The impact of temporal compression and space selection on SVM analysis of single-subject and multi-subject fMRI data

Decoding Single and Paired Phonemes Using 7T Functional MRI

Several studies have shown that mouth movements related to the pronunciation of individual phonemes are represented in the sensorimotor cortex. This would theoretically allow for brain computer interfaces that are capable of decoding continuous speech by training classifiers based on the activity in the sensorimotor cortex related to the production of individual phonemes. To address this, we investigated the decodability of trials with individual and paired phonemes (pronounced consecutively with one second interval) using activity in the sensorimotor cortex. Fifteen participants pronounced 3 different phonemes and 3 combinations of two of the same phonemes in a 7T functional MRI experiment. We confirmed that support vector machine (SVM) classification of single and paired phonemes was possible. Importantly, by combining classifiers trained on single phonemes, we were able to classify paired phonemes with an accuracy of 53% (33% chance level), demonstrating that activity of isolated phonemes is present and distinguishable in combined phonemes. A SVM searchlight analysis showed that the phoneme representations are widely distributed in the ventral sensorimotor cortex. These findings provide insights about the neural representations of single and paired phonemes. Furthermore, it supports the notion that speech BCI may be feasible based on machine learning algorithms trained on individual phonemes using intracranial electrode grids.

Have you been there before? Decoding recognition of spatial scenes from fMRI signals in precuneus

One potential application of forensic "brain reading" is to test whether a suspect has previously experienced a crime scene. Here, we investigated whether it is possible to decode real life autobiographic exposure to spatial locations using fMRI. In the first session, participants visited four out of eight possible rooms on a university campus. During a subsequent scanning session, subjects passively viewed pictures and videos from these eight possible rooms (four old, four novel) without giving any responses. A multivariate searchlight analysis was employed that trained a classifier to distinguish between "seen" versus "unseen" stimuli from a subset of six rooms. We found that bilateral precuneus encoded information that can be used to distinguish between previously seen and unseen rooms and that also generalized to the two stimuli left out from training. We conclude that activity in bilateral precuneus is associated with the memory of previously visited rooms, irrespective of the identity of the room, thus supporting a parietal contribution to episodic memory for spatial locations. Importantly, we could decode whether a room was visited in real life without the need of explicit judgments about the rooms. This suggests that recognition is an automatic response that can be decoded from fMRI data, thus potentially supporting forensic applications of concealed information tests for crime scene recognition.

Functional brain network identification and fMRI augmentation using a VAE-GAN framework

Recently, deep learning models have achieved superior performance for mapping functional brain networks from functional magnetic resonance imaging (fMRI) data compared with traditional methods. However, due to the lack of sufficient data and the high dimensionality of brain volume, deep learning models of fMRI tend to suffer from overfitting. In addition, existing methods rarely studied fMRI data augmentation and its application. To address these issues, we developed a VAE-GAN framework that combined a VAE (variational auto-encoder) with a GAN (generative adversarial net) for functional brain network identification and fMRI augmentation. As a generative model, the VAE-GAN models the distribution of fMRI so that it enables the extraction of more generalized features, and thus relieve the overfitting issue. The VAE-GAN is easier to train on fMRI than a standard GAN since it uses latent variables from VAE to generate fake data rather than relying on random noise that is used in a GAN, and it can generate higher quality of fake data than VAE since the discriminator can promote the training of the generator. In other words, the VAE-GAN inherits the advantages of VAE and GAN and avoids their limitations in modeling of fMRI data. Extensive experiments on task fMRI datasets from HCP have proved the effectiveness and superiority of the proposed VAE-GAN framework for identifying both temporal features and functional brain networks compared with existing models, and the quality of fake data is higher than those from VAE and GAN. The results on resting state fMRI of Attention Deficit Hyperactivity Disorder (ADHD)-200 dataset further demonstrated that the fake data generated by the VAE-GAN can help improve the performance of brain network modeling and ADHD classification.

A deep learning method for autism spectrum disorder identification based on interactions of hierarchical brain networks

BACKGROUND

It has been recently shown that deep learning models exhibited remarkable performance of representing functional Magnetic Resonance Imaging (fMRI) data for the understanding of brain functional activities. With hierarchical structure, deep learning models can infer hierarchical functional brain networks (FBN) from fMRI. However, the applications of the hierarchical FBNs have been rarely studied.

METHODS

In this work, we proposed a hierarchical recurrent variational auto-encoder (HRVAE) to unsupervisedly model the fMRI data. The trained HRVAE encoder can predict hierarchical temporal features from its three hidden layers, and thus can be regarded as a hierarchical feature extractor. Then LASSO (least absolute shrinkage and selection operator) regression was applied to estimate the corresponding hierarchical FBNs. Based on the hierarchical FBNs from each subject, we constructed a novel classification framework for brain disorder identification and test it on the Autism Brain Imaging Data Exchange (ABIDE) dataset, a world-wide multi-site database of autism spectrum disorder (ASD). We analyzed the hierarchy organization of FBNs, and finally used the overlaps of hierarchical FBNs as features to differentiate ASD from typically developing controls (TDC).

RESULTS

The experimental results on 871 subjects from ABIDE dataset showed that the HRVAE model can effectively derive hierarchical FBNs including many well-known resting state networks (RSN). Moreover, the classification result improved the state-of-the-art by achieving a very high accuracy of 82.1 %.

CONCLUSIONS

This work presents a novel data-driven deep learning method using fMRI data for ASD identification, which could provide valuable reference for clinical diagnosis. The classification results suggest that the interactions of hierarchical FBNs have association with brain disorder, which promotes the understanding of FBN hierarchy and could be applied to other brain disorder analysis.

Multi-head attention-based masked sequence model for mapping functional brain networks

The investigation of functional brain networks (FBNs) using task-based functional magnetic resonance imaging (tfMRI) has gained significant attention in the field of neuroimaging. Despite the availability of several methods for constructing FBNs, including traditional methods like GLM and deep learning methods such as spatiotemporal self-attention mechanism (STAAE), these methods have design and training limitations. Specifically, they do not consider the intrinsic characteristics of fMRI data, such as the possibility that the same signal value at different time points could represent different brain states and meanings. Furthermore, they overlook prior knowledge, such as task designs, during training. This study aims to overcome these limitations and develop a more efficient model by drawing inspiration from techniques in the field of natural language processing (NLP). The proposed model, called the Multi-head Attention-based Masked Sequence Model (MAMSM), uses a multi-headed attention mechanism and mask training approach to learn different states corresponding to the same voxel values. Additionally, it combines cosine similarity and task design curves to construct a novel loss function. The MAMSM was applied to seven task state datasets from the Human Connectome Project (HCP) tfMRI dataset. Experimental results showed that the features acquired by the MAMSM model exhibit a Pearson correlation coefficient with the task design curves above 0.95 on average. Moreover, the model can extract more meaningful networks beyond the known task-related brain networks. The experimental results demonstrated that MAMSM has great potential in advancing the understanding of functional brain networks.

