UltraBoneUDF: Self-supervised bone surface reconstruction from ultrasound based on neural unsigned distance functions

Area of Science:

• Medical Imaging
• Computer-Assisted Surgery
• Biomedical Engineering

Background:

• Ultrasound imaging offers a radiation-free alternative for computer-assisted orthopedic surgery (CAOS), but challenges remain in reconstructing partial bone surfaces from 3D ultrasound volumes.
• Existing reconstruction methods struggle with data variability and limitations inherent in ultrasound, leading to artifacts and inaccuracies.
• Accurate reconstruction of open bone surfaces from 3D ultrasound data is crucial for advancing CAOS applications but remains a significant technical challenge.

Purpose of the Study:

• To develop a novel self-supervised framework, UltraBoneUDF, for accurate open bone surface reconstruction from 3D ultrasound data.
• To introduce a new loss function based on local tangent plane optimization to enhance surface reconstruction quality.
• To evaluate the performance of UltraBoneUDF against existing methods on multiple open-source datasets.

Main Methods:

• Proposed UltraBoneUDF, a self-supervised framework that learns unsigned distance functions (UDFs) from 3D ultrasound data.
• Introduced a novel loss function utilizing local tangent plane optimization to improve surface reconstruction fidelity.
• Benchmarked UltraBoneUDF against competing models on three open-source datasets and conducted ablation studies for validation.

Main Results:

• UltraBoneUDF demonstrated superior or comparable performance to state-of-the-art methods across three datasets, achieving lower bi-directional Chamfer distances.
• Achieved significant improvements, including a ≈25.5% enhancement on the UltraBones100k dataset (1.60 mm), 0.21 mm on OpenBoneCT, and 0.18 mm on ClosedBoneCT.
• The proposed method utilizes fewer parameters while delivering high-quality surface reconstructions, outperforming existing techniques qualitatively and quantitatively.

Conclusions:

• UltraBoneUDF offers a robust and effective solution for reconstructing open bone surfaces from 3D ultrasound volumes.
• The framework shows significant potential to improve preoperative planning and intraoperative guidance in computer-assisted orthopedic surgery.
• This advancement addresses critical limitations in current ultrasound-based CAOS technologies, paving the way for broader clinical adoption.