fMRI analysis on the GPU-possibilities and challenges

Quality and denoising in real-time functional magnetic resonance imaging neurofeedback: A methods review

Neurofeedback training using real-time functional magnetic resonance imaging (rtfMRI-NF) allows subjects voluntary control of localised and distributed brain activity. It has sparked increased interest as a promising non-invasive treatment option in neuropsychiatric and neurocognitive disorders, although its efficacy and clinical significance are yet to be determined. In this work, we present the first extensive review of acquisition, processing and quality control methods available to improve the quality of the neurofeedback signal. Furthermore, we investigate the state of denoising and quality control practices in 128 recently published rtfMRI-NF studies. We found: (a) that less than a third of the studies reported implementing standard real-time fMRI denoising steps, (b) significant room for improvement with regards to methods reporting and (c) the need for methodological studies quantifying and comparing the contribution of denoising steps to the neurofeedback signal quality. Advances in rtfMRI-NF research depend on reproducibility of methods and results. Notably, a systematic effort is needed to build up evidence that disentangles the various mechanisms influencing neurofeedback effects. To this end, we recommend that future rtfMRI-NF studies: (a) report implementation of a set of standard real-time fMRI denoising steps according to a proposed COBIDAS-style checklist (https://osf.io/kjwhf/), (b) ensure the quality of the neurofeedback signal by calculating and reporting community-informed quality metrics and applying offline control checks and (c) strive to adopt transparent principles in the form of methods and data sharing and support of open-source rtfMRI-NF software. Code and data for reproducibility, as well as an interactive environment to explore the study data, can be accessed at https://github.com/jsheunis/quality-and-denoising-in-rtfmri-nf.

Physiological Gaussian process priors for the hemodynamics in fMRI analysis

BACKGROUND

Inference from fMRI data faces the challenge that the hemodynamic system that relates neural activity to the observed BOLD fMRI signal is unknown.

NEW METHOD

We propose a new Bayesian model for task fMRI data with the following features: (i) joint estimation of brain activity and the underlying hemodynamics, (ii) the hemodynamics is modeled nonparametrically with a Gaussian process (GP) prior guided by physiological information and (iii) the predicted BOLD is not necessarily generated by a linear time-invariant (LTI) system. We place a GP prior directly on the predicted BOLD response, rather than on the hemodynamic response function as in previous literature. This allows us to incorporate physiological information via the GP prior mean in a flexible way, and simultaneously gives us the nonparametric flexibility of the GP.

RESULTS

Results on simulated data show that the proposed model is able to discriminate between active and non-active voxels also when the GP prior deviates from the true hemodynamics. Our model finds time varying dynamics when applied to real fMRI data.

COMPARISON WITH EXISTING METHOD(S)

The proposed model is better at detecting activity in simulated data than standard models, without inflating the false positive rate. When applied to real fMRI data, our GP model in several cases finds brain activity where previously proposed LTI models does not.

CONCLUSIONS

We have proposed a new non-linear model for the hemodynamics in task fMRI, that is able to detect active voxels, and gives the opportunity to ask new kinds of questions related to hemodynamics.

Instant Feedback Rapid Prototyping for GPU-Accelerated Computation, Manipulation, and Visualization of Multidimensional Data

Objective

We have created an open-source application and framework for rapid GPU-accelerated prototyping, targeting image analysis, including volumetric images such as CT or MRI data.

Methods

A visual graph editor enables the design of processing pipelines without programming. Run-time compiled compute shaders enable prototyping of complex operations in a matter of minutes.

Results

GPU-acceleration increases processing the speed by at least an order of magnitude when compared to traditional multithreaded CPU-based implementations, while offering the flexibility of scripted implementations.

Conclusion

Our framework enables real-time, intuition-guided accelerated algorithm and method development, supported by built-in scriptable visualization.

Significance

This is, to our knowledge, the first tool for medical data analysis that provides both high performance and rapid prototyping. As such, it has the potential to act as a force multiplier for further research, enabling handling of high-resolution datasets while providing quasi-instant feedback and visualization of results.

