Fast semi-dense surface reconstruction from stereoscopic video in laparoscopic surgery

Abstract

Liver resection is the main curative option for liver metastases. While this offers a 5-year survival rate of 50%, only about 20% of all patients are suitable for laparoscopic resection and thus being able to take advantage of minimally invasive surgery. One underlying difficulty is the establishment of a safe resection margin while avoiding critical structures. Intra-operative registration of patient scan data may provide a solution. However, this relies on fast and accurate reconstruction methods to obtain the current shape of the liver. Therefore, this paper presents a method for high-resolution stereoscopic surface reconstruction at interactive rates. To this end, a feature-matching propagation method is adapted to multi-resolution processing to enable parallelisation, remove global synchronisation issues and hence become amenable to a GPU-based implementation. Experiments are conducted on a planar target for reconstruction noise estimation and a visually realistic silicone liver phantom. Results highlight an average reconstruction error of 0.6 mm on the planar target, 2.4-5.7 mm on the phantom and processing times averaging around 370 milliseconds for input images of size 1920 x 540. © 2014 Springer International Publishing Switzerland.