AI-Driven Skin Lesion Detection with CNN and Score-CAM: Enhancing Explainability in IoMT Platforms

Abstract

Using a convolutional neural network (CNN) in conjunction with Score-CAM for visual explanations and integrated into Internet of Medical Things (IoMT) systems, the project aims to improve the explainability and accuracy of skin lesion identification. Through the use of methods such as Canny Edge Detection for border localisation and DF-U-Net for segmentation, the model allows for more accurate diagnosis. The model achieves higher accuracy and is highly effective for real-time skin lesion detection and diagnosis due to the incorporation of clinical metadata, which also improves interpretability.
Objective: The goal of the project is to improve skin lesion identification accuracy and explainability through the use of Score-CAM and convolutional neural networks (CNN) in Internet of Medical Things (IoMT) platforms. Clinicians are now better equipped to understand diagnoses generated by AI.
Methods: The recommended method entails pre-processing images, DF-U-Net segmentation, and Canny Edge Detection border detection. By offering visual explanations for CNN-based predictions, the Score-CAM technique increases transparency in the diagnosing process.
Result: With an accuracy of 99.31%, the model that included CNN, Score-CAM, and clinical information outperformed conventional techniques, making it an excellent choice for real-time diagnosis in IoMT environments.

Conclusion: By combining CNN with Score-CAM, our AI-driven approach improves skin lesion identification while providing more transparent and accurate diagnosis. This is in favour of more dependable and comprehensible clinical judgements in IoMT settings.