Benson Mwangi

A Review of Feature Reduction Techniques in Neuroimaging

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

Cortical abnormalities in adults and adolescents with major depression based on brain scans from 20 cohorts worldwide in the ENIGMA Major Depressive Disorder Working Group

The neuro-anatomical substrates of major depressive disorder (MDD) are still not well understood, despite many neuroimaging studies over the past few decades. Here we present the largest ever worldwide study by the ENIGMA (Enhancing Neuro Imaging Genetics through Meta-Analysis) Major Depressive Disorder Working Group on cortical structural alterations in MDD. Structural T1-weighted brain magnetic resonance imaging (MRI) scans from 2148 MDD patients and 7957 healthy controls were analysed with harmonized protocols at 20 sites around the world. To detect consistent effects of MDD and its modulators on cortical thickness and surface area estimates derived from MRI, statistical effects from sites were meta-analysed separately for adults and adolescents. Adults with MDD had thinner cortical gray matter than controls in the orbitofrontal cortex (OFC), anterior and posterior cingulate, insula and temporal lobes (Cohen’s d effect sizes: −0.10 to −0.14). These effects were most pronounced in first episode and adult-onset patients (>21 years). Compared to matched controls, adolescents with MDD had lower total surface area (but no differences in cortical thickness) and regional reductions in frontal regions (medial OFC and superior frontal gyrus) and primary and higher-order visual, somatosensory and motor areas (d: −0.26 to −0.57). The strongest effects were found in recurrent adolescent patients. This highly powered global effort to identify consistent brain abnormalities showed widespread cortical alterations in MDD patients as compared to controls and suggests that MDD may impact brain structure in a highly dynamic way, with different patterns of alterations at different stages of life.

Subcortical volumetric abnormalities in bipolar disorder.

Considerable uncertainty exists about the defining brain changes associated with bipolar disorder (BD). Understanding and quantifying the sources of uncertainty can help generate novel clinical hypotheses about etiology and assist in the development of biomarkers for indexing disease progression and prognosis. Here we were interested in quantifying case–control differences in intracranial volume (ICV) and each of eight subcortical brain measures: nucleus accumbens, amygdala, caudate, hippocampus, globus pallidus, putamen, thalamus, lateral ventricles. In a large study of 1710 BD patients and 2594 healthy controls, we found consistent volumetric reductions in BD patients for mean hippocampus (Cohen’s d=−0.232; P=3.50 × 10−7) and thalamus (d=−0.148; P=4.27 × 10−3) and enlarged lateral ventricles (d=−0.260; P=3.93 × 10−5) in patients. No significant effect of age at illness onset was detected. Stratifying patients based on clinical subtype (BD type I or type II) revealed that BDI patients had significantly larger lateral ventricles and smaller hippocampus and amygdala than controls. However, when comparing BDI and BDII patients directly, we did not detect any significant differences in brain volume. This likely represents similar etiology between BD subtype classifications. Exploratory analyses revealed significantly larger thalamic volumes in patients taking lithium compared with patients not taking lithium. We detected no significant differences between BDII patients and controls in the largest such comparison to date. Findings in this study should be interpreted with caution and with careful consideration of the limitations inherent to meta-analyzed neuroimaging comparisons.

Brainstem abnormalities in attention deficit hyperactivity disorder support high accuracy individual diagnostic classification

Despite extensive research, psychiatry remains an essentially clinical and, therefore, subjective clinical discipline, with no objective biomarkers to guide clinical practice and research. Development of psychiatric biomarkers is consequently important. A promising approach involves the use of machine learning with neuroimaging, to make predictions of diagnosis and treatment response for individual patients. Herein, we describe predictions of attention deficit hyperactivity disorder (ADHD) diagnosis using structural T1 weighted brain scans obtained from 34 young males with ADHD and 34 controls and a support vector machine. We report 93% accuracy of individual subject diagnostic prediction. Importantly, automated selection of brain regions supporting prediction was used. High accuracy prediction was supported by a region of reduced white matter in the brainstem, associated with a pons volumetric reduction in ADHD, adjacent to the noradrenergic locus coeruleus and dopaminergic ventral tegmental area nuclei. Medications used to treat ADHD modify dopaminergic and noradrenergic function. The white matter brainstem finding raises the possibility of “catecholamine disconnection or dysregulation” contributing to the ADHD syndrome, ameliorated by medication. Hum Brain Mapp 35:5179–5189, 2014.

