Generative-discriminative basis learning for medical imaging

Deep semi-supervised multiple instance learning with self-correction for DME classification from OCT images

Supervised deep learning has achieved prominent success in various diabetic macular edema (DME) recognition tasks from optical coherence tomography (OCT) volumetric images. A common problematic issue that frequently occurs in this field is the shortage of labeled data due to the expensive fine-grained annotations, which increases substantial difficulty in accurate analysis by supervised learning. The morphological changes in the retina caused by DME might be distributed sparsely in B-scan images of the OCT volume, and OCT data is often coarsely labeled at the volume level. Hence, the DME identification task can be formulated as a multiple instance classification problem that could be addressed by multiple instance learning (MIL) techniques. Nevertheless, none of previous studies utilize unlabeled data simultaneously to promote the classification accuracy, which is particularly significant for a high quality of analysis at the minimum annotation cost. To this end, we present a novel deep semi-supervised multiple instance learning framework to explore the feasibility of leveraging a small amount of coarsely labeled data and a large amount of unlabeled data to tackle this problem. Specifically, we come up with several modules to further improve the performance according to the availability and granularity of their labels. To warm up the training, we propagate the bag labels to the corresponding instances as the supervision of training, and propose a self-correction strategy to handle the label noise in the positive bags. This strategy is based on confidence-based pseudo-labeling with consistency regularization. The model uses its prediction to generate the pseudo-label for each weakly augmented input only if it is highly confident about the prediction, which is subsequently used to supervise the same input in a strongly augmented version. This learning scheme is also applicable to unlabeled data. To enhance the discrimination capability of the model, we introduce the Student-Teacher architecture and impose consistency constraints between two models. For demonstration, the proposed approach was evaluated on two large-scale DME OCT image datasets. Extensive results indicate that the proposed method improves DME classification with the incorporation of unlabeled data and outperforms competing MIL methods significantly, which confirm the feasibility of deep semi-supervised multiple instance learning at a low annotation cost.

Machine learning in medical applications: A review of state-of-the-art methods

Applications of machine learning (ML) methods have been used extensively to solve various complex challenges in recent years in various application areas, such as medical, financial, environmental, marketing, security, and industrial applications. ML methods are characterized by their ability to examine many data and discover exciting relationships, provide interpretation, and identify patterns. ML can help enhance the reliability, performance, predictability, and accuracy of diagnostic systems for many diseases. This survey provides a comprehensive review of the use of ML in the medical field highlighting standard technologies and how they affect medical diagnosis. Five major medical applications are deeply discussed, focusing on adapting the ML models to solve the problems in cancer, medical chemistry, brain, medical imaging, and wearable sensors. Finally, this survey provides valuable references and guidance for researchers, practitioners, and decision-makers framing future research and development directions.

Machine learning methods for predicting progression from mild cognitive impairment to Alzheimer's disease dementia: a systematic review

BACKGROUND

An increase in lifespan in our society is a double-edged sword that entails a growing number of patients with neurocognitive disorders, Alzheimer's disease being the most prevalent. Advances in medical imaging and computational power enable new methods for the early detection of neurocognitive disorders with the goal of preventing or reducing cognitive decline. Computer-aided image analysis and early detection of changes in cognition is a promising approach for patients with mild cognitive impairment, sometimes a prodromal stage of Alzheimer's disease dementia.

METHODS

We conducted a systematic review following PRISMA guidelines of studies where machine learning was applied to neuroimaging data in order to predict whether patients with mild cognitive impairment might develop Alzheimer's disease dementia or remain stable. After removing duplicates, we screened 452 studies and selected 116 for qualitative analysis.

RESULTS

Most studies used magnetic resonance image (MRI) and positron emission tomography (PET) data but also magnetoencephalography. The datasets were mainly extracted from the Alzheimer's disease neuroimaging initiative (ADNI) database with some exceptions. Regarding the algorithms used, the most common was support vector machine with a mean accuracy of 75.4%, but convolutional neural networks achieved a higher mean accuracy of 78.5%. Studies combining MRI and PET achieved overall better classification accuracy than studies that only used one neuroimaging technique. In general, the more complex models such as those based on deep learning, combined with multimodal and multidimensional data (neuroimaging, clinical, cognitive, genetic, and behavioral) achieved the best performance.

CONCLUSIONS

Although the performance of the different methods still has room for improvement, the results are promising and this methodology has a great potential as a support tool for clinicians and healthcare professionals.

Fproi-GAN with Fused Regional Features for the Synthesis of High-Quality Paired Medical Images

The use of medical image synthesis with generative adversarial networks (GAN) is effective for expanding medical samples. The structural consistency between the synthesized and actual image is a key indicator of the quality of the synthesized image, and the region of interest (ROI) of the synthesized image is related to its usability, and these parameters are the two key issues in image synthesis. In this paper, the fusion-ROI patch GAN (Fproi-GAN) model was constructed by incorporating a priori regional feature based on the two-stage cycle consistency mechanism of cycleGAN. This model has improved the tissue contrast of ROI and achieved the pairwise synthesis of high-quality medical images and their corresponding ROIs. The quantitative evaluation results in two publicly available datasets, INbreast and BRATS 2017, show that the synthesized ROI images have a DICE coefficient of 0.981 ± 0.11 and a Hausdorff distance of 4.21 ± 2.84 relative to the original images. The classification experimental results show that the synthesized images can effectively assist in the training of machine learning models, improve the generalization performance of prediction models, and improve the classification accuracy by 4% and sensitivity by 5.3% compared with the cycleGAN method. Hence, the paired medical images synthesized using Fproi-GAN have high quality and structural consistency with real medical images.

