concerning health issue globally, necessitating effective methods for early detection and classification. This study proposes an improved method for colorectal polyp classification using a modified VGG19 augmented with transfer learning techniques. Leveraging transfer learning, we integrate the DropBlock regularisation method into the base VGG19, fine-tuning the last few layers for enhanced performance. The study aims to capitalise on the robust feature extraction capabilities of VGG19 while addressing overfitting through structured regularisation. The performance of the proposed model is evaluated on four diverse datasets of colonoscopy images to distinguish between malignant and benign polyps. Experimental findings underscore the efficacy of the proposed method, exhibiting superior performance and, notably, a heightened recall rate across all datasets compared to base VGG19 in detecting polyps, surpassing 86.6% across all datasets. The proposed model offers a promising pathway for enhancing the accuracy of colorectal polyp detection, enabling timely interventions and more effective treatment strategies, thereby reducing the impact on individuals and society.