Non-Invasive In-Situ Rapid Detection of Human Ankles Fractures using the Cole Model and Machine Learning

Abstract

This work presents a pilot study on Electrical Impedance Spectroscopy (EIS) measurements to detect ankle fractures. EIS measurements consist in a frequency sweep from 5 to 100 kHz, performed using a home-made device in bipolar mode. Measurements were performed in the ankles of 36 subjects ageing 37.9 ± 12.1 years (on each malleolus by 2 disposable adhesive electrodes), from which, 18 had a diagnosed fracture in one of the ankles. EIS data were analyzed with the curve fitting from the Cole Model and by a Machine Learning method based on Linear Discriminant Analysis (LDA). From Cole fitting curve, a difference between Rs values with P = 5.2669·10-6 for fractured subjects (highly significant) and for healthy subjects a P=0.3455 or not significant were observed. For this reason, a mean cut-off value was found to distinguish between fractured and healthy subjects of |ΔRs|=60Ω. The LDA-based method obtained acceptable metrics but did not reach the numerical values necessary to validate it as a diagnostic test. These results show the possibility of generating a diagnostic test based on a quantitative parameter to differentiate subjects with a fractured ankle from non-fractured ones by means of electrical bioimpedance measurements.