Stable Anatomy Detection in Multimodal Imaging Through Sparse Group Regularization: A Comparative Study of Iron Accumulation in the Aging Brain

Multimodal neuroimaging provides a rich source of data for identifying brain regions associated with disease progression and aging. However, present studies still typically analyze modalities separately or aggregate voxel-wise measurements and analyses to the structural level, thus reducing statistical power. As a central example, previous works have used two quantitative MRI parameters-R2* and quantitative susceptibility (QS)-to study changes in iron associated with aging in healthy and multiple sclerosis subjects, but failed to simultaneously account for both. In this article, we propose a unified framework that combines information from multiple imaging modalities and regularizes estimates for increased interpretability, generalizability, and stability. Our work focuses on joint region detection problems where overlap between effect supports across modalities is encouraged but not strictly enforced. To achieve this, we combine L ₁ (lasso), total variation (TV), and L ₂ group lasso penalties. While the TV penalty encourages geometric regularization by controlling estimate variability and support boundary geometry, the group lasso penalty accounts for similarities in the support between imaging modalities. We address the computational difficulty in this regularization scheme with an alternating direction method of multipliers (ADMM) optimizer. In a neuroimaging application, we compare our method against independent sparse and joint sparse models using a dataset of R2* and QS maps derived from MRI scans of 113 healthy controls: our method produces clinically-interpretable regions where specific iron changes are associated with healthy aging. Together with results across multiple simulation studies, we conclude that our approach identifies regions that are more strongly associated with the variable of interest (e.g., age), more accurate, and more stable with respect to training data variability. This work makes progress toward a stable and interpretable multimodal imaging analysis framework for studying disease-related changes in brain structure and can be extended for classification and disease prediction tasks.

Deep virtual adversarial self-training with consistency regularization for semi-supervised medical image classification

Convolutional neural networks have achieved prominent success on a variety of medical imaging tasks when a large amount of labeled training data is available. However, the acquisition of expert annotations for medical data is usually expensive and time-consuming, which poses a great challenge for supervised learning approaches. In this work, we proposed a novel semi-supervised deep learning method, i.e., deep virtual adversarial self-training with consistency regularization, for large-scale medical image classification. To effectively exploit useful information from unlabeled data, we leverage self-training and consistency regularization to harness the underlying knowledge, which helps improve the discrimination capability of training models. More concretely, the model first uses its prediction for pseudo-labeling on the weakly-augmented input image. A pseudo-label is kept only if the corresponding class probability is of high confidence. Then the model prediction is encouraged to be consistent with the strongly-augmented version of the same input image. To improve the robustness of the network against virtual adversarial perturbed input, we incorporate virtual adversarial training (VAT) on both labeled and unlabeled data into the course of training. Hence, the network is trained by minimizing a combination of three types of losses, including a standard supervised loss on labeled data, a consistency regularization loss on unlabeled data, and a VAT loss on both labeled and labeled data. We extensively evaluate the proposed semi-supervised deep learning methods on two challenging medical image classification tasks: breast cancer screening from ultrasound images and multi-class ophthalmic disease classification from optical coherence tomography B-scan images. Experimental results demonstrate that the proposed method outperforms both supervised baseline and other state-of-the-art methods by a large margin on all tasks.

FDSR: A new fuzzy discriminative sparse representation method for medical image classification

Recent developments in medical image analysis techniques make them essential tools in medical diagnosis. Medical imaging is always involved with different kinds of uncertainties. Managing these uncertainties has motivated extensive research on medical image classification methods, particularly for the past decade. Despite being a powerful classification tool, the sparse representation suffers from the lack of sufficient discrimination and robustness, which are required to manage the uncertainty and noisiness in medical image classification issues. It is tried to overcome this deficiency by introducing a new fuzzy discriminative robust sparse representation classifier, which benefits from the fuzzy terms in its optimization function of the dictionary learning process. In this work, we present a new medical image classification approach, fuzzy discriminative sparse representation (FDSR). The proposed fuzzy terms increase the inter-class representation difference and the intra-class representation similarity. Also, an adaptive fuzzy dictionary learning approach is used to learn dictionary atoms. FDSR is applied on Magnetic Resonance Images (MRI) from three medical image databases. The comprehensive experimental results clearly show that our approach outperforms its series of rival techniques in terms of accuracy, sensitivity, specificity, and convergence speed.

Region-of-Interest based sparse feature learning method for Alzheimer's disease identification

BACKGROUND AND OBJECTIVE

In recent years, some clinical parameters, such as the volume of gray matter (GM) and cortical thickness, have been used as anatomical features to identify Alzheimer's disease (AD) from Healthy Controls (HC) in some feature-based machine learning methods. However, fewer image-based feature parameters have been proposed, which are equivalent to these clinical parameters, to describe the atrophy of regions-of-interest (ROIs) of the brain. In this study, we aim to extract effective image-based feature parameters to improve the diagnostic performance of AD with magnetic resonance imaging (MRI) data.

METHODS

A new subspace-based sparse feature learning method is proposed, which builds a union-of-subspace representation model to realize feature extraction and disease identification. Specifically, the proposed method estimates feature dimensions reasonably, at the same time, it protects local features for the specified ROIs of the brain, and realizes image-based feature extraction and classification automatically instead of computing the volume of GM or cortical thickness preliminarily.

RESULTS

Experimental results illustrate the effectiveness and robustness of the proposed method on feature extraction and classification, which are based on the sampled clinical dataset from Peking University Third Hospital of China and the Alzheimer's Disease Neuroimaging Initiative (ADNI) dataset. The extracted image-based feature parameters describe the atrophy of ROIs of the brain well as clinical parameters but show better performance in AD identification than clinical parameters. Based on them, the important ROIs for AD identification can be identified even for correlated variables.

CONCLUSION

The extracted features and the proposed identification parameters show high correlation with the volume of GM and the clinical mini-mental state examination (MMSE) score respectively. The proposed method will be useful in denoting the changes of cerebral pathology and cognitive function in AD patients.

