Automated Nail Disease Detection: A Deep Learning Approach to Early Health Assessment

Abstract

This paper describes an intelligent, automated system for the early diagnosis of nail diseases using deep learning techniques. Nail diseases may represent other health conditions, such as fungal infections, skin diseases, and systemic diseases. Existing methods for diagnosing nail diseases still rely on visual, manual inspections or thin nail clippings, which are relative, subjective assessments, manual, and difficult to scale. The proposed system utilizes deep learning, specifically convolutional neural networks, and custom implementation of the well-known VGG16 and GoogLeNet architectures, to classify nail images into eight different diseases. Feature-level and decision level fusion provided a vehicle to improve prediction accuracy by taking advantage of the strengths of each architecture. We trained the mode on a custom dataset of annotated nail images, performing appropriate preprocessing and augmentation, and produced a user-friendly, real-time disease prediction system, using Streamlit, through the provision of a frontend user interface, simply requiring users to upload their nail images. Evaluation metrics such as overall accuracy, precision, recall, and f1-score indicate that the proposed fusion-based architecture performs better than either individual model. Overall, the system provides a good alternative to existing invasive methods for diagnosing nail diseases, especially for healthcare access with certain health challenges such as being in remote locations. The system represents an alternative early diagnosis model, which is automated, non-invasive, and scalable.

Introduction

The text describes the development of an AI-powered automated nail disease detection system designed to support early diagnosis of nail-related health conditions. Nail abnormalities can indicate serious dermatological or systemic diseases such as fungal infections, psoriasis, anemia, and melanoma. Traditional diagnosis relies heavily on visual inspection by medical experts, which is time-consuming and often inaccessible in rural or underserved areas. To address this challenge, the proposed system uses deep learning techniques to provide a non-invasive, accessible, and scalable diagnostic solution.

The system employs two powerful convolutional neural network (CNN) architectures, VGG16 and GoogLeNet, to classify nail images into eight disease categories: Onychogryphosis, Bluish Nail, Clubbing, Koilonychia, Acral Lentiginous Melanoma, Healthy Nail, Onycholysis, and Nail Pitting. To improve performance and robustness, the system uses both feature-level fusion and decision-level fusion, combining the strengths of both models for more accurate predictions.

Several image preprocessing and augmentation techniques are applied to improve model generalization and reliability. These include image resizing, normalization, rotation, flipping, brightness adjustments, cropping, zooming, and shearing. The images are resized to a standard format (32×32×3) before training. The dataset was collected from the Roboflow public dataset and divided into training and validation sets.

The methodology includes modifying pre-trained CNN models to accept the resized images and classify them into eight categories. VGG16 captures fine-grained texture features, while GoogLeNet extracts multi-scale visual patterns through its inception architecture. Fusion strategies combine intermediate features and prediction probabilities from both models, improving overall diagnostic accuracy. The decision-level fusion model achieved the highest accuracy of approximately 89.48%.

The system also includes a user-friendly Streamlit-based graphical user interface (GUI) that allows users to upload nail images and receive instant disease predictions in real time. Performance evaluation uses standard medical metrics such as accuracy, precision, recall, F1-score, sensitivity, specificity, and confusion matrices.

The literature survey compares previous approaches including traditional machine learning methods, lightweight CNNs, MobileNet, SVMs, and ensemble deep learning models. Many earlier systems suffered from limited datasets, low generalization, binary classification limitations, or reduced accuracy in real-world conditions. The proposed system improves upon these limitations by using deep learning ensemble techniques, extensive data augmentation, and multimodal feature fusion.

Conclusion

The project culminated in an Automated Nail Disease Detection System that demonstrated the suitability of deep learning for classifying nail diseases using an image-based diagnostic approach. Specifically, the system utilized VGG16 and GoogLeNet architectures with a late fusion approach to achieve high classification reliability and accuracy. The integrated approach of robust preprocessing, model fusion, and real-time web interface were important contributors to achieving highquality predictions, low latencies, and generalized accuracy performance across all diseases being considered. The moderately-lightweight, modular design of the system makes it deployable in clinical and remote environments, which is useful for screening and health intervention monitoring. The use of Streamlit as a medium for the system\'s web interface makes deployment in surveillance easy and accessible, without the need for dedicated hardware or special technical expertise. In the future enhancements to the system could include attention mechanisms targeted to focus on specific disease related areas of the nail. We could include additional disease classes and expand the dataset to improve classification accuracy and robustness. Additional features such as mobile implementation, cloud-based data logging, and electronic health record (EHR) integration could be employed. Improvements are likely to expand the focus of the system so that it can ultimately develop into a comprehensive and scalable solution for dermatological diagnostics using artificial intelligence.

References

[1] R. Regin, Gautham Reddy G, Sundar Kumar Ch, Jaideep CVN, “Nail Disease Detection and Classification Using Deep Learning,” Central Asian Journal of Medical and Natural Sciences. [2] Shweta Marulkar, Bhawna Narain, “Assessment of Nail Images for Preliminary Disease Detection and Classification based on CNN: The New Horizon in Disease Detection in Humans,” Research Scholar. [3] Jumana Abdulhadi, Ayad Al-Dujaili, Amjad Jaleel Humaidi, Mohammed AbdulRaheem Fadhel, [4] “Human Nail Diseases Classification Based on Transfer Learning,” ICIC Express Letters, DOI: 10.24507/icicel.15.12.1271. [5] Lukas Folle, Pauline Fenzl, Filippo Fagni, Mareike Thies, Vincent Christlein, Christine Eder, David [6] Simon, Ioanna Minopoulou, Michael Sticherling, Georg Schett, Andreas Maier, Arnd Kleyer, “DeepNAPSI: Multi-Reader Nail Psoriasis Prediction Using Deep Learning,” Scientific Reports, https://doi.org/10.1038/41598-023-32440-8. [7] Trupti S. Indi, Yogesh A. Gunge, “EarlyStage Disease Diagnosis System Using Human Nail Image Processing,” International Journal of Information [8] Technology and Computer Science (IJITCS), https://doi.org/10.5815/ijitcs.2016.07.05. [9] Derya Yeliz Co?ar So?ukkuyu, O?uz Ata, “Classification of Melanonychia, Beau’s Lines, and Nail Clubbing Based on Nail Images and Transfer Learning Techniques,” PeerJ Computer Science, https://doi.org/10.7717/peerj-cs.1533. [10] Ima Kurniastuti, Ary Andini, “Determination of RGB in Fingernail Image as Early Detection of Diabetes Mellitus,” IEEE Xplore, DOI: [11] 10.1109/ICOMITEE.2019.8920876. [12] Rahul Nijhawan, Rose Verma, Ayushi, Shashank Bhushan, Rajat Dua, Ankush Mittal, “An Integrated Deep Learning Framework Approach for Nail Disease Identification,” IEEE Xplore, DOI: [13] 10.1109/SITIS.2017.42. [14] K. Dhanashree, P. Jayabal, A. Saran Kumar, S. Logeswari, K. Ranjeetha Priya, “Nail Feature Analysis and Classification Techniques for Disease Detection,” International Journal of Nonlinear Analysis and Applications (IJNAA), DOI: 10.22075/IJNAA.2022.6331. [15] Trupti S. Indi, Dipti D. Patil, “Nail Feature Analysis and Classification Techniques for Disease Detection,” International Journal of Computer Sciences Engineering, https://doi.org/10.26438/ijcse/v7i5.13761383

Copyright

Copyright © 2026 Gangula Padma, M. Chiranjeevi, Giduturi Hima Nandini, Aripilli Jaya Krishna, Abburi Hemanth Kumar. 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.