Optimizing multivariate pattern classification in rapid event-related designs

BACKGROUND

Multivariate pattern analysis (MVPA or pattern decoding) has attracted considerable attention as a sensitive analytic tool for investigations using functional magnetic resonance imaging (fMRI) data. With the introduction of MVPA, however, has come a proliferation of methodological choices confronting the researcher, with few studies to date offering guidance from the vantage point of controlled datasets detached from specific experimental hypotheses.

NEW METHOD

We investigated the impact of four data processing steps on support vector machine (SVM) classification performance aimed at maximizing information capture in the presence of common noise sources. The four techniques included: trial averaging (classifying on separate trial estimates versus condition-based averages), within-run mean centering (centering the data or not), method of cost selection (using a fixed or tuned cost value), and motion-related denoising approach (comparing no denoising versus a variety of nuisance regressions capturing motion-related reference signals). The impact of these approaches was evaluated on real fMRI data from two control ROIs, as well as on simulated pattern data constructed with carefully controlled voxel- and trial-level noise components.

RESULTS

We find significant improvements in classification performance across both real and simulated datasets with run-wise trial averaging and mean centering. When averaging trials within conditions of each run, we note a simultaneous increase in the between-subject variability of SVM classification accuracies which we attribute to the reduced size of the test set used to assess the classifier's prediction error. Therefore, we propose a hybrid technique whereby randomly sampled subsets of trials are averaged per run and demonstrate that it helps mitigate the tradeoff between improving signal-to-noise ratio by averaging and losing exemplars in the test set.

COMPARISON WITH EXISTING METHODS

Though a handful of empirical studies have employed run-based trial averaging, mean centering, or their combination, such studies have done so without theoretical justification or rigorous testing using control ROIs.

CONCLUSIONS

Therefore, we intend this study to serve as a practical guide for researchers wishing to optimize pattern decoding without risk of introducing spurious results.

Learning brain representation using recurrent Wasserstein generative adversarial net

BACKGROUND AND OBJECTIVE

To understand brain cognition and disorders, modeling the mapping between mind and brain has been of great interest to the neuroscience community. The key is the brain representation, including functional brain networks (FBN) and their corresponding temporal features. Recently, it has been proven that deep learning models have superb representation power on functional magnetic resonance imaging (fMRI) over traditional machine learning methods. However, due to the lack of high-quality data and labels, deep learning models tend to suffer from overfitting in the training process.

METHODS

In this work, we applied a recurrent Wasserstein generative adversarial net (RWGAN) to learn brain representation from volumetric fMRI data. Generative adversarial net (GAN) is widely used in natural image generation and is able to capture the distribution of the input data, which enables the extraction of generalized features from fMRI and thus relieves the overfitting issue. The recurrent layers in RWGAN are designed to better model the local temporal features of the fMRI time series. The discriminator of RWGAN works as a deep feature extractor. With LASSO regression, the RWGAN model can decompose the fMRI data into temporal features and spatial features (FBNs). Furthermore, the generator of RWGAN can generate high-quality new data for fMRI augmentation.

RESULTS

The experimental results on seven tasks from the HCP dataset showed that the RWGAN can learn meaningful and interpretable temporal features and FBNs, compared to HCP task designs and general linear model (GLM) derived networks. Besides, the results on different training datasets showed that the RWGAN performed better on small datasets than other deep learning models. Moreover, we used the generator of RWGAN to yield fake subjects. The result showed that the fake data can also be used to learn meaningful representation compared to those learned from real data.

CONCLUSIONS

To our best knowledge, this work is among the earliest attempts of applying generative deep learning for modeling fMRI data. The proposed RWGAN offers a novel methodology for learning brain representation from fMRI, and it can generate high-quality fake data for the potential use of fMRI data augmentation.

Dynamic targeting enables domain-general inhibitory control over action and thought by the prefrontal cortex

Over the last two decades, inhibitory control has featured prominently in accounts of how humans and other organisms regulate their behaviour and thought. Previous work on how the brain stops actions and thoughts, however, has emphasised distinct prefrontal regions supporting these functions, suggesting domain-specific mechanisms. Here we show that stopping actions and thoughts recruits common regions in the right dorsolateral and ventrolateral prefrontal cortex to suppress diverse content, via dynamic targeting. Within each region, classifiers trained to distinguish action-stopping from action-execution also identify when people are suppressing their thoughts (and vice versa). Effective connectivity analysis reveals that both prefrontal regions contribute to action and thought stopping by targeting the motor cortex or the hippocampus, depending on the goal, to suppress their task-specific activity. These findings support the existence of a domain-general system that underlies inhibitory control and establish Dynamic Targeting as a mechanism enabling this ability.

Modeling and augmenting of fMRI data using deep recurrent variational auto-encoder

Objective. Recently, deep learning models have been successfully applied in functional magnetic resonance imaging (fMRI) modeling and associated applications. However, there still exist at least two challenges. Firstly, due to the lack of sufficient data, deep learning models tend to suffer from overfitting in the training process. Secondly, it is still challenging to model the temporal dynamics from fMRI, due to that the brain state is continuously changing over scan time. In addition, existing methods rarely studied and applied fMRI data augmentation.Approach. In this work, we construct a deep recurrent variational auto-encoder (DRVAE) that combined variational auto-encoder and recurrent neural network, aiming to address all of the above mentioned challenges. The encoder of DRVAE can extract more generalized temporal features from assumed Gaussian distribution of input data, and the decoder of DRVAE can generate new data to increase training samples and thus partially relieve the overfitting issue. The recurrent layers in DRVAE are designed to effectively model the temporal dynamics of functional brain activities. LASSO (least absolute shrinkage and selection operator) regression is applied on the temporal features and input fMRI data to estimate the corresponding spatial networks.Main results. Extensive experimental results on seven tasks from HCP dataset showed that the DRVAE and LASSO framework can learn meaningful temporal patterns and spatial networks from both real data and generated data. The results on group-wise data and single subject suggest that the brain activities may follow certain distribution. Moreover, we applied DRVAE on four resting state fMRI datasets from ADHD-200 for data augmentation, and the results showed that the classification performances on augmented datasets have been considerably improved.Significance. The proposed method can not only derive meaningful temporal features and spatial networks from fMRI, but also generate high-quality new data for fMRI data augmentation and associated applications.

Age differences in Neural Activation to Face Trustworthiness: Voxel Pattern and Activation Level Assessments

Judgment of trustworthiness is an important social ability. Many studies show neural activation differences to variations in face trustworthiness in brain reward regions. A previously published analysis of the present fMRI data showed that older adults' (OA) reward region activation responded significantly to trustworthiness in a set of older and younger faces, whereas younger adults' (YA) activation did not-a finding inconsistent with studies that used only younger faces. We hypothesized that voxel pattern analyses would be more sensitive to YA neural responses to trustworthiness in our set of faces, replicating YA neural discrimination in prior literature. Based on evidence for OA neural dedifferentiation, we also hypothesized that voxel pattern analyses would more accurately classify YA than OA neural responses to face trustworthiness. We reanalyzed the data with two pattern classification models and evaluated the models' performance with permutation testing. Voxel patterns discriminated face trustworthiness levels in both YA and OA reward regions, while allowing better classification of face trustworthiness for YA than OA, the reverse of previous results for neural activation levels. The moderation of age differences by analytic method shines a light on the possibility that voxel patterns uniquely index neural representations of the stimulus information content, consistent with findings of impaired representation with age.

