An Intelligent Deep Learning Framework for Automated Nail Disease Identification from Images

The nail abnormalities usually indicate theunderlying dermatological or systemic related healthconditions, but they are mostly diagnosed through expertnon-objective visual examination, which is not alwaysaccurate. In this paper, an intelligent deep learningsystem of automatic detection of nail diseases isintroduced through image analysis. In contrast to thecurrent methods that are based on the traditional CNNpipelines, in the proposed framework, multi-stage image enhancement, region-centered feature learning, and light-weight transfer learning are incorporated to enhance robustness in the real world. The system in question issupposed to identify and label nail conditions throughlearning fine-grained visual features like chromaticvariations, texture discontinuities andstructuraldisfigurement. Empirical testing shows that theframework works with credible classification capabilitiesand yet is computationally efficient, which is appropriatewhen the framework can be used in clinical support andtele-dermatology.