Real-Time MRI With Deep Learning for Efficient Evaluation of Neuromuscular Breathing Impairment

Efficient detection of breathing impairment is critical for treatment and prognosis in neuromuscular disorders. However, standard pulmonary function tests often yield ambiguous results. This prospective study evaluates whether advanced real-time MRI (RT-MRI) combined with deep learning-based image segmentation provides sensitive outcome measures for respiratory dysfunction in late-onset Pompe disease (LOPD), a model disease for diaphragmatic weakness. Eleven Pompe patients (mean age 52.2 years; 55% female) and 11 controls (mean age 50.9 years; 55% female) were included. RT-MRI with a temporal resolution of 50 ms, combined with U-Net-supported lung segmentation, revealed significantly reduced diaphragmatic motion in Pompe patients compared to controls and unmasked paradoxical diaphragmatic motion in Pompe patients (7 of 11). Reduced diaphragmatic sniff velocity and pathological diaphragmatic/thoracic synchronicity were detected in Pompe patients with still normal results in standard pulmonary function tests. Fatty involution of the diaphragm as quantified by fast T1 mapping correlated significantly with functional parameters from RT-MRI and pulmonary function tests. RT-MRI combined with deep learning-based lung segmentation offers novel biomarkers for early detection of respiratory muscle weakness. This new technique provides useful outcome measures for clinical care as well as treatment studies in patients with neuromuscular breathing impairment. The technique can also be used to characterize physiologic breathing patterns in healthy individuals.