Alexander D Cohen, Laura Glass Umfleet, Shawn Obarski, Halle Berres, Kelly Ristow, Malgorzata Franczak, John R Kirby, Yang Wang
Mean regional AUC values across the 10 highest-performing regions ranged from 0.740 to 0.766 for CBF, 0.753 to 0.805 for CVR, and 0.816 to 0.830 for the combined CBF + CVR model. Compared with the better-performing single-metric model, the combined model improved regional classification performance by ΔAUC = 0.020-0.062. Five regions showed nominal paired-permutation significance, although none remained significant after false discovery rate correction. Higher decision scores were associated with poorer delayed recall (partial r = -0.353 to -0.422) and poorer TMT Parts A and B performance (partial r = -0.243 to -0.390).
BACKGROUND: Cerebrovascular dysfunction contributes to mild cognitive impairment (MCI), but the complementary value of arterial spin labeling (ASL)-derived cerebral blood flow (CBF) and breath-hold cerebrovascular reactivity (CVR) for multivariate MRI classification remains unclear.
PURPOSE/HYPOTHESIS: To determine whether combining CBF and CVR improves multivariate classification of clinically defined MCI compared with either metric alone and whether classifier-derived scores relate to cognitive performance.
STUDY TYPE: Retrospective case-control study.
POPULATION/SUBJECTS: Eighty-five older adults, including 37 participants with clinically defined MCI (73.8 ± 5.1 years; 21 male) and 48 cognitively unimpaired (CU) controls (69.2 ± 7.1 years; 14 male).
FIELD STRENGTH/SEQUENCE: 3 T MRI; 3D pseudo-continuous arterial spin labeling (pCASL), multiband multi-echo (MBME) echo planar imaging (EPI) blood oxygenation level dependent (BOLD) functional MRI breath-hold CVR imaging, 3D T1-weighted magnetization-prepared rapid gradient-echo, and 3D T2-weighted fast spin-echo (FSE).
ASSESSMENT: Whole-brain searchlight multivariate pattern analysis using a linear logistic regression classifier evaluated MCI versus CU subjects from CBF-only, CVR-only, and combined CBF + CVR features. Regional performance was summarized using the 400-region Schaefer atlas. Decision scores from the combined model were correlated with delayed recall and Trail Making Test (TMT) Part A and B performance.
STATISTICAL TESTS: Permutation testing assessed classifier significance. Linear regression evaluated cognitive associations. False discovery rate correction was applied for multiple comparisons. Statistical significance was defined as p < 0.05.
RESULTS: Mean regional AUC values across the 10 highest-performing regions ranged from 0.740 to 0.766 for CBF, 0.753 to 0.805 for CVR, and 0.816 to 0.830 for the combined CBF + CVR model. Compared with the better-performing single-metric model, the combined model improved regional classification performance by ΔAUC = 0.020-0.062. Five regions showed nominal paired-permutation significance, although none remained significant after false discovery rate correction. Higher decision scores were associated with poorer delayed recall (partial r = -0.353 to -0.422) and poorer TMT Parts A and B performance (partial r = -0.243 to -0.390).
DATA CONCLUSION: Combining CBF and CVR improved multivariate discrimination of clinically defined MCI relative to either metric alone and identified clinically relevant posterior cortical neurovascular patterns associated with memory and executive dysfunction.
EVIDENCE LEVEL: 3.
TECHNICAL EFFICACY: Stage 2.