A joint network optimization framework to predict clinical severity from resting state functional MRI data

We propose a novel optimization framework to predict clinical severity from resting state fMRI (rs-fMRI) data. Our model consists of two coupled terms. The first term decomposes the correlation matrices into a sparse set of representative subnetworks that define a network manifold. These subnetworks are modeled as rank-one outer-products which correspond to the elemental patterns of co-activation across the brain; the subnetworks are combined via patient-specific non-negative coefficients. The second term is a linear regression model that uses the patient-specific coefficients to predict a measure of clinical severity. We validate our framework on two separate datasets in a ten fold cross validation setting. The first is a cohort of fifty-eight patients diagnosed with Autism Spectrum Disorder (ASD). The second dataset consists of sixty three patients from a publicly available ASD database. Our method outperforms standard semi-supervised frameworks, which employ conventional graph theoretic and statistical representation learning techniques to relate the rs-fMRI correlations to behavior. In contrast, our joint network optimization framework exploits the structure of the rs-fMRI correlation matrices to simultaneously capture group level effects and patient heterogeneity. Finally, we demonstrate that our proposed framework robustly identifies clinically relevant networks characteristic of ASD.

Refined measure of functional connectomes for improved identifiability and prediction

Brain functional connectome analysis is commonly based on population-wise inference. However, in this way precious information provided at the individual subject level may be overlooked. Recently, several studies have shown that individual differences contribute strongly to the functional connectivity patterns. In particular, functional connectomes have been proven to offer a fingerprint measure, which can reliably identify a given individual from a pool of participants. In this work, we propose to refine the standard measure of individual functional connectomes using dictionary learning. More specifically, we rely on the assumption that each functional connectivity is dominated by stable group and individual factors. By subtracting population-wise contributions from connectivity patterns facilitated by dictionary representation, intersubject variability should be increased within the group. We validate our approach using several types of analyses. For example, we observe that refined connectivity profiles significantly increase subject-specific identifiability across functional magnetic resonance imaging (fMRI) session combinations. Besides, refined connectomes can also improve the prediction power for cognitive behaviors. In accordance with results from the literature, we find that individual distinctiveness is closely linked with differences in neurocognitive activity within the brain. In summary, our results indicate that individual connectivity analysis benefits from the group-wise inferences and refined connectomes are indeed desirable for brain mapping.

Machine learning in major depression: From classification to treatment outcome prediction

AIMS

Major depression disorder (MDD) is the single greatest cause of disability and morbidity, and affects about 10% of the population worldwide. Currently, there are no clinically useful diagnostic biomarkers that are able to confirm a diagnosis of MDD from bipolar disorder (BD) in the early depressive episode. Therefore, exploring translational biomarkers of mood disorders based on machine learning is in pressing need, though it is challenging, but with great potential to improve our understanding of these disorders.

DISCUSSIONS

In this study, we review popular machine-learning methods used for brain imaging classification and predictions, and provide an overview of studies, specifically for MDD, that have used magnetic resonance imaging data to either (a) classify MDDs from controls or other mood disorders or (b) investigate treatment outcome predictors for individual patients. Finally, challenges, future directions, and potential limitations related to MDD biomarker identification are also discussed, with a goal of offering a comprehensive overview that may help readers to better understand the applications of neuroimaging data mining in depression.

CONCLUSIONS

We hope such efforts may highlight the need for an urgently needed paradigm shift in treatment, to guide personalized optimal clinical care.

Significant Anatomy Detection Through Sparse Classification: A Comparative Study

We present a comparative study for discriminative anatomy detection in high dimensional neuroimaging data. While most studies solve this problem using mass univariate approaches, recent works show better accuracy and variable selection using a sparse classification model. Two types of image-based regularization methods have been proposed in the literature based on either a Graph Net (GN) model or a total variation (TV) model. These studies showed increased classification accuracy and interpretability of results when using image-based regularization, but did not look at the accuracy and quality of the recovered significant regions. In this paper, we theoretically prove bounds on the recovered sparse coefficients and the corresponding selected image regions in four models (two based on GN penalty and two based on TV penalty). Practically, we confirm the theoretical findings by measuring the accuracy of selected regions compared with ground truth on simulated data. We also evaluate the stability of recovered regions over cross-validation folds using real MRI data. Our findings show that the TV penalty is superior to the GN model. In addition, we showed that adding an l₂ penalty improves the accuracy of estimated coefficients and selected significant regions for the both types of models.

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

Neuroimaging-based single subject prediction of brain disorders has gained increasing attention in recent years. Using a variety of neuroimaging modalities such as structural, functional and diffusion MRI, along with machine learning techniques, hundreds of studies have been carried out for accurate classification of patients with heterogeneous mental and neurodegenerative disorders such as schizophrenia and Alzheimer's disease. More than 500 studies have been published during the past quarter century on single subject prediction focused on a multiple brain disorders. In the first part of this study, we provide a survey of more than 200 reports in this field with a focus on schizophrenia, mild cognitive impairment (MCI), Alzheimer's disease (AD), depressive disorders, autism spectrum disease (ASD) and attention-deficit hyperactivity disorder (ADHD). Detailed information about those studies such as sample size, type and number of extracted features and reported accuracy are summarized and discussed. To our knowledge, this is by far the most comprehensive review of neuroimaging-based single subject prediction of brain disorders. In the second part, we present our opinion on major pitfalls of those studies from a machine learning point of view. Common biases are discussed and suggestions are provided. Moreover, emerging trends such as decentralized data sharing, multimodal brain imaging, differential diagnosis, disease subtype classification and deep learning are also discussed. Based on this survey, there is extensive evidence showing the great potential of neuroimaging data for single subject prediction of various disorders. However, the main bottleneck of this exciting field is still the limited sample size, which could be potentially addressed by modern data sharing models such as the ones discussed in this paper. Emerging big data technologies and advanced data-intensive machine learning methodologies such as deep learning have coincided with an increasing need for accurate, robust and generalizable single subject prediction of brain disorders during an exciting time. In this report, we survey the past and offer some opinions regarding the road ahead.