Analysis of MRI Data in Diagnostic Neuroradiology

Magnetic resonance imaging (MRI) is a noninvasive imaging tool for neuroradiological diagnosis. Numerous concepts of automated MRI analysis and the use of machine learning have been proposed to assist diagnosis and prognosis. While these academic innovations have proven effective in principle within controlled environments, their application to clinical practice has faced unmet requirements, such as the ability to perform reliably across a heterogeneous population, to work robustly in the presence of comorbidities, and to be invariant to scanner hardware and image quality. The lack of realistic confidence bounds and the inability to handle missing data have also reduced the application of most of these methods outside of academic studies. Mastering the complex challenges in the diagnostic process may help researchers discover novel biological constructs in multimodal data and improve stratification for clinical trials, paving the way for precision medicine. This review presents the state of the art of computerized brain MRI analysis for diagnostic purposes. We critically evaluate the current clinical usefulness of the methods and highlight challenges and future perspectives of the field.

Decoding Changes of Mind in Voluntary Action-Dynamics of Intentional Choice Representations

Voluntary actions rely on appropriate flexibility of intentions. Usually, we should pursue our goals, but sometimes we should change goals if they become too costly to achieve. Using functional magnetic resonance imaging, we investigated the neural dynamics underlying the capacity to change one's mind based on new information after action onset. Multivariate pattern analyses revealed that in visual areas, neural representations of intentional choice between 2 visual stimuli were unchanged by additional decision-relevant information. However, in fronto-parietal cortex, representations changed dynamically as decisions evolved. Precuneus, angular gyrus, and dorsolateral prefrontal cortex encoded new externally cued rewards/costs that guided subsequent changes of mind. Activity in medial frontal cortex predicted changes of mind when participants detached from externally cued evidence, suggesting a role in endogenous decision updates. Finally, trials with changes of mind were associated with an increase in functional connectivity between fronto-parietal areas, allowing for integration of various endogenous and exogenous decision components to generate a distributed consensus about whether to pursue or abandon an initial intention. In conclusion, local and global dynamics of choice representations in fronto-parietal cortex allow agents to maintain the balance between adapting to changing environments versus pursuing internal goals.

Evidence for allocentric boundary and goal direction information in the human entorhinal cortex and subiculum

In rodents, cells in the medial entorhinal cortex (EC) and subiculum code for the allocentric direction to environment boundaries, which is an important prerequisite for accurate positional coding. Although in humans boundary-related signals have been reported, there is no evidence that they contain allocentric direction information. Furthermore, it has not been possible to separate boundary versus goal direction signals in the EC/subiculum. Here, to address these questions, we had participants learn a virtual environment containing four unique boundaries. Participants then underwent fMRI scanning where they made judgements about the allocentric direction of a cue object. Using multivariate decoding, we found information regarding allocentric boundary direction in posterior EC and subiculum, whereas allocentric goal direction was decodable from anterior EC and subiculum. These data provide the first evidence of allocentric boundary coding in humans, and are consistent with recent conceptualisations of a division of labour within the EC.

Impaired Odor Perception in Autism Spectrum Disorder Is Associated with Decreased Activity in Olfactory Cortex

Autism Spectrum Disorders (ASDs) are characterized by atypical sensory functioning in the visual, tactile, and auditory systems. Although less explored, olfactory changes have been reported in ASD patients. To explore these changes on a neural level, 18 adults with ASD and 18 healthy neurotypical controls were examined in a 2-phase study. Participants were first tested for odor threshold and odor identification. Then, (i) structural magnetic resonance (MR) images of the olfactory bulb were acquired, and (ii) a functional MR imaging olfaction study was conducted. ASD patients exhibited decreased function for odor thresholds and odor identification; this was accompanied by a relatively decreased activation in the piriform cortex. In conclusion, these findings suggest, that the known alterations in olfaction in ASD are rooted in the primary olfactory cortex.

Cultural specialization of visual cortex

A growing body of evidence suggests culture influences how individuals perceive the world around them. This study investigates whether these cultural differences extend to a simple object viewing task and visual cortex by examining voxel pattern representations with multi-voxel pattern analysis (MVPA). During functional magnetic resonance imaging scanning, 20 East Asian and 20 American participants viewed photos of everyday items, equated for familiarity and conceptual agreement across cultures. Whole brain searchlight mapping with non-parametric statistical evaluation tested whether these stimuli evoked multi-voxel patterns that were distinct between cultural groups. We found that participants' cultural identities were successfully predicted from stimuli representations in visual cortex Brodmann areas 18 and 19. This result demonstrates culturally specialized visual cortex during a basic perceptual task ubiquitous to everyday life.

Feature Selection and Transfer Learning for Alzheimer’s Disease Clinical Diagnosis

Background and Purpose: A majority studies on diagnosis of Alzheimer’s Disease (AD) are based on an assumption: the training and testing data are drawn from the same distribution. However, in the diagnosis of AD and mild cognitive impairment (MCI), this identical-distribution assumption may not hold. To solve this problem, we utilize the transfer learning method into the diagnosis of AD. Methods: The MR (Magnetic Resonance) images were segmented using spm-Dartel toolbox and registrated with Automatic Anatomical Labeling (AAL) atlas, then the gray matter (GM) tissue volume of the anatomical region were computed as characteristic parameter. The information gain was introduced for feature selection. The TrAdaboost algorithm was used to classify AD, MCI, and normal controls (NC) data from Alzheimer’s Disease Neuroimaging Initiative (ADNI) database, meanwhile, the “knowledge” learned from ADNI was transferred to AD samples from local hospital. The classification accuracy, sensitivity and specificity were calculated and compared with four classical algorithms. Results: In the experiment of transfer task: AD to MCI, 177 AD and 40NC subjects were grouped as training data; 245 MCI and 45 remaining NC subjects were combined as testing data, the highest accuracy achieved 85.4%, higher than the other four classical algorithms. Meanwhile, feature selection that is based on information gain reduced the features from 90 to 7, controlled the redundancy efficiently. In the experiment of transfer task: ADNI to local hospital data, the highest accuracy achieved 93.7%, and the specificity achieved 100%. Conclusions: The experimental results showed that our algorithm has a clear advantage over classic classification methods with higher accuracy and less fluctuation.

