Fast 2-D ultrasound strain imaging: the benefits of using a GPU

Abstract

Deformation of tissue can be accurately estimated from radio-frequency ultrasound data using a 2-dimensional normalized cross correlation (NCC)-based algorithm. This procedure, however, is very computationally time-consuming. A major time reduction can be achieved by parallelizing the numerous computations of NCC. In this paper, two approaches for parallelization have been investigated: the OpenMP interface on a multi-CPU system and Compute Unified Device Architecture (CUDA) on a graphics processing unit (GPU). The performance of the OpenMP and GPU approaches were compared with a conventional Matlab implementation of NCC. The OpenMP approach with 8 threads achieved a maximum speed-up factor of 132 on the computing of NCC, whereas the GPU approach on an Nvidia Tesla K20 achieved a maximum speed-up factor of 376. Neither parallelization approach resulted in a significant loss in image quality of the elastograms. Parallelization of the NCC computations using the GPU, therefore, significantly reduces the computation time and increases the frame rate for motion estimation.

A GPU-Accelerated 3-D Coupled Subsample Estimation Algorithm for Volumetric Breast Strain Elastography

“…1–2%), while early work focused on either 2D SE [17], [19]–[21] or tracking thermal expansion [22], [23]. Particularly, we explored the utility of on-chip memory to further accelerate motion tracking.…”

“…Fortunately, this process can be massively parallelized using GPUs [17], [19]–[21]. Typically, such a parallelization is done using a single instruction, multiple thread (SIMT) mode.…”

“…In order to do so, a graphic processing unit (GPU) was chosen to improve its computational efficiency through massive parallelization. It is worth noting that GPU-based elastography has drawn interests in recent years, through prior work was concentrated on 2D SE applications [17], [19]–[21] or accumulation of fairly small echo strain due to thermal expansion [22], [23]. The work reported here is the first to use a GPU card to accelerate 3D strain imaging formation in breast applications where the tissue volume underwent modest (1–2%) average strain from frame/volume to frame/volume.…”

Ultrasound Open Platforms for Next-Generation Imaging Technique Development

“…Note that GPU processing is not limited to beamforming operations. Various postbeamforming signal processing operations may also be performed using the GPU, such as Doppler imaging [75] and related adaptive clutter filtering operations [76] , motion estimation in elastography [77] , [78] , temperature mapping for therapeutic monitoring [79] , as well as image filtering [80] . It is also possible to integrate different GPU processing modules to realize more advanced algorithms such as high frame-rate vector flow estimation [81] and color-encoded speckle imaging [82] .…”

Utilización de GPU-CUDA en el Procesamiento Digital de Imágenes

“…El registro de imágenes, es una técnica para definir la relación geométrica entre cada punto de dos imágenes, esto es de gran ayuda en las cirugías ayudadas por computadora, pero es indispensable que el tiempo de procesamiento de un valor pequeño, por lo que Kei Ikeda y Fumihiko Ino proponen un método eficiente para acelerar el registro no rígido de información mutua con CUDA (Ikeda, Ino and Hagihara, 2014).Otro tipo de imágenes que es necesario analizar son las visualizaciones tridimensionales de vasos cerebrales, que sirven para diagnosticar enfermedades, sin embargo, cuando se analizan varios vasos cerebrales es difícil decidir el orden de profundidad de manera clara; como solución en (Luo, 2013), (Won et al, 2013), se plantea la combinación de colores a distancia y la mejora de la profundidad estereoscópica se combinan con la reproducción de volúmenes pre integrados, basada en CUDA y su función avanzada de transferencia para una mejor percepción de la profundidad. En (Idzenga et al, 2014) se mencionan las ventajas de utilizar una GPU en la imagenología de ultrasonidos de dos dimensiones, mientras que en (Xanthis et al, 2014), se presenta un simulador (228 veces más rápido que la implementación serial basada en CPU) de imagenología de resonancia magnética (MRI), con ayuda de un ambiente basado en GPU.…”