Infrared Thermal Imaging and Artificial Neural Networks to Screen for Wrist Fractures in Pediatrics

Abstract

Paediatric wrist fractures are commonly seen injuries at emergency departments. Around 50% of the X-rays taken to identify these injuries indicate no fracture. The aim of this study was to develop a model using infrared thermal imaging (IRTI) data and multilayer perceptron (MLP) neural networks as a screening tool to assist clinicians in deciding which patients require X-ray imaging to diagnose a fracture. Forty participants with wrist injury (19 with a fracture, 21 without, X-ray confirmed), mean age 10.50 years, were included. IRTI of both wrists was performed with the contralateral as reference. The injured wrist region of interest (ROI) was segmented and represented by the means of cells of 10 × 10 pixels. The fifty largest means were selected, the mean temperature of the contralateral ROI was subtracted, and they were expressed by their standard deviation, kurtosis, and interquartile range for MLP processing. Training and test files were created, consisting of randomly split 2/3 and 1/3 of the participants, respectively. To avoid bias of participant inclusion in the two files, the experiments were repeated 100 times, and the MLP outputs were averaged. The model’s sensitivity and specificity were 84.2% and 71.4%, respectively. Further work involves a larger sample size, adults, and other bone fractures.