ResNet18-Based Deep Learning Approach for Efficient and Accurate Nail Disease Detection

Nail diseases are often overlooked in early stages, but timely detection is crucial for effective treatment and preventing further complications. This study presents an efficient and accurate deep learning approach using the ResNet18 architecture to detect various nail diseases. Leveraging the power of ResNet18, a lightweight yet robust convolutional neural network, we train our model to classify nail images into six categories: Acral Lentiginous Melanoma, Healthy Nail, Onychogryphosis, Blue Finger, Clubbing, and Pitting. To enhance model performance, we incorporate data augmentation techniques such as color jitter, random affine transformations, and horizontal and vertical flipping, enabling the model to learn complex and varied visual patterns associated with each condition. The ResNet18 model achieves a classification accuracy of over 91% on a validated dataset, with high precision, recall, and F1-scores across all classes, proving its ability to handle real-world variability in nail images. The approach addresses the computational challenges in medical image processing by utilizing an optimized ResNet18 model, which ensures quick inference suitable for clinical and remote applications. The confusion matrix analysis reveals the model’s strong performance, with minimal misclassifications between categories, even in visually similar cases. This research offers a scalable solution for nail disease detection, demonstrating the potential for deep learning to enhance dermatological diagnostics, particularly in resource-limited settings. By automating nail disease detection, this study contributes to improved accessibility and early intervention in nail health.