Using artificial neural networks for anomaly detection in infrared thermography images for rapid diagnosis in an emergency care unit

Infrared thermography (IRT) has emerged as an affordable, rapid and noninvasive complement to widely adopted yet resource-demanding medical imaging techniques such as MRI, CT scans and X-rays, offering diverse applications in the medical field. While IRT hardware is well established and capable of providing high-quality thermographic images, the analysis of such images often requires well-trained experts. Current state-of-the art methods for computer-aided IRT analysis rely on statistical tests of temperature gradients between control points, which are suboptimal because they do not fully exploit the available information regarding spatial temperature distributions.

This paper addresses this issue by incorporating artificial neural networks (ANN) into the IRT analysis workflow. We focused on a particular case in which the IRT was utilized in the emergency department (ED) for predicting 30-day mortality, thereby contributing to improved diagnosis and patient care in emergency medicine. In total, the IRT images of 214 patients were analyzed. Various ANN-based approaches were considered in this study, and the best results were obtained using an anomaly detection model based on a variational autoencoder (VAE), which achieved promising results for detecting abnormal images. This paper comprehensively presents all the analysis details as well as recommendations regarding image preprocessing, augmentation, and potential enhancements of the models.