Sparse Representation with Residual Learning Model for Medical Image Classification

In the context of medical image analysis and computer-aided diagnosis, accurate classification of medical images is crucial. Deep learning methods have become immensely popular in comparison to traditional techniques. Consequently, medical image classification for computer-aided diagnosis is highly challenging due to factors like imaging modalities, clinical conditions, and the complex aspects of medical images. We introduce Sparse Representation with Residual Learning Model (SR2LM) to address these challenges effectively. Our approach combines the strengths of sparse representation and deep residual networks to capture the essential image characteristics. We achieve this by learning sparse codes from a dictionary and then integrating these codes with the output of deep residual networks. The resulting combined input is processed by dense layers for image classification. Our SR2LM model improves classification accuracy by leveraging sparse representation and deep residual networks. Importantly, it can be trained end-to-end, benefiting from the advantages of both techniques within a unified framework. The SR2LM model outperforms existing benchmarks in medical image classification, with an average area under the curve (AUC) of 0.902 and an average accuracy (ACC) of 0.816 across the five datasets taken from MedMNIST dataset.