Single subject prediction of brain disorders in neuroimaging: Promises and pitfalls

Multi-modal, Multi-measure, and Multi-class Discrimination of ADHD with Hierarchical Feature Extraction and Extreme Learning Machine Using Structural and Functional Brain MRI

Structural and functional MRI unveil many hidden properties of the human brain. We performed this multi-class classification study on selected subjects from the publically available attention deficit hyperactivity disorder ADHD-200 dataset of patients and healthy children. The dataset has three groups, namely, ADHD inattentive, ADHD combined, and typically developing. We calculated the global averaged functional connectivity maps across the whole cortex to extract anatomical atlas parcellation based features from the resting-state fMRI (rs-fMRI) data and cortical parcellation based features from the structural MRI (sMRI) data. In addition, the preprocessed image volumes from both of these modalities followed an ANOVA analysis separately using all the voxels. This study utilized the average measure from the most significant regions acquired from ANOVA as features for classification in addition to the multi-modal and multi-measure features of structural and functional MRI data. We extracted most discriminative features by hierarchical sparse feature elimination and selection algorithm. These features include cortical thickness, image intensity, volume, cortical thickness standard deviation, surface area, and ANOVA based features respectively. An extreme learning machine performed both the binary and multi-class classifications in comparison with support vector machines. This article reports prediction accuracy of both unimodal and multi-modal features from test data. We achieved 76.190% (p < 0.0001) classification accuracy in multi-class settings as well as 92.857% (p < 0.0001) classification accuracy in binary settings. In addition, we found ANOVA-based significant regions of the brain that also play a vital role in the classification of ADHD. Thus, from a clinical perspective, this multi-modal group analysis approach with multi-measure features may improve the accuracy of the ADHD differential diagnosis.

Computational neuroscience approach to biomarkers and treatments for mental disorders

Psychiatry research has long experienced a stagnation stemming from a lack of understanding of the neurobiological underpinnings of phenomenologically defined mental disorders. Recently, the application of computational neuroscience to psychiatry research has shown great promise in establishing a link between phenomenological and pathophysiological aspects of mental disorders, thereby recasting current nosology in more biologically meaningful dimensions. In this review, we highlight recent investigations into computational neuroscience that have undertaken either theory- or data-driven approaches to quantitatively delineate the mechanisms of mental disorders. The theory-driven approach, including reinforcement learning models, plays an integrative role in this process by enabling correspondence between behavior and disorder-specific alterations at multiple levels of brain organization, ranging from molecules to cells to circuits. Previous studies have explicated a plethora of defining symptoms of mental disorders, including anhedonia, inattention, and poor executive function. The data-driven approach, on the other hand, is an emerging field in computational neuroscience seeking to identify disorder-specific features among high-dimensional big data. Remarkably, various machine-learning techniques have been applied to neuroimaging data, and the extracted disorder-specific features have been used for automatic case-control classification. For many disorders, the reported accuracies have reached 90% or more. However, we note that rigorous tests on independent cohorts are critically required to translate this research into clinical applications. Finally, we discuss the utility of the disorder-specific features found by the data-driven approach to psychiatric therapies, including neurofeedback. Such developments will allow simultaneous diagnosis and treatment of mental disorders using neuroimaging, thereby establishing 'theranostics' for the first time in clinical psychiatry.

Multi-View Ensemble Classification of Brain Connectivity Images for Neurodegeneration Type Discrimination

Brain connectivity analyses using voxels as features are not robust enough for single-patient classification because of the inter-subject anatomical and functional variability. To construct more robust features, voxels can be aggregated into clusters that are maximally coherent across subjects. Moreover, combining multi-modal neuroimaging and multi-view data integration techniques allows generating multiple independent connectivity features for the same patient. Structural and functional connectivity features were extracted from multi-modal MRI images with a clustering technique, and used for the multi-view classification of different phenotypes of neurodegeneration by an ensemble learning method (random forest). Two different multi-view models (intermediate and late data integration) were trained on, and tested for the classification of, individual whole-brain default-mode network (DMN) and fractional anisotropy (FA) maps, from 41 amyotrophic lateral sclerosis (ALS) patients, 37 Parkinson's disease (PD) patients and 43 healthy control (HC) subjects. Both multi-view data models exhibited ensemble classification accuracies significantly above chance. In ALS patients, multi-view models exhibited the best performances (intermediate: 82.9%, late: 80.5% correct classification) and were more discriminative than each single-view model. In PD patients and controls, multi-view models' performances were lower (PD: 59.5%, 62.2%; HC: 56.8%, 59.1%) but higher than at least one single-view model. Training the models only on patients, produced more than 85% patients correctly discriminated as ALS or PD type and maximal performances for multi-view models. These results highlight the potentials of mining complementary information from the integration of multiple data views in the classification of connectivity patterns from multi-modal brain images in the study of neurodegenerative diseases.

