Developing and testing an algorithm for automatic segmentation of the fetal face from three-dimensional ultrasound...

Area of Science:

• Medical imaging
• Computer-aided diagnosis
• Fetal medicine

Background:

• Prenatal ultrasound (US) faces challenges in detecting fetal craniofacial abnormalities.
• Three-dimensional US volume analysis offers objective quantification of fetal facial features.
• Current atlas-based automated segmentation requires time-intensive manual segmentation (MS).

Purpose of the Study:

• To assess inter- and intra-operator variability of manual segmentation (MS) for fetal faces.
• To test an optimized version of an atlas-based automated segmentation (AS) algorithm.
• To improve algorithmic performance and move towards a fully automated system for fetal facial analysis.

Main Methods:

• Manual refinements of 15 fetal faces by three operators, repeated by one, were assessed using Dice score, surface distance, and volume difference.
• The partially automated algorithm's performance was evaluated with varying atlas sizes using Dice score and computational time.

Main Results:

• Manual segmentation demonstrated low inter- and intra-operator variability, indicating its reliability.
• The automated segmentation algorithm's performance improved with larger atlas sizes.
• Increased atlas size reduced the need for manual refinement in the automated segmentation process.

Conclusions:

• Manual segmentation is a reliable method suitable for optimizing automated fetal face segmentation.
• The developed automated segmentation algorithm shows promise for clinical adoption by reducing manual intervention.
• Further development using larger atlases can lead to a fully automated system for fetal craniofacial abnormality detection.