Test results for comparing frame rate performance of multi-stream GPU elastography (threaded) with single-stream (streamed) and non-stream (normal) GPU elastography.

This table reports average frames per second (with standard deviation in brackets) of images generated by various versions of the elastography program. The term normal-N indicates the basic GPU implementation of NCC elastography, streamed-N indicates the streamed GPU implementation, and threaded-N indicates the multi-streamed GPU implementation, where N indicates the number of RF lines in each RF image. Four test cases were performed at different NCC window sizes, NCC maximum search distances (displacements), and NCC search step sizes (specified as percentage of window overlap). The computational load increases with larger window size, displacement, and percent overlap. As seen in the results, the highest speed obtained is 78 frames per second (fps) running the multi-streamed GPU implementation.Test results for comparing frame rate performance of multi-stream GPU elastography (threaded) with single-stream (streamed) and non-stream (normal) GPU elastography.本表格给出了不同版本弹性成像程序生成图像的平均每秒帧数（fps，括号内标注标准差）。其中，normal-N 指代归一化互相关（Normalized Cross-Correlation，NCC）弹性成像的基础图形处理器（Graphics Processing Unit，GPU）实现，streamed-N 指代流式GPU实现，threaded-N 指代多流式GPU实现，此处的N代表每张射频图像（RF image）中的射频线（RF lines）数量。本次测试共设置四组工况，分别在不同的NCC窗口尺寸、NCC最大搜索距离（位移量）以及NCC搜索步长（以窗口重叠百分比指定）下开展。计算负载随窗口尺寸、位移量及重叠百分比的增大而提升。测试结果显示，采用多流式GPU实现时可获得的最高帧率为78帧每秒（fps）。本测试同时对比了多流式GPU弹性成像（threaded模式）与单流式（streamed模式）、非流式（normal模式）GPU弹性成像的帧率性能。