Copyright © 2026 Authors retain the copyright of this article. This article is an open access article distributed under the Creative Commons Attribution License which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

@article{193245,
        author = {Niraj N. Patil and Nayan S. Sushir and Prerana B. Waghode and Prof. Tushant A. Tayde},
        title = {Vision-Based Deep Learning Framework for Early Skin Disorder Recognition},
        journal = {International Journal of Innovative Research in Technology},
        year = {2026},
        volume = {12},
        number = {9},
        pages = {4073-4078},
        issn = {2349-6002},
        url = {https://ijirt.org/article?manuscript=193245},
        abstract = {Skin ailments constitute individual of ultimate widespread classifications of strength disorders globally, moving heaps of individuals although age, neuter, or region. Early and correct discovery is essential to avoid confusions, including diseased metamorphosis, perm anent skin damage, and raised healthcare costs. Traditional disease relies heavily on dermatologists’ happening and dermoscopic equipment, that are frequently unavailable in country or source limited scenes. In addition, the ocular similarity betwixt lesions and the internal of human observation can bring about misdiagnosis.
In current age, machine education—particularly deep convolutional affecting animate nerve organs networks (CNNs)—has shown unusual advance in medical concept study. These models can automatically gain discriminating patterns from large-scale commented representation datasets, achieving levels of veracity corresponding to prepared dermatologists. In this research, we present a complete automated passage for rash classification from representations utilizing transfer-learning located CNN architectures. The system combines preprocessing methods in the way that hairstyle removal, representation normalization, color thickness correction, and dossier improving to embellish image condition and strength. A fine-tuned ResNet 50 model prepared on the HAM10000 and ISIC datasets illustrates strong categorization efficiency with a test veracity of ~92%, large-F1 score of ~0.89, and macro-ROC AUC of ~0.95 across seven affliction types.
This work likewise evaluates the model’s interpretability utilizing Grad CAM visualizations and discusses arrangement concerns for tele dermatology principles. The experimental results climax the important potential of figure- based ML methods to supplement dermatological disease, specifically in low-capital surroundings.},
        keywords = {Skin Disease Identification, Machine Learning, Deep Learning, CNN, Image Classification, Dermatology, Medical AI, Computer Vision, Flask, OpenCV.},
        month = {February},
        }