A SYSTEMATIC ANALYSIS OF GASTROINTESTINAL DISEASE CLASSIFICATION USING HYBRID ATTENTION CONVOLUTIONAL NEURAL NETWORK

Timely intervention of gastrointestinal (GI) diseases, namely polyps, ulcerative colitis, and esophagitis are critical for improving the quality of human life and reducing the mortality rate associated with these conditions. Hence, early detection and diagnosis of GI disease are essential because they can reduce the severity of the disease. Traditional medical imaging techniques are time-intensive, labor-intensive, and susceptible to human error. Recently, deep learning models have been extensively used for image classification tasks, and they are consistently achieving promising results in real-time decision-making. However, the conventional deep learning models struggle with overfitting and poor generalization on medical imaging datasets because of the wide variability in disease types. To address this issue, a Hybrid Attention Convolutional Neural Network (HA-CNN) is proposed in this analysis. This proposed model integrates the strength of the convolutional operation and attention mechanism to focus on discriminative regions and features in medical images. The hybrid model is designed for high variability and complex features in the medical images. This model can accurately recognise lesion regions and detect types of diseases, and avoids overfitting. The effectiveness of the proposed HA-CNN is evaluated using a benchmark dataset, namely the Kvasir dataset, using 5-fold stratified cross-validation. The model achieves a mean classification accuracy of 94.43% ± 0.58, outperforming existing comparative methods. Moreover, the integration of empirical mode decomposition and dynamic scaling enhanced the quality of training data by improving the generalization ability of the model. By overcoming the existing challenges, this framework focuses on improving the diagnostic process in medical imaging, resulting in the precise detection of GI diseases