Recognition of Colon Neoplasm using Contrast-Aware Capsule Network

Colon neoplasms are the second leading cause of mortality worldwide, and despite advances in endoscopy, even experienced endoscopists miss 22-28% of cases. Intelligent systems can bridge this gap, but current approaches face significant challenges. For instance, Convolutional Neural Networks (CNNs) excel in image recognition but require large datasets, which are often scarce in healthcare. On the other hand, Capsule Networks, suitable for smaller datasets, struggle with complex images. To address these limitations, we propose a novel contrast-aware Capsule Network for colon neoplasm recognition. By incorporating a novel contrast-aware feature transformation algorithm, our proposed model enhances extracted features and focuses on processing relevant features, achieving remarkable recognition accuracies of 96.70%, 93.68%, and 99.99% on PolyGen, ETIS-Larib, and CVC-ClinicDB datasets, respectively. Our approach outperforms state-of-the-art Capsule Network algorithms and offers a promising solution to support endoscopists in accurate colon neoplasm detection, reducing the burden of image analysis and improving patient outcomes.