Physics informed model for classification of dry skin using THz-TDS signals

Current techniques for reliable dry skin assessment involve histopathology on biopsies, making it an invasive and time consuming procedure. Terahertz (THz) paves a way for in vivo characterisation of skin, making it possible for an assessment to be made non-invasively within minutes. However, a meth...

Full description

Saved in:
Bibliographic Details
Main Authors: Agarwal, Agrima, Minhas, Fayyaz, Pickwell-MacPherson, Emma
Format: Conference Proceeding
Language:English
Subjects:
In vivo
Machine learning
Physics
Receivers
Reliability
Skin
Spectroscopy
Time measurement
Time-domain analysis
Online Access:Request full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Current techniques for reliable dry skin assessment involve histopathology on biopsies, making it an invasive and time consuming procedure. Terahertz (THz) paves a way for in vivo characterisation of skin, making it possible for an assessment to be made non-invasively within minutes. However, a method is needed to map the features derived from Terahertz Time Domain Spectroscopy (THz-TDS) measurements to dry skin conditions. This study explores a machine learning technique to classify dry and healthy skin types of patients based on their THz-TDS measurements. The technique uses a physics informed model of the hydration profile of skin to derive hydration parameters which determine the hydration profile of the skin with depth. This profile is further used to reconstruct a signal. Since the model is only expected to work for healthy skin, the error between actual and reconstructed signal is larger for dry skin than healthy skin. Using the reconstruction error as a predictor, a mean Area Under the Curve Receiver Operating Characteristics (AUCROC) value of 0.69 with a standard deviation of 0.08 is obtained, indicating that the predictor has a good performance over the dataset.
ISSN:2162-2035
DOI:10.1109/IRMMW-THz60956.2024.10697654