Application of machine learning for fast prediction of MRI-induced RF heating in patients with implanted conductive leads

点击翻译按钮获取中文摘要

Project Summary There is a steady growth in the use of conductive medical implants in the US and globally. Currently, more than 12 million Americans carry a form of orthopedic, cardiac, or neuromodulation device, and the number grows by 100,000 annually. It is estimated that 50-75% of patients with implants would benefit from magnetic resonance imaging (MRI) during their lifetime, some with repeated examinations. Unfortunately, the interaction between MRI's radiofrequency (RF) fields and conductive implants have led to fatal injuries due to RF heating of implants, making MRI inaccessible to most patients. In response, extensive effort has been dedicated to quantifying and mitigating the problem of MR-induced RF heating. Following regulatory recommendations, these efforts heavily rely on full-wave electromagnetic (EM) simulations that model details of MRI RF coils, human body, and implant, and as such are notoriously cumbersome. Even taking advantage of today's high- power computing clusters it typically takes tens of hours to complete a single simulation. Our long-term goal is to enable application of in-silico medicine for RF heating assessment of implants in real time and on a patient- by-patient basis. Our main hypothesis is to test whether advanced deep learning (DL) methods can rapidly and accurately predict RF heating of elongated implants (such as leads), when only the background electric field of the MRI RF coil and the implant's trajectory are in hand. The background RF field is the field that exists in the body in the absence of the implanted device and can be easily calculated in advance for any known MRI coil. Similarly, the implant's trajectory can be extracted from routine medical images in only a few minutes. Herein, we propose to develop, optimize, and experimentally validate a deep learning approach that predicts RF heating of DBS systems during MRI with body coils at both 1.5 T and 3 T with <2℃ error. We will build training datasets from 500 patient-derived DBS lead models, apply EM simulations to calculate ground truth RF heating using vendor-provided models of MRI RF coils, and develop deep learning algorithms to predict the RF heating with 2℃ accuracy with knowledge of only the implant's trajectory (CT-based) and the coil's features (vendor- specific). If successful, our work will introduce a paradigm shift in the practice of MRI RF heating assessment, reducing simulation times from tens of hours to a few minutes. This will democratize a practice that is currently afforded by only a handful of well-resourced companies and opens the door to a plethora of novel implant designs and patient-specific safety guidelines. Importantly, the knowledge gained in this innovative work can be translated to patients with other types of implants, especially those with cardiac implantable electronic devices and spinal cord stimulators.