Gastrointestinal Tract Lesion Classification Using Deep Involution Neural Networks

Abstract

Classification of gastrointestinal (GI) tract lesions from endoscopic images is one of the challenging medical image analysis. Conventionally, the automation techniques that leverage very large and complex deep learning methods, that are very difficult to be deployed. This study proposes a lightweight four-layer Deep Involution Neural Network for automated classification of GI tract lesions into four classes: Normal, Ulcerative Colitis, Polyps, and Esophagitis. The proposed model uses involution operations to efficiently extract essential features while reducing computational complexity. The model has been evaluated on a publicly available dataset that consists of 6000 endoscopic images. After training the model on the dataset, it achieves an overall accuracy of 91.13%. These results validate that the proposed model is an effective and computationally efficient approach for computer-aided GI tract lesion classification.