Graphics Processing Units Can Speed Up Medical Image Reconstruction
Optimizing parallel computation reduces the time needed to process ultrasound computed tomography data.
Ultrasound computed tomography is a promising imaging tool for the early detection of breast cancer. But the synthetic aperture focusing technique (SAFT) widely used for image reconstruction is highly compute-intensive. A recent study showed that SAFT reconstruction algorithms can be accelerated with graphics processing units (GPUs). A GPU designed with a massively parallel structure to efficiently manipulate computer graphics is more efficient than a general-purpose central processing unit (CPU) for processing large blocks of data in parallel.
The researchers analyzed the speed of SAFT reconstruction algorithms for USCT images of a breast phantom. The processing time for a 128 x 128 pixel image was 42 seconds for the GPU and 79 seconds for the CPU. When the image size increased to 1024 x 1024, the processing time for the GPU rose to 2 minutes 38 seconds, compared with more than 67 minutes for the CPU. The processing time for the GPU was nearly twice as fast for the smaller image size and almost 26 times faster for the larger image size, suggesting that the acceleration effect is more notable with a greater number of pixels.
According to the authors, GPUs offer a simple, low-cost approach for speeding up USCT image reconstruction. In the future, they will test the method in vivo and work on further reducing the SAFT algorithm runtime.
Their findings were presented at the SPIE Medical Imaging conference held in Houston, Texas, this February.