Multi-scale RANSAC algorithm for needle localization in 3D ultrasound guided puncture surgery

Correct localization of the needle is of vital importance to guarantee successful puncture. The complexity of real US data increases the difficulties. A multi-scale random sample consensus (MS-RANSAC) algorithm is proposed in this paper to locate the needle in complicated 3D US data. The algorithm uses the radius difference between needle and other tubular human tissues to extract the correct needle location. The performance of classic RANSAC and MS-RANSAC are compared using three different datasets. Results show that MS-RANSAC can locate needle correctly in complicated condition where classic RANSAC cannot. A parallel framework of the algorithm is designed and implemented using CUDA, making it usable in real time and online.