Design and Implementation of a Real-Time CNN-Based Web Platform for Automated Skin Disease Diagnosis

Abstract

Skin diseases continue to be one of the most frequently reported health concerns worldwide, significantly impacting quality of life and incurring high healthcare costs. In rural and underserved regions, the shortage of dermatologists and diagnostic infrastructure compounds the burden. The proposed research presents an advanced, real-time, web-based diagnostic system leveraging Convolutional Neural Networks (CNNs) for automated classification of common dermatological conditions. Using a robust dataset of over 8,000 labeled skin disease images across nine disease classes, the ResNet50 CNN model was trained and optimized with modern preprocessing and augmentation techniques. The model was converted to TensorFlow Lite (TFLite) for enhanced portability and faster inference. The system is deployed via a Flask-based REST API, containerized using Docker, and exposed through a mobile-responsive frontend with webcam support and image upload features. Special emphasis is placed on accessibility, speed, privacy, and offline capabilities. Usability studies with non-technical users showed strong acceptance and fast learning curves, affirming the system\'s potential in telehealth and public health settings. This paper makes a strong case for AI-driven dermatological diagnostics, extending machine learning beyond theoretical models into actionable healthcare tools.

Introduction

Background and Motivation

• Skin disorders (e.g., eczema, psoriasis, impetigo) are globally prevalent, especially in rural India where access to dermatologists is limited.

• Deep learning, particularly Convolutional Neural Networks (CNNs), shows strong performance in skin lesion classification due to their ability to detect subtle visual patterns.

• Despite academic success, real-world adoption is limited by deployment and usability challenges.

2. Objective

To develop a CNN-based web platform for real-time skin disease diagnosis that:

• Is accurate

• Runs on low-resource devices (offline-compatible)

• Is user-friendly and deployable in rural/field settings

3. Literature Review Highlights

• Esteva et al. (2017): AI matched dermatologists in skin cancer detection.

• Gururaj et al. (2023): DeepSkin (VGG16-based) achieved 91.2% accuracy.

• Patel et al. (2021): Focused on mobile deployment and quantization.

• Emphasis is now shifting toward explainable AI (XAI) and usability.

4. Dataset and Preprocessing

• Data Sources: DermNet, HAM10000, ISIC, and clinical images (~8,000 images).

• Diseases Covered: Chickenpox, Shingles, Psoriasis, Eczema, etc. (9 classes total).

• Preprocessing: Image resizing, normalization, contrast enhancement (CLAHE), denoising, and heavy augmentation.

• Split: 70% training, 15% validation, 15% test (stratified and weighted).

5. Model Architecture

• Base Model: ResNet50 (robust and deep architecture with skip connections)

• Added layers: Global Average Pooling, Dropout (0.3), Dense layers, Softmax output

• Training Setup: TensorFlow + Keras, Adam optimizer, 25 epochs, batch size 32

• Performance:

  • Accuracy: 92.4%

  • Precision: 91.7%

  • Recall: 93.2%

  • F1-Score: 92.3%

  • ROC-AUC: 0.96

  • Outperformed MobileNetV2 and baseline CNN models

6. System Deployment

• Model Optimization: Converted to TensorFlow Lite, reducing size to ~24MB

• Backend: Flask API + Docker + Gunicorn + NGINX

• Frontend: Responsive HTML5/Bootstrap UI with webcam support, dark mode, offline capabilities (PWA, IndexedDB)

• Offline Deployment: Fully functional offline via service workers

7. Usability Evaluation

• Test Group: 20 non-technical users

• Avg. Time to Prediction: 11.5 seconds

• System Usability Scale (SUS): 87.4/100

• Satisfaction Rate: 93.5%

• User Feedback:

  • Interface is intuitive and quick

  • Requested features: Grad-CAM visualizations, multilingual support, downloadable reports