Probabilistic Modeling of Imaging, Genetics and Diagnosis

We propose a unified Bayesian framework for detecting genetic variants associated with disease by exploiting image-based features as an intermediate phenotype. The use of imaging data for examining genetic associations promises new directions of analysis, but currently the most widely used methods make sub-optimal use of the richness that these data types can offer. Currently, image features are most commonly selected based on their relevance to the disease phenotype. Then, in a separate step, a set of genetic variants is identified to explain the selected features. In contrast, our method performs these tasks simultaneously in order to jointly exploit information in both data types. The analysis yields probabilistic measures of clinical relevance for both imaging and genetic markers. We derive an efficient approximate inference algorithm that handles the high dimensionality of image and genetic data. We evaluate the algorithm on synthetic data and demonstrate that it outperforms traditional models. We also illustrate our method on Alzheimer's Disease Neuroimaging Initiative data.

Computational neuroanatomy using brain deformations: From brain parcellation to multivariate pattern analysis and machine learning

The past 20 years have seen a mushrooming growth of the field of computational neuroanatomy. Much of this work has been enabled by the development and refinement of powerful, high-dimensional image warping methods, which have enabled detailed brain parcellation, voxel-based morphometric analyses, and multivariate pattern analyses using machine learning approaches. The evolution of these 3 types of analyses over the years has overcome many challenges. We present the evolution of our work in these 3 directions, which largely follows the evolution of this field. We discuss the progression from single-atlas, single-registration brain parcellation work to current ensemble-based parcellation; from relatively basic mass-univariate t-tests to optimized regional pattern analyses combining deformations and residuals; and from basic application of support vector machines to generative-discriminative formulations of multivariate pattern analyses, and to methods dealing with heterogeneity of neuroanatomical patterns. We conclude with discussion of some of the future directions and challenges.

A Generative Probabilistic Model and Discriminative Extensions for Brain Lesion Segmentation--With Application to Tumor and Stroke

We introduce a generative probabilistic model for segmentation of brain lesions in multi-dimensional images that generalizes the EM segmenter, a common approach for modelling brain images using Gaussian mixtures and a probabilistic tissue atlas that employs expectation-maximization (EM), to estimate the label map for a new image. Our model augments the probabilistic atlas of the healthy tissues with a latent atlas of the lesion. We derive an estimation algorithm with closed-form EM update equations. The method extracts a latent atlas prior distribution and the lesion posterior distributions jointly from the image data. It delineates lesion areas individually in each channel, allowing for differences in lesion appearance across modalities, an important feature of many brain tumor imaging sequences. We also propose discriminative model extensions to map the output of the generative model to arbitrary labels with semantic and biological meaning, such as "tumor core" or "fluid-filled structure", but without a one-to-one correspondence to the hypo- or hyper-intense lesion areas identified by the generative model. We test the approach in two image sets: the publicly available BRATS set of glioma patient scans, and multimodal brain images of patients with acute and subacute ischemic stroke. We find the generative model that has been designed for tumor lesions to generalize well to stroke images, and the extended discriminative -discriminative model to be one of the top ranking methods in the BRATS evaluation.

Landmark Based Shape Analysis for Cerebellar Ataxia Classification and Cerebellar Atrophy Pattern Visualization

Cerebellar dysfunction can lead to a wide range of movement disorders. Studying the cerebellar atrophy pattern associated with different cerebellar disease types can potentially help in diagnosis, prognosis, and treatment planning. In this paper, we present a landmark based shape analysis pipeline to classify healthy control and different ataxia types and to visualize the characteristic cerebellar atrophy patterns associated with different types. A highly informative feature representation of the cerebellar structure is constructed by extracting dense homologous landmarks on the boundary surfaces of cerebellar sub-structures. A diagnosis group classifier based on this representation is built using partial least square dimension reduction and regularized linear discriminant analysis. The characteristic atrophy pattern for an ataxia type is visualized by sampling along the discriminant direction between healthy controls and the ataxia type. Experimental results show that the proposed method can successfully classify healthy controls and different ataxia types. The visualized cerebellar atrophy patterns were consistent with the regional volume decreases observed in previous studies, but the proposed method provides intuitive and detailed understanding about changes of overall size and shape of the cerebellum, as well as that of individual lobules.

On characterizing population commonalities and subject variations in brain networks

Brain networks based on resting state connectivity as well as inter-regional anatomical pathways obtained using diffusion imaging have provided insight into pathology and development. Such work has underscored the need for methods that can extract sub-networks that can accurately capture the connectivity patterns of the underlying population while simultaneously describing the variation of sub-networks at the subject level. We have designed a multi-layer graph clustering method that extracts clusters of nodes, called 'network hubs', which display higher levels of connectivity within the cluster than to the rest of the brain. The method determines an atlas of network hubs that describes the population, as well as weights that characterize subject-wise variation in terms of within- and between-hub connectivity. This lowers the dimensionality of brain networks, thereby providing a representation amenable to statistical analyses. The applicability of the proposed technique is demonstrated by extracting an atlas of network hubs for a population of typically developing controls (TDCs) as well as children with autism spectrum disorder (ASD), and using the structural and functional networks of a population to determine the subject-level variation of these hubs and their inter-connectivity. These hubs are then used to compare ASD and TDCs. Our method is generalizable to any population whose connectivity (structural or functional) can be captured via non-negative network graphs.

