Distinguishing bipolar and major depressive disorders by brain structural morphometry: a pilot study

Abstract

BackgroundThe clinical presentation of common symptoms during depressive episodes in bipolar disorder (BD) and major depressive disorder (MDD) poses challenges for accurate diagnosis. Disorder-specific neuroanatomical features may aid the development of reliable discrimination between these two clinical conditions.MethodsFor our sample of 16 BD patients, 19 MDD patients and 29 healthy volunteers, we adopted vertex-wise cortical based brain imaging techniques to examine cortical thickness and surface area, two components of cortical volume with distinct genetic determinants. Based on specific characteristics of neuroanatomical features, we then used support vector machine (SVM) algorithm to discriminate between patients with BD and MDD.ResultsCompared to MDD patients, BD patients showed significantly larger cortical surface area in the left bankssts, precuneus, precentral, inferior parietal, superior parietal and the right middle temporal gyri. In addition, larger volumes of subcortical regions were found in BD patients. In SVM discriminative analyses, the overall accuracy was 74.3 %, with a sensitivity of 62.5 % and a specificity of 84.2 % (p = 0.028). Compared to controls, larger surface area in the temporo-parietal regions were observed in BD patients, and thinner cortices in fronto-temporal regions were observed in MDD patients, especially in the medial orbito-frontal area.ConclusionsThese findings have demonstrated distinct spatially distributed variations in cortical thickness and surface area in patients with BD and MDD, suggesting potentially varying etiological and neuropathological processes in these two conditions. The employment of multimodal classification on disorder-specific biological features has shed light to the development of potential classification tools that could aid diagnostic decisions.

Neurobiology of bipolar disorders: a review of genetic components, signaling pathways, biochemical changes, and neuroimaging findings

“…239 This potential impact of machine learning techniques in the evaluation of individuals with BD was extensively explored in a systematic review by Librenza-Garcia et al 240 Of 51 studies included, 38 applied machine learning to discriminate between BD and healthy controls or other psychiatric disorders, especially with neuroimaging data. 240 For instance, Fung et al 241 investigated psychiatric diagnosis accuracy using a support vector machine (SVM) algorithm with brain cortical thickness and surface area data. The authors found that structural brain differences between individuals with BD and major depressive disorder were able to discriminate these psychiatric disorders with 74.3% (adequate) accuracy (sensitivity: 62.5%; specificity: 84.2%).…”

“…The authors found that structural brain differences between individuals with BD and major depressive disorder were able to discriminate these psychiatric disorders with 74.3% (adequate) accuracy (sensitivity: 62.5%; specificity: 84.2%). 241 More recently, a Big Data Task Force from the International Society for Bipolar Disorders has expanded this literature review and discussed issues to be addressed in machine learning-based studies, including the main barriers for applying these techniques and strategies to approach them. 242 Since human behavior, including cognition, emotion, and social interaction, reflects complex neural circuit communication, 243 the signs and symptoms we observe in individuals with psychiatric disorders could be understood as the manifestation of different brain circuitry dysfunction.…”

Will machine learning applied to neuroimaging in bipolar disorder help the clinician? A critical review and methodological suggestions

“…Reported ACC for discriminating BD vs MDD ranged between 49.5% and 93.1% (Figure 4). Six studies used only sMRI as input features, 44,48,50,54,66,67 5 studies used grey matter input features reaching 59.45%‐75.9% ACC 44,48,50,66,67 and 1 used both grey and white matter input features reaching 54.76% ACC 54 …”

Using structural MRI to identify bipolar disorders – 13 site machine learning study in 3020 individuals from the ENIGMA Bipolar Disorders Working Group

“…Small studies may also yield a wide range of classification performances and inconsistencies in regions, which contribute to the overall classification [25][26][27]. Previous ML structural MRI studies in BD have typically included <50 BD participants recruited in a single site [23,[28][29][30][31][32][33][34]. The largest currently available neurostructural ML studies investigated 128-190 BD and 127-284 control participants [35][36][37], from up to two sites [22,23,38].…”