Search for a command to run...
Accurate diagnosis of nail diseases such as onychomycosis and nail psoriasis remains a clinical challenge due to overlapping visual symptoms and limited diagnostic resources. This study investigates the potential of deep learning models, MobileNetV2 and CNN-LSTM, for image-based classification of these conditions. Publicly available datasets were collected, preprocessed, and expanded through augmentation to enhance robustness. Both models were trained and evaluated under multiple hyperparameter configurations. MobileNetV2 consistently achieved higher performance, with a reported peak test accuracy of 99.90%, compared to CNN-LSTM’s maximum of 92.08%. To assess clinical applicability, a prototype system was developed using Streamlit, enabling users to upload nail images for automated classification. Validation against a set of dermatologist-labeled clinical images yielded a real-world accuracy of 70%. This outcome highlights both the feasibility of AI-driven nail disease detection and the limitations of dataset size, diversity, and preparation protocols. Importantly, the analysis of misclassified images revealed challenges such as lighting variability, multiple nails in a single frame, and symptom overlap, underscoring the need for more representative datasets. The contribution of this work is twofold: first, it demonstrates the promise of lightweight architectures like MobileNetV2 for nail disease detection; second, it presents a deployable prototype that bridges machine learning research with accessible healthcare applications. While the current accuracy is constrained by dataset limitations, this research provides a proof-of-concept foundation. Future work will adopt best-practice augmentation protocols, expand dataset diversity, and incorporate additional nail conditions to strengthen clinical robustness.