Enhancing Medical Image Classification with Streamlined-EvidNet: A Robust Neural Network Architecture

Accurate classification of medical images is crucial for automated analysis and disease diagnosis. Traditional deep learning methods often struggle with challenges such as variations in image quality and imbalanced datasets, leading to diagnostic errors. This study introduces an advanced framework that integrates contrastive learning, ResNet architectures, and Bayesian methods to enhance the accuracy and reliability of medical image classification, with a focus on heart disease detection. Contrastive learning enhances the model's capability to differentiate between similar and dissimilar image pairs, thereby reducing the challenges posed by imbalanced datasets. ResNet architectures provide robust feature extraction, while Bayesian methods offer probabilistic interpretations, addressing uncertainty in predictions. Our comprehensive evaluation demonstrates significant improvements in classification performance, positioning this approach as a promising tool for clinical applications in radiology. The proposed framework effectively overcomes challenges associated with variable image quality and imbalanced datasets, contributing to improved diagnostic accuracy and reliability in cardiovascular disease detection.