PCReg: a coarse-to-fine registration framework using point cloud completion for intraoperative liver deformation...

Area of Science:

• Medical Imaging
• Computer-Aided Surgery
• Geometric Deep Learning

Background:

• Image-guided liver surgery (IGLS) relies on accurate registration of preoperative models with intraoperative data.
• Tissue deformation and limited views in laparoscopy pose significant challenges to current registration methods.
• Precise tumor localization and resection are critical for successful liver surgery outcomes.

Purpose of the Study:

• To introduce a novel coarse-to-fine non-rigid registration framework, PCReg, for improved accuracy in IGLS.
• To address challenges of tissue deformation and partial overlaps in laparoscopic views.
• To enhance the precision of tumor localization and resection in liver surgery.

Main Methods:

• PCReg employs a three-stage approach: intraoperative point cloud completion, coarse registration, and fine registration.
• A point cloud completion network (PCN) is utilized for intraoperative data completion.
• An improved optimal transport (OT)-based method predicts initial displacement fields, followed by refinement.

Main Results:

• PCReg demonstrated state-of-the-art performance on simulated and real-world datasets.
• The framework significantly outperformed existing methods in handling complex deformations and varying overlap ratios.
• Superior registration accuracy was achieved, particularly in low-overlapping scenarios.

Conclusions:

• PCReg offers a novel and effective framework for point cloud registration in IGLS.
• The method provides a promising solution for accurate registration despite challenging intraoperative conditions.
• This advancement can lead to more precise tumor localization and safer liver resections.