UC San Diego

Liver cancer is a life threatening disease that must be treated urgently once detected. A partial hepatectomy to remove cancerous lesions has become the mainstay of treatment. A crucial part of the surgery is to identify where the tumors, vessels, and other important landmarks are located.

Surgeons require years of training and practice to perfect the art of image guided surgery. The liver is soft and readily changes shape during a procedure. This means the surgeon must mentally map landmarks between preoperative scans and the surgical view of a liver by modeling the rotation, scaling and distortion of the liver shape in their mind's eye. Only then can they estimate where landmarks are relative to their tools.

Computer-aided Image Guidance (CAIG) fuses preoperative scans with intraoperative images to provide more detailed information about the surgical site. During surgery, CAIG can merge preoperative data directly into the surgeon's view as a visual overlay on top of the intraoperative video feed.

The goal is to display the precise location of vessels, ducts and tumors that are hidden beneath the surface shown by the camera.This type of CAIG allows a surgeon to avoid damaging important vessels while excising a tumor, reducing patient risk. Although accurate real-time CAIG is a valuable and important surgical tool, no system exists that is efficient enough to maintain clinically acceptable accuracy and frame rate.

This work explores the idea that a given CAIG system can be optimized for the clinical requirements of each surgical case. Optimization is done using patient specific preoperative data to tune algorithm and hardware configurations. The thesis statement is that:

“Preoperative and intraoperative image data can be used to instruct clinically significant efficiency optimizations in a Computer Aided Image Guidance pipeline.”

We demonstrate the thesis statement by discussing data driven efficiency improvements we have contributed to the canonical CAIG pipeline. Specifically, our methods improve efficiency of “non-rigid 3D registration” and physical simulation” which are online steps used to track surgical landmarks at video rate in the setting of liver surgery image guidance.

Taken together our contributions collated in the thesis build a suite of new analysis and implementation approaches that have clinically significant impact on the efficiency of CAIG pipelines.