Assessing the performance of atlas-based prefrontal brain parcellation in an aging cohort

OBJECTIVE: It is unclear whether atlas-based parcellation is suitable in aging cohorts because age-related brain changes confound the performance of automatic methods. We assessed atlas-based parcellation of the prefrontal lobe in an aging population using visual assessment and volumetric and spatial concordance. METHODS: We used an atlas-based approach to parcellate brain MR images of 90 non-demented healthy adults, aged 72.7 +/- 0.7 years, and assessed performance. RESULTS: Volumetric assessment showed that both single-atlas- and multi-atlas-based methods performed acceptably (intraclass correlation coefficient [ICC], 0.74-0.76). Spatial overlap measurements showed that multi-atlas (dice coefficient [DC], 0.84) offered an improvement over the single-atlas (DC, 0.75-0.78) approach. Visual assessment also showed that multi-atlas outperformed single atlas and identified an additional postprocessing step of cerebrospinal fluid removal, enhancing concordance (intraclass correlation coefficient, 0.86; DC, 0.89). CONCLUSIONS: Atlas-based parcellation performed reasonably well in the aging population. Rigorous performance assessment aided method refinement and emphasizes the importance of age matching and postprocessing. Further work is required in more varied subjects.