Breaking the Spatiotemporal Barriers of MR Spectroscopic Imaging: A Marriage of Spin Physics and Machine Learning for Label-Free Molecular Imaging

Abstract

Molecular imaging has been a dream of biomedical imaging scientists for decades, and governments and industries around the world have invested billions into this area. However, most existing molecular imaging techniques (e.g., PET and SPECT) require exogenous molecular probes or reporters to be introduced into a subject in order to obtain molecule-specific information, thereby limiting their practical utility. Magnetic resonance spectroscopic imaging (MRSI) has long been recognized as a powerful tool for non-invasive, label-free molecular imaging and a lot of outstanding work has been done over the past three decades, resulting in significant advances in MRSI data acquisition, pulse sequences, data processing, and image reconstruction. However, in spite of these enormous progresses, current MRSI technology still falls short of providing adequate spatial resolution, speed, and signal-to-noise ratio (SNR) for routine clinical and research applications.

The talk will discuss our recent advances in overcoming the long-standing technical barriers for label-free molecular imaging using intrinsic MR signals. This ultrafast MRSI technology, resulting from many years of research efforts, is based on a new approach to spatiospectral imaging, which includes rapid data acquisition, sparse sampling of (k, t)-space, constrained image reconstruction, and learning-based spectral quantification using spectral basis from quantum simulation. This technology has demonstrated an unprecedented combination of resolution, speed and SNR for MRSI. In this talk, I will discuss this new ultrafast MRSI technology and show some exciting experimental results we have obtained.