Crossing the divide: a longitudinal study of effective treatments for people with autism and attention deficit hyperactivity disorder across the lifespan

Background

Autism spectrum disorder (ASD) and attention deficit hyperactivity disorder (ADHD) frequently persist into adolescence and young adulthood. However, there are few clinical services that support those with these disorders through adulthood.

Objective

Our aim was to determine if clinical services meet the needs of people with ASD and ADHD, who are ‘at transition’ from childhood to adulthood.

Design

A longitudinal study of individuals with ASD and ADHD, the impact of services and treatments.

Methods

Our research methods included (1) interviewing > 180 affected individuals (and their families) with a confirmed diagnosis of ASD and/or ADHD, (2) screening for ASD and ADHD in approximately 1600 patients and (3) surveying general practitioner prescribing to 5651 ASD individuals across the UK. In addition, we tested the effectiveness of (1) new ASD diagnostic interview measures in 169 twins, 145 familes and 150 non-twins, (2) a magnetic resonance imaging-based diagnostic aid in 40 ASD individuals, (3) psychological treatments in 46 ASD individuals and (4) the feasability of e-learning in 28 clinicians.

Setting

NHS clinical services and prisons.

Participants

Focus – young people with ASD and ADHD as they ‘transition’ from childhood and adolescence into early adulthood.

Interventions

Testing the utility of diagnostic measures and services, web-based learning interventions, pharmacological prescribing and cognitive–behavioural treatments.

Main outcome measures

Symptom severity, service provision and met/unmet need.

Results

People with ASD and ADHD have very significant unmet needs as they transition through adolescence and young adulthood. A major contributor to this is the presence of associated mental health symptoms. However, these are mostly undiagnosed (and untreated) by clinical services. Furthermore, the largest determinant of service provision was age and not severity of symptoms. We provide new tools to help diagnose both the core disorders and their associated symptoms. We also provide proof of concept for the effectiveness of simple psychological interventions to treat obsessional symptoms, the potential to run treatment trials in prisons and training interventions.

Limitations

Our findings only apply to clinical service settings.

Conclusions

As individuals ‘transition’ their contact with treatment and support services reduces significantly. Needs-led services are required, which can both identify individuals with the ‘core symptoms’ of ASD and ADHD and treat their residual symptoms and associated conditions.

Future work

To test our new diagnostic measures and treatment approaches in larger controlled trials.

Trial registration

Current Controlled Trials ISRCTN87114880.

Funding

The National Institute for Health Research Programme Grants for Applied Research programme.

Patch-based local learning method for cerebral blood flow quantification with arterial spin-labeling MRI

Arterial spin-labeling (ASL) perfusion MRI is a non-invasive method for quantifying cerebral blood flow (CBF). Standard ASL CBF calibration mainly relies on pair-wise subtraction of the spin-labeled images and controls images at each voxel separately, ignoring the abundant spatial correlations in ASL data. To address this issue, we previously proposed a multivariate support vector machine (SVM) learning-based algorithm for ASL CBF quantification (SVMASLQ). But the original SVMASLQ was designed to do CBF quantification for all image voxels simultaneously, which is not ideal for considering local signal and noise variations. To fix this problem, we here in this paper extended SVMASLQ into a patch-wise method by using a patch-wise classification kernel. At each voxel, an image patch centered at that voxel was extracted from both the control images and labeled images, which was then input into SVMASLQ to find the corresponding patch of the surrogate perfusion map using a non-linear SVM classifier. Those patches were eventually combined into the final perfusion map. Method evaluations were performed using ASL data from 30 young healthy subjects. The results showed that the patch-wise SVMASLQ increased perfusion map SNR by 6.6% compared to the non-patch-wise SVMASLQ.

Sparsity Is Better with Stability: Combining Accuracy and Stability for Model Selection in Brain Decoding

Structured sparse methods have received significant attention in neuroimaging. These methods allow the incorporation of domain knowledge through additional spatial and temporal constraints in the predictive model and carry the promise of being more interpretable than non-structured sparse methods, such as LASSO or Elastic Net methods. However, although sparsity has often been advocated as leading to more interpretable models it can also lead to unstable models under subsampling or slight changes of the experimental conditions. In the present work we investigate the impact of using stability/reproducibility as an additional model selection criterion¹ on several different sparse (and structured sparse) methods that have been recently applied for fMRI brain decoding. We compare three different model selection criteria: (i) classification accuracy alone; (ii) classification accuracy and overlap between the solutions; (iii) classification accuracy and correlation between the solutions. The methods we consider include LASSO, Elastic Net, Total Variation, sparse Total Variation, Laplacian and Graph Laplacian Elastic Net (GraphNET). Our results show that explicitly accounting for stability/reproducibility during the model optimization can mitigate some of the instability inherent in sparse methods. In particular, using accuracy and overlap between the solutions as a joint optimization criterion can lead to solutions that are more similar in terms of accuracy, sparsity levels and coefficient maps even when different sparsity methods are considered.

Real-Time fMRI in Neuroscience Research and Its Use in Studying the Aging Brain

Cognitive decline is a major concern in the aging population. It is normative to experience some deterioration in cognitive abilities with advanced age such as related to memory performance, attention distraction to interference, task switching, and processing speed. However, intact cognitive functioning in old age is important for leading an independent day-to-day life. Thus, studying ways to counteract or delay the onset of cognitive decline in aging is crucial. The literature offers various explanations for the decline in cognitive performance in aging; among those are age-related gray and white matter atrophy, synaptic degeneration, blood flow reduction, neurochemical alterations, and change in connectivity patterns with advanced age. An emerging literature on neurofeedback and Brain Computer Interface (BCI) reports exciting results supporting the benefits of volitional modulation of brain activity on cognition and behavior. Neurofeedback studies based on real-time functional magnetic resonance imaging (rtfMRI) have shown behavioral changes in schizophrenia and behavioral benefits in nicotine addiction. This article integrates research on cognitive and brain aging with evidence of brain and behavioral modification due to rtfMRI neurofeedback. We offer a state-of-the-art description of the rtfMRI technique with an eye towards its application in aging. We present preliminary results of a feasibility study exploring the possibility of using rtfMRI to train older adults to volitionally control brain activity. Based on these first findings, we discuss possible implementations of rtfMRI neurofeedback as a novel technique to study and alleviate cognitive decline in healthy and pathological aging.

Anomalous gray matter patterns in specific reading comprehension deficit are independent of dyslexia

Specific reading comprehension deficit (SRCD) affects up to 10 % of all children. SRCD is distinct from dyslexia (DYS) in that individuals with SRCD show poor comprehension despite adequate decoding skills. Despite its prevalence and considerable behavioral research, there is not yet a unified cognitive profile of SRCD. While its neuroanatomical basis is unknown, SRCD could be anomalous in regions subserving their commonly reported cognitive weaknesses in semantic processing or executive function. Here we investigated, for the first time, patterns of gray matter volume difference in SRCD as compared to DYS and typical developing (TD) adolescent readers (N = 41). A linear support vector machine algorithm was applied to whole brain gray matter volumes generated through voxel-based morphometry. As expected, DYS differed significantly from TD in a pattern that included features from left fusiform and supramarginal gyri (DYS vs. TD: 80.0 %, p < 0.01). SRCD was well differentiated not only from TD (92.5 %, p < 0.001) but also from DYS (88.0 %, p < 0.001). Of particular interest were findings of reduced gray matter volume in right frontal areas that were also supported by univariate analysis. These areas are thought to subserve executive processes relevant for reading, such as monitoring and manipulating mental representations. Thus, preliminary analyses suggest that SRCD readers possess a distinct neural profile compared to both TD and DYS readers and that these differences might be linked to domain-general abilities. This work provides a foundation for further investigation into variants of reading disability beyond DYS.