Discriminating cognitive status in Parkinson's disease through functional connectomics and machine learning

There is growing interest in the potential of neuroimaging to help develop non-invasive biomarkers in neurodegenerative diseases. In this study, connection-wise patterns of functional connectivity were used to distinguish Parkinson’s disease patients according to cognitive status using machine learning. Two independent subject samples were assessed with resting-state fMRI. The first (training) sample comprised 38 healthy controls and 70 Parkinson’s disease patients (27 with mild cognitive impairment). The second (validation) sample included 25 patients (8 with mild cognitive impairment). The Brainnetome atlas was used to reconstruct the functional connectomes. Using a support vector machine trained on features selected through randomized logistic regression with leave-one-out cross-validation, a mean accuracy of 82.6% (p < 0.002) was achieved in separating patients with mild cognitive impairment from those without it in the training sample. The model trained on the whole training sample achieved an accuracy of 80.0% when used to classify the validation sample (p = 0.006). Correlation analyses showed that the connectivity level in the edges most consistently selected as features was associated with memory and executive function performance in the patient group. Our results demonstrate that connection-wise patterns of functional connectivity may be useful for discriminating Parkinson’s disease patients according to the presence of cognitive deficits.

Functional connectomics from a "big data" perspective

In the last decade, explosive growth regarding functional connectome studies has been observed. Accumulating knowledge has significantly contributed to our understanding of the brain's functional network architectures in health and disease. With the development of innovative neuroimaging techniques, the establishment of large brain datasets and the increasing accumulation of published findings, functional connectomic research has begun to move into the era of "big data", which generates unprecedented opportunities for discovery in brain science and simultaneously encounters various challenging issues, such as data acquisition, management and analyses. Big data on the functional connectome exhibits several critical features: high spatial and/or temporal precision, large sample sizes, long-term recording of brain activity, multidimensional biological variables (e.g., imaging, genetic, demographic, cognitive and clinic) and/or vast quantities of existing findings. We review studies regarding functional connectomics from a big data perspective, with a focus on recent methodological advances in state-of-the-art image acquisition (e.g., multiband imaging), analysis approaches and statistical strategies (e.g., graph theoretical analysis, dynamic network analysis, independent component analysis, multivariate pattern analysis and machine learning), as well as reliability and reproducibility validations. We highlight the novel findings in the application of functional connectomic big data to the exploration of the biological mechanisms of cognitive functions, normal development and aging and of neurological and psychiatric disorders. We advocate the urgent need to expand efforts directed at the methodological challenges and discuss the direction of applications in this field.

3D scattering transforms for disease classification in neuroimaging

Classifying neurodegenerative brain diseases in MRI aims at correctly assigning discrete labels to MRI scans. Such labels usually refer to a diagnostic decision a learner infers based on what it has learned from a training sample of MRI scans. Classification from MRI voxels separately typically does not provide independent evidence towards or against a class; the information relevant for classification is only present in the form of complicated multivariate patterns (or “features”). Deep learning solves this problem by learning a sequence of non-linear transformations that result in feature representations that are better suited to classification. Such learned features have been shown to drastically outperform hand-engineered features in computer vision and audio analysis domains. However, applying the deep learning approach to the task of MRI classification is extremely challenging, because it requires a very large amount of data which is currently not available. We propose to instead use a three dimensional scattering transform, which resembles a deep convolutional neural network but has no learnable parameters. Furthermore, the scattering transform linearizes diffeomorphisms (due to e.g. residual anatomical variability in MRI scans), making the different disease states more easily separable using a linear classifier. In experiments on brain morphometry in Alzheimer's disease, and on white matter microstructural damage in HIV, scattering representations are shown to be highly effective for the task of disease classification. For instance, in semi-supervised learning of progressive versus stable MCI, we reach an accuracy of 82.7%. We also present a visualization method to highlight areas that provide evidence for or against a certain class, both on an individual and group level.

•

We have developed and implemented a feature extraction method based on the three dimensional (3D) scattering transform.

•

We tested it for its ability to discriminate diseased from healthy subjects and subjects with mild cognitive impairment.

