Project Grant F30HL189213

ADAPTIVE IMAGE RECONSTRUCTION AND DEVICE TRACKING IN INTERVENTIONAL CARDIOVASCULAR MRI VIA DEEP LEARNING - PROJECT SUMMARY INTERVENTIONAL CARDIOVASCULAR MRI (ICMR) PRESENTS A RADIATION-FREE ALTERNATIVE FOR PEDIATRIC CARDIAC CATHETERIZATIONS, PROVIDING SUPERIOR SOFT TISSUE CONTRAST AND HEMODYNAMIC ASSESSMENTS COMPARED TO TRADITIONAL FLUOROSCOPY. DESPITE ADVANCEMENTS IN MRI-COMPATIBLE DEVICES AND SEQUENCES, ICMR ADOPTION REMAINS CONSTRAINED DUE TO POOR REAL-TIME IMAGE QUALITY FROM SPARSE MRI DATA AND UNRELIABLE REAL-TIME PASSIVE DEVICE TRACKING. CURRENT ADVANCED RECONSTRUCTION METHODS ARE COMPUTATIONALLY INTENSIVE, PREVENTING THEIR REAL-TIME USE IN INTERVENTIONAL SETTINGS. PASSIVE CATHETER TRACKING, WHILE WIDELY APPLICABLE, RELIES HEAVILY ON MANUAL PLANE ADJUSTMENTS AND LACKS AUTOMATION. THIS PROPOSAL INNOVATIVELY APPLIES DEEP LEARNING TO OVERCOME SUBOPTIMAL IMAGE QUALITY AND UNRELIABLE REAL-TIME TRACKING OF PASSIVE DEVICES IN ICMR. OUR TEAM WILL DEVELOP AND INTEGRATE A RAPID, ADAPTIVE MRI RECONSTRUCTION FRAMEWORK USING TEMPORALLY-AWARE RESIDUAL NEURAL NETWORKS (RNN) TO RECONSTRUCT HIGH-QUALITY IMAGES FROM UNDERSAMPLED DATA. A NEURAL NETWORK-BASED IMAGE QUALITY ASSESSMENT METHOD WILL BE EMBEDDED FOR REAL-TIME OPTIMIZATION. FOR DEVICE TRACKING, WE WILL CREATE NEURAL NETWORK-BASED ALGORITHMS TO SEGMENT AND PREDICT THE CATHETER POSITION DYNAMICALLY ACROSS IMAGING PLANES AND ACQUISITION STRATEGIES, COMPARING MULTIPLE 3D FRAMEWORK APPROACHES. THE FRAMEWORK WILL UNDERGO EVALUATION IN SIMULATED HEART PHANTOMS AND BOTH PEDIATRIC AND ADULT PATIENT STUDIES, VALIDATING IMPROVEMENTS IN TEMPORAL RESOLUTION, IMAGE QUALITY, AND TRACKING ACCURACY. SUCCESSFUL IMPLEMENTATION WILL STREAMLINE ICMR PROCEDURES, IMPROVE CLINICAL DECISION-MAKING, REDUCE MANUAL INTERVENTION, AND ENHANCE PATIENT SAFETY AND OUTCOMES.