LSTM-based hand thermal recovery analysis using infrared and RGB-visual imagery

Analysis of hand thermal recovery is a valuable clinical method regularly used in the diagnosis and follow up of a number of pathologies such as Raynaud Syndrome (RS), Systemic Sclerosis (SS), and Rheumatoid Arthritis (RA). This paper presents an ongoing project on hand thermal recovery analysis using infrared and visual video signals simultaneously acquired, with potential application in the clinic evaluation of such diseases. Morphological image pre-processing on frame sequences allows the user to define the desired Regions of Interest (ROI) as required in the corresponding analysis. A collection of time series obtained during a lapse of 10 minutes provides information about the evolution of temperature acquired with an infrared camera, and video signals on three color planes. The collection of signals is fed into a Long Short-Time Memory neural network (LSTM-NN) structured and implemented in classification mode, which is aimed to capture the temporal evolution of the time series for classification purposes. Data was collected from a group of 48 healthy volunteers ranging from 24– 32 years old. The obtained results come from a set of experiments aimed to distinguish between quick, medium, and slow thermal recovery in several cases using some predefined ROI.