Dynamic infrared thermography (DIRT) is a noninvasive method that provides real-time insights into skin perfusion by tracking thermal variations alongside temperature images. In this study, we explored the possibility of using DIRT imaging data to differentiate selected diagnostic features of examined patients. We employed a combination of preprocessing algorithms and neural models built on the acquired series of full-body DIRT scans to enable accurate discrimination between patients after myocardial infarction and patients after being diagnosed with COVID-19 infection. Training dedicated convolutional neural networks (CNN) on the full thermogram scans yielded promising results, with an overall accuracy of 85%. These findings underline the value of preserving comprehensive thermal information in medical imaging for CNN classification, especially in applications where subtle thermal variations are critical to diagnosis. Our work, as preliminary research, highlights that full-body DIRT imaging data can serve as a meaningful source of information about patients’ health and the presented results clearly show that the application of machine learning techniques can allow for the creation of an automated framework for medical thermogram analysis.