Synthetic Map of the Cerebral Metabolic Rate of Oxygen Based on Quantitative MRI

The cerebral metabolic rate of oxygen CMRO2 quantifies oxygen consumption in brain tissue, serving as an indicator of neuronal activity and metabolic demand. It is considered a valuable biomarker for assessing neurovascular health and monitoring neurological diseases. While the gold standard for CMRO2 measurement is O-15 PET imaging, it is invasive and expensive. An alternative MRI-based method, called multiparametric quantitative BOLD (mqBOLD), estimates CMRO2 using multiple quantitative MRI (qMRI) modalities such as T2, T2*, cerebral blood flow (CBF), and cerebral blood volume (CBV). However, this method has several limitations: it requires multiple image acquisitions, and CBV estimation typically involves the use of contrast agents, introducing potential health risks and increased costs. This research proposes a deep learning approach to overcome these limitations by generating synthetic CMRO2 maps from a reduced set of qMRI inputs. By training a conditional generative adversarial network to replicate the CMRO2 image using fewer and less invasive MRI sequences, the method aims to decrease acquisition time, eliminate the need for contrast agents, and maintain high image quality. The quality of the generated maps was assessed not only through model performance metrics but also by their practical applicability in real-world scenarios, such as their capacity to detect functional alterations and the ability to generalize to different datasets. Among all combinations, the contrast-free CBF and T2* input set delivered the best balance of minimal inputs and performance. The results indicate that synthetic CMRO2 maps can be generated from a subset of the mqBOLD inputs and generalize to similar datasets, while cross-scanner robustness requires further validation.