Deep learning of magnetic resonance fingerprints for the quantitative imaging of apoptosis following oncolytic virotherapy

Oncolytic virotherapy is a promising anticancer strategy, however, non-invasive imaging methods that detect intratumoral viral spread and an individual’s response to therapy are either slow, lack specificity or require the use of radioactive or metal-based contrast agents, hindering their clinical translation. Here, we describe a fast chemical-exchange saturation transfer magnetic resonance fingerprinting (CEST-MRF) approach that selectively labels the exchangeable amide protons of endogenous proteins and the exchangeable macromolecule protons of lipids, without requiring an exogenous contrast agent. A deep neural network, previously trained with simulated MR-fingerprints, then generates, using CEST-MRF image data, quantitative maps of intratumoral pH, and protein and lipid concentrations. In a mouse model of glioblastoma multiforme, the maps enabled the detection of the early apoptotic response to oncolytic virotherapy, characterized by decreased cytosolic pH and reduced protein synthesis. Imaging of a healthy human volunteer yields molecular parameters in good agreement with literature values, demonstrating the clinical potential of artificial intelligence-based CEST-MRF, which may be further applicable to the imaging of a wide range of pathologies, including stroke, cancer, and neurological disorders.