Non-rigid image-volume registration for human livers in laparoscopic surgery

Area of Science:

• Medical Imaging
• Computer-Assisted Surgery
• Surgical Robotics

Background:

• Minimally invasive surgery benefits from fusing 2D laparoscopic images with 3D scans for improved spatial awareness.
• Accurate 2D-3D registration is crucial for augmented reality in surgery, but challenges include organ deformation and camera parameter estimation.
• Real-time and precise alignment of deformable organs remains a significant hurdle in current surgical navigation systems.

Purpose of the Study:

• To develop a non-rigid image-volume registration (NRIVR) framework for deformable organs.
• To enable accurate real-time 2D-3D registration for augmented reality in laparoscopic surgery.
• To overcome challenges in organ deformation, occlusion, and camera parameter estimation from monocular images.

Main Methods:

• A novel long short-term memory-based camera estimation neural network (LCENN) predicts camera poses from 2D anatomical contours.
• A differentiable mapping from 2D boundaries to camera parameters allows for real-time inference.
• Non-rigid registration is performed in 2D space using projected mesh and estimated deformation fields for consistent alignment.

Main Results:

• The LCENN efficiently predicts camera pose from 2D boundaries with minimal rotational (0.35±0.44°) and translational (0.51±0.31 mm) errors.
• The proposed framework achieved effective 2D-3D registration on a clinical dataset.
• An average target registration error of 2.74±1.51 mm was recorded, demonstrating high accuracy.

Conclusions:

• The study validates the feasibility and effectiveness of the proposed method for real-time 2D-3D registration.
• This approach enhances image guidance in laparoscopic surgery workflows.
• The developed framework paves the way for improved augmented reality applications in clinical settings.