Machine Learning Approaches to Gastrointestinal Disease Diagnosis: An Experimental Study with Endoscopic Images

Abstract

Automatic detection of gastrointestinal (GI) diseases using endoscopic images is a critical and emerging field of research with significant implications for healthcare. This study leverages the Kvasir dataset, available on Kaggle, to develop a machine learning model for disease detection and classification. The dataset, consisting of annotated images from the GI tract, was pre-processed using Canny edge detection for segmentation and Hu Moments for feature extraction. The images were divided into training (80%) and testing (20%) sets. A Random Forest Classifier was employed to classify three specific classes: dyed lifted polyps, dyed resection margins, and esophagitis. The performance of the classifier was evaluated using metrics such as accuracy, precision, recall, and F1-score. The results showed moderate performance with accuracy ranging from 39.00% to 44.67%, precision from 40.68% to 47.34%, recall from 39.00% to 44.67%, and F1-scores from 38.09% to 45.07%. These findings indicate that while the Random Forest Classifier demonstrates potential, there is room for improvement in the model and pre-processing techniques. The study contributes to the field by providing a comprehensive evaluation of a machine learning approach for GI disease detection and highlights the need for further research using more advanced models and diverse datasets. Future research should focus on optimizing pre-processing methods, exploring convolutional neural networks, and expanding the dataset to improve classification performance and clinical applicability.