2022 Undergraduate Research Symposium

Samaah Saifullah (psychology)

Graduate co-author: Kelsey L. Canada

Faculty mentor: Ana Daugherty

Abstract

Automated hippocampal subfield segmentation methods have led to exponential growth in the literature. Although automated segmentation is perfectly reliable, deviations from anatomical definitions are common and weaken their validity.

To address the lack of guidance for quality control (QC) of automated segmentation methods, we developed a taxonomy of errors and two-step protocol for 1) identification and 2) manual correction. Raters examined segmentations from a customized atlas to identify errors in six categories with severity ratings; only major severity errors are manually corrected in order to minimize introducing human error.

The QC procedure was developed and cross-validated with an independent sample on two MRI scanners. Raters had excellent agreement in QC decisions between- and within-raters on all regions, manual corrections introduced less than 3% error and were unrelated to participant demographic.

The feasible and reliable QC procedure allows confident interpretation of hippocampal subfield segmentations, and potential applications to other brain regions.

Poster pitch

Poster

Rate this presentation

↠Return to symposium