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

Abstract

Bipolar Disorders. 2020;22:334-355. wileyonlinelibrary.com/journal/bdi | INTRODUC TI ONBipolar disorder (BD) is a severe chronic mood disorder affecting more than 1% of the general adult population 1 and is associated with a high socio-economic burden. 2,3 Many challenges persist regarding BD management and growing evidence suggests that BD is under-recognized in clinical practice. Currently, the establishment of the diagnosis is solely based on clinical assessments. Therefore, the Abstract Objectives: The existence of anatomofunctional brain abnormalities in bipolar disorder (BD) is now well established by magnetic resonance imaging (MRI) studies. To create diagnostic and prognostic tools, as well as identifying biologically valid subtypes of BD, research has recently turned towards the use of machine learning (ML) techniques. We assessed both supervised ML and unsupervised ML studies in BD to evaluate their robustness, reproducibility and the potential need for improvement. Method:We systematically searched for studies using ML algorithms based on MRI data of patients with BD until February 2019. Result:We identified 47 studies, 45 using supervised ML techniques and 2 including unsupervised ML analyses. Among supervised studies, 43 focused on diagnostic classification. The reported accuracies for classification of BD ranged between (a) 57% and 100%, for BD vs healthy controls; (b) 49.5% and 93.1% for BD vs patients with major depressive disorder; and (c) 50% and 96.2% for BD vs patients with schizophrenia. Reported accuracies for discriminating subjects genetically at risk for BD (either from control or from patients with BD) ranged between 64.3% and 88.93%. Conclusions:Although there are strong methodological limitations in previous studies and an important need for replication in large multicentric samples, the conclusions of our review bring hope of future computer-aided diagnosis of BD and pave the way for other applications, such as treatment response prediction. To reinforce the reliability of future results we provide methodological suggestions for good practice in conducting and reporting MRI-based ML studies in BD. K E Y W O R D Sbipolar disorders, machine learning, magnetic resonance imaging, precision medicine | 335 CLAUDE Et AL.

Using Minimal-Redundant and Maximal-Relevant Whole-Brain Functional Connectivity to Classify Bipolar Disorder

“…When discriminating BD patients from HCs, previous studies have shown medium to high levels of accuracy when a variety of features were used including voxel-based morphometry (Mwangi et al, 2016), cortical thickness and skewedness (Squarcina et al, 2019), and functional connectivity (Roberts et al, 2017;Wang et al, 2019). Moreover, as a review study (Claude et al, 2020), which added some more recent studies (Squarcina et al, 2019;Wang et al, 2019), demonstrated that more than half of the studies classifying BD and HC used structural MRI, and, furthermore, the number of studies that used functional MRI was greater than the ones using diffusion tensor images. In general, the classification performances of the studies using functional MRI outperformed those of studies using other modalities.…”

“…For example, one reviewed study discriminated between 12 patients with BD and 25 HCs with 100% accuracy using white matter integrity as the features (Besga et al, 2012). Moreover, as the sample sizes became larger, the classification performance levels were reduced (Claude et al, 2020). Notwithstanding the above, the present study, which has a relatively large sample size, has been able to achieve a high accuracy; this accuracy is higher than the median accuracy of the above previous studies as well as being closed to the minimum threshold of clinical relevance (i.e., 80%).…”

<p>Machine Learning Methods to Evaluate the Depression Status of Chinese Recruits: A Diagnostic Study</p>

“…A recent systematic review has pointed out that the current available studies have methodological limitations, but that research is on the right way. 56 In particular, there is a need to recapitulate the results in multiple centers. Another research avenue could also be the inclusion of neuroimaging biomarkers, which have been identified for PTSD.…”

Untitled

“…73 In light of those limitations, machine learning (ML) combines neuroimaging with pattern identification approaches and tests predictive models in large samples providing individual-level results (supervised learning) and defines homogenous groups among heterogeneous samples (unsupervised learning). 74 By implementing multivariate approaches and intergrading different levels of data (neuroimaging, neurocognitive, genetic, demographics), ML suggests a predictive tool with the potential to deliver outcome biomarkers of PBD and ADHD. 75 Recently, several studies have demonstrated the ability of ML for differentiating unique BD and ADHD subgroups from control participants, based on neurocognitive and neuroimaging data.…”