Fast-GPU-PCC: A GPU-Based Technique to Compute Pairwise Pearson's Correlation Coefficients for Time Series Data-fMRI Study

Functional magnetic resonance imaging (fMRI) is a non-invasive brain imaging technique, which has been regularly used for studying brain’s functional activities in the past few years. A very well-used measure for capturing functional associations in brain is Pearson’s correlation coefficient. Pearson’s correlation is widely used for constructing functional network and studying dynamic functional connectivity of the brain. These are useful measures for understanding the effects of brain disorders on connectivities among brain regions. The fMRI scanners produce huge number of voxels and using traditional central processing unit (CPU)-based techniques for computing pairwise correlations is very time consuming especially when large number of subjects are being studied. In this paper, we propose a graphics processing unit (GPU)-based algorithm called Fast-GPU-PCC for computing pairwise Pearson’s correlation coefficient. Based on the symmetric property of Pearson’s correlation, this approach returns N ( N − 1 ) / 2 correlation coefficients located at strictly upper triangle part of the correlation matrix. Storing correlations in a one-dimensional array with the order as proposed in this paper is useful for further usage. Our experiments on real and synthetic fMRI data for different number of voxels and varying length of time series show that the proposed approach outperformed state of the art GPU-based techniques as well as the sequential CPU-based versions. We show that Fast-GPU-PCC runs 62 times faster than CPU-based version and about 2 to 3 times faster than two other state of the art GPU-based methods.

A multi-variate blind source separation algorithm

BACKGROUND AND OBJECTIVE

The study follows the proposal of decomposing a given data matrix into a product of independent spatial and temporal component matrices. A multi-variate decomposition approach is presented, based on an approximate diagonalization of a set of matrices computed using a latent space representation.

METHODS

The proposed methodology follows an algebraic approach, which is common to space, temporal or spatiotemporal blind source separation algorithms. More specifically, the algebraic approach relies on singular value decomposition techniques, which avoids computationally costly and numerically instable matrix inversion. The method is equally applicable to correlation matrices determined from second order correlations or by considering fourth order correlations.

RESULTS

The resulting algorithms are applied to fMRI data sets either to extract the underlying fMRI components or to extract connectivity maps from resting state fMRI data collected for a dynamic functional connectivity analysis. Intriguingly, our algorithm shows increased spatial specificity compared to common approaches, while temporal precision stays similar.

CONCLUSION

The study presents a novel spatiotemporal blind source separation algorithm, which is both robust and avoids parameters that are difficult to fine tune. Applied on experimental data sets, the new method yields highly confined and focused areas with least spatial extent in the retinotopy case, and similar results in the dynamic functional connectivity analyses compared to other blind source separation algorithms. Therefore, we conclude that our novel algorithm is highly competitive and yields results, which are superior or at least similar to existing approaches.

Accelerating permutation testing in voxel-wise analysis through subspace tracking: A new plugin for SnPM

Permutation testing is a non-parametric method for obtaining the max null distribution used to compute corrected p-values that provide strong control of false positives. In neuroimaging, however, the computational burden of running such an algorithm can be significant. We find that by viewing the permutation testing procedure as the construction of a very large permutation testing matrix, T, one can exploit structural properties derived from the data and the test statistics to reduce the runtime under certain conditions. In particular, we see that T is low-rank plus a low-variance residual. This makes T a good candidate for low-rank matrix completion, where only a very small number of entries of T (∼0.35% of all entries in our experiments) have to be computed to obtain a good estimate. Based on this observation, we present RapidPT, an algorithm that efficiently recovers the max null distribution commonly obtained through regular permutation testing in voxel-wise analysis. We present an extensive validation on a synthetic dataset and four varying sized datasets against two baselines: Statistical NonParametric Mapping (SnPM13) and a standard permutation testing implementation (referred as NaivePT). We find that RapidPT achieves its best runtime performance on medium sized datasets (50≤n≤200), with speedups of 1.5× - 38× (vs. SnPM13) and 20x-1000× (vs. NaivePT). For larger datasets (n≥200) RapidPT outperforms NaivePT (6× - 200×) on all datasets, and provides large speedups over SnPM13 when more than 10000 permutations (2× - 15×) are needed. The implementation is a standalone toolbox and also integrated within SnPM13, able to leverage multi-core architectures when available.

Clinically useful brain imaging for neuropsychiatry: How can we get there?

