Automated Medical Image Analysis for Colorectal Polyp Detection and Classification Using a Two-Stage ResNet-18–YOLOv8 Deep Learning Framework

Authors: Sowmyashree M R, Supreetha Gowda H D

DOI: https://doi.org/10.37082/IJIRMPS.v14.i3.233168

Short DOI: https://doi.org/

Country: India

Full-text Research PDF File:   View   |   Download

Abstract: Colorectal cancer is among the leading causes of cancer-related mortality worldwide, and the vast majority of cases originate from precursor lesions known as colorectal polyps. Colonoscopy remains the principal screening tool for detecting these lesions, but manual visual interpretation of endoscopic frames is laborious and susceptible to observer fatigue, leading to missed or inconsistent diagnoses. This paper presents a two-stage computer-aided diagnostic pipeline that first screens colonoscopy images using a ResNet-18 convolutional neural network to distinguish normal mucosa from polyp-bearing frames, and subsequently localizes confirmed polyp regions using the YOLOv8 object detector. Restricting detection to images already flagged as abnormal avoids unnecessary computation on normal frames and improves overall throughput. The framework was trained and evaluated on a balanced corpus of 2,000 endoscopic images-1,000 polyp frames drawn from the public Kvasir-SEG dataset and 1,000 normal frames assembled from Z-line, pylorus, and cecum endoscopic captures-using an 80:20 train-test split. Experimental results show that the classification stage attains an overall accuracy of 99.2%, precision of 99.1%, recall of 99.1%, and an F1-score of 99.1%, while the detection stage produces tight bounding boxes around polyp regions with confidence scores ranging from 0.83 to 0.96. The combined pipeline demonstrates that cascading a lightweight classifier with a dedicated detector yields an efficient and accurate decision-support tool that can assist gastroenterologists in early colorectal cancer screening.

Keywords: Colorectal Cancer; Polyp Detection; Computer-Aided Diagnosis; Deep Learning; Convolutional Neural Network; ResNet-18; YOLOv8; Medical Image Classification.

Paper Id: 233168

Published On: 2026-06-30

Published In: Volume 14, Issue 3, May-June 2026

Share this