The impact of version update of an automated brain volumetry software: Possibility of mixed use and repeatability of clinical findings

Abstract Background As a tool of automated brain volumetry, AccuBrain has been widely used to identify brain structural changes in different neurological diseases with the older version (IV1.2). As AccuBrain underwent a substantial revision update (IV2.0) for segmentation algorithms recently, a comparison of the two versions would be necessary to identify and characterize any differences that could affect the interpretation and clinical translation of the older research findings. Method We included 200 Alzheimer’s disease (AD) patients (75.3±7.6 years old, 44.0% women) and 200 normal control (NC) subjects (75.2±5.6 years old, 41.5% women) with 3D T1‐weighted scans processed with AccuBrain IV1.2 and IV2.0. To measure the volumetric difference or agreement of the two versions, we calculated the mean relative absolute error (MRAE, i.e., the mean of |(V2‐V1)/V1| for each volumetric measure) and intraclass correlation coefficient (ICC) between two versions regarding the volumes of typical brain structures, lobar atrophy indices and the quantitative medial temporal lobe atrophy (QMTA) score. In addition, we investigated the difference of the two versions regarding volumetric associations with age using ANCOVA, with the volumetric measures as the dependent variable, the version as fixed variable and age as the covariate. An interaction term between version and age was additionally included in the model to determine whether these two versions were statistically different in this respect. Result The ICC between the volumetric measures of the two versions were >0.85 for absolute volumes of typical brain structures (e.g., hippocampus) and >0.60 for lobar atrophy indices and QMTA (Table 1), indicating moderate to good agreement between versions. Furthermore, the MRAE indicated substantial difference between versions with >5% for half of the measures, and >10% for QMTA and most lobar atrophy indices, where the MRAE was slightly larger in AD than NC. There were no significant differences between versions regarding their volumetric associations with age (Table 2) for either NC or AD group (regression lines shown for NC in Figure 1∼2). Conclusion The two versions of AccuBrain are not suggested for a mixed use. Previous research findings with older version may be repeatable with the new version of AccuBrain.