Grey matter changes in BD are generally subtle as their average effect size is 0.5 or less (Kempton et al. 2008). There is little evidence for global grey matter abnormalities especially in first-episode patients (Vita et al. 2009). Regional grey matter volume reductions have been consistently reported in BD patients compared to healthy individuals in the VPFC, insula and ACC (Bora et al. 2010; Ellison-Wright and Bullmore 2010; Selvaraj et al. 2012). Despite initial reports implicating the amygdala, meta-analyses of sMRI data from voxel brain morphology (VBM) studies in adult BD patients have not found consistent alterations in this structure. Similarly, meta-analyses of region-of-interest (ROI) studies report minimal differences in amygdala volume between adult BD patients and healthy individuals (Hedges’ g −0.04 to −0.07) (Kempton et al. 2008). However, amygdala volume reduction has emerged as a consistent finding in paediatric BD (Terry et al. 2009). Evidence from longitudinal studies indicate that the smaller size of the amygdala early in life in BD may be accounted for by delayed or abnormal developmental trajectories in patients compared to healthy youth (Bitter et al. 2011).

Grey matter reductions in the VPFC and the ACC represent a robust correlate of disease expression in BD and are present regardless the age of onset. Studies that have examined the relevance of these findings to other disorders have reported overlap but also differences. Grey matter volume reductions in SZ are generally more extensive and more severe, but there is substantial overlap between the two disorders in the insula and ACC (Ellison-Wright and Bullmore 2010). Patients with SZ also appear to have greater ventricular enlargement and smaller amygdala volumes than those with BD (Arnone et al. 2009). A substantial overlap has also been found in brain structural changes in BD and MDD (Kempton et al. 2011). However, MDD patients may have smaller hippocampal and basal ganglia volume and larger corpus callosum areas compared to BD patients (Kempton et al. 2011).

Sex does not appear to have a major moderator role in brain morphology in BD (Jogia et al. 2012). Age and duration of illness are highly collinear. As there is a paucity of longitudinal studies in BD, the effects of these two factors cannot be easily disentangled. Sarnicola et al. (2009) did not find differential age-related changes in brain morphology in BD patients and healthy individuals over an age span of 40 years. However, multi-episode (as opposed to first episode) BD patients may have smaller VPFC and larger striatum and amygdala (Bora et al. 2010). The latter observations may reflect chronic exposure to medication and particularly antipsychotics and lithium. The association between antipsychotics and increased striatal volume is well replicated and is considered to reflect diagnosis-independent microstructural changes in response to long-term reduction in dopamine neurotransmission (Navari and Dazzan 2009). Moore et al. (2000) were the first to report that treatment with lithium increased global grey matter volume in BD patients. This observation has been robustly replicated with regard to global (Bearden et al. 2007; Lyoo et al. 2010) and regional grey matter volume increases particularly in the amygdala, rostral ACC and PFC (Bearden et al. 2007; Germaná et al. 2010; Moore et al. 2009). This brain volume “response” to lithium is of therapeutic significance as it has been linked to symptomatic improvement (Lyoo et al. 2010; Moore et al. 2009). Although the precise mechanism underlying lithium-induced brain structural changes in BD remains to be determined, the drug’s primary target, glycogen synthase kinase-3, is involved in pathways with neurotrophic, neurogenetic and neuroprotective functions (Machado-Vieira et al. 2009). Clinical symptoms at the time of scanning may also impact neuroimaging results. This is mostly the case for functional rather than anatomical studies. In general, acute symptoms, either depressive or manic, appear to increase fMRI signal in subcortical regions (hippocampus, amygdala) and reduce PFC engagement (Chen et al. 2011; Delvecchio et al. 2013).

The findings described above demonstrate the contribution of neuroimaging to our understanding of the neural correlates in BD. However, their impact on clinical practice has been negligible. This is primarily because conventional data analysis computes differences between groups either within predefined anatomical regions of interest (ROI) or throughout the brain using univariate mass analyses such as voxel-based morphometry (VBM). Both types of analyses are not able to differentiate between groups where mean differences are small to moderate. The ROI approach is statistically more powerful but restricted to regions that can be reliably anatomically defined. VBM requires correction for multiple comparisons that limits its statistical power to detect subtle changes. Additionally, neither approach utilises information about the spatial distribution of case-control differences. Current research interest is focusing on maximising the translational value of neuroimaging by developing new computational tools for data analysis that could assist with the clinical evaluation of patients.

Calhoun and colleagues (2008) used independent component analysis to extract mode images of brain activity during resting state and during performance on a selective attention task in patients with BD or SZ and healthy individuals. A multistage classification algorithm was then applied to concatenate each participant’s mode images into a single image. A mean image was computed for each group, and the Euclidean distance between an individual’s brain image and each group’s images was computed. Using this approach, the authors were able to classify participants according to diagnostic group with an average sensitivity of 90 % and specificity of 95 %.

Another computational approach involves the use of supervised machine learning algorithms to predict diagnostic classification. Costafreda and colleagues (2011) applied support vector machine (SVM) to functional imaging data derived from patients with BD or SZ and healthy individuals while performing a verbal fluency task. Classification accuracy for SZ was 92 % (91 % sensitivity and 92 % specificity) and for BD 79 % (56 % sensitivity and 89 % specificity); misclassification occurred mostly because of the wrong categorization of BD patients as healthy individuals. Rocha-Rego and colleagues (2013) applied Gaussian Process Classifiers (GPCs) to sMRI data to evaluate the feasibility of using pattern recognition techniques for the diagnostic differentiation of patients with BD from healthy individuals. They chose to focus on sMRI because of its high acceptability by patients and wide availability in clinical settings. GPCs represent a significant advance as they combine equivalent predictive performance to SVM with the additional benefit of probabilistic classification (Marquand et al. 2010). Grey and white matter classifiers correctly assigned patients to the appropriate diagnostic category with respective accuracy levels of 73 % (sensitivity 69 %, specificity 77 %) and 69 % (sensitivity 69 %, specificity 69 %). Grey matter discriminative clusters were localised within cortical and subcortical structures implicated in BD, including the VPFC, ACC, parahippocampal gyrus, insula, thalamus and striatum (Fig. 16.2a). White matter discriminative clusters were identified mainly within the cingulum, occipital regions and the genu of the corpus callosum (Fig. 16.2b).