Poster B93, Sunday, March 26, 8:00 – 10:00 am, Pacific Concourse
Developmental Difference in Hippocampal Segmentation using FreeSurfer Compared with Manual Demarcation
Qijing Yu1, Roya Homayouni1, Andrea Shafer1, Naftali Raz1, Noa Ofen1; 1Wayne State University
There is growing interest in measuring the structural properties of the human hippocampus. In many investigations, researchers rely on automatic segmentation to calculate hippocampal volume. The validity of automatic hippocampal segmentation methods has recently been challenged, particularly when compared against manual demarcation in studies in older adults and young children. We assessed the agreement between hippocampal volumes estimated by manual morphometry and FreeSurfer—one of the most popular automatic segmentation software packages. For that purpose, we used intra-class correlation (ICC) in a sample of 5-25-year old participants (N=102). We found a modest agreement between manually measured volumes and the FreeSurfer output: ICC(3) left: 0.58-0.82, right: 0.44-0.84, total: 0.64-0.84. Consistent with prior findings, FreeSurfer overestimation was marginally more severe in younger compared to older participants: F(1,99)=3.84, p=0.053. In additional analyses, we truncated the range of the FreeSurfer segmentation to precisely match the range defined in the manual demarcation protocol, which is limited because of the poor visualization of the most anterior and posterior hippocampal boundaries. In this comparison, the ICC estimates were higher (ICC(3) left: 0.80-0.90, right: 0.78-0.91, total: 0.84-0.92), and FreeSurfer overestimation was no longer significant. However, the difference between hippocampal volumes estimated by the two methods became significantly age-dependent: F(1,99)=5.33, p=0.02. Thus, our findings provide additional evidence of the poor agreement between hippocampal volumes obtained by manual demarcation and FreeSurfer segmentation. Moreover, our results confirm the overestimation and age-related bias in FreeSurfer segmentation and call for caution in the interpretation of findings based on hippocampal FreeSurfer segmentation in developmental research.
Topic Area: LONG-TERM MEMORY: Development & aging