Detecting Polyps in the Colon Using Machine Learning

Overview

Polyps, a small cluster of cells that grow on the lining of the colon, are becoming more common, with over 2.4 billion people who have polyps. Polyps are linked with conditions like cancer and inflammatory bowel diseases, which both have severe effects if not detected during their early stages. With over 30.9% of polyps being missed during colonoscopies, being able to detect these polyps is crucial.The goal of this project was to use a Convolutional Neural Network (CNN) and a YOLO model to detect, localize and classify polyps. The CNN classifies the images as having a polyp or not, while the YOLO detects its location and if it is an adenomatous or hyperplastic polyp (the two major classes of polyps). In the experiment, four different CNN models were tested. Each model had different hyperparameters. It was determined that Model 1 had the highest validation accuracy while Models 2 and 3 experienced overfitting. During the image classification tests, Model 1 had a 92% accuracy, Model 0 had a 82% accuracy, and Models 2 and 3 both had an accuracy of 42%. Two YOLO models were tested: a YOLOv5 and a YOLOv7. The YOLOv5 had the highest mean average precision (mAP) of 99.3%, recall of 98.1%, and precision of 98%. Model 1 and YOLOv5 were converted into a website called PolypDetect, which has life saving consequences, as it can detect, localize, and classify polyps from both images and videos, allowing for early treatment. Mississippi

Competition history

• JSHS 2025 Category not listed

Resources

• National JSHS abstract book ↗

Related projects

Closest projects by meaning, across every fair and year in the corpus.