Texture Feature-Based Colonic Polyp Detection Using Deep Learning Techniques

Colorectal cancer is a serious worldwide health issue that leads to high mortality rates. Early detection of colorectal polyps is crucial for timely intervention and improved patient outcomes. However, a manual clinical inspection of colonoscopy images and videos can result in missed or false diagnoses. To increase the accuracy of diagnosis and help medical professionals make wise decisions on patient treatment, a few CAD (Computer-Aided Design) algorithms have been published in the literature. These algorithms use deep learning techniques and grayscale photos or videos. However, existing CAD methods face challenges such as overfitting, gradient vanishing, and difficulty in distinguishing between different polyp types. To overcome these challenges, hyperparameter optimization is essential for accurate diagnosis. In this study, we established two models: Deeplab3+ and VGG16. We have compared and analyzed the performance of these models. In our evaluation, the DeepLab3+ model outperformed VGG16 in segmenting and identifying polyps in colonoscopy videos.