Manual and semiautomated measurement of cerebellar subregions on MR images.

Previous structural and functional imaging studies suggest that the corticocerebellar-thalamic-cortical circuit is dysfunctional in schizophrenia. Accurate identification and volumetric measurement of cerebellar subregions are essential to the assessment of the cerebellum's role in healthy and disease states. Manual parcellation of the cerebellum on MR images was performed with the use of guide traces. Guide traces identified relevant fissures and borders in several planes, and their intersections with the primary tracing plane were used to maintain consistency and accuracy during the parcellation. The cerebellum was parcellated into right and left anterior lobes, superior posterior lobes, inferior posterior lobes, and corpus medullare. A systematic review of the final traces ensured their accuracy. An artificial neural network was trained using a novel landmark-warped method to help account for wide variability in structure size and location. Overlaps of the manually traced lobes (intersection/union) ranged from 0.78 to 0.85 and intraclass correlations (r2) ranged from 0.82 to 0.94. In a comparison of the semiautomated method with the manual method overlaps ranged from 0.83 to 0.88 and intraclass correlations ranged from 0.92 to 0.97. For two raters using the semiautomated method overlaps ranged from 0.83 to 0.88 and intraclass correlations ranged from 0.97 to 0.99. The semiautomated method was built on the groundwork of the manual method to produce more reliable results in a fraction of the time, making valid measurements possible on a large number of subjects.