Real-Time Surface Deformation Recovery from Stereo Videos

Area of Science:

• Medical Imaging
• Computer Vision
• Surgical Technology

Background:

• Surgical navigation systems rely on accurate anatomical models.
• Tissue deformation during surgery can significantly reduce navigation accuracy.

Purpose of the Study:

• To develop a real-time approach for estimating surgical tissue surface deformation from stereo videos.
• To improve the accuracy and robustness of surgical navigation systems.

Main Methods:

• Real-time depth extraction from stereo videos using stereo matching.
• Deformation estimation via Iterative Closest Point (ICP), ORB feature matching, and as-rigid-as-possible (ARAP) costs.
• Novel 1-point RANSAC with reweighting for outlier removal and a new ARAP cost function for dense connections.

Main Results:

• The proposed method achieves an update rate of 15 Hz.
• Accuracy of less than 2.5 mm was demonstrated on ex vivo and in vivo data.
• The approach effectively handles smooth surfaces, occlusions, and fast deformations.

Conclusions:

• The developed method provides accurate and real-time estimation of tissue deformation.
• This technique has the potential to significantly enhance surgical navigation accuracy.
• GPU parallel computing enables efficient implementation for clinical use.