DLOCPD (Deep Learning on Colorectal Polyps Detection): Development of Colorectal Polyps Detection Using Deep Learning

This study focuses on the development and evaluation of a deep learning model for the early detection of colon polyps, a precancerous condition that often progresses asymptomatically in its early stages. Although epidemiological data in Indonesia remain limited, it is estimated that more than 400,000 new cancer cases are detected annually, underscoring the urgent need for effective diagnostic technologies. A quantitative descriptive method with a dry lab approach was employed, using Google Colaboratory and the Python programming language. The dataset consisted of 6,000 endoscopic images representing multiple gastrointestinal conditions. The ResNet-50 architecture was utilized to construct the classification model. Training showed a steady reduction in loss values, reaching 0.0524 by the 10th epoch, demonstrating strong learning convergence. The model achieved an overall accuracy of 93.25%, a precision of 94%, a recall of 93%, and a weighted F1-score of 0.9317 (93.17%), indicating a robust balance between sensitivity and specificity. Class-wise F1-scores further highlighted excellent detection performance with Normal (98%), Ulcerative Colitis (88%), Polyps (88%), and Esophagitis (99%). These findings confirm that the proposed model can accurately distinguish gastrointestinal disease classes, with strong potential for early polyp detection and clinical application in preventing colorectal cancer.