Explainable artificial intelligence and deep transfer learning for skin disease diagnosis

Abstract

This Report aims at describing a tool that can potentially be used to support institutions, research, and patients in the implementation of focused remedies, based on quantifiable and Prescaler elements, for interpretable diagnosis of skin diseases. We envisioned models to include specific neural networks which were also trained to learn the features residing in imagery data, related to the case interest: skin diseases. The tool designed shall serve as a decision support system (DSS) for various skin diseases, when fed imagery data and also usher in an understanding of which variables or feature interactions impacted model predictions, and the procedures taken to make particular decisions. Research from IDHC[[6, 7]], unmasks the fact that errors arising from misdiagnosis or impeded diagnosis carry on all around the healthcare environment and keep tormenting an intolerable quantity of victims. This misdiagnosis is a result of inadequate victim assessment, failure to order appropriate diagnostic tests, negligence in terms of referring the su↵ering party to a specialist, and failure to investigate more on a patient’s a✏iction. Researchers have suggested computer-based diagnostic decision support tools to supplement medical specialist decisions in an attempt to reduce misdiagnosis. A recent study from the Johns Hopkins School of Medicine shows, from a sample space of 10,000 diseases identified by doctors in patients, 15 stand out as the most misdiagnosed: among these being skin cancers, thus the need for accurate/ precise diagnosis is critical to prevent wrong course of treatments being prescribed as well as time wastage for that matter. Proper diagnosis can therefore avoid long-term complications for the infected patients and improve the e↵ectiveness of treatments.