•Overview of deep learning basic concepts: architecture, learning and testing.•Literature review of deep learning in neuroimaging studies of brain-based disorders.•Discussion about future research ...and challenges of deep learning in neuroimaging.
Deep learning (DL) is a family of machine learning methods that has gained considerable attention in the scientific community, breaking benchmark records in areas such as speech and visual recognition. DL differs from conventional machine learning methods by virtue of its ability to learn the optimal representation from the raw data through consecutive nonlinear transformations, achieving increasingly higher levels of abstraction and complexity. Given its ability to detect abstract and complex patterns, DL has been applied in neuroimaging studies of psychiatric and neurological disorders, which are characterised by subtle and diffuse alterations. Here we introduce the underlying concepts of DL and review studies that have used this approach to classify brain-based disorders. The results of these studies indicate that DL could be a powerful tool in the current search for biomarkers of psychiatric and neurologic disease. We conclude our review by discussing the main promises and challenges of using DL to elucidate brain-based disorders, as well as possible directions for future research.
Cognitive dysfunction is one of the most prominent characteristics of psychiatric disorders. Currently, the neural correlates of cognitive dysfunction across psychiatric disorders are poorly ...understood. The aim of this study was to investigate functional connectivity and structural perturbations across psychiatric diagnoses in three neurocognitive networks of interest: the default mode network (DMN), the frontoparietal network (FPN), and the salience network (SN).
We performed meta-analyses of resting-state functional magnetic resonance imaging whole-brain seed-based functional connectivity in 8298 patients (involving eight disorders) and 8165 healthy control subjects and a voxel-based morphometry analysis of structural magnetic resonance imaging data in 14,027 patients (involving eight disorders) and 14,504 healthy control subjects. To aid the interpretation of the results, we examined neurocognitive function in 776 healthy participants from the Human Connectome Project.
We found that the three neurocognitive networks of interest were characterized by shared alterations of functional connectivity architecture across psychiatric disorders. More specifically, hypoconnectivity was expressed between the DMN and ventral SN and between the SN and FPN, whereas hyperconnectivity was evident between the DMN and FPN and between the DMN and dorsal SN. This pattern of network alterations was associated with gray matter reductions in patients and was localized in regions that subserve general cognitive performance.
This study is the first to provide meta-analytic evidence of common alterations of functional connectivity within and between neurocognitive networks. The findings suggest a shared mechanism of network interactions that may associate with the generalized cognitive deficits observed in psychiatric disorders.
Abstract
Objective
Increasingly, studies have identified abnormalities in the functional connectivity (FC) of large-scale neural networks in early psychosis, but the findings thus far have been ...inconclusive. Therefore, the aim of this study was to identify robust alterations in FC of the default mode (DMN), salience (SN), and central executive networks (CEN), in patients with first-episode psychosis (FEP) using a meta-analytic approach.
Methods
Included studies were required to be resting-state, seed-to-whole brain, FC neuroimaging studies, comparing FEP patients to healthy controls (HC), with seeds within the boundaries of the region-of-interest networks. Peak effect coordinates and peak t, z, or p values were meta-analyzed using Seed-based d Mapping software.
Results
The DMN seeds primarily displayed within-network hypoconnectivity (largest clusters including the middle orbital gyrus; and ventral anterior cingulate gyrus). The SN seeds displayed hypoconnectivity with regions in the DMN and CEN (largest clusters located in the bilateral middle temporal gyri). Review of the limited CEN data revealed hypo- and hyperconnectivity across the networks. Negative symptoms were positively correlated with all DMN FC abnormalities in the FEP group. Antipsychotic-treated patients displayed greater hypoconnectivity than antipsychotic-naïve patients between both the DMN/SN seeds and prefrontal regions.
Conclusions
These findings provide substantial evidence of widespread resting-state FC abnormalities of the DMN, SN, and CEN in early psychosis; particularly implicating DMN and SN dysconnectivity as a core deficit underlying the psychopathology of psychosis. Additionally, we highlight the importance of disentangling connectivity abnormalities resulting from disease processes, from those that result from antipsychotic treatment.
The disconnection hypothesis suggests that the core symptoms of schizophrenia (SZ) are related to aberrant, or 'dys-', connectivity between distinct brain regions. A proliferation of functional and ...structural neuroimaging studies have been conducted to investigate this hypothesis, across the full course of the disorder; from people at Ultra-High-Risk of developing psychosis to patients with chronic SZ. However the results of these studies have not always been consistent, and to date, there have been no attempts to summarise the results of both methodologies in conjunction. In this article, we systematically review both the structural and functional connectivity literature in SZ. The main trends to emerge are that schizophrenia is associated with connectivity reductions, as opposed to increases, relative to healthy controls, and that this is particularly evident in the connections involving the frontal lobe. These two trends appear to apply across all stages of the disorder, and to be independent of the neuroimaging methodology employed. We discuss the potential implications of these trends, and identify possible future investigative directions.
