GPU Parallel Computing And Its Application In Photoacoustic Image Reconstruction

As a parallel computing technology, GPU has been widely used in many fields to provide fast computing framework for large data processing, such as mathematics, weather forecasting and data mining. In today’s era of huge information, the effective way of dealing with them can not only save cost but also improve the work efficiency. Photoacoustic imaging is an non-invasive imaging modality, which has the high optical contrast of the pure optical imaging method and the deep tissue imaging ability of ultrasound imaging. As a result, it has become one of the research highlights in the biomedical imaging fields. In fact, high-speed signal processing can accelerate the imaging process of PACT, provide the significance information of patient in time, and help doctor to make a treatment plan early. However, the large amount of data acquisition and processing in the photoacoustic computed tomography(PACT) are the key factors limiting its imaging speed. In recent years, the rapid development of GPU parallel computing give us a chance to solve this problem. The goal of this article is to implement the fast / real-time image display by accelerating the image reconstruction process using GPU parallel computing technology. First, based on high frequency array photoacoustic imaging system, we have reconstructed the vascular image of a rat in vivo. In the process of reconstruction,CUDA have created two kernel functions, which named kernel1 and kernel2. The kernel1 is designed to complete the reconstruction of pixel points, while kernel2 is designed to complete accumulation of pixel values. After the experiment, we tested the imaging speed of single frame vascular image on CPU and GPU. Secondly, based on low frequency array(128-element commercial ultrasonic array) photoacoustic imaging system, we have reconstructed the cross sectional images of hair in phantom. In the process of reconstruction, texture memory is used to store the raw data to further improve the speed of data reading. After the experiment, we tested the imaging speed of single frame hair image on CPU and GPU. Through two experimental results, we can verify the advantages of GPU in speeding up the image rebuilding. In other words, GPU can effectively improve the speed of photoacoustic image reconstruction, and realize its fast/real-time display. At last, based on low frequency array photoacoustic imaging system with a 128-element commercial ultrasonic array, the experiments of inserting the needle into water were performed to image the complete process of pinning. Results of the experiment showed that when images are reconstructed using GPU, the imaging speed is greatly improved. For example, when the image including 1100*640 pixels, the reconstruction process need about 50 ms. Compared with the serial imaging method, the imaging speed was improved about 20 times, and the frame rate of real-time display can be achieved. In short, GPU parallel computing can effectively improve the speed of photacoustic image reconstruction, and push the wide application of photoacoustic imaging technology in monitoring of hemodynamics, clinical disease diagnosis and treatment, as well as the drug research.