Interpreting support vector machine models for multivariate group wise analysis in neuroimaging

Machine learning based classification algorithms like support vector machines (SVMs) have shown great promise for turning a high dimensional neuroimaging data into clinically useful decision criteria. However, tracing imaging based patterns that contribute significantly to classifier decisions remains an open problem. This is an issue of critical importance in imaging studies seeking to determine which anatomical or physiological imaging features contribute to the classifier's decision, thereby allowing users to critically evaluate the findings of such machine learning methods and to understand disease mechanisms. The majority of published work addresses the question of statistical inference for support vector classification using permutation tests based on SVM weight vectors. Such permutation testing ignores the SVM margin, which is critical in SVM theory. In this work we emphasize the use of a statistic that explicitly accounts for the SVM margin and show that the null distributions associated with this statistic are asymptotically normal. Further, our experiments show that this statistic is a lot less conservative as compared to weight based permutation tests and yet specific enough to tease out multivariate patterns in the data. Thus, we can better understand the multivariate patterns that the SVM uses for neuroimaging based classification.

Randomized structural sparsity via constrained block subsampling for improved sensitivity of discriminative voxel identification

In this paper, we consider voxel selection for functional Magnetic Resonance Imaging (fMRI) brain data with the aim of finding a more complete set of probably correlated discriminative voxels, thus improving interpretation of the discovered potential biomarkers. The main difficulty in doing this is an extremely high dimensional voxel space and few training samples, resulting in unreliable feature selection. In order to deal with the difficulty, stability selection has received a great deal of attention lately, especially due to its finite sample control of false discoveries and transparent principle for choosing a proper amount of regularization. However, it fails to make explicit use of the correlation property or structural information of these discriminative features and leads to large false negative rates. In other words, many relevant but probably correlated discriminative voxels are missed. Thus, we propose a new variant on stability selection "randomized structural sparsity", which incorporates the idea of structural sparsity. Numerical experiments demonstrate that our method can be superior in controlling for false negatives while also keeping the control of false positives inherited from stability selection.

Morphological covariance in anatomical MRI scans can identify discrete neural pathways in the brain and their disturbances in persons with neuropsychiatric disorders

We hypothesize that coordinated functional activity within discrete neural circuits induces morphological organization and plasticity within those circuits. Identifying regions of morphological covariation that are independent of morphological covariation in other regions therefore may therefore allow us to identify discrete neural systems within the brain. Comparing the magnitude of these variations in individuals who have psychiatric disorders with the magnitude of variations in healthy controls may allow us to identify aberrant neural pathways in psychiatric illnesses. We measured surface morphological features by applying nonlinear, high-dimensional warping algorithms to manually defined brain regions. We transferred those measures onto the surface of a unit sphere via conformal mapping and then used spherical wavelets and their scaling coefficients to simplify the data structure representing these surface morphological features of each brain region. We used principal component analysis (PCA) to calculate covariation in these morphological measures, as represented by their scaling coefficients, across several brain regions. We then assessed whether brain subregions that covaried in morphology, as identified by large eigenvalues in the PCA, identified specific neural pathways of the brain. To do so, we spatially registered the subnuclei for each eigenvector into the coordinate space of a Diffusion Tensor Imaging dataset; we used these subnuclei as seed regions to track and compare fiber pathways with known fiber pathways identified in neuroanatomical atlases. We applied these procedures to anatomical MRI data in a cohort of 82 healthy participants (42 children, 18 males, age 10.5 ± 2.43 years; 40 adults, 22 males, age 32.42 ± 10.7 years) and 107 participants with Tourette's Syndrome (TS) (71 children, 59 males, age 11.19 ± 2.2 years; 36 adults, 21 males, age 37.34 ± 10.9 years). We evaluated the construct validity of the identified covariation in morphology using DTI data from a different set of 20 healthy adults (10 males, mean age 29.7 ± 7.7 years). The PCA identified portions of structures that covaried across the brain, the eigenvalues measuring the magnitude of the covariation in morphology along the respective eigenvectors. Our results showed that the eigenvectors, and the DTI fibers tracked from their associated brain regions, corresponded with known neural pathways in the brain. In addition, the eigenvectors that captured morphological covariation across regions, and the principal components along those eigenvectors, identified neural pathways with aberrant morphological features associated with TS. These findings suggest that covariations in brain morphology can identify aberrant neural pathways in specific neuropsychiatric disorders.

Supervised block sparse dictionary learning for simultaneous clustering and classification in computational anatomy

An important prerequisite for computational neuroanatomy is the spatial normalization of the data. Despite its importance for the success of the subsequent statistical analysis, image alignment is dealt with from the perspective of image matching, while its influence on the group analysis is neglected. The choice of the template, the registration algorithm as well as the registration parameters, all confound group differences and impact the outcome of the analysis. In order to limit their influence, we perform multiple registrations by varying these parameters, resulting in multiple instances for each sample. In order to harness the high dimensionality of the data and emphasize the group differences, we propose a supervised dimensionality reduction technique that takes into account the organization of the data. This is achieved by solving a supervised dictionary learning problem for block-sparse signals. Structured sparsity allows the grouping of instances across different independent samples, while label supervision allows for discriminative dictionaries. The block structure of dictionaries allows constructing multiple classifiers that treat each dictionary block as a basis of a subspace that spans a separate band of information. We formulate this problem as a convex optimization problem with a geometric programming (GP) component. Promising results that demonstrate the potential of the proposed approach are shown for an MR image dataset of Autism subjects.

Mapping brains on grids of features for schizophrenia analysis

This paper exploits the embedding provided by the counting grid model and proposes a framework for the classification and the analysis of brain MRI images. Each brain, encoded by a count of local features, is mapped into a window on a grid of feature distributions. Similar sample are mapped in close proximity on the grid and their commonalities in their feature distributions are reflected in the overlap of windows on the grid. Here we exploited these properties to design a novel kernel and a visualization strategy which we applied to the analysis of schizophrenic patients. Experiments report a clear improvement in classification accuracy as compared with similar methods. Moreover, our visualizations are able to highlight brain clusters and to obtain a visual interpretation of the features related to the disease.