Reward Motivation Enhances Task Coding in Frontoparietal Cortex

Reward motivation often enhances task performance, but the neural mechanisms underlying such cognitive enhancement remain unclear. Here, we used a multivariate pattern analysis (MVPA) approach to test the hypothesis that motivation-related enhancement of cognitive control results from improved encoding and representation of task set information. Participants underwent two fMRI sessions of cued task switching, the first under baseline conditions, and the second with randomly intermixed reward incentive and no-incentive trials. Information about the upcoming task could be successfully decoded from cue-related activation patterns in a set of frontoparietal regions typically associated with task control. More critically, MVPA classifiers trained on the baseline session had significantly higher decoding accuracy on incentive than non-incentive trials, with decoding improvement mediating reward-related enhancement of behavioral performance. These results strongly support the hypothesis that reward motivation enhances cognitive control, by improving the discriminability of task-relevant information coded and maintained in frontoparietal brain regions.

Successful retrieval of competing spatial environments in humans involves hippocampal pattern separation mechanisms

The rodent hippocampus represents different spatial environments distinctly via changes in the pattern of “place cell” firing. It remains unclear, though, how spatial remapping in rodents relates more generally to human memory. Here participants retrieved four virtual reality environments with repeating or novel landmarks and configurations during high-resolution functional magnetic resonance imaging (fMRI). Both neural decoding performance and neural pattern similarity measures revealed environment-specific hippocampal neural codes. Conversely, an interfering spatial environment did not elicit neural codes specific to that environment, with neural activity patterns instead resembling those of competing environments, an effect linked to lower retrieval performance. We find that orthogonalized neural patterns accompany successful disambiguation of spatial environments while erroneous reinstatement of competing patterns characterized interference errors. These results provide the first evidence for environment-specific neural codes in the human hippocampus, suggesting that pattern separation/completion mechanisms play an important role in how we successfully retrieve memories.

DOI:http://dx.doi.org/10.7554/eLife.10499.001

How do we remember the different places that we have visited during our day? Studies of brain activity in rats suggest that each place a rat visits is represented by a different pattern of activity in a region of the brain called hippocampus. However, it is not clear what role this “spatial remapping” plays in the formation of memories in humans.

Kyle et al. used a technique called functional magnetic resonance imaging (fMRI) to investigate how our brain represents the different places we’ve visited, and how this is linked to how well we can remember these locations. For the experiments, human volunteers played a video game where they visited four virtual environments in a different order. Later, the volunteers were asked to remember details about the virtual environments they had visited while their brain activity was monitored using fMRI.

The experiments show that in order to remember distinct locations – even if they have some features in common – the hippocampus produces patterns of activity that have very little overlap with each other. This process is termed “pattern separation”. Sometimes our memory confuses different locations, which could be due to a failure of the brain to distinguish between the patterns of brain activity that represent these locations.

Kyle et al.’s findings provide the first evidence for spatial remapping in the human hippocampus and its importance in forming memories of locations. The next steps are to find out how many different environments our brain might be able to store at the same time, and to identify factors that could aid in our memory for spatial locations.

DOI:http://dx.doi.org/10.7554/eLife.10499.002

A Window into the Brain: Advances in Psychiatric fMRI

Functional magnetic resonance imaging (fMRI) plays a key role in modern psychiatric research. It provides a means to assay differences in brain systems that underlie psychiatric illness, treatment response, and properties of brain structure and function that convey risk factor for mental diseases. Here we review recent advances in fMRI methods in general use and progress made in understanding the neural basis of mental illness. Drawing on concepts and findings from psychiatric fMRI, we propose that mental illness may not be associated with abnormalities in specific local regions but rather corresponds to variation in the overall organization of functional communication throughout the brain network. Future research may need to integrate neuroimaging information drawn from different analysis methods and delineate spatial and temporal patterns of brain responses that are specific to certain types of psychiatric disorders.

Discriminative analysis of Parkinson's disease based on whole-brain functional connectivity

Recently, there has been an increasing emphasis on applications of pattern recognition and neuroimaging techniques in the effective and accurate diagnosis of psychiatric or neurological disorders. In the present study, we investigated the whole-brain resting-state functional connectivity patterns of Parkinson's disease (PD), which are expected to provide additional information for the clinical diagnosis and treatment of this disease. First, we computed the functional connectivity between each pair of 116 regions of interest derived from a prior atlas. The most discriminative features based on Kendall tau correlation coefficient were then selected. A support vector machine classifier was employed to classify 21 PD patients with 26 demographically matched healthy controls. This method achieved a classification accuracy of 93.62% using leave-one-out cross-validation, with a sensitivity of 90.47% and a specificity of 96.15%. The majority of the most discriminative functional connections were located within or across the default mode, cingulo-opercular and frontal-parietal networks and the cerebellum. These disease-related resting-state network alterations might play important roles in the pathophysiology of this disease. Our results suggest that analyses of whole-brain resting-state functional connectivity patterns have the potential to improve the clinical diagnosis and treatment evaluation of PD.

MANIA-a pattern classification toolbox for neuroimaging data

Conventional univariate statistics are common and widespread in neuroimaging research. However, functional and structural MRI data reveal a multivariate nature, since neighboring voxels are highly correlated and different localized brain regions activate interdependently. Multivariate pattern classification techniques are capable of overcoming shortcomings of univariate statistics. A rising interest in such approaches on neuroimaging data leads to an increasing demand of appropriate software and tools in this field. Here, we introduce and release MANIA-Machine learning Application for NeuroImaging Analyses. MANIA is a Matlab based software toolbox enabling easy pattern classification of neuroimaging data and offering a broad assortment of machine learning algorithms and feature selection methods. Between groups classifications are the main scope of this software, for instance the differentiation between patients and controls. A special emphasis was placed on an intuitive and easy to use graphical user interface allowing quick implementation and guidance also for clinically oriented researchers. MANIA is free and open source, published under GPL3 license. This work will give an overview regarding the functionality and the modular software architecture as well as a comparison between other software packages.

