Onychomycosis: a Portable Nail Infection Detection System Using Advanced Image Processing

Onychomycosis, or fungal infection of the nail, is a highly prevalent disease with a difficult diagnosis owing to the limitations of traditional diagnostic methods (microscopic examination, fungal culture, and biopsy) often tedious, invasive, and subjective. We describe in this study a new, automated diagnostic tool that utilizes a custom-built Convolutional Neural Network (CNN) to classify nail images as either healthy or fungalinfected. Using a curated dataset consisting of 899 images, which included healthy and fungal-infected samples, sophisticated image processing, including spectral transformations (HSV) and morphological operations, along with data augmentation (flips, rotations, etc.), was applied with the aim of maximizing overall model performance. The architecture of the CNN was finalized to maximize accuracy while minimizing computation, resulting in a training accuracy of 98.44 %. When evaluated on a separate test dataset, an accuracy of 95.2 % was obtained, along with precision and recall exceeding 94 %. The final trained model was converted to TensorFlow Lite format for execution on low-power computing, such as Raspberry Pi. The model operated in $<4$ seconds to perform a diagnosis that is non-invasive, rapid, and accessible, ideal for resource-poor environments, remote environments, and low-resource settings. The automation of the diagnostic process has great potential to help support clinical activities, reduce diagnosis time, and improve patient care in the treatment of nail fungal infections.