Despite decades of research, visions of transforming neuropsychiatry through the development of brain imaging-based "growth charts" or "lab tests" have remained out of reach. In recent years, there is renewed enthusiasm about the prospect of achieving clinically useful tools capable of aiding the diagnosis and management of neuropsychiatric disorders. The present work explores the basis for this enthusiasm. We assert that there is no single advance that currently has the potential to drive the field of clinical brain imaging forward. Instead, there has been a constellation of advances that, if combined, could lead to the identification of objective brain imaging-based markers of illness. In particular, we focus on advances that are helping to (1) elucidate the research agenda for biological psychiatry (e.g., neuroscience focus, precision medicine), (2) shift research models for clinical brain imaging (e.g., big data exploration, standardization), (3) break down research silos (e.g., open science, calls for reproducibility and transparency), and (4) improve imaging technologies and methods. Although an arduous road remains ahead, these advances are repositioning the brain imaging community for long-term success.

Robust activation detection methods for real-time and offline fMRI analysis

We propose two contributions with novel approaches to fMRI activation analysis. The first is to apply confidence intervals to locate activations in real-time, and second is a new metric based on robust regression of fMRI signals. These contributions are implemented in our four proposed methods; Instantaneous Activation Method (IAM), Instantaneous Activation Method with Past Blocks (IAMP) for real-time analysis, Task Robust Regression Distance Method (TRRD) for the new metric with robust regression and Instantaneous Robust Regression Distance Method (IRRD) for both contributions. For comparison, a statistical offline method called Task Activation Method (TAM) and a correlation analysis method are also implemented. The methods are initially evaluated with synthetic data generated using two different approaches; first using varying hemodynamic response function signals to simulate a wide range of stimuli responses, along with a Gaussian white noise, and second using no activity state data of a real fMRI experiment, which removes the need to generate noise. The methods are also tested with real fMRI experiments and compared with the results obtained by the widely used SPM tool. The results show that instantaneous methods reveal activations that are lost statistically in an offline analysis. They also reveal further improvements by robust fitting application, which minimizes the outlier effect. TRRD has an area under the ROC curve of 0,7127 for very noisy synthetic images, is reaching up to 0,9608 as the noise decreases, while the instantaneous score is in the range of 0,6124 to 0,8019 in the same noise levels.

Fast and Parallel Summed Area Table for Fabric Defect Detection

Automating fabric defect detection has a significant role in fabric industries. However, the existing fabric defect detection algorithms lack the real-time performance that is required in real applications due to their high demanding computation. To ensure real time, high accuracy and reliable fabric defect detection this paper developed a fast and parallel normalized cross-correlation algorithm based on summed-area table technique called PFDD-SAT. To meet real-time requirements, extensive use of the NVIDIA CUDA framework for Graphical Processing Unit (GPU) computing is made. The detailed implementation steps of the PFDD-SAT are illustrated in this paper. Several experiments have been carried out to evaluate the detection time and accuracy and then the robustness to illumination and Gaussian noises. The results show that the PFDD-SAT has robustness to noise and speeds the defect detection process more than 200 times than normal required time and that greatly met the needs for real-time automatic fabric defect detection.

Faster permutation inference in brain imaging

Permutation tests are increasingly being used as a reliable method for inference in neuroimaging analysis. However, they are computationally intensive. For small, non-imaging datasets, recomputing a model thousands of times is seldom a problem, but for large, complex models this can be prohibitively slow, even with the availability of inexpensive computing power. Here we exploit properties of statistics used with the general linear model (GLM) and their distributions to obtain accelerations irrespective of generic software or hardware improvements. We compare the following approaches: (i) performing a small number of permutations; (ii) estimating the p-value as a parameter of a negative binomial distribution; (iii) fitting a generalised Pareto distribution to the tail of the permutation distribution; (iv) computing p-values based on the expected moments of the permutation distribution, approximated from a gamma distribution; (v) direct fitting of a gamma distribution to the empirical permutation distribution; and (vi) permuting a reduced number of voxels, with completion of the remainder using low rank matrix theory. Using synthetic data we assessed the different methods in terms of their error rates, power, agreement with a reference result, and the risk of taking a different decision regarding the rejection of the null hypotheses (known as the resampling risk). We also conducted a re-analysis of a voxel-based morphometry study as a real-data example. All methods yielded exact error rates. Likewise, power was similar across methods. Resampling risk was higher for methods (i), (iii) and (v). For comparable resampling risks, the method in which no permutations are done (iv) was the absolute fastest. All methods produced visually similar maps for the real data, with stronger effects being detected in the family-wise error rate corrected maps by (iii) and (v), and generally similar to the results seen in the reference set. Overall, for uncorrected p-values, method (iv) was found the best as long as symmetric errors can be assumed. In all other settings, including for familywise error corrected p-values, we recommend the tail approximation (iii). The methods considered are freely available in the tool PALM - Permutation Analysis of Linear Models.