Prediction of individual subject's age across the human lifespan using diffusion tensor imaging: A machine learning approach

Diffusion tensor imaging has the potential to be used as a neuroimaging marker of natural ageing and assist in elucidating trajectories of cerebral maturation and ageing. In this study, we applied a multivariate technique relevance vector regression (RVR) to predict individual subjects age using whole brain fractional anisotropy (FA), mean diffusivity (MD), axial diffusivity (AD) and radial diffusivity (RD) from a cohort of 188 subjects aged 4-85 years. High prediction accuracy as derived from Pearson correlation coefficient of actual versus predicted age (FA - r=0.870 p<0.0001; MD - r=0.896 p<0.0001; AD - r=0.895 p<0.0001; RD - r=0.899 p<0.0001) was achieved. Cerebral white-matter regions that contributed to these predictions include; corpus callosum, cingulum bundles, posterior longitudinal fasciculus and the cerebral peduncle. A post-hoc analysis of these regions showed that FA follows a nonlinear rational-quadratic trajectory across the lifespan peaking at approximately 21.8 years. The MD, RD and AD volumes were particularly useful for making predictions using grey matter cerebral regions. These results suggest that diffusion tensor imaging measurements can reliably predict individual subjects age and demonstrate that FA cerebral maturation and ageing patterns follow a non-linear trajectory with a noteworthy peaking age. These data will contribute to the understanding of neurobiology of cerebral maturation and ageing. Most notably, from a neuropsychiatric perspective our results may allow differentiation of cerebral changes that may occur due to natural maturation and ageing, and those due to developmental or neuropsychiatric disorders.

Areas of controversy in neuroprogression in bipolar disorder.

We aimed to review clinical features and biological underpinnings related to neuroprogression in bipolar disorder (BD). Also, we discussed areas of controversy and future research in the field.

Big data analytics and machine learning: 2015 and beyond.

www.thelancet.com/psychiatry Vol 3 January 2016 13 defi cit). These fi ndings, like those highlighted above, have drawn into serious question the usefulness of symptom-based diagnostic constructs if they fail to provide information or insight about the underlying psycho patholgical mechanisms. Another long-standing challenge has been the limited contact between neuroimaging and other areas of psychiatric neuroscience, especially genetic and molecular investigations. This too has seen progress in 2015, in part related to larger neuroimaging sample sizes that can allow such interfaces. Findings from a large twin study of white matter structure in schizophrenia showed a strong association between white matter integrity and genetic risk for schizophrenia, suggesting a potential common causal genetic mechanism. Another study of amygdala volume and emotion recognition task performance in a very large sample (n=858) showed a common relationship of both to a polymorphism in the PDE5 gene using a genome-wide quantitative trait locus analysis. Particularly exciting is that this association suggests potential pharmacological treatment insights for many neuropsychiatric disorders in which emotion recognition is impaired, as drugs that target PDE5 (a phosphodiesterase) already exist. Finally, even the organising theme of anatomically distributed, large-scale connectivity networks, which has been central to many neuroimaging studies during the past decade, has made contact with genomics. Specifi cally, using a post-mortem gene expression dataset drawn from many regions across the human brain, a relationship has been found between spatial patterns of functional connectivity and gene expression . Perhaps the biggest obstacle to the clinical utility of neuroimaging, however, has been the dependence on group-level analyses, which generally preclude insights about individual patients. This too has seen striking progress, both with respect to methods that can readily parcellate functional regions within individual brains, and demonstrations that functional connectivity-based analyses are powerful and specifi c enough to reliably identify an individual from a group. Thus, looking forward to 2016, it would seem that a greater focus on individuals will be a crucial area of progress that can fi nally bring neuroimaging into the clinical sphere in psychiatry. We must therefore also demand more of study design so that it can be relevant to individuals, such as use of control arms when studying interventions, which still remains rare in neuroimaging.

The medial forebrain bundle as a deep brain stimulation target for treatment resistant depression: A review of published data.

