Improved assessment of longitudinal spinal cord atrophy in multiple sclerosis using a registration-based approach: relevance for clinical studies

BACKGROUND:Reliable measurements of cervical cord atrophy progression may be useful for monitoring neurodegeneration in multiple sclerosis (MS). PURPOSE:To compare a new, registration-based (Reg) method with two existing methods (active surface [AS] and propagation segmentation [PropSeg]) to measure cord atrophy changes over time in MS. STUDY TYPE:Retrospective. SUBJECTS:Cohort I: Eight healthy controls (HC) and 28 MS patients enrolled at a single institution, and cohort II: 25 HC and 63 MS patients enrolled at three European sites. FIELD STRENGTH/SEQUENCE:3D T1-weighted gradient echo sequence, acquired at 1.5 T (cohort I) and 3.0 T (cohort II). ASSESSMENT:Percentage cord area changes (PCACs) between baseline and follow-up (cohort I: 2.34 years [interquartile range = 2.00-2.55 years], cohort II: 1.05 years [interquartile range = 1.01-1.18 years]) were evaluated for all subjects using Reg, AS, and PropSeg. Reg included an accurate registration of baseline and follow-up straightened cord images, followed by AS-based optimized cord segmentation. A subset of studies was analyzed twice by two observers. STATISTICAL TESTS:Linear regression models were used to estimate annualized PCAC, and effect sizes expressed as the ratio between the estimated differences and HC error term (P < 0.05). Reproducibility was assessed by linear mixed-effect models. Annualized PCACs were used for sample size calculations (significance: α = 0.05, power: 1 - β = 0.80). RESULTS:Annualized PCACs and related standard errors (SEs) were lower with Reg than with other methods: PCAC in MS patients at 1.5 T was -1.12% (SE = 0.22) with Reg, -1.32% (SE = 0.30) with AS, and -1.40% (SE = 0.33) with PropSeg, while at 3.0 T PCAC was -0.83% (SE = 0.25) with Reg, -0.92% (SE = 0.32) with AS, and -1.18 (SE = 0.53) with PropSeg. This was reflected in larger effect sizes and lower sample sizes. Intra- and inter-observer agreement range was 0.72-0.91 with AS, and it was >0.96 with Reg. DATA CONCLUSION:The results support the use of the registration method to measure cervical cord atrophy progression in future MS clinical studies. LEVEL OF EVIDENCE:3 TECHNICAL EFFICACY STAGE: 2.