•

We have clearly shown that our proposed methodology achieves higher accuracy than the best competing methods.

•

The scattering transform linearizes diffeomorphisms leading to more separable disease states using a linear classifier.

•

Scattering representations are shown to be highly effective for the task of disease classification.

•

We present a visualization method to highlight areas that provide evidence for or against a certain class.

Support Vector Machine Analysis of Functional Magnetic Resonance Imaging of Interoception Does Not Reliably Predict Individual Outcomes of Cognitive Behavioral Therapy in Panic Disorder with Agoraphobia

BACKGROUND

The approach to apply multivariate pattern analyses based on neuro imaging data for outcome prediction holds out the prospect to improve therapeutic decisions in mental disorders. Patients suffering from panic disorder with agoraphobia (PD/AG) often exhibit an increased perception of bodily sensations. The purpose of this investigation was to assess whether multivariate classification applied to a functional magnetic resonance imaging (fMRI) interoception paradigm can predict individual responses to cognitive behavioral therapy (CBT) in PD/AG.

METHODS

This analysis is based on pretreatment fMRI data during an interoceptive challenge from a multicenter trial of the German PANIC-NET. Patients with DSM-IV PD/AG were dichotomized as responders (n = 30) or non-responders (n = 29) based on the primary outcome (Hamilton Anxiety Scale Reduction ≥50%) after 6 weeks of CBT (2 h/week). fMRI parametric maps were used as features for response classification with linear support vector machines (SVM) with or without automated feature selection. Predictive accuracies were assessed using cross validation and permutation testing. The influence of methodological parameters and the predictive ability for specific interoception-related symptom reduction were further evaluated.

RESULTS

SVM did not reach sufficient overall predictive accuracies (38.0-54.2%) for anxiety reduction in the primary outcome. In the exploratory analyses, better accuracies (66.7%) were achieved for predicting interoception-specific symptom relief as an alternative outcome domain. Subtle information regarding this alternative response criterion but not the primary outcome was revealed by post hoc univariate comparisons.

CONCLUSION

In contrast to reports on other neurofunctional probes, SVM based on an interoception paradigm was not able to reliably predict individual response to CBT. Results speak against the clinical applicability of this technique.

Diagnostic classification of unipolar depression based on resting-state functional connectivity MRI: effects of generalization to a diverse sample

In small, selected samples, an approach combining resting-state functional connectivity MRI and multivariate pattern analysis has been able to successfully classify patients diagnosed with unipolar depression. Purposes of this investigation were to assess the generalizability of this approach to a large clinically more realistic sample and secondarily to assess the replicability of previously reported methodological feasibility in a more homogeneous subgroup with pronounced depressive symptoms. Two independent subsets were drawn from the depression and control cohorts of the BiDirect study, each with 180 patients with and 180 controls without depression. Functional connectivity either among regions covering the gray matter or selected regions with known alterations in depression was assessed by resting-state fMRI. Support vector machines with and without automated feature selection were used to train classifiers differentiating between individual patients and controls in the entire first subset as well as in the subgroup. Model parameters were explored systematically. The second independent subset was used for validation of successful models. Classification accuracies in the large, heterogeneous sample ranged from 45.0 to 56.1% (chance level 50.0%). In the subgroup with higher depression severity, three out of 90 models performed significantly above chance (60.8-61.7% at independent validation). In conclusion, common classification methods previously successful in small homogenous depression samples do not immediately translate to a more realistic population. Future research to develop diagnostic classification approaches in depression should focus on more specific clinical questions and consider heterogeneity, including symptom severity as an important factor.

Selection bias in the reported performances of AD classification pipelines

The last decade has seen a great proliferation of supervised learning pipelines for individual diagnosis and prognosis in Alzheimer's disease. As more pipelines are developed and evaluated in the search for greater performance, only those results that are relatively impressive will be selected for publication. We present an empirical study to evaluate the potential for optimistic bias in classification performance results as a result of this selection. This is achieved using a novel, resampling-based experiment design that effectively simulates the optimisation of pipeline specifications by individuals or collectives of researchers using cross validation with limited data. Our findings indicate that bias can plausibly account for an appreciable fraction (often greater than half) of the apparent performance improvement associated with the pipeline optimisation, particularly in small samples. We discuss the consistency of our findings with patterns observed in the literature and consider strategies for bias reduction and mitigation.

