Enhancing MRI Safety: Real-Time Thermal Imaging Integrated with Deep Learning for Burn Prevention

Radiofrequency (RF)-induced burns are the most common MRI-related adverse event. Standard safety practices such as visual checks and patient communication are often insufficient, especially for anesthetized or incapacitated patients. To evaluate the feasibility of combining thermal infrared imaging with a convolutional neural network (CNN) for detecting abnormal heating in real time during MRI. Phantom and human subject experiments were performed on 3T MRI systems. A thermal camera captured images from front and back scanner placements during controlled heating. Images were labeled, preprocessed, and used to train CNNs in MATLAB. Model performance was assessed by accuracy, sensitivity, specificity, and area under the curve (AUC). Phantom testing demonstrated linear temperature rise with constant specific absorption rate (SAR) scanning. In human subject testing, CNNs achieved accuracies between 78–100%. Front placement yielded higher performance, likely due to larger datasets and improved visibility. A radar plot summarizing network performance demonstrated robust classification. This study demonstrates the feasibility of CNN-enhanced infrared monitoring to detect abnormal heating during MRI, providing a potential path toward automated, real-time burn prevention systems.