Artificial intelligence with deep learning driven entropy-curvature attention mechanism for detection and segmentation of skin lesions using biomedical images.
Authors
Affiliations (7)
Affiliations (7)
- Department of Information Systems, Faculty of Computing and Information Technology, King Abdulaziz University, Jeddah, Saudi Arabia.
- Department of Computer Science, College of Computer and Information Sciences, King Saud University, P.O.Box 51178, Riyadh, 11543, Saudi Arabia.
- Department of Information Systems, College of Computer and Information Sciences, Princess Nourah Bint Abdulrahman University, P.O. Box 84428, Riyadh, 11671, Saudi Arabia.
- Department of Computer Science, Applied College at Mahayil, King Khalid University, Abha, Saudi Arabia.
- Engineering and Technology Unit, Applied College, Majmaah University, Al Majmaah, 11952, Saudi Arabia.
- Department of Cybersecurity, College of Computer Science and Engineering, University of Jeddah, Jeddah, 21493, Saudi Arabia.
- Department of Information Technology, College of Computing and Information Technology, Northern Border University, Arar, Saudi Arabia. [email protected].
Abstract
In medical image analysis, skin lesion diagnosis remains a complicated task. Skin lesions are a prevalent form of skin disease that exists globally. A malignant skin cancer, otherwise called melanoma, usually starts after melanocyte cells begin to grow uncontrollably. Among several forms of cancer, malignant melanoma is the deadliest form as it has a high vulnerability to spread into deeper tissues and because of its high death rate as well. Early identification of skin tumours is essential to permit advanced treatment. In general, the manual inspection of dermatologists is time-consuming, subjective, and tedious, resulting in dissimilar recognition precision based on their experience. In recent times, many investigators have studied the application of deep learning (DL) methods to skin lesion segmentation. In this study, the DL-driven entropy-curvature attention mechanism for enhanced segmentation and classification of skin lesions (DLECAM-ESCSL) model is proposed in medical imaging techniques. The primary purpose of the DLECAM-ESCSL model is to develop an effective approach for precise skin lesion segmentation to assist in early and reliable skin disease diagnosis using advanced techniques. At first, the image pre-processing step involves various levels, such as image resizing, hair removal, and noise removal, to improve the quality of raw images by eliminating the noise. Furthermore, the attention mechanism-based entropy-curvature (ECA) method is employed for the segmentation process. For the feature extraction process, the DLECAM-ESCSL model utilises the vision transformer (ViT) model to recognise and isolate the most relevant information from raw data. Finally, the Wasserstein autoencoder (WAE) model is used for classification. The performance valuation of the DLECAM-ESCSL approach portrayed a superior accuracy value of 99.16% over existing methods under the Skin Cancer ISIC dataset.