Deep CNN Framework for Multi-Class Skin Identification Using Datasets

Abstract

This study introduces a deep learning–based framework for automatic skin disease classification using a custom-designed Convolutional Neural Network (CNN). The proposed model is trained on a combined dataset of 20,466 images collected from four publicly available sources: HAM10000, ISIC 2018, PAD-UFES-20, and SD-198. To ensure consistency, the original 216 disease categories were reorganized into 22 clinically meaningful classes, including Acne, Melanoma, Psoriasis, and others. The dataset reflects real-world imbalance, with class sizes ranging from 49 to 8,900 samples.The developed system achieves an overall classification accuracy of approximately 70% and is integrated into a multi-user web platform. The platform enables patients to upload images, obtain predictions with confidence visualization, store reports, and share results with medical professionals. Additionally, doctors can review cases and provide feedback, while administrators manage system operations. This work contributes toward bridging the gap between AI-based diagnosis and practical healthcare applications by offering an accessible preliminary screening tool.

Introduction

Skin disorders affect around 1.9 billion people worldwide and represent a major global health burden. Access to dermatologists remains limited, particularly in rural and developing regions, making early detection of serious diseases such as melanoma essential for improving patient outcomes.

Recent advances in deep learning, especially Convolutional Neural Networks (CNNs), have significantly improved medical image analysis by automatically learning features from skin images. Previous research has shown that CNN-based models can achieve dermatologist-level performance in certain classification tasks. However, many existing systems are limited by small datasets, class imbalance, and lack of practical deployment.

The proposed Skin Disease Detection System addresses these limitations by combining multiple public datasets (HAM10000, ISIC 2018, PAD-UFES-20, and SD-198), resulting in a dataset of 20,466 images covering 22 skin disease classes. Images undergo preprocessing steps such as resizing, normalization, augmentation, and class balancing through weighted sampling.

A custom CNN architecture with five convolutional layers, dropout regularization, and transfer learning from ImageNet is used for classification. The model is trained using categorical cross-entropy loss and the Adam optimizer, with performance evaluated through accuracy, precision, recall, F1-score, and confusion matrix metrics.

The trained model is deployed as a web-based platform supporting three user roles:

• Patients: Upload skin images, view results, save reports, and share them with doctors.
• Doctors: Review reports and provide feedback.
• Admins: Manage users and monitor the system.

The platform provides diagnostic predictions, confidence scores, and visualizations, making skin disease screening more accessible, efficient, and scalable for real-world use.

Conclusion

This paper presented a comprehensive deep learning system for skin disease detection using a custom CNN trained on 20,466 images across 22 clinically relevant disease classes. The model achieved 70% classification accuracy with 27.8 million parameters, 89 MB model size, and 0.35 second inference time on GPU. The multi-role healthcare platform enables patients to upload or capture skin images, receive AI predictions with confidence visualization, save PDF reports, and share with doctors. Clinicians can validate predictions and provide feedback, while administrators manage users. Future work includes expanding to more classes, improving minority class performance using GAN-based synthetic data generation, developing a dedicated mobile app, and conducting prospective clinical validation studies.

References

[1] G. Ramella and L. Serino, \"CNN Issues in Skin Lesion Classification: Data Distribution and Quantity,\" IEEE Access, vol. 13, 2025. DOI: 10.1109/ACCESS.2025.3556479 [2] P. Tschandl, C. Rosendahl, and H. Kittler, \"The HAM10000 dataset,\" Scientific Data, vol. 5, p. 180161, 2018. [3] A. Esteva et al., \"Dermatologist-level classification of skin cancer with deep neural networks,\" Nature, vol. 542, no. 7639, pp. 115-118, 2017. [4] N. C. F. Codella et al., \"Skin lesion analysis toward melanoma detection 2018,\" arXiv:1902.03368, 2019. [5] R. C. Maron et al., \"Systematic outperformance of 112 dermatologists,\" European Journal of Cancer, vol. 119, pp. 57-65, 2019. [6] L. C. P. P. P. P. Mendes, \"PAD-UFES-20,\" Data in Brief, vol. 39, p. 107570, 2021. [7] SD-198 Dataset, Figshare, 2023. [8] P. Gupta, J. Nirmal, and N. Mehendale, \"Custom CNN architectures for skin disease classification,\" Multimedia Tools and Applications, 2024. [9] R. Yadav and A. Bhat, \"Systematic literature survey on skin disease detection,\" Multimedia Tools and Applications, 2024. [10] T. J. Brinker et al., \"CNN trained with dermoscopic images performed on par with 145 dermatologists,\" European Journal of Cancer, vol. 111, pp. 148-154, 2019. [11] K. Likhar and S. Ridhorkar, \"Skin Cancer Diagnosis with a Customized CNN Model using Deep Learning Approaches,\" Communications on Applied Nonlinear Analysis, vol. 31, no. 6s, 2024. [12] S. P. Choy et al., \"Systematic review of deep learning image analyses for the diagnosis and monitoring of skin disease,\" npj Digital Medicine, vol. 6, Article 180, 2023. DOI: 10.1038/s41746-023-00914-8 [13] \"Categorical Weighting Domination for Imbalanced Classification With Skin Cancer in Intelligent Healthcare Systems,\" IEEE Access, vol. 11, pp. 105170-105181, 2023. [14] M. Subramanian, K. Shanmugavadivel, S. Thangaraj, J. Cho, and S. Ve, \"Attention-aware Deep Learning Models for Dermoscopic Image Classification for Skin Disease Diagnosis,\" Current Medical Imaging, 2023. [15] A. S. Jaradat et al., \"Automated Monkeypox Skin Lesion Detection Using Deep Learning and Transfer Learning Techniques,\" International Journal of Environmental Research and Public Health, vol. 20, no. 5, p. 4422, Mar. 2023

Copyright

Copyright © 2026 Gaurav Dhoble, Dhiraj Wanjari, Ayush Kamble, Prajwal Kumbhalkar, Jayant Rajurkar. This 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.