Tissue Deformation Recovery with Gaussian Mixture Model Based Structure from Motion

Accurate 3D reconstruction of the surgical scene is important in intra-operative guidance. Existing methods are often based on the assumption that the camera is static or the tissue is deforming with periodic motion. In minimally invasive surgery, these assumptions do not always hold due to free-form tissue deformation induced by instrumenttissue interaction and camera motion required for continuous exploration of the surgical scene, particularly for intraluminal procedures. The aim of this work is to propose a novel framework for intra-operative freeform deformation recovery. The proposed method builds on a compact scene representation scheme that is suitable for both surgical episode identification and instrument-tissue motion modeling. Unlike previous approaches, it does not impose explicit models on tissue deformation, allowing realistic free-form deformation recovery. Validation is provided on both synthetic and phantom data. The practical value of the method is further demonstrated by deformation recovery on in vivo data recorded from a robotic assisted minimally invasive surgical procedure.