•

Demonstration and measurement of selection bias in AD classification experiments

•

Bias accounts for much of the performance improvement seen with pipeline optimisation.

•

Assessment of key risk factors and guidance on best research practices

•

Evidence of selection bias in results collected by a recent literature review

Structural brain imaging correlates of ASD and ADHD across the lifespan: a hypothesis-generating review on developmental ASD-ADHD subtypes

We hypothesize that it is plausible that biologically distinct developmental ASD-ADHD subtypes are present, each characterized by a distinct time of onset of symptoms, progression and combination of symptoms. The aim of the present narrative review was to explore if structural brain imaging studies may shed light on key brain areas that are linked to both ASD and ADHD symptoms and undergo significant changes during development. These findings may possibly pinpoint to brain mechanisms underlying differential developmental ASD-ADHD subtypes. To this end we brought together the literature on ASD and ADHD structural brain imaging symptoms and particularly highlight the adolescent years and beyond. Findings indicate that the vast majority of existing MRI studies has been cross-sectional and conducted in children, and sometimes did include adolescents as well, but without explicitly documenting on this age group. MRI studies documenting on age effects in adults with ASD and/or ADHD are rare, and if age is taken into account, only linear effects are examined. Data from various studies suggest that a crucial distinctive feature underlying different developmental ASD-ADHD subtypes may be the differential developmental thinning patterns of the anterior cingulate cortex and related connections towards other prefrontal regions. These regions are crucial for the development of cognitive/effortful control and socio-emotional functioning, with impairments in these features as key to both ASD and ADHD.

Promises, Pitfalls, and Basic Guidelines for Applying Machine Learning Classifiers to Psychiatric Imaging Data, with Autism as an Example

Most psychiatric disorders are associated with subtle alterations in brain function and are subject to large interindividual differences. Typically, the diagnosis of these disorders requires time-consuming behavioral assessments administered by a multidisciplinary team with extensive experience. While the application of Machine Learning classification methods (ML classifiers) to neuroimaging data has the potential to speed and simplify diagnosis of psychiatric disorders, the methods, assumptions, and analytical steps are currently opaque and not accessible to researchers and clinicians outside the field. In this paper, we describe potential classification pipelines for autism spectrum disorder, as an example of a psychiatric disorder. The analyses are based on resting-state fMRI data derived from a multisite data repository (ABIDE). We compare several popular ML classifiers such as support vector machines, neural networks, and regression approaches, among others. In a tutorial style, written to be equally accessible for researchers and clinicians, we explain the rationale of each classification approach, clarify the underlying assumptions, and discuss possible pitfalls and challenges. We also provide the data as well as the MATLAB code we used to achieve our results. We show that out-of-the-box ML classifiers can yield classification accuracies of about 60-70%. Finally, we discuss how classification accuracy can be further improved, and we mention methodological developments that are needed to pave the way for the use of ML classifiers in clinical practice.

Development of an Automated MRI-Based Diagnostic Protocol for Amyotrophic Lateral Sclerosis Using Disease-Specific Pathognomonic Features: A Quantitative Disease-State Classification Study

BACKGROUND

Despite significant advances in quantitative neuroimaging, the diagnosis of ALS remains clinical and MRI-based biomarkers are not currently used to aid the diagnosis. The objective of this study is to develop a robust, disease-specific, multimodal classification protocol and validate its diagnostic accuracy in independent, early-stage and follow-up data sets.

METHODS

147 participants (81 ALS patients and 66 healthy controls) were divided into a training sample and a validation sample. Patients in the validation sample underwent follow-up imaging longitudinally. After removing age-related variability, indices of grey and white matter integrity in ALS-specific pathognomonic brain regions were included in a cross-validated binary logistic regression model to determine the probability of individual scans indicating ALS. The following anatomical regions were assessed for diagnostic classification: average grey matter density of the left and right precentral gyrus, the average fractional anisotropy and radial diffusivity of the left and right superior corona radiata, inferior corona radiata, internal capsule, mesencephalic crus of the cerebral peduncles, pontine segment of the corticospinal tract, and the average diffusivity values of the genu, corpus and splenium of the corpus callosum.

RESULTS

Using a 50% probability cut-off value of suffering from ALS, the model was able to discriminate ALS patients and HC with good sensitivity (80.0%) and moderate accuracy (70.0%) in the training sample and superior sensitivity (85.7%) and accuracy (78.4%) in the independent validation sample.

CONCLUSIONS

This diagnostic classification study endeavours to advance ALS biomarker research towards pragmatic clinical applications by providing an approach of automated individual-data interpretation based on group-level observations.

