Quantifying Nocturnal Itch And Its Impact On Sleep Using Machine Learning And Radio Signals

Abstract

Today, chronic itch affects up to 15% of the population, and is associated with over $90 Billion in annual population-expenditures in the US. Despite all the interest around this area, there’s still no solution for quantifying nocturnal scratching and its impact on patients’ sleep quality in an objective, sensitive and privacy preserving way. In this work we collect large nocturnal scratching dataset, consisting of 370 nights of infrared footage, radio-frequency (RF) data, and human annotations of scratching. Using this data, we develop a neural network model that can detect occurrences of nocturnal scratching using only radio signals. The developed model can achieve very high accuracy in measuring meaningful scratching metrics, across a diverse population of patients. Additionally, by utilizing prior art on extracting sleep stages from radio signals, we can gain insights about the effect of itch on the sleep quality of a chronic itch patients, especially relative to healthy individuals.