3D Reconstruction of Brain Tumors from Endoscopic and Ultrasound Images

Abstract

1.1 Statement of the Problem Endo-neuro-sonography (ENS) has gained special attention in recent years. Its major advantages over traditional brain surgery are that it is a minimally invasive surgical technique and the endoscopic camera and ultrasound images provide useful information. Ultrasound images are inexpensive compared to tomographic and resonance magnetic images (which are very hard to obtain in an intraoperative setting) and allow surgeons to see beyond the tissues within the brain. Another way would be to extract three-dimensional (3D) information from the combined endoscopic and ultrasound images to help surgeons better locate brain structures (such as tumors). Some work has been done in this direction, mainly in the replacement of classic ultrasound (2D imaging methodology) by 3D ultrasound equipment (Unsgaard et al., 2006). We have focused our attention on using classic ultrasound techniques and endoscopic images to extract 3D information. We propose tracking the ultrasound probe in the endoscopic images and then computing the ultrasound probe’s pose in 3D space without an external method (optical or magnetic). We tested two alternative methods to track the ultrasound probe in endoscopic camera images as well as two methods to segment brain structures in ultrasound images, and then we compared the latter two types. We used conformal geometric algebra for the necessary geometric calculations and to put the results in 3D space.