Multivariate and repeated measures (MRM): A new toolbox for dependent and multimodal group-level neuroimaging data

Repeated measurements and multimodal data are common in neuroimaging research. Despite this, conventional approaches to group level analysis ignore these repeated measurements in favour of multiple between-subject models using contrasts of interest. This approach has a number of drawbacks as certain designs and comparisons of interest are either not possible or complex to implement. Unfortunately, even when attempting to analyse group level data within a repeated-measures framework, the methods implemented in popular software packages make potentially unrealistic assumptions about the covariance structure across the brain. In this paper, we describe how this issue can be addressed in a simple and efficient manner using the multivariate form of the familiar general linear model (GLM), as implemented in a new MATLAB toolbox. This multivariate framework is discussed, paying particular attention to methods of inference by permutation. Comparisons with existing approaches and software packages for dependent group-level neuroimaging data are made. We also demonstrate how this method is easily adapted for dependency at the group level when multiple modalities of imaging are collected from the same individuals. Follow-up of these multimodal models using linear discriminant functions (LDA) is also discussed, with applications to future studies wishing to integrate multiple scanning techniques into investigating populations of interest.

Big Data Approaches for the Analysis of Large-Scale fMRI Data Using Apache Spark and GPU Processing: A Demonstration on Resting-State fMRI Data from the Human Connectome Project

Technologies for scalable analysis of very large datasets have emerged in the domain of internet computing, but are still rarely used in neuroimaging despite the existence of data and research questions in need of efficient computation tools especially in fMRI. In this work, we present software tools for the application of Apache Spark and Graphics Processing Units (GPUs) to neuroimaging datasets, in particular providing distributed file input for 4D NIfTI fMRI datasets in Scala for use in an Apache Spark environment. Examples for using this Big Data platform in graph analysis of fMRI datasets are shown to illustrate how processing pipelines employing it can be developed. With more tools for the convenient integration of neuroimaging file formats and typical processing steps, big data technologies could find wider endorsement in the community, leading to a range of potentially useful applications especially in view of the current collaborative creation of a wealth of large data repositories including thousands of individual fMRI datasets.

Genetic algorithm supported by graphical processing unit improves the exploration of effective connectivity in functional brain imaging

Brain regions of human subjects exhibit certain levels of associated activation upon specific environmental stimuli. Functional Magnetic Resonance Imaging (fMRI) detects regional signals, based on which we could infer the direct or indirect neuronal connectivity between the regions. Structural Equation Modeling (SEM) is an appropriate mathematical approach for analyzing the effective connectivity using fMRI data. A maximum likelihood (ML) discrepancy function is minimized against some constrained coefficients of a path model. The minimization is an iterative process. The computing time is very long as the number of iterations increases geometrically with the number of path coefficients. Using regular Quad-Core Central Processing Unit (CPU) platform, duration up to 3 months is required for the iterations from 0 to 30 path coefficients. This study demonstrates the application of Graphical Processing Unit (GPU) with the parallel Genetic Algorithm (GA) that replaces the Powell minimization in the standard program code of the analysis software package. It was found in the same example that GA under GPU reduced the duration to 20 h and provided more accurate solution when compared with standard program code under CPU.

GPU-based parallel group ICA for functional magnetic resonance data

The goal of our study is to develop a fast parallel implementation of group independent component analysis (ICA) for functional magnetic resonance imaging (fMRI) data using graphics processing units (GPU). Though ICA has become a standard method to identify brain functional connectivity of the fMRI data, it is computationally intensive, especially has a huge cost for the group data analysis. GPU with higher parallel computation power and lower cost are used for general purpose computing, which could contribute to fMRI data analysis significantly. In this study, a parallel group ICA (PGICA) on GPU, mainly consisting of GPU-based PCA using SVD and Infomax-ICA, is presented. In comparison to the serial group ICA, the proposed method demonstrated both significant speedup with 6-11 times and comparable accuracy of functional networks in our experiments. This proposed method is expected to perform the real-time post-processing for fMRI data analysis.

