Medical image processing on the GPU – Past, present and future

Abstract

Graphics processing units (GPUs) are used today in a wide range of applications, mainly because they can dramatically accelerate parallel computing, are affordable and energy efficient. In the field of medical imaging, GPUs are in some cases crucial for enabling practical use of computationally demanding algorithms. This review presents the past and present work on GPU accelerated medical image processing, and is meant to serve as an overview and introduction to existing GPU implementations. The review covers GPU acceleration of basic image processing operations (filtering, interpolation, histogram estimation and distance transforms), the most commonly used algorithms in medical imaging (image registration, image segmentation and image denoising) and algorithms that are specific to individual modalities (CT, PET, SPECT, MRI, fMRI, DTI, ultrasound, optical imaging and microscopy). The review ends by highlighting some future possibilities and challenges.

Empirically investigating the statistical validity of SPM, FSL and AFNI for single subject fMRI analysis

“…methods have become more flexible [15], that multivariate statistical tests have more complicated null distributions than univariate tests [16,17,18] and that more computing power is available [19,20], it may be time for the fMRI community to consider non-parametric statistical methods. A permutation test, for example, does not assume Gaussian data, a constant noise variance or a constant smoothness (and the smoothness does not need to be estimated from the data).…”

Anatomically-adapted graph wavelets for improved group-level fMRI activation mapping

“…Although this computational burden can be reduced by pre-computing and storing the absolute-value-wavelets f; jψ s;l j; jϕ l jg S s¼1 n o N cbl l¼1 for G cbl , this should still be done for each studydependent G cbr . Another possibility would be to perform the reconstructions in parallel on the computer's graphic card, as such applicability has been shown for other fMRI analysis procedures (Eklund et al, 2013).…”

Cache-Aware GPU Optimization for Out-of-Core Cone Beam CT Reconstruction of High-Resolution Volumes

“…With respect to the filtering stage, i.e., steps (1)-(4), the previous method had little advantage over the non-pipelined implementation in that only data transfer stages (2) and (3) were partially overlapped with filtering stage (3). In contrast, our proposed method realized a full overlap, including file I/O stage (1), such that the execution time was reduced from 59.6 s to 38.5 s.…”

“…The Feldkamp-Davis-Kress (FDK) algorithm [1] is the de facto standard for cone beam CT reconstruction and is widely adopted in medical and industrial applications [2]- [4]. Because reconstruction time is critical, especially for real-time medical applications such as imageguided surgery [5], [6], research activities on accelerating the FDK algorithm have been ongoing ever since its advent in 1984.…”

“…In particular, utilizing the compute unified device architecture (CUDA) compatible GPUs [12], to parallelize FDK computation has gained popularity due to its high-performance and low-cost implementation as compared to other devices [2], [13], [14]. In general, using implementations based on CUDA offloads the performance bottleneck of a CPU-based sequential code.…”