A Scalable, Annotation-Free Pipeline for Automated CT Pelvimetry: A Validation Study with Landmark Uncertainty...

Area of Science:

• Medical Imaging
• Radiology
• Computational Anatomy

Background:

• Accurate pelvic measurements (pelvimetry) are crucial for clinical decisions.
• Manual CT pelvimetry is time-consuming and subject to inter-observer variability.
• Automated methods require robust validation and clinical assessment.

Purpose of the Study:

• To validate a fully automated, annotation-free CT pelvimetry pipeline using a public segmentation backend.
• To assess the agreement between automated and manual pelvimetry measurements.
• To explore the clinical utility of automated pelvimetry in relation to intraoperative blood loss.

Main Methods:

• Development of an open-source, rule-based CT pelvimetry pipeline utilizing TotalSegmentator.
• Technical validation using intraclass correlation coefficients (ICC), Bland-Altman analysis, and 3D landmark displacement in 60 patients.
• Exploratory clinical evaluation in 106 patients assessing associations between pelvic dimensions and intraoperative blood loss.

Main Results:

• The automated pipeline processed 99.5% of CT datasets successfully.
• Moderate-to-excellent agreement (ICC 0.66-0.97) was observed between automated and manual measurements.
• Coordinate-level analysis identified specific landmarks contributing to variability, offering diagnostic insights.

Conclusions:

• The validated open-source pipeline provides scalable, annotation-free CT pelvimetry.
• The system demonstrates good agreement with manual methods, with potential for broader applications in geometric parameter derivation.
• Automated pelvimetry shows promise for clinical correlation, such as predicting intraoperative blood loss.