Spinal Cord Gray Matter Atrophy in Amyotrophic Lateral Sclerosis

Abstract

Background and purpose: There is an emerging need for biomarkers to better categorize clinical phenotypes and predict progression in amyotrophic lateral sclerosis. This study aimed to quantify cervical spinal gray matter atrophy in amyotrophic lateral sclerosis and investigate its association with clinical disability at baseline and after 1 year.

Materials and methods: Twenty-nine patients with amyotrophic lateral sclerosis and 22 healthy controls were scanned with 3T MR imaging. Standard functional scale was recorded at the time of MR imaging and after 1 year. MR imaging data were processed automatically to measure the spinal cord, gray matter, and white matter cross-sectional areas. A statistical analysis assessed the difference in cross-sectional areas between patients with amyotrophic lateral sclerosis and controls, correlations between spinal cord and gray matter atrophy to clinical disability at baseline and at 1 year, and prediction of clinical disability at 1 year.

Results: Gray matter atrophy was more sensitive to discriminate patients with amyotrophic lateral sclerosis from controls (P = .004) compared with spinal cord atrophy (P = .02). Gray matter and spinal cord cross-sectional areas showed good correlations with clinical scores at baseline (R = 0.56 for gray matter and R = 0.55 for spinal cord; P < .01). Prediction at 1 year with clinical scores (R² = 0.54) was improved when including a combination of gray matter and white matter cross-sectional areas (R² = 0.74).

Conclusions: Although improvements over spinal cord cross-sectional areas were modest, this study suggests the potential use of gray matter cross-sectional areas as an MR imaging structural biomarker to monitor the evolution of amyotrophic lateral sclerosis.

Fig 1.

Processing pipeline for the GM segmentation and computation of the GMCSA.

Fig 2.

GM automatic segmentation and manual delineation patients with ALS. Manual delineation of the GM is displayed with the blue line, automatic probabilistic segmentation is shown in red-to-yellow. Dice coefficient comparing the automatic and manual segmentation is shown on the bottom line.

Fig 3.

GMCSA and SCCSA measured on controls and patients with ALS between the C6–C3 vertebral levels. GMCSA (A) and SCCSA (B) averaged within group and plot against the cervical SC axis. Overall, a stronger intergroup difference can be observed for GMCSA. Asterisk (P ≤ .05) and double asterisk (P ≤ .01) at specific vertebral levels indicate significant differences between patients with ALS and controls according to Student t test P values representing control-to-patient differences in GMCSA and SCCSA for each cervical level between C6 and C3 and across levels.

Fig 4.

Boxplot distribution of GMCSA (A) and SCCSA (B) averaged between the C4–C6 vertebral levels. Each dark point represents an individual value. The median is represented as a thick horizontal line and the interquartile range as a light rectangle. The horizontal bar at both extremities of the whiskers represent the 5th and 95th percentiles. The 2 patients presenting the SOD1 gene are identified in the plot.

Fig 5.

Prediction error on the ALSFRS-R at 1 year, from a leave-one-out cross-validation with regression trees. Results are compared between the regression model including clinical predictors (left distribution plot), clinical predictors + SCCSA (middle distribution plot), and clinical predictors + GMCSA + WM/GMCSA (right distribution plot), where each point represents 1 iteration of the leave-one-out cross-validation. The best value is at 0.