Multimodal Classification of Schizophrenia Patients with MEG and fMRI Data Using Static and Dynamic Connectivity Measures

Mental disorders like schizophrenia are currently diagnosed by physicians/psychiatrists through clinical assessment and their evaluation of patient's self-reported experiences as the illness emerges. There is great interest in identifying biological markers of prognosis at the onset of illness, rather than relying on the evolution of symptoms across time. Functional network connectivity, which indicates a subject's overall level of "synchronicity" of activity between brain regions, demonstrates promise in providing individual subject predictive power. Many previous studies reported functional connectivity changes during resting-state using only functional magnetic resonance imaging (fMRI). Nevertheless, exclusive reliance on fMRI to generate such networks may limit the inference of the underlying dysfunctional connectivity, which is hypothesized to be a factor in patient symptoms, as fMRI measures connectivity via hemodynamics. Therefore, combination of connectivity assessments using fMRI and magnetoencephalography (MEG), which more directly measures neuronal activity, may provide improved classification of schizophrenia than either modality alone. Moreover, recent evidence indicates that metrics of dynamic connectivity may also be critical for understanding pathology in schizophrenia. In this work, we propose a new framework for extraction of important disease related features and classification of patients with schizophrenia based on using both fMRI and MEG to investigate functional network components in the resting state. Results of this study show that the integration of fMRI and MEG provides important information that captures fundamental characteristics of functional network connectivity in schizophrenia and is helpful for prediction of schizophrenia patient group membership. Combined fMRI/MEG methods, using static functional network connectivity analyses, improved classification accuracy relative to use of fMRI or MEG methods alone (by 15 and 12.45%, respectively), while combined fMRI/MEG methods using dynamic functional network connectivity analyses improved classification up to 5.12% relative to use of fMRI alone and up to 17.21% relative to use of MEG alone.

Predicting individualized clinical measures by a generalized prediction framework and multimodal fusion of MRI data

Neuroimaging techniques have greatly enhanced the understanding of neurodiversity (human brain variation across individuals) in both health and disease. The ultimate goal of using brain imaging biomarkers is to perform individualized predictions. Here we proposed a generalized framework that can predict explicit values of the targeted measures by taking advantage of joint information from multiple modalities. This framework also enables whole brain voxel-wise searching by combining multivariate techniques such as ReliefF, clustering, correlation-based feature selection and multiple regression models, which is more flexible and can achieve better prediction performance than alternative atlas-based methods. For 50 healthy controls and 47 schizophrenia patients, three kinds of features derived from resting-state fMRI (fALFF), sMRI (gray matter) and DTI (fractional anisotropy) were extracted and fed into a regression model, achieving high prediction for both cognitive scores (MCCB composite r=0.7033, MCCB social cognition r=0.7084) and symptomatic scores (positive and negative syndrome scale [PANSS] positive r=0.7785, PANSS negative r=0.7804). Moreover, the brain areas likely responsible for cognitive deficits of schizophrenia, including middle temporal gyrus, dorsolateral prefrontal cortex, striatum, cuneus and cerebellum, were located with different weights, as well as regions predicting PANSS symptoms, including thalamus, striatum and inferior parietal lobule, pinpointing the potential neuromarkers. Finally, compared to a single modality, multimodal combination achieves higher prediction accuracy and enables individualized prediction on multiple clinical measures. There is more work to be done, but the current results highlight the potential utility of multimodal brain imaging biomarkers to eventually inform clinical decision-making.

Brain CT and MRI: differential diagnosis of imaging findings

Following a traditional approach, in Chapters 5 and 14-29 in the previous volume, diverse brain diseases are listed and their imaging findings described in detail. In this chapter the approach is from the imaging finding to the disease: for instance, what list of diseases can give rise to a contrast-enhancing mass in the cerebellopontine angle? Imaging findings that are reviewed in succession include the location of the lesion, its multiplicity and symmetry, its volume, ranging from atrophy to mass effect, its homogeneity, its density, measurable by computed tomography (CT), its appearance on T1, T2, and diffusion magnetic resonance imaging (MRI), and, finally, its characteristics after the infusion of intravenous contrast. A differential diagnosis for each finding is provided. While the approach adopted in this chapter is unconventional, we hope that it will be most helpful to anyone reading images. Furthermore, it could serve as the basis to create or complete image databases to guide in the interpretation of brain CT and MRI.