The chemical exchange relayed nuclear Overhauser effect (rNOE) is a molecular MRI contrast mechanism associated with nonexchangeable carbon-bound protons in mobile macromolecules (such as aliphatic and aromatic protons)1. By exploiting the rNOE sensitivity to membrane lipids and proteins, several research groups have demonstrated its potential for the detection and characterization of a variety of pathologies, including: brain cancer2–5, stroke6, liver disease7,8, and spinal cord injury9.
In the typical settings, rNOE contrast-weighted images are generated following a full Z-spectrum acquisition, as traditionally performed in chemical exchange saturation transfer (CEST) MRI10,11. However, a straightforward computation of the rNOE-weighted signal via the magnetization transfer ratio (MTR) metric will inherently convolve the desired effects with the much stronger magnetization transfer (MT) contrast stemming from semisolid macromolecules12. Moreover, the signal will be scaled by the longitudinal water relaxation and diluted by the water proton direct saturation (spillover)13.
Several previous works were able to mitigate or eliminate the water T1, T2, and semisolid MT contributions from the rNOE signals12,14–16. However, these approaches still require a lengthy full Z-spectrum acquisition or are directly affected by the saturation pulse settings employed2,17, limiting the ability to compare image findings across different sites. In any case, rNOE-weighted images represent a combined contribution from the aliphatic proton volume fraction and exchange rate, hindering the direct evaluation of the compound of interest concentration.
Magnetic resonance fingerprinting (MRF) is an emerging approach for quantitative MRI18, which was recently FDA-approved for some indications19. It combines non-steady state signals acquisition with model-based image reconstruction, for the rapid and simultaneous quantification of multiple magnetic properties20. While originally suggested for water-pool relaxation mapping, MRF was later expanded and modified for a variety of other contrast mechanisms, including CEST and semisolid MT21–24. Preliminary animal and human studies have demonstrated that CEST-MRF of the amide proton and the semisolid MT exchange parameters can accurately detect the treatment response to oncolytic virotherapy25 and distinguish between different tumor regions in brain metastasis26.
While the first CEST-MRF reports used correlation-based pattern recognition for quantitative image reconstruction, the complexity of the in-vivo multi-pool environment, which translates into impractical parameter quantification times, has motivated the pursuit of alternative and faster reconstruction methods. Specifically, a variety of neural network architectures were designed and validated for ultra-short (~ a few seconds-long or less) reconstruction of the proton exchange parameter maps25–30.
Here, we designed a deep molecular MRF method for the quantitative and rapid mapping of the aliphatic rNOE proton volume fraction and exchange rate. Moreover, considering the “cross-talk” and potential MT contributions affecting the rNOE signal interpretation in-vivo, a dual-purpose serial acquisition and reconstruction framework was utilized (Fig. 1), which extracts both proton pool characteristics. The method was systematically validated using rNOE phantoms, in-vivo wild-type mice, and human volunteers, yielding excellent agreement with ground truth and previous literature values.