GPU accelerated dynamic functional connectivity analysis for functional MRI data

Recent advances in multi-core processors and graphics card based computational technologies have paved the way for an improved and dynamic utilization of parallel computing techniques. Numerous applications have been implemented for the acceleration of computationally-intensive problems in various computational science fields including bioinformatics, in which big data problems are prevalent. In neuroimaging, dynamic functional connectivity (DFC) analysis is a computationally demanding method used to investigate dynamic functional interactions among different brain regions or networks identified with functional magnetic resonance imaging (fMRI) data. In this study, we implemented and analyzed a parallel DFC algorithm based on thread-based and block-based approaches. The thread-based approach was designed to parallelize DFC computations and was implemented in both Open Multi-Processing (OpenMP) and Compute Unified Device Architecture (CUDA) programming platforms. Another approach developed in this study to better utilize CUDA architecture is the block-based approach, where parallelization involves smaller parts of fMRI time-courses obtained by sliding-windows. Experimental results showed that the proposed parallel design solutions enabled by the GPUs significantly reduce the computation time for DFC analysis. Multicore implementation using OpenMP on 8-core processor provides up to 7.7× speed-up. GPU implementation using CUDA yielded substantial accelerations ranging from 18.5× to 157× speed-up once thread-based and block-based approaches were combined in the analysis. Proposed parallel programming solutions showed that multi-core processor and CUDA-supported GPU implementations accelerated the DFC analyses significantly. Developed algorithms make the DFC analyses more practical for multi-subject studies with more dynamic analyses.

Connectomics and new approaches for analyzing human brain functional connectivity

Estimating the functional interactions between brain regions and mapping those connections to corresponding inter-individual differences in cognitive, behavioral and psychiatric domains are central pursuits for understanding the human connectome. The number and complexity of functional interactions within the connectome and the large amounts of data required to study them position functional connectivity research as a “big data” problem. Maximizing the degree to which knowledge about human brain function can be extracted from the connectome will require developing a new generation of neuroimaging analysis algorithms and tools. This review describes several outstanding problems in brain functional connectomics with the goal of engaging researchers from a broad spectrum of data sciences to help solve these problems. Additionally it provides information about open science resources consisting of raw and preprocessed data to help interested researchers get started.

Cerebellar cortex and cerebellar nuclei are concomitantly activated during eyeblink conditioning: a 7T fMRI study in humans

There are controversies whether learning of conditioned eyeblink responses primarily takes place within the cerebellar cortex, the interposed nuclei, or both. It has also been suggested that the cerebellar cortex may be important during early stages of learning, and that there is a shift to the cerebellar nuclei during later stages. As yet, human studies have provided little to resolve this question. In the present study, we established a setup that allows ultra-high-field 7T functional magnetic resonance imaging (fMRI) of the cerebellar cortex and interposed cerebellar nuclei simultaneously during delay eyeblink conditioning in humans. Event-related fMRI signals increased concomitantly in the cerebellar cortex and nuclei during early acquisition of conditioned eyeblink responses in 20 healthy human subjects. ANOVAs with repeated-measures showed significant effects of time across five blocks of 20 conditioning trials in the cortex and nuclei (p < 0.05, permutation corrected). Activations were most pronounced in, but not limited to, lobules VI and interposed nuclei. Increased activations were most prominent at the first time the maximum number of conditioned responses was achieved. Our data are consistent with a simultaneous and synergistic two-site model of learning during acquisition of classically conditioned eyeblinks. Because increased MRI signal reflects synaptic activity, concomitantly increased signals in the cerebellar nuclei and cortex are consistent with findings of learning related potentiation at the mossy fiber to nuclear cell synapse and mossy fiber to granule cell synapse. Activity related to the expression of conditioned responses, however, cannot be excluded.

Augment low-field intra-operative MRI with preoperative MRI using a hybrid non-rigid registration method

BACKGROUND

Preoperatively acquired diffusion tensor image (DTI) and blood oxygen level dependent (BOLD) have been proved to be effective in providing more anatomical and functional information; however, the brain deformation induced by brain shift and tumor resection severely impairs the correspondence between the image space and the patient space in image-guided neurosurgery.

