点击次数：

论文类型：会议论文

发表时间：2015-01-01

收录刊物：CPCI-S

页面范围：1-4

关键字：CUDA; shortest path algorithm; medical image segmentation

摘要：Dijkstra algorithm can be adopted to distinguish different parts of boundaries in medical image segmentation problem, which can be a reference for further segmentation operation. However, classical Dijkstra algorithm can hardly adapt to real time image segmentation problem owing to its exponentially O(n(2)) computing complexity, especially for the increasing number of nodes. In this paper, we designed and implemented a parallel shortest path algorithm accelerated by GPU for medical image segmentation problem. A dynamic relax approach is presented to optimize classical Dijkstra algorithm. Therefore, the new parallel Dijkstra algorithm can be easily applied to CUDA parallel framework without concerning about GPU hardware and CUDA optimize details. Two experiments have been conducted to evaluate the algorithm performance. The results show that our new Dijkstra algorithm can get 8 speedup for 4096 points compared with the classical Dijkstra algorithm. Besides, an impressing result with two speed up for 128*128 points problems is demonstrated in parallel Dijkstra compared with parallel Moore. In conclusion, the new parallel Dijkstra algorithm can significantly improve the real-time performance of image segmentation.