Optimized Atlas-Based Auto-Segmentation of Bony Structures from Whole-Body Computed Tomography

Area of Science:

• Medical Imaging
• Radiology
• Computational Anatomy

Background:

• Accurate segmentation of bony structures in whole-body CT scans is crucial for various clinical applications.
• Manual segmentation is time-consuming and prone to inter-observer variability.
• Automated methods are needed to improve efficiency and consistency.

Purpose of the Study:

• To develop and validate a fully automated workflow for bony structure segmentation from whole-body CT.
• To evaluate the performance of the automated method against manual segmentation.
• To assess the impact of a postprocessing module on segmentation accuracy.

Main Methods:

• An atlas-based segmentation (ABS) workflow was developed using MIM MAESTRO software.
• A postprocessing module (ABS_PP) was integrated to enhance segmentation accuracy.
• The workflow was trained on 52 CT scans and tested on 29 scans, comparing ABS_PP with ABS without postprocessing (ABS_NPP) against manual segmentation.

Main Results:

• The ABS_PP method demonstrated significantly improved segmentation accuracy (Dice Similarity Coefficient [DSC] range, 0.85-0.98) compared to ABS_NPP (DSC range, 0.55-0.87; P < .001).
• High agreement was observed between ABS_PP and manual delineations (mean distance to agreement range, 0.11-1.56 mm), outperforming ABS_NPP.
• Relative volume errors were significantly lower for ABS_PP compared to ABS_NPP for most bony structures.

Conclusions:

• A fully automated MIM workflow for whole-body CT bony structure segmentation was successfully developed.
• The automated workflow achieved high accuracy comparable to manual delineation.
• The integrated postprocessing module significantly enhanced the overall performance of the segmentation workflow.