The Decoding Toolbox (TDT): a versatile software package for multivariate analyses of functional imaging data

The multivariate analysis of brain signals has recently sparked a great amount of interest, yet accessible and versatile tools to carry out decoding analyses are scarce. Here we introduce The Decoding Toolbox (TDT) which represents a user-friendly, powerful and flexible package for multivariate analysis of functional brain imaging data. TDT is written in Matlab and equipped with an interface to the widely used brain data analysis package SPM. The toolbox allows running fast whole-brain analyses, region-of-interest analyses and searchlight analyses, using machine learning classifiers, pattern correlation analysis, or representational similarity analysis. It offers automatic creation and visualization of diverse cross-validation schemes, feature scaling, nested parameter selection, a variety of feature selection methods, multiclass capabilities, and pattern reconstruction from classifier weights. While basic users can implement a generic analysis in one line of code, advanced users can extend the toolbox to their needs or exploit the structure to combine it with external high-performance classification toolboxes. The toolbox comes with an example data set which can be used to try out the various analysis methods. Taken together, TDT offers a promising option for researchers who want to employ multivariate analyses of brain activity patterns.

A review of feature reduction techniques in neuroimaging

Machine learning techniques are increasingly being used in making relevant predictions and inferences on individual subjects neuroimaging scan data. Previous studies have mostly focused on categorical discrimination of patients and matched healthy controls and more recently, on prediction of individual continuous variables such as clinical scores or age. However, these studies are greatly hampered by the large number of predictor variables (voxels) and low observations (subjects) also known as the curse-of-dimensionality or small-n-large-p problem. As a result, feature reduction techniques such as feature subset selection and dimensionality reduction are used to remove redundant predictor variables and experimental noise, a process which mitigates the curse-of-dimensionality and small-n-large-p effects. Feature reduction is an essential step before training a machine learning model to avoid overfitting and therefore improving model prediction accuracy and generalization ability. In this review, we discuss feature reduction techniques used with machine learning in neuroimaging studies.

Beyond BOLD: optimizing functional imaging in stroke populations

Blood oxygenation level-dependent (BOLD) signal changes are often assumed to directly reflect neural activity changes. Yet the real relationship is indirect, reliant on numerous assumptions, and subject to several sources of noise. Deviations from the core assumptions of BOLD contrast functional magnetic resonance imaging (fMRI), and their implications, have been well characterized in healthy populations, but are frequently neglected in stroke populations. In addition to conspicuous local structural and vascular changes after stroke, there are many less obvious challenges in the imaging of stroke populations. Perilesional ischemic changes, remodeling in regions distant to lesion sites, and diffuse perfusion changes all complicate interpretation of BOLD signal changes in standard fMRI protocols. Most stroke patients are also older than the young populations on which assumptions of neurovascular coupling and the typical analysis pipelines are based. We present a review of the evidence to show that the basic assumption of neurovascular coupling on which BOLD-fMRI relies does not capture the complex changes arising from stroke, both pathological and recovery related. As a result, estimating neural activity using the canonical hemodynamic response function is inappropriate in a number of contexts. We review methods designed to better estimate neural activity in stroke populations. One promising alternative to event-related fMRI is a resting-state-derived functional connectivity approach. Resting-state fMRI is well suited to stroke populations because it makes no performance demands on patients and is capable of revealing network-based pathology beyond the lesion site.

Abnormal changes of multidimensional surface features using multivariate pattern classification in amnestic mild cognitive impairment patients

Previous studies have suggested that amnestic mild cognitive impairment (aMCI) is associated with changes in cortical morphological features, such as cortical thickness, sulcal depth, surface area, gray matter volume, metric distortion, and mean curvature. These features have been proven to have specific neuropathological and genetic underpinnings. However, most studies primarily focused on mass-univariate methods, and cortical features were generally explored in isolation. Here, we used a multivariate method to characterize the complex and subtle structural changing pattern of cortical anatomy in 24 aMCI human participants and 26 normal human controls. Six cortical features were extracted for each participant, and the spatial patterns of brain abnormities in aMCI were identified by high classification weights using a support vector machine method. The classification accuracy in discriminating the two groups was 76% in the left hemisphere and 80% in the right hemisphere when all six cortical features were used. Regions showing high weights were subtle, spatially complex, and predominately located in the left medial temporal lobe and the supramarginal and right inferior parietal lobes. In addition, we also found that the six morphological features had different contributions in discriminating the two groups even for the same region. Our results indicated that the neuroanatomical patterns that discriminated individuals with aMCI from controls were truly multidimensional and had different effects on the morphological features. Furthermore, the regions identified by our method could potentially be useful for clinical diagnosis.

Feature interactions enable decoding of sensorimotor transformations for goal-directed movement

Neurophysiology and neuroimaging evidence shows that the brain represents multiple environmental and body-related features to compute transformations from sensory input to motor output. However, it is unclear how these features interact during goal-directed movement. To investigate this issue, we examined the representations of sensory and motor features of human hand movements within the left-hemisphere motor network. In a rapid event-related fMRI design, we measured cortical activity as participants performed right-handed movements at the wrist, with either of two postures and two amplitudes, to move a cursor to targets at different locations. Using a multivoxel analysis technique with rigorous generalization tests, we reliably distinguished representations of task-related features (primarily target location, movement direction, and posture) in multiple regions. In particular, we identified an interaction between target location and movement direction in the superior parietal lobule, which may underlie a transformation from the location of the target in space to a movement vector. In addition, we found an influence of posture on primary motor, premotor, and parietal regions. Together, these results reveal the complex interactions between different sensory and motor features that drive the computation of sensorimotor transformations.

Voxel selection framework in multi-voxel pattern analysis of FMRI data for prediction of neural response to visual stimuli

Multi-voxel pattern analysis (MVPA) of functional magnetic resonance imaging (fMRI) data is an emerging approach for probing the neural correlates of cognition. MVPA allows cognitive states to be modeled as distributed patterns of neural activity and classified according to stimulus conditions. In practice, building a robust, generalizable classification model can be challenging because the number of voxels (features) far exceeds the number of stimulus instances/data observations. To avoid model overfitting, there is a need to select informative voxels before building a classification model. In this paper, we propose a robust feature (voxel) selection framework using mutual information (MI) and partial least square regression (PLS) to establish an informativeness index for prioritizing selection of voxels based on the degree of their association to the experimental conditions. We evaluated the robustness of our proposed framework by assessing performance of standard classification algorithms, when combined with our feature selection approach, in a publicly-available fMRI dataset of object-level representation widely used to benchmark MVPA performance (Haxby, 2001). The computational results suggest that our feature selection framework based on MI and PLS drastically improves the classification accuracy relative to those previously reported in the literature. Our results also suggest that highly informative voxels may provide meaningful insight into the functional-anatomic relationship of brain activity and stimulus conditions.

