Deep hybrid model for Mpox disease diagnosis from skin lesion images

Abstract This research presents DNLR‐NET, a novel model designed for automated and accurate diagnosis of MPox disease. The model's performance is constructed and validated using a carefully collected MPox dataset from online repositories. DNLR‐NET begins by extracting deep features from the DenseNet201 pre‐trained model, which exhibited superior performance compared to other models during the comparison. The deep features obtained from each dense layer are then used to train six classifiers, among which logistic regression showcases the best performance with the extracted deep, dense feature. A comparative study with earlier advanced CNN models classifying the same dataset demonstrates that DNLR‐NET achieves an impressive accuracy of 97.55%, outperforming the base DenseNet201 model, which only attains 95.91% accuracy. This accuracy emphasizes the efficacy of combining deep features with logistic regression. A Grid Search algorithm is employed for optimal hyperparameter extraction, creating multiple unified deep feature sets and achieving the highest classification accuracy. The fusion of deep features with logistic regression yields superior results compared to ensemble techniques such as random forest and support vector machines and also reduces training time complexity. DNLR‐NET surpasses existing models, ML classifiers, and pre‐trained models, demonstrating its effectiveness and potential for clinical implementation in diagnosing MPox. The promising outcomes of advantage deep learning algorithms, particularly DenseNet201 transfer learning, highlight the significance of adopting transfer learning methodologies for CNN‐based MPox diagnosis in clinical settings. Researchers and clinicians are strongly encouraged to explore and implement these techniques to improve the accuracy and efficiency of MPox diagnosis.