Cortical surface biomarkers for predicting cognitive outcomes using group l2,1 norm

Regression models have been widely studied to investigate the prediction power of neuroimaging measures as biomarkers for inferring cognitive outcomes in the Alzheimer's disease study. Most of these models ignore the interrelated structures either within neuroimaging measures or between cognitive outcomes, and thus may have limited power to yield optimal solutions. To address this issue, we propose to use a new sparse multitask learning model called Group-Sparse Multi-task Regression and Feature Selection (G-SMuRFS) and demonstrate its effectiveness by examining the predictive power of detailed cortical thickness measures toward 3 types of cognitive scores in a large cohort. G-SMuRFS proposes a group-level l2,1-norm strategy to group relevant features together in an anatomically meaningful manner and use this prior knowledge to guide the learning process. This approach also takes into account the correlation among cognitive outcomes for building a more appropriate predictive model. Compared with traditional methods, G-SMuRFS not only demonstrates a superior performance but also identifies a small set of surface markers that are biologically meaningful.

Finding imaging patterns of structural covariance via Non-Negative Matrix Factorization

In this paper, we investigate the use of Non-Negative Matrix Factorization (NNMF) for the analysis of structural neuroimaging data. The goal is to identify the brain regions that co-vary across individuals in a consistent way, hence potentially being part of underlying brain networks or otherwise influenced by underlying common mechanisms such as genetics and pathologies. NNMF offers a directly data-driven way of extracting relatively localized co-varying structural regions, thereby transcending limitations of Principal Component Analysis (PCA), Independent Component Analysis (ICA) and other related methods that tend to produce dispersed components of positive and negative loadings. In particular, leveraging upon the well known ability of NNMF to produce parts-based representations of image data, we derive decompositions that partition the brain into regions that vary in consistent ways across individuals. Importantly, these decompositions achieve dimensionality reduction via highly interpretable ways and generalize well to new data as shown via split-sample experiments. We empirically validate NNMF in two data sets: i) a Diffusion Tensor (DT) mouse brain development study, and ii) a structural Magnetic Resonance (sMR) study of human brain aging. We demonstrate the ability of NNMF to produce sparse parts-based representations of the data at various resolutions. These representations seem to follow what we know about the underlying functional organization of the brain and also capture some pathological processes. Moreover, we show that these low dimensional representations favorably compare to descriptions obtained with more commonly used matrix factorization methods like PCA and ICA.

Identifying group discriminative and age regressive sub-networks from DTI-based connectivity via a unified framework of non-negative matrix factorization and graph embedding

Diffusion tensor imaging (DTI) offers rich insights into the physical characteristics of white matter (WM) fiber tracts and their development in the brain, facilitating a network representation of brain's traffic pathways. Such a network representation of brain connectivity has provided a novel means of investigating brain changes arising from pathology, development or aging. The high dimensionality of these connectivity networks necessitates the development of methods that identify the connectivity building blocks or sub-network components that characterize the underlying variation in the population. In addition, the projection of the subject networks into the basis set provides a low dimensional representation of it, that teases apart different sources of variation in the sample, facilitating variation-specific statistical analysis. We propose a unified framework of non-negative matrix factorization and graph embedding for learning sub-network patterns of connectivity by their projective non-negative decomposition into a reconstructive basis set, as well as, additional basis sets representing variational sources in the population like age and pathology. The proposed framework is applied to a study of diffusion-based connectivity in subjects with autism that shows localized sparse sub-networks which mostly capture the changes related to pathology and developmental variations.

Eigenanatomy: sparse dimensionality reduction for multi-modal medical image analysis

Rigorous statistical analysis of multimodal imaging datasets is challenging. Mass-univariate methods for extracting correlations between image voxels and outcome measurements are not ideal for multimodal datasets, as they do not account for interactions between the different modalities. The extremely high dimensionality of medical images necessitates dimensionality reduction, such as principal component analysis (PCA) or independent component analysis (ICA). These dimensionality reduction techniques, however, consist of contributions from every region in the brain and are therefore difficult to interpret. Recent advances in sparse dimensionality reduction have enabled construction of a set of image regions that explain the variance of the images while still maintaining anatomical interpretability. The projections of the original data on the sparse eigenvectors, however, are highly collinear and therefore difficult to incorporate into multi-modal image analysis pipelines. We propose here a method for clustering sparse eigenvectors and selecting a subset of the eigenvectors to make interpretable predictions from a multi-modal dataset. Evaluation on a publicly available dataset shows that the proposed method outperforms PCA and ICA-based regressions while still maintaining anatomical meaning. To facilitate reproducibility, the complete dataset used and all source code is publicly available.

Identifying Sparse Connectivity Patterns in the brain using resting-state fMRI

The human brain processes information via multiple distributed networks. An accurate model of the brain's functional connectome is critical for understanding both normal brain function as well as the dysfunction present in neuropsychiatric illnesses. Current methodologies that attempt to discover the organization of the functional connectome typically assume spatial or temporal separation of the underlying networks. This assumption deviates from an intuitive understanding of brain function, which is that of multiple, inter-dependent spatially overlapping brain networks that efficiently integrate information pertinent to diverse brain functions. It is now increasingly evident that neural systems use parsimonious formations and functional representations to efficiently process information while minimizing redundancy. Hence we exploit recent advances in the mathematics of sparse modeling to develop a methodological framework aiming to understand complex resting-state fMRI connectivity data. By favoring networks that explain the data via a relatively small number of participating brain regions, we obtain a parsimonious representation of brain function in terms of multiple "Sparse Connectivity Patterns" (SCPs), such that differential presence of these SCPs explains inter-subject variability. In this manner the sparsity-based framework can effectively capture the heterogeneity of functional activity patterns across individuals while potentially highlighting multiple sub-populations within the data that display similar patterns. Our results from simulated as well as real resting state fMRI data show that SCPs are accurate and reproducible between sub-samples as well as across datasets. These findings substantiate existing knowledge of intrinsic functional connectivity and provide novel insights into the functional organization of the human brain.

