In vivo imaging of phosphocreatine with artificial neural networks

Phosphocreatine (PCr) plays a vital role in neuron and myocyte energy homeostasis. Currently, there are no routine diagnostic tests to noninvasively map PCr distribution with clinically relevant spatial resolution and scan time. Here, we demonstrate that artificial neural network-based chemical exchange saturation transfer (ANNCEST) can be used to rapidly quantify PCr concentration with robust immunity to commonly seen MRI interferences. High-quality PCr mapping of human skeletal muscle, as well as the information of exchange rate, magnetic field and radio-frequency transmission inhomogeneities, can be obtained within 1.5 min on a 3 T standard MRI scanner using ANNCEST. For further validation, we apply ANNCEST to measure the PCr concentrations in exercised skeletal muscle. The ANNCEST outcomes strongly correlate with those from 31P magnetic resonance spectroscopy (R = 0.813, p < 0.001, t test). These results suggest that ANNCEST has potential as a cost-effective and widely available method for measuring PCr and diagnosing related diseases. Phosphocreatine plays a vital role in cellular energetic homeostasis, but there are no routine diagnostic tests to noninvasively map the distribution with clinically relevant spatial resolution. Here, the authors develop and validate a noninvasive approach for quantifying and imaging phosphocreatine, without contrast agents, on widely available clinical MRI scanners with artificial neural networks.