Resolving the Ambiguity of Complete-to-Partial Point Cloud Registration for Image-Guided Liver Surgery With...

Area of Science:

• Medical image analysis
• Surgical navigation
• Computer-assisted surgery

Background:

• Image-guided liver surgery relies on accurate alignment of preoperative (CT/MRI) and intraoperative data (point clouds).
• Current semi-automatic alignment methods are prone to errors, necessitating manual correction.
• Correspondence-based point cloud registration offers a fully automatic solution but struggles with limited intraoperative visibility (complete-to-partial ambiguity).

Purpose of the Study:

• To evaluate the performance of state-of-the-art learning-based point cloud registration methods in scenarios with limited intraoperative visibility.
• To propose a novel module to address the complete-to-partial ambiguity in liver surgery registration.
• To improve the accuracy and robustness of point cloud registration for image-guided liver surgery.

Main Methods:

• Evaluation of existing learning-based point cloud registration methods on in silico and in vitro datasets simulating low-visibility conditions.
• Development of a patches-to-partial matching strategy as a plug-and-play module for existing registration frameworks.
• Integration and testing of the proposed module within learning-based registration pipelines.

Main Results:

• State-of-the-art methods showed performance degradation under low-visibility conditions.
• The proposed patches-to-partial matching module significantly reduced registration errors: 6.7 mm (-29%) in silico and 12.5 mm (-40%) in vitro.
• The module outperformed the Lepard method, achieving lower registration errors (9.5 mm in silico, 20.7 mm in vitro).

Conclusions:

• The complete-to-partial ambiguity poses a significant challenge for point cloud registration in image-guided liver surgery.
• The proposed patches-to-partial matching strategy effectively resolves this ambiguity, enhancing registration accuracy in low-visibility scenarios.
• The developed benchmark and module provide a foundation for advancing automated registration in image-guided liver surgery.