Gastrointestinal Abnormality Detection and Classification using Empirical Wavelet Transform and Deep CNN

Abstract

This research proposes a novel automated system for detecting and classifying gastrointestinal (GI) abnormalities from endoscopic images. The system leverages the Empirical Wavelet Transform (EWT) for effective feature extraction, enhancing the discriminative power of textural information within endoscopic images, which is then fed into a Deep Convolutional Neural Network (CNN) for accurate classification. The EWT pre-processing step effectively addresses the challenges posed by variations in image quality and illumination commonly found in endoscopic procedures. Our results demonstrate improved accuracy and robustness compared to existing methods, paving the way for faster, more reliable GI abnormality diagnosis and aiding clinicians in improving patient care.

Introduction

Accurate and timely diagnosis of gastrointestinal abnormalities is crucial for effective treatment and improved patient outcomes. Traditional methods rely heavily on the expertise of gastroenterologists, which can be subjective and time-consuming. Endoscopy provides detailed visual information, but manual analysis is prone to human error and inter-observer variability. Automated image analysis offers a promising solution by leveraging computer vision techniques. However, the significant variations in image quality, illumination, and the subtle nature of certain abnormalities pose significant challenges. This research aims to address these challenges by combining the power of EWT for robust feature extraction with the high classification accuracy of deep CNNs.

Objectives

• To develop an automated system for detecting GI abnormalities from endoscopic images with high accuracy and robustness.
• To improve the accuracy of GI abnormality classification by incorporating EWT for enhanced feature extraction.
• To evaluate the performance of the proposed system using a comprehensive dataset and compare it with existing state-of-the-art methods.