PRoNTo: pattern recognition for neuroimaging toolbox

In the past years, mass univariate statistical analyses of neuroimaging data have been complemented by the use of multivariate pattern analyses, especially based on machine learning models. While these allow an increased sensitivity for the detection of spatially distributed effects compared to univariate techniques, they lack an established and accessible software framework. The goal of this work was to build a toolbox comprising all the necessary functionalities for multivariate analyses of neuroimaging data, based on machine learning models. The “Pattern Recognition for Neuroimaging Toolbox” (PRoNTo) is open-source, cross-platform, MATLAB-based and SPM compatible, therefore being suitable for both cognitive and clinical neuroscience research. In addition, it is designed to facilitate novel contributions from developers, aiming to improve the interaction between the neuroimaging and machine learning communities. Here, we introduce PRoNTo by presenting examples of possible research questions that can be addressed with the machine learning framework implemented in PRoNTo, and cannot be easily investigated with mass univariate statistical analysis.

Multiclass fMRI data decoding and visualization using supervised self-organizing maps

When multivariate pattern decoding is applied to fMRI studies entailing more than two experimental conditions, a most common approach is to transform the multiclass classification problem into a series of binary problems. Furthermore, for decoding analyses, classification accuracy is often the only outcome reported although the topology of activation patterns in the high-dimensional features space may provide additional insights into underlying brain representations. Here we propose to decode and visualize voxel patterns of fMRI datasets consisting of multiple conditions with a supervised variant of self-organizing maps (SSOMs). Using simulations and real fMRI data, we evaluated the performance of our SSOM-based approach. Specifically, the analysis of simulated fMRI data with varying signal-to-noise and contrast-to-noise ratio suggested that SSOMs perform better than a k-nearest-neighbor classifier for medium and large numbers of features (i.e. 250 to 1000 or more voxels) and similar to support vector machines (SVMs) for small and medium numbers of features (i.e. 100 to 600voxels). However, for a larger number of features (>800voxels), SSOMs performed worse than SVMs. When applied to a challenging 3-class fMRI classification problem with datasets collected to examine the neural representation of three human voices at individual speaker level, the SSOM-based algorithm was able to decode speaker identity from auditory cortical activation patterns. Classification performances were similar between SSOMs and other decoding algorithms; however, the ability to visualize decoding models and underlying data topology of SSOMs promotes a more comprehensive understanding of classification outcomes. We further illustrated this visualization ability of SSOMs with a re-analysis of a dataset examining the representation of visual categories in the ventral visual cortex (Haxby et al., 2001). This analysis showed that SSOMs could retrieve and visualize topography and neighborhood relations of the brain representation of eight visual categories. We conclude that SSOMs are particularly suited for decoding datasets consisting of more than two classes and are optimally combined with approaches that reduce the number of voxels used for classification (e.g. region-of-interest or searchlight approaches).

Representations of Invariant Musical Categories Are Decodable by Pattern Analysis of Locally Distributed BOLD Responses in Superior Temporal and Intraparietal Sulci

In categorical perception (CP), continuous physical signals are mapped to discrete perceptual bins: mental categories not found in the physical world. CP has been demonstrated across multiple sensory modalities and, in audition, for certain over-learned speech and musical sounds. The neural basis of auditory CP, however, remains ambiguous, including its robustness in nonspeech processes and the relative roles of left/right hemispheres; primary/nonprimary cortices; and ventral/dorsal perceptual processing streams. Here, highly trained musicians listened to 2-tone musical intervals, which they perceive categorically while undergoing functional magnetic resonance imaging. Multivariate pattern analyses were performed after grouping sounds by interval quality (determined by frequency ratio between tones) or pitch height (perceived noncategorically, frequency ratios remain constant). Distributed activity patterns in spheres of voxels were used to determine sound sample identities. For intervals, significant decoding accuracy was observed in the right superior temporal and left intraparietal sulci, with smaller peaks observed homologously in contralateral hemispheres. For pitch height, no significant decoding accuracy was observed, consistent with the non-CP of this dimension. These results suggest that similar mechanisms are operative for nonspeech categories as for speech; espouse roles for 2 segregated processing streams; and support hierarchical processing models for CP.

SCoRS--A Method Based on Stability for Feature Selection and Mapping inNeuroimaging [corrected]

Feature selection (FS) methods play two important roles in the context of neuroimaging based classification: potentially increase classification accuracy by eliminating irrelevant features from the model and facilitate interpretation by identifying sets of meaningful features that best discriminate the classes. Although the development of FS techniques specifically tuned for neuroimaging data is an active area of research, up to date most of the studies have focused on finding a subset of features that maximizes accuracy. However, maximizing accuracy does not guarantee reliable interpretation as similar accuracies can be obtained from distinct sets of features. In the current paper we propose a new approach for selecting features: SCoRS (survival count on random subsamples) based on a recently proposed Stability Selection theory. SCoRS relies on the idea of choosing relevant features that are stable under data perturbation. Data are perturbed by iteratively sub-sampling both features (subspaces) and examples. We demonstrate the potential of the proposed method in a clinical application to classify depressed patients versus healthy individuals based on functional magnetic resonance imaging data acquired during visualization of happy faces.

Single trial decoding of belief decision making from EEG and fMRI data using independent components features

The complex task of assessing the veracity of a statement is thought to activate uniquely distributed brain regions based on whether a subject believes or disbelieves a given assertion. In the current work, we present parallel machine learning methods for predicting a subject's decision response to a given propositional statement based on independent component (IC) features derived from EEG and fMRI data. Our results demonstrate that IC features outperformed features derived from event related spectral perturbations derived from any single spectral band, yet were similar to accuracy across all spectral bands combined. We compared our diagnostic IC spatial maps with our conventional general linear model (GLM) results, and found that informative ICs had significant spatial overlap with our GLM results, yet also revealed unique regions like amygdala that were not statistically significant in GLM analyses. Overall, these results suggest that ICs may yield a parsimonious feature set that can be used along with a decision tree structure for interpretation of features used in classifying complex cognitive processes such as belief and disbelief across both fMRI and EEG neuroimaging modalities.

Identifying neural patterns of functional dyspepsia using multivariate pattern analysis: a resting-state FMRI study

BACKGROUND

Previous imaging studies on functional dyspepsia (FD) have focused on abnormal brain functions during special tasks, while few studies concentrated on the resting-state abnormalities of FD patients, which might be potentially valuable to provide us with direct information about the neural basis of FD. The main purpose of the current study was thereby to characterize the distinct patterns of resting-state function between FD patients and healthy controls (HCs).

METHODOLOGY/PRINCIPAL FINDINGS

Thirty FD patients and thirty HCs were enrolled and experienced 5-mintue resting-state scanning. Based on the support vector machine (SVM), we applied multivariate pattern analysis (MVPA) to investigate the differences of resting-state function mapped by regional homogeneity (ReHo). A classifier was designed by using the principal component analysis and the linear SVM. Permutation test was then employed to identify the significant contribution to the final discrimination. The results displayed that the mean classifier accuracy was 86.67%, and highly discriminative brain regions mainly included the prefrontal cortex (PFC), orbitofrontal cortex (OFC), supplementary motor area (SMA), temporal pole (TP), insula, anterior/middle cingulate cortex (ACC/MCC), thalamus, hippocampus (HIPP)/parahippocamus (ParaHIPP) and cerebellum. Correlation analysis revealed significant correlations between ReHo values in certain regions of interest (ROI) and the FD symptom severity and/or duration, including the positive correlations between the dmPFC, pACC and the symptom severity; whereas, the positive correlations between the MCC, OFC, insula, TP and FD duration.

