Development and validation of a quality control procedure for automatic segmentation of hippocampal subfields

Area of Science:

• Neuroimaging
• Medical Image Analysis
• Computational Neuroscience

Background:

• Automatic segmentation in in vivo magnetic resonance imaging (MRI) offers efficiency and reproducibility.
• However, the accuracy of automatic segmentation methods requires rigorous validation, as they can be consistently erroneous.
• Current quality control (QC) practices in neuroimaging research are insufficient for applied studies.

Purpose of the Study:

• To present a detailed, validated quality control (QC) and correction procedure for automatic hippocampal subfield segmentation.
• To establish a reliable method for ensuring the validity of automatic neuroimaging measurements.
• To provide a framework for implementing and testing QC procedures in research.

Main Methods:

• Development of a two-step QC procedure for identifying segmentation errors.
• Creation of a taxonomy of errors and an error severity rating scale for systematic assessment.
• Cross-validation of the QC procedure on independent datasets with varying imaging parameters and by independent raters.

Main Results:

• The QC procedure demonstrated high between-rater reliability for error identification and manual correction.
• Manual correction introduced a maximum of 3% error variance in volume measurements.
• Analysis of error frequency showed no evidence of systematic bias across different datasets and raters.

Conclusions:

• The presented QC procedure is efficient, reliable, and prioritizes measurement validity for automatic atlases.
• This method enhances the trustworthiness of automatic segmentation in neuroimaging research.
• Recommendations for implementation and hypothesis testing strategies are provided for broader adoption.