Artificial neural networks and statistical classification applied to Electrical Impedance Spectroscopy data for Melanoma diagnosis in Dermatology (DermaSense)

Abstract

The gold standard for diagnosis of cutaneous melanoma is excisional biopsy and histological examination, however dermoscopy remains the most common examination during a patient’s first visit to the dermatologist. Dermoscopy is non-invasive but relies on a clinician’s subjective pattern recognition skills and experience. We designed and built a novel, non-invasive dermatological scanner based on electrical impedance spectroscopy, which can complement dermoscopy by providing objective measurement data within seconds, at the point-of-examination. In this paper we present the DermaSense scanner as well as associated statistical and neural network classification algorithms aimed at correlating acquired with verified diagnostic data. In addition, we present an initial pilot study on measurements of melanoma against nevi as well as against clear patches of skin.