Functional Magnetic Resonance Imaging and Deep Learning to Improve Deep Brain Stimulation Therapy

点击翻译按钮获取中文摘要

PROJECT SUMMARY/ABSTRACT Successful treatment of Parkinson's disease (PD) using deep brain stimulation (DBS) therapy requires an optimal setting of stimulation parameters to correct brain function anomalies. The commonly employed DBS 1.0 electrodes have only four contact locations (with no stimulation directionality) that are used to electric pulses to a target volume of the brain. DBS 1.0 electrodes require the optimization of four stimulation parameters: signal frequency, voltage, pulse width, and contact location. In current standard-of-care optimization protocol, the DBS parameters are adjusted (via trial and error) until the physician determines an optimal set of parameters. This empirical optimization protocol requires numerous clinical visits (~6 weeks interval) that substantially increases the time to optimization (TTO) per patient (~1 year), financial burden, and ultimately limits the number of patients that can have access to DBS therapy. Even though there are more effective electrodes, DBS 1.0 electrodes are mostly used by clinicians because their smaller parameter space pose less difficulty during manual clinical optimization. However, DBS 1.0 electrodes cannot be directed to stimulate a smaller volume of tissue, which can lead to extraneous stimulations that can reduce patient clinical benefits and increase side effects. By contrast, the newer DBS electrodes (dubbed DBS 2.0) have a greater number of contact locations and can be programmed to stimulate a smaller volume of tissue at multiple levels and directions. Several published reports have shown that DBS 2.0 electrodes (compared to DBS 1.0) are more energy-efficient and improve patient outcomes with lesser side-effects and a wider therapeutic window. However, the expanded DBS 2.0 parameter space has made empirical programming of the electrodes difficult as the TTO per patient is beyond acceptable clinical timeframes. This increased difficulty has hindered adoption of DBS 2.0 electrodes by clinicians. To significantly shorten and simplify DBS 2.0 parameter optimization—thus enabling its wider adoption for more precise therapy—a uniquely qualified multi- disciplinary team of magnetic resonance imaging (MRI) physicists, artificial intelligence (AI) engineers, and clinicians from GE Research and the University Health Network propose to: 1) develop a semi-automated functional MRI (fMRI) and deep learning (DL)-based system for rapid optimization of DBS 2.0 parameters; 2) demonstrate its clinical benefit in the treatment of PD patients using bilateral stimulation of the sub-thalamic nucleus with DBS 2.0 electrodes in a pilot study. Success of this program will decrease the TTO per patient for PD patients with DBS 2.0 implants to ~1 hour, and will improve patient throughput and outcomes in the treatment of PD. The proposed fMRI-DL-based optimization method may also improve access by making it possible for non-expert centers (without highly specialized clinicians) to carry out stimulation parameters optimization in patients after the electrode insertion surgery have been completed in expert centers.