Unsupervised learning of functional network dynamics in resting state fMRI

Research in recent years has provided some evidence of temporal non-stationarity of functional connectivity in resting state fMRI. In this paper, we present a novel methodology that can decode connectivity dynamics into a temporal sequence of hidden network "states" for each subject, using a Hidden Markov Modeling (HMM) framework. Each state is characterized by a unique covariance matrix or whole-brain network. Our model generates these covariance matrices from a common but unknown set of sparse basis networks, which capture the range of functional activity co-variations of regions of interest (ROIs). Distinct hidden states arise due to a variation in the strengths of these basis networks. Thus, our generative model combines a HMM framework with sparse basis learning of positive definite matrices. Results on simulated fMRI data show that our method can effectively recover underlying basis networks as well as hidden states. We apply this method on a normative dataset of resting state fMRI scans. Results indicate that the functional activity of a subject at any point during the scan is composed of combinations of overlapping task-positive/negative pairs of networks as revealed by our basis. Distinct hidden temporal states are produced due to a different set of basis networks dominating the covariance pattern in each state.

Hierarchical manifold learning for regional image analysis

We present a novel method of hierarchical manifold learning which aims to automatically discover regional properties of image datasets. While traditional manifold learning methods have become widely used for dimensionality reduction in medical imaging, they suffer from only being able to consider whole images as single data points. We extend conventional techniques by additionally examining local variations, in order to produce spatially-varying manifold embeddings that characterize a given dataset. This involves constructing manifolds in a hierarchy of image patches of increasing granularity, while ensuring consistency between hierarchy levels. We demonstrate the utility of our method in two very different settings: 1) to learn the regional correlations in motion within a sequence of time-resolved MR images of the thoracic cavity; 2) to find discriminative regions of 3-D brain MR images associated with neurodegenerative disease.

Discriminative sparse connectivity patterns for classification of fMRI Data

Functional connectivity using resting-state fMRI has emerged as an important research tool for understanding normal brain function as well as changes occurring during brain development and in various brain disorders. Most prior work has examined changes in pairwise functional connectivity values using a multi-variate classification approach, such as Support Vector Machines (SVM). While it is powerful, SVMs produce a dense set of high-dimensional weight vectors as output, which are difficult to interpret, and require additional post-processing to relate to known functional networks. In this paper, we propose a joint framework that combines network identification and classification, resulting in a set of networks, or Sparse Connectivity Patterns (SCPs) which are functionally interpretable as well as highly discriminative of the two groups. Applied to a study of normal development classifying children vs. adults, the proposed method provided accuracy of 76%(AUC= 0.85), comparable to SVM (79%,AUC=0.87), but with dramatically fewer number of features (50 features vs. 34716 for the SVM). More importantly, this leads to a tremendous improvement in neuro-scientific interpretability, which is specially advantageous in such a study where the group differences are wide-spread throughout the brain. Highest-ranked discriminative SCPs reflect increases in long-range connectivity in adults between the frontal areas and posterior cingulate regions. In contrast, connectivity between the bilateral parahippocampal gyri was decreased in adults compared to children.

Deep Learning for Cerebellar Ataxia Classification and Functional Score Regression

Cerebellar ataxia is a progressive neuro-degenerative disease that has multiple genetic versions, each with a characteristic pattern of anatomical degeneration that yields distinctive motor and cognitive problems. Studying this pattern of degeneration can help with the diagnosis of disease subtypes, evaluation of disease stage, and treatment planning. In this work, we propose a learning framework using MR image data for discriminating a set of cerebellar ataxia types and predicting a disease related functional score. We address the difficulty in analyzing high-dimensional image data with limited training subjects by: 1) training weak classifiers/regressors on a set of image subdomains separately, and combining the weak classifier/regressor outputs to make the decision; 2) perturbing the image subdomain to increase the training samples; 3) using a deep learning technique called the stacked auto-encoder to develop highly representative feature vectors of the input data. Experiments show that our approach can reliably classify between one of four categories (healthy control and three types of ataxia), and predict the functional staging score for ataxia.

Analytic estimation of statistical significance maps for support vector machine based multi-variate image analysis and classification

Multivariate pattern analysis (MVPA) methods such as support vector machines (SVMs) have been increasingly applied to fMRI and sMRI analyses, enabling the detection of distinctive imaging patterns. However, identifying brain regions that significantly contribute to the classification/group separation requires computationally expensive permutation testing. In this paper we show that the results of SVM-permutation testing can be analytically approximated. This approximation leads to more than a thousandfold speedup of the permutation testing procedure, thereby rendering it feasible to perform such tests on standard computers. The speedup achieved makes SVM based group difference analysis competitive with standard univariate group difference analysis methods.

IDENTIFYING PATTERNS IN TEMPORAL VARIATION OF FUNCTIONAL CONNECTIVITY USING RESTING STATE FMRI

Estimating functional brain networks from fMRI data has been the focus of much research in recent years. Low sample sizes (time-points) and high dimensionality of fMRI has restricted estimation to a temporally averaged connectivity matrix per subject, due to which the dynamics of functional connectivity is largely unknown. In this paper, we propose a novel method based on constrained matrix factorization that addresses two major issues. Firstly, it finds a set of basis networks that are the semantic parts of the time-varying whole-brain functional networks. The whole-brain network at any point in time, for any subject, is a non-negative combination of these basis networks. Secondly, significant dimensionality reduction is achieved by projecting the data onto this basis, facilitating subsequent analysis of temporal dynamics. Results on simulated fMRI data show that our method can effectively recover underlying basis networks. We apply this method on a normative dataset of resting state fMRI scans. Results indicate that the functional connectivity of a subject at any point during the scan is composed of combinations of overlapping task-positive/negative pairs of sub-networks.

