Automated Detection of Skin Disorders Using Convolutional Neural Networks

Abstract

Skin conditions that affect quality of life, like acne and pigmentation problems, must be detected early and accurately. This study uses deep learning to overcome the shortcomings of conventional diagnostic techniques by presenting a Convolutional Neural Network (CNN)-based system for automated acne and pigmentation diagnosis. Using a curated dataset of 1,159 clinical photos from Kaggle and supplemented with publicly available data, the suggested methodology incorporates MobileNetV2 architecture for feature extraction and classification. For both the acne and pigmentation classes, the model\'s precision and recall surpass 90%, and its total accuracy on test data is 97%. By providing scalable solutions for remote diagnosis and lowering clinical burdens, this research advances AI-driven dermatological tools. Future research should incorporate real-time adaptive learning and broaden the scope of diseases.

Introduction

Global Burden of Skin Diseases:

• Skin diseases affect about 1.9 billion people globally, ranking fourth in non-fatal disability burden.

• Conditions like acne vulgaris and melasma severely impact quality of life due to psychological and social effects.

• Infectious skin diseases, such as pyoderma, are highly prevalent, especially in low-income regions (e.g., 17 times higher in Southern Sub-Saharan Africa).

Challenges in Skin Disease Diagnosis:

1. Lack of Specialists: In 40% of low-income countries, there is less than 1 dermatologist per 100,000 people.

2. Diagnostic Subjectivity: Acne severity scoring shows 23–41% disagreement between experts.

3. Inadequate Technology: Current teledermatology software has only 68–72% accuracy and does not solve rural healthcare issues.

Objectives of This Work:

• Develop a lightweight CNN model (MobileNetV2-based) for:

  • Acne subtype classification (papules, pustules, nodules)

  • Pigmentation disorder recognition

• Ensure mobile compatibility with fast (<200ms) inference on mid-range smartphones.

• Provide treatment guidance aligned with WHO guidelines, especially for low-resource settings.

Key Statistics Supporting the Study:

• Dermatologic DALYs grew by 7.38% (1990–2019).

• Racial disparities exist in acne prevalence (e.g., 30.5% in Chinese infants vs. 11% in US Caucasians).

• Mobile health solutions could reduce diagnosis delays by 83% in low- and middle-income countries.

Literature Gaps Addressed:

• Limited datasets lacking ethnic and disease diversity.

• Lack of multi-task models capable of both classification and localization.

• Rare real-world testing on varied mobile devices.

Methodology:

A. Dataset Collection & Preprocessing

• 1,159 images from Kaggle and real smartphone submissions.

• Covers diverse skin types (Fitzpatrick III–VI).

• Images resized, normalized, augmented (rotation, brightness, flipping).

• Additional steps: histogram equalization and HSV normalization.

B. Model Architecture

• Base model: MobileNetV2 (efficient for mobile use).

• Transfer Learning with ImageNet weights.

• Custom layers for:

  • Feature extraction

  • Acne and pigmentation classification

  • Lesion localization (bounding box regression)

• Loss functions: categorical cross-entropy and Smooth L1 loss.

C. Evaluation Metrics & Testing

• Used accuracy, precision, recall, and F1-score.

• Achieved 97% test accuracy; 84% real-world accuracy on Android devices.

• Inference time: 184ms on Snapdragon 7 Gen 1 chip.

Deployment & Impact:

• Embedded into an Android application that:

  • Captures user skin images.

  • Offers instant diagnostic feedback.

  • Delivers treatment recommendations in line with global clinical guidelines.

Results & Insights:

• Accuracy varies with disease type due to differences in data size and image quality.

• Melanoma image processing included contrast enhancement and grayscale conversion.

• The model shows high potential for use in remote and underserved areas for early diagnosis and treatment support.

Conclusion

This research proposes an AI-powered, mobile-friendly dermatology tool that addresses key gaps in global skin disease diagnosis—combining accuracy, accessibility, and real-world utility. It targets a critical healthcare need using a CNN model optimized for mobile deployment and clinical integration.

References

[1] Furqon et al., \"Detection of Eight Skin Diseases Using Convolutional Neural Network with MobileNetV2 Architecture,\" J. IlmiahTek. ElektroKomputerdanInformatika, vol. 10, no. 2, pp. 373–384, Jun. 2024. [2] L. Zhang et al., \"Advancements in Acne Detection: Application of the CenterNet Framework,\" PMC, Mar. 2024. [3] T. V. Reddy et al., \"Design and Analysis of CNN-Based Skin Disease Detection System,\" Proc. ICETE 2023, pp. 334–346, 2023. [4] Deep learning for AI-based diagnosis of skin-related neglected tropical diseases K. O. A. Kouadio, et al., PMC, 2023. [5] Deep skin diseases diagnostic system with Dual-channel Image and AI H. He, et al., Frontiers in Artificial Intelligence, 2023. [6] The Use of Artificial Intelligence for Skin Disease Diagnosis in Primary Care A. M. A. Alshahrani, et al., PMC, 2024. [7] Artificial Intelligence in Infectious Skin Disease M. E. S. Dyer, et al., Wiley Online Library, 2024. [8] Skin disease detection using artificial intelligence R. Smith, et al., AIP Publishing, 2023.

Copyright

Copyright © 2025 Sachin More, Aditya Parekar, Yusuf Ansari, Tanay Jajoo, Hemant Chaudhary. 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.