CONCLUSIONS

These findings indicated that significantly distinct patterns existed between FD patients and HCs during the resting-state, which could expand our understanding of the neural basis of FD. Meanwhile, our results possibly showed potential feasibility of functional magnetic resonance imaging diagnostic assay for FD.

The statistical analysis of multi-voxel patterns in functional imaging

The goal of multi-voxel pattern analysis (MVPA) in BOLD imaging is to determine whether patterns of activation across multiple voxels change with experimental conditions. MVPA is a powerful technique, its use is rapidly growing, but it poses serious statistical challenges. For instance, it is well-known that the slow nature of the BOLD response can lead to greatly exaggerated performance estimates. Methods are available to avoid this overestimation, and we present those here in tutorial fashion. We go on to show that, even with these methods, standard tests of significance such as Students' T and the binomial tests are invalid in typical MRI experiments. Only a carefully constructed permutation test correctly assesses statistical significance. Furthermore, our simulations show that performance estimates increase with both temporal as well as spatial signal correlations among multiple voxels. This dependence implies that a comparison of MVPA performance between areas, between subjects, or even between BOLD signals that have been preprocessed in different ways needs great care.

Multiclass classification of FDG PET scans for the distinction between Parkinson's disease and atypical parkinsonian syndromes

Most available pattern recognition methods in neuroimaging address binary classification problems. Here, we used relevance vector machine (RVM) in combination with booststrap resampling (‘bagging’) for non-hierarchical multiclass classification. The method was tested on 120 cerebral ¹⁸fluorodeoxyglucose (FDG) positron emission tomography (PET) scans performed in patients who exhibited parkinsonian clinical features for 3.5 years on average but that were outside the prevailing perception for Parkinson's disease (PD). A radiological diagnosis of PD was suggested for 30 patients at the time of PET imaging. However, at follow-up several years after PET imaging, 42 of them finally received a clinical diagnosis of PD. The remaining 78 APS patients were diagnosed with multiple system atrophy (MSA, N = 31), progressive supranuclear palsy (PSP, N = 26) and corticobasal syndrome (CBS, N = 21), respectively. With respect to this standard of truth, classification sensitivity, specificity, positive and negative predictive values for PD were 93% 83% 75% and 96%, respectively using binary RVM (PD vs. APS) and 90%, 87%, 79% and 94%, respectively, using multiclass RVM (PD vs. MSA vs. PSP vs. CBS). Multiclass RVM achieved 45%, 55% and 62% classification accuracy for, MSA, PSP and CBS, respectively. Finally, a majority confidence ratio was computed for each scan on the basis of class pairs that were the most frequently assigned by RVM. Altogether, the results suggest that automatic multiclass RVM classification of FDG PET scans achieves adequate performance for the early differentiation between PD and APS on the basis of cerebral FDG uptake patterns when the clinical diagnosis is felt uncertain. This approach cannot be recommended yet as an aid for distinction between the three APS classes under consideration.

•

Multiclass classification is one of the challenges of computer-aided diagnosis.

•

This was addressed here using relevance vector machine and bootstrap aggregation.

•

Performance was tested on FDG-PET scans from 120 parkinsonian patients.

•

Four diagnostic classes under consideration as defined on average 3.5 years after PET.

•

Confusion matrices, majority confidence ratio and discriminant maps were computed.

Multivariate decoding of brain images using ordinal regression

Neuroimaging data are increasingly being used to predict potential outcomes or groupings, such as clinical severity, drug dose response, and transitional illness states. In these examples, the variable (target) we want to predict is ordinal in nature. Conventional classification schemes assume that the targets are nominal and hence ignore their ranked nature, whereas parametric and/or non-parametric regression models enforce a metric notion of distance between classes. Here, we propose a novel, alternative multivariate approach that overcomes these limitations — whole brain probabilistic ordinal regression using a Gaussian process framework. We applied this technique to two data sets of pharmacological neuroimaging data from healthy volunteers. The first study was designed to investigate the effect of ketamine on brain activity and its subsequent modulation with two compounds — lamotrigine and risperidone. The second study investigates the effect of scopolamine on cerebral blood flow and its modulation using donepezil. We compared ordinal regression to multi-class classification schemes and metric regression. Considering the modulation of ketamine with lamotrigine, we found that ordinal regression significantly outperformed multi-class classification and metric regression in terms of accuracy and mean absolute error. However, for risperidone ordinal regression significantly outperformed metric regression but performed similarly to multi-class classification both in terms of accuracy and mean absolute error. For the scopolamine data set, ordinal regression was found to outperform both multi-class and metric regression techniques considering the regional cerebral blood flow in the anterior cingulate cortex. Ordinal regression was thus the only method that performed well in all cases. Our results indicate the potential of an ordinal regression approach for neuroimaging data while providing a fully probabilistic framework with elegant approaches for model selection.

•

Often in neuroimaging the independent variables are ranked or ordered.

•

Classification and regression models cannot explicitly model an ordinal target.

•

We present a novel multivariate ordinal regression approach for neuroimaging data.

•

Our results show that ordinal regression is a powerful method for ranking data.

Confounds in multivariate pattern analysis: Theory and rule representation case study

Multivariate pattern analysis (MVPA) is a relatively recent innovation in functional magnetic resonance imaging (fMRI) methods. MVPA is increasingly widely used, as it is apparently more effective than classical general linear model analysis (GLMA) for detecting response patterns or representations that are distributed at a fine spatial scale. However, we demonstrate that widely used approaches to MVPA can systematically admit certain confounds that are appropriately eliminated by GLMA. Thus confounds rather than distributed representations may explain some cases in which MVPA produced positive results but GLMA did not. The issue is that it is common practice in MVPA to conduct group tests on single-subject summary statistics that discard the sign or direction of underlying effects, whereas GLMA group tests are conducted directly on single-subject effects themselves. We describe how this common MVPA practice undermines standard experiment design logic that is intended to control at the group level for certain types of confounds, such as time on task and individual differences. Furthermore, we note that a simple application of linear regression can restore experimental control when using MVPA in many situations. Finally, we present a case study with novel fMRI data in the domain of rule representations, or flexible stimulus-response mappings, which has seen several recent MVPA publications. In our new dataset, as with recent reports, standard MVPA appears to reveal rule representations in prefrontal cortex regions, whereas GLMA produces null results. However, controlling for a variable that is confounded with rule at the individual-subject level but not the group level (reaction time differences across rules) eliminates the MVPA results. This raises the question of whether recently reported results truly reflect rule representations, or rather the effects of confounds such as reaction time, difficulty, or other variables of no interest.