METHOD

To address the brain deformation, we developed a hybrid non-rigid registration method to register high-field preoperative MRI with low-field intra-operative MRI in order to recover the deformation induced by brain shift and tumor resection. The registered DTI and BOLD are fused with low-field intra-operative MRI for image-guided neurosurgery.

RESULTS

The proposed hybrid registration method was evaluated by comparing the landmarks predicted by the hybrid registration method with the landmarks identified in the low-field intra-operative MRI for 10 patients. The prediction error of the hybrid method is 1.92±0.54 mm, and the compensation accuracy is 74.3±5.0%. Compared to the landmarks far from the resection region, those near the resection region demonstrated a higher compensation accuracy (P-value=.003) although these landmarks had larger initial displacements.

CONCLUSIONS

The proposed hybrid registration method is able to bring preoperatively acquired BOLD and DTI into the operating room and compensate for the deformation to augment low-field intra-operative MRI with rich anatomical and functional information.

Harnessing graphics processing units for improved neuroimaging statistics

Simple models and algorithms based on restrictive assumptions are often used in the field of neuroimaging for studies involving functional magnetic resonance imaging, voxel based morphometry, and diffusion tensor imaging. Nonparametric statistical methods or flexible Bayesian models can be applied rather easily to yield more trustworthy results. The spatial normalization step required for multisubject studies can also be improved by taking advantage of more robust algorithms for image registration. A common drawback of algorithms based on weaker assumptions, however, is the increase in computational complexity. In this short overview, we will therefore present some examples of how inexpensive PC graphics hardware, normally used for demanding computer games, can be used to enable practical use of more realistic models and accurate algorithms, such that the outcome of neuroimaging studies really can be trusted.

Fast ray-tracing of human eye optics on Graphics Processing Units

We present a new technique for simulating retinal image formation by tracing a large number of rays from objects in three dimensions as they pass through the optic apparatus of the eye to objects. Simulating human optics is useful for understanding basic questions of vision science and for studying vision defects and their corrections. Because of the complexity of computing such simulations accurately, most previous efforts used simplified analytical models of the normal eye. This makes them less effective in modeling vision disorders associated with abnormal shapes of the ocular structures which are hard to be precisely represented by analytical surfaces. We have developed a computer simulator that can simulate ocular structures of arbitrary shapes, for instance represented by polygon meshes. Topographic and geometric measurements of the cornea, lens, and retina from keratometer or medical imaging data can be integrated for individualized examination. We utilize parallel processing using modern Graphics Processing Units (GPUs) to efficiently compute retinal images by tracing millions of rays. A stable retinal image can be generated within minutes. We simulated depth-of-field, accommodation, chromatic aberrations, as well as astigmatism and correction. We also show application of the technique in patient specific vision correction by incorporating geometric models of the orbit reconstructed from clinical medical images.

BROCCOLI: Software for fast fMRI analysis on many-core CPUs and GPUs

Analysis of functional magnetic resonance imaging (fMRI) data is becoming ever more computationally demanding as temporal and spatial resolutions improve, and large, publicly available data sets proliferate. Moreover, methodological improvements in the neuroimaging pipeline, such as non-linear spatial normalization, non-parametric permutation tests and Bayesian Markov Chain Monte Carlo approaches, can dramatically increase the computational burden. Despite these challenges, there do not yet exist any fMRI software packages which leverage inexpensive and powerful graphics processing units (GPUs) to perform these analyses. Here, we therefore present BROCCOLI, a free software package written in OpenCL (Open Computing Language) that can be used for parallel analysis of fMRI data on a large variety of hardware configurations. BROCCOLI has, for example, been tested with an Intel CPU, an Nvidia GPU, and an AMD GPU. These tests show that parallel processing of fMRI data can lead to significantly faster analysis pipelines. This speedup can be achieved on relatively standard hardware, but further, dramatic speed improvements require only a modest investment in GPU hardware. BROCCOLI (running on a GPU) can perform non-linear spatial normalization to a 1 mm³ brain template in 4–6 s, and run a second level permutation test with 10,000 permutations in about a minute. These non-parametric tests are generally more robust than their parametric counterparts, and can also enable more sophisticated analyses by estimating complicated null distributions. Additionally, BROCCOLI includes support for Bayesian first-level fMRI analysis using a Gibbs sampler. The new software is freely available under GNU GPL3 and can be downloaded from github (https://github.com/wanderine/BROCCOLI/).

Medical image processing on the GPU - past, present and future

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.

