Deep Learning-Driven Computational Imaging for Noninvasive Monitoring System of Brain Temperature and Metabolism: A Hypothermia Validation for Acute Ischemic Stroke.

Brain temperature (BT) is a critical physiological indicator closely associated with neurological function and disease progression. However, real-time, noninvasive monitoring of BT remains a challenge due to the limitations of current technologies. Here, we present a novel multimodal framework combining bioheat transfer modeling, deep learning, and computational thermography for accurate BT prediction and imaging. A one-dimensional convolutional neural network was trained on multimodal clinical data, integrating cerebral blood flow, tissue oxygen saturation, and intracranial pressure, achieving a mean absolute error of 0.31°C in BT prediction. The framework incorporates finite element analysis to generate 3D thermographic maps of brain tissue with a spatial resolution of 0.4 mm, validated using MRI-derived data. This approach demonstrated robust performance in predicting localized temperature variations in acute ischemic stroke patients undergoing therapeutic hypothermia, with deviations below 0.45°C. Our findings highlight the potential of this system to enable precise BT monitoring, bridging the gap between computational modeling and clinical neuro-thermometry, and paving the way for advanced diagnostic and therapeutic interventions.