Skin Disease Detection Using Deep Learning

Abstract

Skin diseases are among the most widespread health problems globally, impacting hundreds of millions of people every year. In many countries, the shortage of trained dermatologists means that patients wait weeks or even months before receiving a diagnosis — a delay that can have serious consequences when dealing with conditions like melanoma. Manual diagnosis through visual examination, while still the clinical gold standard, is inherently subjective and difficult to scale. Over the past decade, deep learning has reshaped the landscape of medical image analysis, and dermoscopic skin lesion classification is no exception. This paper presents a comprehensive deep learning framework for automated multi-class skin disease detection from dermoscopic images. We make use of transfer learning by adapting three well-established convolutional neural network architectures — ResNet-50, VGG-16, and EfficientNet-B0 — that were originally trained on ImageNet and fine-tuned on the HAM10000 dermoscopy dataset. The HAM10000 dataset contains 10,015 labelled images spanning seven clinically relevant categories: Melanocytic Nevi, Melanoma, Benign Keratosis-Like Lesions, Basal Cell Carcinoma, Actinic Keratosis, Vascular Lesions, and Dermatofibroma. A significant practical challenge in this dataset is severe class imbalance — Melanocytic Nevi alone accounts for nearly 67% of all samples — which we address through targeted data augmentation and class-weighted loss functions. Among the three architectures evaluated, the fine-tuned EfficientNet-B0 model achieves the highest overall classification accuracy of 92.4%, with an AUC-ROC of 0.961. The system is deployed via a lightweight Flask web application that allows clinicians or patients to upload a skin image and receive a real-time prediction. We believe this work represents a meaningful step toward accessible, AI-assisted dermatology, particularly in settings where specialist care is difficult to access.

Introduction

a deep learning-based system for multi-class skin disease classification using dermoscopic images (HAM10000 dataset) and its end-to-end development.

It explains that skin diseases are extremely common and difficult to diagnose because many conditions look similar, vary across patients, and suffer from limited expert availability in many regions. To address this, the study applies convolutional neural networks (CNNs) and transfer learning to build an automated diagnostic tool.

The proposed system classifies seven types of skin lesions using models such as VGG-16, ResNet-50, and EfficientNet-B0, with EfficientNet-B0 performing best. It tackles key challenges like class imbalance (rare diseases being underrepresented) using data augmentation and class-weighted training.

The workflow includes:

• Image preprocessing (resize, normalization)
• Feature extraction using CNNs
• Classification into 7 disease categories using a softmax layer
• Deployment through a Flask web application for real-time prediction

The system is trained and evaluated on the HAM10000 dataset, achieving strong performance (about 92.4% accuracy and 0.961 AUC-ROC).

The study also highlights:

• Existing systems mostly focus on binary classification (melanoma vs non-melanoma), which is not clinically sufficient.
• Major research gaps include class imbalance, lack of real-world deployment, and limited multi-class capability.
• The proposed work addresses these by building a deployable, multi-class, clinically useful decision-support tool.

Conclusion

This paper has presented a comprehensive deep learning-based system for multi-class skin disease detection, addressing one of the most pressing needs in accessible dermatological care. We designed and evaluated a transfer learning framework leveraging three pre-trained CNN architectures — VGG-16, ResNet-50, and EfficientNet-B0 — fine-tuned on the HAM10000 dermoscopic benchmark dataset covering seven clinically relevant skin lesion categories. A principled data augmentation and class weighting strategy was applied to address the severe class imbalance inherent in the dataset, resulting in meaningful and consistent performance improvements across all disease categories, particularly the clinically important minority classes. The fine-tuned EfficientNet-B0 model achieved the best overall performance, with a classification accuracy of 92.4%, macro F1-score of 91.5%, and AUC-ROC of 0.961 on the held-out test set — results that compare favourably with the current state of the art in multi-class dermoscopic classification while being computationally efficient enough to support real-time web deployment. The system was integrated into a Flask-based web application that provides an accessible interface for uploading skin images and receiving instantaneous, confidence-calibrated predictions. We believe this work makes a meaningful contribution both to the technical literature on deep learning for medical image analysis and to the broader goal of democratising access to dermatological expertise. With thoughtful further development — particularly the incorporation of patient metadata, broader disease coverage, and clinical validation — systems of this kind have genuine potential to reduce diagnostic delays, improve outcomes for patients in underserved regions, and serve as an intelligent decision-support layer within the clinical workflow. We hope this research serves as a foundation and an invitation for the community to continue advancing AI-assisted dermatology.

