• We developed and demonstrated a MobileNetV3-based mobile application for detecting human nail diseases. • We trained and evaluated multiple models on nail image datasets, achieving high accuracy and precision. • We also emphasized the suitability of MobileNetV3 (proposed model) because of its lightweight design • Complex disease cases were analyzed to improve robustness and test it in complex environments Nail abnormalities can provide visual cues associated with various health conditions; however, their assessment in routine clinical practice is often subjective and inconsistent. This study presents a lightweight MobileNetV3-based transfer learning approach for the automated classification of nail conditions using deep learning. A dataset consisting of 280 nail images across four classes, normal, hyperpigmentation, clubbing, and fungal infection, was utilized. To ensure reliable evaluation, the dataset was split at the original image level prior to training, and data augmentation was applied exclusively to the training set. Six transfer learning models were evaluated under a unified experimental protocol, including VGG16, ResNet50, DenseNet201, InceptionV3, Vision Transformer, and lightweight MobileNetV3. Experimental results demonstrate that the proposed lightweight MobileNetV3 achieved the highest performance on an independent test set, attaining up to 99% accuracy while maintaining a lightweight architecture. These findings highlight the effectiveness of the proposed lightweight MobileNetV3 for image-based nail condition classification under limited data constraints and suggest its potential suitability for resource-efficient, screening-oriented applications. The proposed system is intended as a decision-support tool rather than a diagnostic replacement.
Alnafisah et al. (Sun,) studied this question.