Automatic Rejection based on Tissue Signal (ARTS) for motion-corrected quantification of cerebral venous oxygenation in neonates and older adults.

Objective: Cerebral venous oxygenation (Y) is a key parameter for the brain's oxygen utilization and has been suggested to be a valuable biomarker in various brain diseases including hypoxic ischemic encephalopathy in neonates and Alzheimer's disease in older adults. T-Relaxation-Under-Spin-Tagging (TRUST) MRI is a widely used technique to measure global Y level and has been validated against gold-standard PET. However, subject motion during TRUST MRI scan can introduce considerable errors in Y quantification, especially for noncompliant subjects. The aim of this study was to develop an Automatic Rejection based on Tissue Signal (ARTS) algorithm for automatic detection and exclusion of motion-contaminated images to improve the precision of Y quantification.

Methods: TRUST MRI data were collected from a neonatal cohort (N = 37, 16 females, gestational age = 39.12 ± 1.11 weeks, postnatal age = 1.89 ± 0.74 days) and an older adult cohort (N = 223, 134 females, age = 68.02 ± 9.01 years). Manual identification of motion-corrupted images was conducted for both cohorts to serve as a gold-standard. 9.3% of the images in the neonatal datasets and 0.4% of the images in the older adult datasets were manually identified as motion-contaminated. The ARTS algorithm was trained using the neonatal datasets. TRUST Y values, as well as the estimation uncertainty (ΔR) and test-retest coefficient-of-variation (CoV) of Y, were calculated with and without ARTS motion exclusion. The ARTS algorithm was tested on datasets of older adults: first on the original adult datasets with little motion, and then on simulated adult datasets where the percentage of motion-corrupted images matched that of the neonatal datasets.

Results: In the neonatal datasets, the ARTS algorithm exhibited a sensitivity of 0.95 and a specificity of 0.97 in detecting motion-contaminated images. Compared to no motion exclusion, ARTS significantly reduced the ΔR (median = 3.68 Hz vs. 4.89 Hz, P = 0.0002) and CoV (median = 2.57% vs. 6.87%, P = 0.0005) of Y measurements. In the original older adult datasets, the sensitivity and specificity of ARTS were 0.70 and 1.00, respectively. In the simulated adult datasets, ARTS demonstrated a sensitivity of 0.91 and a specificity of 1.00. Additionally, ARTS significantly reduced the ΔR compared to no motion exclusion (median = 2.15 Hz vs. 3.54 Hz, P < 0.0001).

Conclusion: ARTS can improve the reliability of Y estimation in noncompliant subjects, which may enhance the utility of Y as a biomarker for brain diseases.