Assessing Consistency between Manual and Automated Human Hippocampal Segmentations from MRI Images
Mentor 1
Caitlin Bowman
Start Date
28-4-2023 12:00 AM
Description
Ensuring that automated data analysis tools work efficiently and precisely is paramount to furthering scientific advancements, as well as reducing error. Many brain imaging researchers use the program Freesurfer to automatically detect anatomical boundaries from MRI images in individual subjects. The objective of this project is to assess the validity of Freesurfer’s automated segmentation of the head, body and tail of human hippocampus, and to assess the reliability of the same segmentation by human raters. Members of our lab manually noted the anterior and posterior boundaries of the hippocampal head, body and tail from MRI images in 650 participants ages 18-85. We then compared these boundaries to the values that were produced by automated segmentation with Freesurfer. We plan to run statistical tests to assess how closely the human raters’ hippocampi boundaries resemble those generated by the program. In addition, due to the large participant age range, age-related differences will also be assessed. Based on a prior analysis in a small training sample of participants, we expect that Freesurfer will provide valid boundaries, and that human ratings may not always be reliable. We also expect that human raters may diverge from one another more when identifying boundaries in older brains that have more variable underlying anatomy.
Assessing Consistency between Manual and Automated Human Hippocampal Segmentations from MRI Images
Ensuring that automated data analysis tools work efficiently and precisely is paramount to furthering scientific advancements, as well as reducing error. Many brain imaging researchers use the program Freesurfer to automatically detect anatomical boundaries from MRI images in individual subjects. The objective of this project is to assess the validity of Freesurfer’s automated segmentation of the head, body and tail of human hippocampus, and to assess the reliability of the same segmentation by human raters. Members of our lab manually noted the anterior and posterior boundaries of the hippocampal head, body and tail from MRI images in 650 participants ages 18-85. We then compared these boundaries to the values that were produced by automated segmentation with Freesurfer. We plan to run statistical tests to assess how closely the human raters’ hippocampi boundaries resemble those generated by the program. In addition, due to the large participant age range, age-related differences will also be assessed. Based on a prior analysis in a small training sample of participants, we expect that Freesurfer will provide valid boundaries, and that human ratings may not always be reliable. We also expect that human raters may diverge from one another more when identifying boundaries in older brains that have more variable underlying anatomy.
