Fast fetal head compounding from multi-view 3D ultrasound

Area of Science:

• Medical Imaging
• Computational Anatomy
• Obstetric Ultrasound

Background:

• Ultrasound imaging artifacts (acoustic shadows, signal dropout) limit diagnostic accuracy.
• Partial anatomical views from varying probe orientations hinder comprehensive assessment.
• Existing methods struggle with fetal and probe motion during scans.

Purpose of the Study:

• To develop a novel method for fusing multiple 2D ultrasound views into a coherent 3D fetal head volume.
• To improve the quality and robustness of fetal head imaging for clinical applications.
• To enable detailed visualization of fetal facial, skull, and brain structures.

Main Methods:

• Acquisition of freehand 3D ultrasound (US) image streams from diverse fetal head views.
• Independent alignment of each US image to a canonical pose using a recurrent spatial transformer network.
• Fusion of aligned images by averaging consistent features to minimize artifacts and enhance detail.

Main Results:

• Achieved state-of-the-art performance in image quality and robustness to misalignment.
• Demonstrated superior detail and artifact reduction compared to traditional methods.
• Validated through quantitative metrics and expert qualitative ratings.

Conclusions:

• The proposed online, fast, and automated method effectively fuses ultrasound data into high-quality 3D fetal head volumes.
• Shows significant promise for real-time clinical deployment in fetal screening.
• Offers unprecedented insight into fetal craniofacial and brain development.