Conclusion

We have successfully designed and deployed a complete real-time diagnostic tool for skin disease detection using CNNs. The system bridges the research-to-real-world gap by combining robust deep learning with practical deployment infrastructure. With high accuracy, fast performance, and offline capabilities, this solution has real-world applicability in health camps, rural clinics, and low-infrastructure telehealth setups. A. Conclusions The version of this template is V2. Most of the formatting instructions in this document have been compiled by Causal Productions from the IEEE LaTeX style files. Causal Productions offers both A4 templates and US Letter templates for LaTeX and Microsoft Word. The LaTeX templates depend on the official IEEEtran.cls and IEEEtran.bst files, whereas the Microsoft Word templates are self-contained. Causal Productions has used its best efforts to ensure that the templates have the same appearance. Causal Productions permits the distribution and revision of these templates on the condition that Causal Productions is credited in the revised template as follows: “original version of this template was provided by courtesy of Causal Productions (www.causalproductions.com)”. B. Future Work Includes: • Adding Grad-CAM for visual explanations • Expanding dataset to include pediatric/rare conditions • Integration with EHR systems and government APIs • Developing a clinician dashboard for logging predictions • Supporting voice-based interaction

References

[1] Esteva, A. et al. (2017). Dermatologist-level classification of skin cancer with deep neural networks. Nature. [2] Gururaj, H.L. et al. (2023). DeepSkin: A Deep Learning Approach for Skin Cancer Classification. IEEE Access. [3] Codella, N. et al. (2019). Skin lesion analysis toward melanoma detection. IEEE J. Biomedical Health Informatics. [4] Tschandl, P. et al. (2018). HAM10000 dataset: A large collection of multi-source dermatoscopic images. Scientific Data. [5] TensorFlow Lite Documentation – tensorflow.org/lite [6] Selvaraju, R.R. et al. (2017). Grad-CAM: Visual Explanations from Deep Networks. ICCV. [7] Ronneberger, O. et al. (2015). U-Net: Convolutional Networks for Biomedical Image Segmentation. MICCAI. [8] Patel, R. et al. (2021). Deploying Quantized CNNs for Edge AI Healthcare. Elsevier Procedia Computer Science. [9] Simonyan, K. & Zisserman, A. (2015). Very Deep Convolutional Networks for Large-Scale Image Recognition. ICLR. [10] Redmon, J. et al. (2018). YOLOv3: An Incremental Improvement. arXiv preprint arXiv:1804.02767. [11] S. M. Metev and V. P. Veiko, Laser Assisted Microtechnology, 2nd ed., R. M. Osgood, Jr., Ed. Berlin, Germany: Springer-Verlag, 1998. [12] J. Breckling, Ed., The Analysis of Directional Time Series: Applications to Wind Speed and Direction, ser. Lecture Notes in Statistics. Berlin, Germany: Springer, 1989, vol. 61. [13] S. Zhang, C. Zhu, J. K. O. Sin, and P. K. T. Mok, “A novel ultrathin elevated channel low-temperature poly-Si TFT,” IEEE Electron Device Lett., vol. 20, pp. 569–571, Nov. 1999. [14] M. Wegmuller, J. P. von der Weid, P. Oberson, and N. Gisin, “High resolution fiber distributed measurements with coherent OFDR,” in Proc. ECOC’00, 2000, paper 11.3.4, p. 109. [15] R. E. Sorace, V. S. Reinhardt, and S. A. Vaughn, “High-speed digital-to-RF converter,” U.S. Patent 5 668 842, Sept. 16, 1997. [16] (2002) The IEEE website. [Online]. Available: http://www.ieee.org/ [17] M. Shell. (2002) IEEEtran homepage on CTAN. [Online]. Available: http://www.ctan.org/tex-archive/macros/latex/contrib/supported/IEEEtran/ [18] FLEXChip Signal Processor (MC68175/D), Motorola, 1996. [19] “PDCA12-70 data sheet,” Opto Speed SA, Mezzovico, Switzerland. [20] A. Karnik, “Performance of TCP congestion control with rate feedback: TCP/ABR and rate adaptive TCP/IP,” M. Eng. thesis, Indian Institute of Science, Bangalore, India, Jan. 1999. [21] J. Padhye, V. Firoiu, and D. Towsley, “A stochastic model of TCP Reno congestion avoidance and control,” Univ. of Massachusetts, Amherst, MA, CMPSCI Tech. Rep. 99-02, 1999. [22] Wireless LAN Medium Access Control (MAC) and Physical Layer (PHY) Specification, IEEE Std. 802.11, 1997.

Copyright

Copyright © 2025 Pratik Nagargoje, Onkar Mhamane, Anushka Kale, Abhijit Patil. 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.