Enhancing Skin Disease Classification with MetaCLIP Models: A Multi-Dataset Analysis Across 15 Sources
Year: 2025
Type: preprint
Abstract: Early detection of skin conditions is essential for prompt treatment, which improves patient outcomes and minimizes complications. Identifying the initial stage of skin diagnosis reduces discomfort, pain, or other symptoms and significantly improves individuals’ quality of life. Furthermore, it prevents the spread of contagious skin conditions such as fungal infections and scabies, thereby protecting public health. This highlights developments in the diagnostic tool, especially AI-based, in the field of health care. In this study, a new deep learning MetaCLIP model is introduced for efficient classification of skin disease images with the help of Vision Language Models (VLM). Features are extracted from pre-trained MetaCLIP models and given to classifiers such as SVC, RidgeClassifierCV, and LinearDiscriminantAnalysis. The method is validated on 15 datasets using various classifiers, applying evaluation metrics comprising precision, recall, F1 score, and accuracy. Datasets include the HAM10000 Hair Removal Dataset (7 classes, 10015 images), the Melanoma skin cancer dataset containing 10000 images (2 classes, 10605 images), and Skin.Cancer.Data (2 classes, 3297 images), 5-Common-Human-Skin-Disease (6 classes, 3295 images). On three datasets, the algorithm produces exceptional results: The 10,000-image melanoma skin cancer dataset has a 93% accuracy and 94% across all metrics. Monkeypox Skin Images Dataset (MSID) obtains 95% accuracy, 98% across all metrics. While the 5-Common-Human-Skin-Disease dataset 99% accuracy, 99% Precision, recall and f1-score. The efficiency of the proposed approach is supported by a comparison of the simulation results with those of the traditional algorithms.
Authors: Naga Raju challa, Bhuma Chandra Mohan, Tanguturi Naga Malleswari, Olutayo Oyerinde, Kalapraveen Bagadi, Visalakshi Annepu, Lingisetty Naga Pavan Kumar
Language: English
Subfield: Oncology
Field: Medicine
Domain: Health Sciences
Open Access status: gold
OpenAlex ID: W4413514121