INTRODUCTION Despite a wide variety of therapeutic interventions for major depressive disorder (MDD), treatment resistant depression (TRD) remains to be prevalent and troublesome in clinical practice. In recent years, deep brain stimulation (DBS) has emerged as an alternative for individuals suffering from TRD not responding to combining antidepressants, multiple adjunctive strategies and electroconvulsive therapy (ECT). Although the best site for TRD-DBS is still unclear, pilot data suggests that the medial forebrain bundle (MFB) might be a key target to accomplish therapeutic efficacy in TRD patients. OBJECTIVE To explore the anatomic, electrophysiologic, neurocognitive and treatment data supporting the MFB as a target for TRD-DBS. RESULTS The MFB connects multiple targets involved in motivated behavior, mood regulation and antidepressant response. Specific phenomenology associated with TRD can be linked specifically to the superolateral branch (sl) of the MFB (slMFB). TRD patients who received DBS-slMFB reported high response/remission rates with an improvement in functioning and no significant adverse outcomes in their physical health or neurocognitive performance. DISCUSSION The slMFB is an essential component of a network of structural and functional pathways connecting different areas possibly involved in the pathogenesis of mood disorders. Therefore, the slMFB should be considered as an exciting therapeutic target for DBS therapy to achieve a sustained relief in TRD patients. CONCLUSION There is an urgent need for clinical trials exploring DBS-slMFB in TRD. Further efforts should pursue measuring baseline pro-inflammatory cytokines, oxidative stress, and cognition as possible biomarkers of DBS-slMFB response in order to aid clinicians in better patient selection.

Hippocampal subfield volumes in mood disorders

Volume reduction and shape abnormality of the hippocampus have been associated with mood disorders. However, the hippocampus is not a uniform structure and consists of several subfields, such as the cornu ammonis (CA) subfields CA1–4, the dentate gyrus (DG) including a granule cell layer (GCL) and a molecular layer (ML) that continuously crosses adjacent subiculum (Sub) and CA fields. It is known that cellular and molecular mechanisms associated with mood disorders may be localized to specific hippocampal subfields. Thus, it is necessary to investigate the link between the in vivo hippocampal subfield volumes and specific mood disorders, such as bipolar disorder (BD) and major depressive disorder (MDD). In the present study, we used a state-of-the-art hippocampal segmentation approach, and we found that patients with BD had reduced volumes of hippocampal subfields, specifically in the left CA4, GCL, ML and both sides of the hippocampal tail, compared with healthy subjects and patients with MDD. The volume reduction was especially severe in patients with bipolar I disorder (BD-I). We also demonstrated that hippocampal subfield volume reduction was associated with the progression of the illness. For patients with BD-I, the volumes of the right CA1, ML and Sub decreased as the illness duration increased, and the volumes of both sides of the CA2/3, CA4 and hippocampal tail had negative correlations with the number of manic episodes. These results indicated that among the mood disorders the hippocampal subfields were more affected in BD-I compared with BD-II and MDD, and manic episodes had focused progressive effect on the CA2/3 and CA4 and hippocampal tail.

Hippocampal volume and verbal memory performance in late-stage bipolar disorder

Studies about changes in hippocampal volumes in subjects with bipolar disorder (BD) have been contradictory. Since the number of manic episodes and hospitalization has been associated with brain changes and poor cognitive outcomes among BD patients, we have hypothesized that these variables could clarify this issue. We stratified subjects with BD in early (BD-Early), intermediate (BD-intermediate) and late (BD-Late) stages as a function of number of manic episodes and prior hospitalization. Then, we compared their hippocampal volumes and California Verbal Learning Test-II (CVLT-II) scores with healthy controls (HC) using the general linear model. A total of 173 subjects were included in the study (112 HC, 15 BD-Early, 30 BD-Intermediate, and 16 BD-Late). We found a significant group effect on hippocampus volume (F(3,167) = 3.227, p = 0.024). Post-hoc analysis showed that BD-Late subjects had smaller hippocampus than HC (p = 0.017). BD-Early and BD-Intermediate subjects showed no significant difference in hippocampus volume compared to HC and BD-Late subjects. The CVLT trial 1 to 5 scores were significantly different across the groups (F(3,167) = 6.371, p < 0.001). Post-hoc analysis showed that BD-Intermediate (p = 0.006) and BD-Late (p = 0.017) subjects had worse memory performance during immediate recall than HC, while the performance difference between BD-Early subjects and HC was not significant (p = 0.208). These findings add to the notion that BD is a neuroprogressive disorder with brain changes and cognitive impairment according to prior morbidity (number of manic episodes and hospitalization). Also, they suggest that hippocampus is a brain marker and a potential therapeutic target for patients at late stage.