Brain networks or 'connectomes' include a minority of highly connected hub nodes that are functionally valuable, because their topological centrality supports integrative processing and adaptive ...behaviours. Recent studies also suggest that hubs have higher metabolic demands and longer-distance connections than other brain regions, and therefore could be considered biologically costly. Assuming that hubs thus normally combine both high topological value and high biological cost, we predicted that pathological brain lesions would be concentrated in hub regions. To test this general hypothesis, we first identified the hubs of brain anatomical networks estimated from diffusion tensor imaging data on healthy volunteers (n = 56), and showed that computational attacks targeted on hubs disproportionally degraded the efficiency of brain networks compared to random attacks. We then prepared grey matter lesion maps, based on meta-analyses of published magnetic resonance imaging data on more than 20 000 subjects and 26 different brain disorders. Magnetic resonance imaging lesions that were common across all brain disorders were more likely to be located in hubs of the normal brain connectome (P < 10(-4), permutation test). Specifically, nine brain disorders had lesions that were significantly more likely to be located in hubs (P < 0.05, permutation test), including schizophrenia and Alzheimer's disease. Both these disorders had significantly hub-concentrated lesion distributions, although (almost completely) distinct subsets of cortical hubs were lesioned in each disorder: temporal lobe hubs specifically were associated with higher lesion probability in Alzheimer's disease, whereas in schizophrenia lesions were concentrated in both frontal and temporal cortical hubs. These results linking pathological lesions to the topological centrality of nodes in the normal diffusion tensor imaging connectome were generally replicated when hubs were defined instead by the meta-analysis of more than 1500 task-related functional neuroimaging studies of healthy volunteers to create a normative functional co-activation network. We conclude that the high cost/high value hubs of human brain networks are more likely to be anatomically abnormal than non-hubs in many (if not all) brain disorders.
Standard univariate analysis of neuroimaging data has revealed a host of neuroanatomical and functional differences between healthy individuals and patients suffering a wide range of neurological and ...psychiatric disorders. Significant only at group level however these findings have had limited clinical translation, and recent attention has turned toward alternative forms of analysis, including Support-Vector-Machine (SVM). A type of machine learning, SVM allows categorisation of an individual's previously unseen data into a predefined group using a classification algorithm, developed on a training data set. In recent years, SVM has been successfully applied in the context of disease diagnosis, transition prediction and treatment prognosis, using both structural and functional neuroimaging data. Here we provide a brief overview of the method and review those studies that applied it to the investigation of Alzheimer's disease, schizophrenia, major depression, bipolar disorder, presymptomatic Huntington's disease, Parkinson's disease and autistic spectrum disorder. We conclude by discussing the main theoretical and practical challenges associated with the implementation of this method into the clinic and possible future directions.
Despite five decades of search for clinically meaningful 'biomarkers' in schizophrenia there are still no common tests to inform diagnosis or treatment. Our aim was to understand why it has been so ...difficult to convert biological findings into clinical tests. We categorized all PubMed-indexed articles investigating psychosis-related biomarkers to date (over 3200). Studies showed an evident publication bias, a confusing array of terminology, and few systematic efforts at longitudinal evaluation or external validation. Fewer than 200 studies investigated biomarkers, longitudinally, for prediction of illness course and treatment response. These biomarkers were then evaluated in terms of their statistical reliability and clinical effect size. Only one passed our a priori threshold for clinical applicability. This is a modest record. In order to promote real progress, the field needs: (a) consistent use of terminology so that studies can be compared; (b) a system of standardized universal reporting to overcome the existing publication bias; and (c) practical criteria a prototype is suggested here for assessing the clinical applicability of the findings.
Machine learning is becoming an increasingly popular approach for investigating spatially distributed and subtle neuroanatomical alterations in brain‐based disorders. However, some machine learning ...models have been criticized for requiring a large number of cases in each experimental group, and for resembling a “black box” that provides little or no insight into the nature of the data. In this article, we propose an alternative conceptual and practical approach for investigating brain‐based disorders which aim to overcome these limitations. We used an artificial neural network known as “deep autoencoder” to create a normative model using structural magnetic resonance imaging data from 1,113 healthy people. We then used this model to estimate total and regional neuroanatomical deviation in individual patients with schizophrenia and autism spectrum disorder using two independent data sets (n = 263). We report that the model was able to generate different values of total neuroanatomical deviation for each disease under investigation relative to their control group (p < .005). Furthermore, the model revealed distinct patterns of neuroanatomical deviations for the two diseases, consistent with the existing neuroimaging literature. We conclude that the deep autoencoder provides a flexible and promising framework for assessing total and regional neuroanatomical deviations in neuropsychiatric populations.
The rise of machine learning has unlocked new ways of analysing structural neuroimaging data, including brain age prediction. In this state-of-the-art review, we provide an introduction to the ...methods and potential clinical applications of brain age prediction. Studies on brain age typically involve the creation of a regression machine learning model of age-related neuroanatomical changes in healthy people. This model is then applied to new subjects to predict their brain age. The difference between predicted brain age and chronological age in a given individual is known as ‘brain-age gap’. This value is thought to reflect neuroanatomical abnormalities and may be a marker of overall brain health. It may aid early detection of brain-based disorders and support differential diagnosis, prognosis, and treatment choices. These applications could lead to more timely and more targeted interventions in age-related disorders.
There is growing interest in the complex topology of human brain functional networks, often measured using resting-state functional MRI (fMRI). Here, we used a meta-analysis of the large primary ...literature that used fMRI or PET to measure task-related activation (>1,600 studies; 1985-2010). We estimated the similarity (Jaccard index) of the activation patterns across experimental tasks between each pair of 638 brain regions. This continuous coactivation matrix was used to build a weighted graph to characterize network topology. The coactivation network was modular, with occipital, central, and default-mode modules predominantly coactivated by specific cognitive domains (perception, action, and emotion, respectively). It also included a rich club of hub nodes, located in parietal and prefrontal cortex and often connected over long distances, which were coactivated by a diverse range of experimental tasks. Investigating the topological role of edges between a deactivated and an activated node, we found that such competitive interactions were most frequent between nodes in different modules or between an activated rich-club node and a deactivated peripheral node. Many aspects of the coactivation network were convergent with a connectivity network derived from resting state fMRI data (n = 27, healthy volunteers); although the connectivity network was more parsimoniously connected and differed in the anatomical locations of some hubs. We conclude that the community structure of human brain networks is relevant to cognitive function. Deactivations may play a role in flexible reconfiguration of the network according to cognitive demand, varying the integration between modules, and between the periphery and a central rich club.