Applying tensor-based morphometry to parametric surfaces can improve MRI-based disease diagnosis

Many methods have been proposed for computer-assisted diagnostic classification. Full tensor information and machine learning with 3D maps derived from brain images may help detect subtle differences or classify subjects into different groups. Here we develop a new approach to apply tensor-based morphometry to parametric surface models for diagnostic classification. We use this approach to identify cortical surface features for use in diagnostic classifiers. First, with holomorphic 1-forms, we compute an efficient and accurate conformal mapping from a multiply connected mesh to the so-called slit domain. Next, the surface parameterization approach provides a natural way to register anatomical surfaces across subjects using a constrained harmonic map. To analyze anatomical differences, we then analyze the full Riemannian surface metric tensors, which retain multivariate information on local surface geometry. As the number of voxels in a 3D image is large, sparse learning is a promising method to select a subset of imaging features and to improve classification accuracy. Focusing on vertices with greatest effect sizes, we train a diagnostic classifier using the surface features selected by an L1-norm based sparse learning method. Stability selection is applied to validate the selected feature sets. We tested the algorithm on MRI-derived cortical surfaces from 42 subjects with genetically confirmed Williams syndrome and 40 age-matched controls, multivariate statistics on the local tensors gave greater effect sizes for detecting group differences relative to other TBM-based statistics including analysis of the Jacobian determinant and the largest eigenvalue of the surface metric. Our method also gave reasonable classification results relative to the Jacobian determinant, the pair of eigenvalues of the Jacobian matrix and volume features. This analysis pipeline may boost the power of morphometry studies, and may assist with image-based classification.

Bridging paradigms: hybrid mechanistic-discriminative predictive models

Many disease processes are extremely complex and characterized by multiple stochastic processes interacting simultaneously. Current analytical approaches have included mechanistic models and machine learning (ML), which are often treated as orthogonal viewpoints. However, to facilitate truly personalized medicine, new perspectives may be required. This paper reviews the use of both mechanistic models and ML in healthcare as well as emerging hybrid methods, which are an exciting and promising approach for biologically based, yet data-driven advanced intelligent systems.

Locality preserving non-negative basis learning with graph embedding

The high dimensionality of connectivity networks necessitates the development of methods identifying the connectivity building blocks that not only characterize the patterns of brain pathology but also reveal representative population patterns. In this paper, we present a non-negative component analysis framework for learning localized and sparse sub-network patterns of connectivity matrices by decomposing them into two sets of discriminative and reconstructive bases. In order to obtain components that are designed towards extracting population differences, we exploit the geometry of the population by using a graphtheoretical scheme that imposes locality-preserving properties as well as maintaining the underlying distance between distant nodes in the original and the projected space. The effectiveness of the proposed framework is demonstrated by applying it to two clinical studies using connectivity matrices derived from DTI to study a population of subjects with ASD, as well as a developmental study of structural brain connectivity that extracts gender differences.

Joint modeling of imaging and genetics

We propose a unified Bayesian framework for detecting genetic variants associated with a disease while exploiting image-based features as an intermediate phenotype. Traditionally, imaging genetics methods comprise two separate steps. First, image features are selected based on their relevance to the disease phenotype. Second, a set of genetic variants are identified to explain the selected features. In contrast, our method performs these tasks simultaneously to ultimately assign probabilistic measures of relevance to both genetic and imaging markers. We derive an efficient approximate inference algorithm that handles high dimensionality of imaging genetic data. We evaluate the algorithm on synthetic data and show that it outperforms traditional models. We also illustrate the application of the method on ADNI data.

The relevance voxel machine (RVoxM): a self-tuning Bayesian model for informative image-based prediction

This paper presents the relevance voxel machine (RVoxM), a dedicated Bayesian model for making predictions based on medical imaging data. In contrast to the generic machine learning algorithms that have often been used for this purpose, the method is designed to utilize a small number of spatially clustered sets of voxels that are particularly suited for clinical interpretation. RVoxM automatically tunes all its free parameters during the training phase, and offers the additional advantage of producing probabilistic prediction outcomes. We demonstrate RVoxM as a regression model by predicting age from volumetric gray matter segmentations, and as a classification model by distinguishing patients with Alzheimer's disease from healthy controls using surface-based cortical thickness data. Our results indicate that RVoxM yields biologically meaningful models, while providing state-of-the-art predictive accuracy.

Dominant component analysis of electrophysiological connectivity networks

Connectivity matrices obtained from various modalities (DTI, MEG and fMRI) provide a unique insight into brain processes. Their high dimensionality necessitates the development of methods for population-based statistics, in the face of small sample sizes. In this paper, we present such a method applicable to functional connectivity networks, based on identifying the basis of dominant connectivity components that characterize the patterns of brain pathology and population variation. Projection of individual connectivity matrices into this basis allows for dimensionality reduction, facilitating subsequent statistical analysis. We find dominant components for a collection of connectivity matrices by using the projective non-negative component analysis technique which ensures that the components have non-negative elements and are non-negatively combined to obtain individual subject networks, facilitating interpretation. We demonstrate the feasibility of our novel framework by applying it to simulated connectivity matrices as well as to a clinical study using connectivity matrices derived from resting state magnetoencephalography (MEG) data in a population of subjects diagnosed with autism spectrum disorder (ASD).