References

[1] A. Esteva, B. Kuprel, R. A. Novoa, J. Ko, S. M. Swetter, H. M. Blau, and S. Thrun, \"Dermatologist-level classification of skin cancer with deep neural networks,\" Nature, vol. 542, no. 7639, pp. 115–118, 2017. [2] P . Tschandl, C. Rosendahl, and H. Kittler, \"The HAM10000 dataset: A large collection of multi-source dermatoscopic images of common pigmented skin lesions,\" Scientific Data, vol. 5, pp. 180161, 2018. [3] N. Codella, D. Gutman, M. E. Celebi, B. Helba, M. Marchetti, S. Dusza, A. Kalloo, K. Liopyris, N. Mishra, H. Kittler, and A. Halpern, \"Skin lesion analysis toward melanoma detection: ISIC 2017 challenge,\" in Proc. IEEE ISBI, 2018, pp. 168–172. [4] S. S. Han, M. S. Kim, W. Lim, G. H. Park, I. Park, and S. E. Chang, \"Classification of the clinical images for benign and malignant cutaneous tumors using a deep learning algorithm,\" Journal of Investigative Dermatology, vol. 138, no. 7, pp. 1529–1538, 2018. [5] M. Tan and Q. V. Le, \"EfficientNet: Rethinking model scaling for convolutional neural networks,\" in Proc. 36th International Conference on Machine Learning (ICML), 2019, pp. 6105–6114. [6] K. He, X. Zhang, S. Ren, and J. Sun, \"Deep residual learning for image recognition,\" in Proc. IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2016, pp. 770–778. [7] K. Simonyan and A. Zisserman, \"Very deep convolutional networks for large-scale image recognition,\" in Proc. International Conference on Learning Representations (ICLR), 2015. [8] A. G. Howard, M. Zhu, B. Chen, D. Kalenichenko, W. Wang, T. Weyand, M. Andreetto, and H. Adam, \"MobileNets: Efficient convolutional neural networks for mobile vision applications,\" arXiv preprint arXiv:1704.04861, 2017. [9] Y. LeCun, Y. Bengio, and G. Hinton, \"Deep learning,\" Nature, vol. 521, no. 7553, pp. 436–444, 2015. [10] C. Szegedy, V. Vanhoucke, S. Ioffe, J. Shlens, and Z. Wojna, \"Rethinking the inception architecture for computer vision,\" in Proc. IEEE CVPR, 2016, pp. 2818–2826. [11] J. Kawahara, S. Daneshvar, G. Argenziano, and G. Hamarneh, \"Seven-point checklist and skin lesion classification using multitask multimodal neural nets,\" IEEE Journal of Biomedical and Health Informatics, vol. 23, no. 2, pp. 538–546, 2019. [12] D. A. Gutman, N. Codella, E. Celebi, B. Helba, M. Marchetti, N. Mishra, and A. Halpern, \"Skin lesion analysis toward melanoma detection: ISIC 2016 challenge,\" arXiv preprint arXiv:1605.01397, 2016. [13] F. Chollet, Deep Learning with Python, 2nd ed., Shelter Island, NY, USA: Manning Publications, 2021. [14] I. Goodfellow, Y. Bengio, and A. Courville, Deep Learning, Cambridge, MA, USA: MIT Press, 2016. [15] R. Elmasri and S. B. Navathe, Fundamentals of Database Systems, 7th ed., Boston, MA, USA

Copyright

Copyright © 2026 Sameer Yadav, Ansh Dullat, Arpit Kumar Arya, Arnav Sharma. 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.