Development of Computer Aided System for Classification of Gastrointestinal Lesions

Abstract

Colorectal cancer is a major global health problem and is one of the major contributors to deaths worldwide. Because malignant transformations are rare, endoscopic resection of hyperplastic polyps exacerbates medical costs, including those for resection and unnecessary pathological assessment. The proposed work is based on assessment of exploratory data analysis and visualization, initial pre-processing step followed by selection of attributes and performance assessment of proposed supervised machine learning algorithms. For features selection, two method were implemented (Boruta and SelectFromModel (SFM) Random Forest) to compare the performances of the models. The comparative analysis of machine learning included Random Forest (RF), Support Vector Machines (SVM), Deep Neural Network (DNN). For Boruta algorithm, it is shown that RF has the highest accuracy of 90.32%, sensitivity of 93.05% and specificity of 95.95%. Therefore, the adenoma detection rate (ADR) is desirable for improvement. A computer-aided system is developed which offers the opportunity to evaluate the presence of colorectal polyps objectively during colonoscopy.