Magnetic resonance imaging and graph theoretical analysis of complex brain networks in neuropsychiatric disorders

Neurological and psychiatric disorders disturb higher cognitive functions and are accompanied by aberrant cortico-cortical axonal pathways or synchronizations of neural activity. A large proportion of neuroimaging studies have focused on examining the focal morphological abnormalities of various gray and white matter structures or the functional activities of brain areas during goal-directed tasks or the resting state, which provides vast quantities of information on both the structural and functional alterations in the patients' brain. However, these studies often ignore the interactions among multiple brain regions that constitute complex brain networks underlying higher cognitive function. Information derived from recent advances of noninvasive magnetic resonance imaging (MRI) techniques and computational methodologies such as graph theory have allowed researchers to explore the patterns of structural and functional connectivity of healthy and diseased brains in vivo. In this article, we summarize the recent advances made in the studies of both structural (gray matter morphology and white matter fibers) and functional (synchronized neural activity) brain networks based on human MRI data pertaining to neuropsychiatric disorders. These studies bring a systems-level perspective to the alterations of the topological organization of complex brain networks and the underlying pathophysiological mechanisms. Specifically, noninvasive imaging of structural and functional brain networks and follow-up graph-theoretical analyses demonstrate the potential to establish systems-level biomarkers for clinical diagnosis, progression monitoring, and treatment effects evaluation for neuropsychiatric disorders.

Fast box-counting algorithm on GPU

The box-counting algorithm is one of the most widely used methods for calculating the fractal dimension (FD). The FD has many image analysis applications in the biomedical field, where it has been used extensively to characterize a wide range of medical signals. However, computing the FD for large images, especially in 3D, is a time consuming process. In this paper we present a fast parallel version of the box-counting algorithm, which has been coded in CUDA for execution on the Graphic Processing Unit (GPU). The optimized GPU implementation achieved an average speedup of 28 times (28×) compared to a mono-threaded CPU implementation, and an average speedup of 7 times (7×) compared to a multi-threaded CPU implementation. The performance of our improved box-counting algorithm has been tested with 3D models with different complexity, features and sizes. The validity and accuracy of the algorithm has been confirmed using models with well-known FD values. As a case study, a 3D FD analysis of several brain tissues has been performed using our GPU box-counting algorithm.

Fast random permutation tests enable objective evaluation of methods for single-subject FMRI analysis

Parametric statistical methods, such as Z-, t-, and F-values, are traditionally employed in functional magnetic resonance imaging (fMRI) for identifying areas in the brain that are active with a certain degree of statistical significance. These parametric methods, however, have two major drawbacks. First, it is assumed that the observed data are Gaussian distributed and independent; assumptions that generally are not valid for fMRI data. Second, the statistical test distribution can be derived theoretically only for very simple linear detection statistics. With nonparametric statistical methods, the two limitations described above can be overcome. The major drawback of non-parametric methods is the computational burden with processing times ranging from hours to days, which so far have made them impractical for routine use in single-subject fMRI analysis. In this work, it is shown how the computational power of cost-efficient graphics processing units (GPUs) can be used to speed up random permutation tests. A test with 10000 permutations takes less than a minute, making statistical analysis of advanced detection methods in fMRI practically feasible. To exemplify the permutation-based approach, brain activity maps generated by the general linear model (GLM) and canonical correlation analysis (CCA) are compared at the same significance level.

True 4D Image Denoising on the GPU

The use of image denoising techniques is an important part of many medical imaging applications. One common application is to improve the image quality of low-dose (noisy) computed tomography (CT) data. While 3D image denoising previously has been applied to several volumes independently, there has not been much work done on true 4D image denoising, where the algorithm considers several volumes at the same time. The problem with 4D image denoising, compared to 2D and 3D denoising, is that the computational complexity increases exponentially. In this paper we describe a novel algorithm for true 4D image denoising, based on local adaptive filtering, and how to implement it on the graphics processing unit (GPU). The algorithm was applied to a 4D CT heart dataset of the resolution 512  × 512  × 445  × 20. The result is that the GPU can complete the denoising in about 25 minutes if spatial filtering is used and in about 8 minutes if FFT-based filtering is used. The CPU implementation requires several days of processing time for spatial filtering and about 50 minutes for FFT-based filtering. The short processing time increases the clinical value of true 4D image denoising significantly.