Artificial Intelligence in Dermatology: Improving Diagnostic Accuracy in Nail Disorders Through Deep Learning Algorithms

This study presents the development and evaluation of an AI-based system for detecting nail diseases using two models: Of them one was trained with Roboflow, and the other one was trained with YOLO. The Roboflow model was a multi-label classification of 7264 images of 11 nail conditions, and the accuracy of the model was 98.8%. In the YOLO model using medium and large variations, six particular nail diseases were distinguished with a total of 3,081 images. For the medium YOLO model, if each of the model sizes is trained for 20 epochs, the top-1 accuracy was 78.075% and the top-5 accuracy of 96.791%. At the same time, the large-sized model had a top 1 accuracy of 76.471% and a top 5 accuracy of 96.791%. Both models were trained based on the augmented datasets which assisted in improving the performance of the models. The Roboflow model was very good for some diseases, such as Acral Lentiginous Melanoma and Terry’s Nails but was not so good for diseases like Lindsay’s Nails, which are not very common. This research is in harmony with the current development in AI in dermatology and can be used to diagnose nail diseases at an early stage and a cheaper cost.