A Multi-Scale Attention Efficient Network (Mae-Net) For Gastrointestinal Disease Detection and Classification

Abstract

Gastrointestinal diseases (GIDs) are increasingly prevalent worldwide, necessitating robust methods for their early detection and diagnosis. This study proposes a novel Multi scale Attention Efficient Network (MAE Net) for identifying abnormalities in the gastric region using time–frequency analysis. The Kvasir dataset is employed as the primary source of input images. The methodology begins with a preprocessing phase where the input images are resized, converted to grayscale, and enhanced for texture. This is followed by segmentation using approximate spatial fuzzy c means clustering. Features are extracted using wavelet transform coefficients to analyze frequency and time series data. The decomposed image classes are then input into the MAE Net model for training and testing, enabling accurate classification and prediction. The performance of the proposed model is evaluated using standard metrics such as accuracy, precision, recall, and F1 score. Comparative analysis with existing methods on similar datasets demonstrates the superiority of the proposed approach in terms of classification performance. This work highlights the potential of MAE Net in advancing the detection and diagnosis of gastrointestinal abnormalities.