Hybrid Convolutional Network Fusion: Enhanced Medical Image Classification with Dual-Pathway Learning from Raw and Enhanced Visual Features

This study presents a novel dual-pathway deep learning framework for diabetic retinopathy (DR) classification using the APTOS 2019 Blindness Detection dataset. Our approach combines the global feature detection capabilities of ResNet50 with the localized, intensity-focused analysis of MobileNetV2. This innovative method processes both raw and selectively enhanced visual features of medical images, aiming to improve the accuracy of automated DR detection and classification. We introduce an original preprocessing technique that enhances pixel intensities (interval of Red, Green, and Blue color channels of the image), allowing the model to learn from both unaltered and filtered image data. The performance of this hybrid model is evaluated against several baseline models, including ensemble methods and the ConvMixer model. Our results demonstrate a significant improvement in classification accuracy, precision, recall, and F1 score, outperforming the benchmarks. In addition, our research contributes to the field of medical imaging by introducing a robust preprocessing method that highlights critical features in the images, thereby enhancing classification performance. This study not only offers a novel approach to DR detection but also sets the stage for future advancements in automated medical image analysis.