Towards individualized cortical thickness assessment for clinical routine

Abstract

Background: Cortical thickness measures the width of gray matter of the human cortex. It can be calculated from T1-weighted magnetic resonance images (MRI). In group studies, this measure has been shown to correlate with the diagnosis/prognosis of a number of neurologic and psychiatric conditions, but has not been widely adapted for clinical routine. One of the reasons for this might be that there is no reference system which allows to rate individual cortical thickness data with respect to a control population. Methods: To address this problem, this study compared different methods to assess statistical significance of cortical thinning, i.e. atrophy. All compared methods were nonparametric and encompassed rating an individual subject's data set with respect to a control data population. Null distributions were calculated using data from the Human Connectome Project (HCP, n = 1000), and an additional HCP data set (n = 113) was used to calculate sensitivity and specificity to compare the different methods, whereas atrophy was simulated for sensitivity assessment. Validation measures were calculated for the entire cortex ("cumulative") and distinct brain regions ("regional") where possible. Results: The approach yielding the highest combination of specificity and sensitivity implemented generating null distributions for anatomically distinct brain regions, based on the most extreme values observed in the population. With that method, while regional variations were observed, cumulative specificity of 98.9% and cumulative sensitivity at 80% was achieved for simulated atrophy of 23%. Conclusions: This study shows that validated rating of individual cortical thickness measures is possible, which can help clinicians in their daily routine to discover signs of atrophy before they become visually apparent on an unprocessed MRI. Furthermore, given different pathologies present with distinct atrophy patterns, the regional validation proposed here allows to detect distinct patterns of atrophy, which can further enhance differential diagnosis/prognosis.