The Ten-Classification of Whole-body Parts from Infrared Thermal Images

The human body is a natural source of biological infrared radiation. Infrared thermal imaging began to be applied in clinical research and used as diagnostic applications in head, neck, cardiovascular, lung, breast, gastrointestinal, liver, gallbladder, prostate, spine, limbs, and blood vessels. Whole-body thermal imaging can comprehensively conduct early warning analysis for various diseases in the whole body. This study investigates the automatic detection of posture and body parts in medical thermal images. The dataset used in this research comprises 12,282 infrared thermal images from 600 individual cases. This research presents a novel approach to identify and categorize ten specific body parts depicted in twenty infrared thermal images. It involves designing a deep learning-based model that outputs ten categories of human body parts by processing twenty infrared thermal images at a time. The results indicate that the models that utilized the transfer learning technique achieved a classification accuracy above 97.6%, while the model trained from scratch with the best results of 86.2%.