Automated classification of depression from structural brain measures across two independent community‐based cohorts

Abstract

Major depressive disorder (MDD) has been the subject of many neuroimaging case–control classification studies. Although some studies report accuracies ≥80%, most have investigated relatively small samples of clinically‐ascertained, currently symptomatic cases, and did not attempt replication in larger samples. We here first aimed to replicate previously reported classification accuracies in a small, well‐phenotyped community‐based group of current MDD cases with clinical interview‐based diagnoses (from STratifying Resilience and Depression Longitudinally cohort, ‘STRADL’). We performed a set of exploratory predictive classification analyses with measures related to brain morphometry and white matter integrity. We applied three classifier types—SVM, penalised logistic regression or decision tree—either with or without optimisation, and with or without feature selection. We then determined whether similar accuracies could be replicated in a larger independent population‐based sample with self‐reported current depression (UK Biobank cohort). Additional analyses extended to lifetime MDD diagnoses—remitted MDD in STRADL, and lifetime‐experienced MDD in UK Biobank. The highest cross‐validation accuracy (75%) was achieved in the initial current MDD sample with a decision tree classifier and cortical surface area features. The most frequently selected decision tree split variables included surface areas of bilateral caudal anterior cingulate, left lingual gyrus, left superior frontal, right precentral and paracentral regions. High accuracy was not achieved in the larger samples with self‐reported current depression (53.73%), with remitted MDD (57.48%), or with lifetime‐experienced MDD (52.68–60.29%). Our results indicate that high predictive classification accuracies may not immediately translate to larger samples with broader criteria for depression, and may not be robust across different classification approaches.

The cingulum in very preterm infants relates to language and social-emotional impairment at 2 years of term-equivalent age

“…The model equation is as follows: where is a vector of the size n 1 containing the group label (n = 132; sample was labeled “1″ for preterm and “-1” for full-term) of the subjects, is a matrix of the size n 104 containing the DTI metrics calculated from the 26 WM pathway regions, C is the inverse of the regularization strength, and ρ controls the balance of the and regularizations. The 10-fold cross-validation and grid search strategy was performed to choose the optimal regularization parameters C and ρ ( Casanova et al, 2011 , Stolicyn et al, 2020 ). The mean misclassification rate for each pair of C and ρ was calculated in the training set over a range of different values between 0 and 1 with increments of 0.01 (C, total of 100 values) and 0.05 (ρ, total of 20 values).…”

Spectral clustering based on structural magnetic resonance imaging and its relationship with major depressive disorder and cognitive ability

“…GS (subsample) is a community‐based dataset with imaging data, reported previously (Habota et al, 2019; Navrady et al, 2018; Romaniuk et al, 2019; Rupprechter et al, 2020; Smith et al, 2012; Stolicyn et al, 2020). Demographic details of these participants and for the replication cohort (UKB) are presented in Table 1.…”

“…Diagnosis was established using the Structured Clinical Interview for DSM‐IV Disorders (SCID) (First, 1997) and was based on criteria from the Diagnostic and Statistical Manual of Mental Disorders (DSM) (American Psychiatric Association, 2000) (Section S1.2.1). Participants were classed as currently depressed if they had an ongoing depressive episode, and as past MDD if they were not depressed at the time of MRI scan but had at least one depressive episode previously (Stolicyn et al, 2020). Participants were classed as recurrent if they had had more than one depressive episode.…”

“…However, the ability of ML to determine case–control status using such features remains uncertain, especially when most existing studies have been conducted on relatively small datasets ( N < 100), with limited independent replication. In addition, the majority of existing studies focus on clinically ascertained cases, and therefore the results may not be generalisable to population or community‐based samples (Stolicyn et al, 2020).…”

DNA methylation signatures of C-reactive protein associations with structural neuroimaging measures and major depressive disorder

“…The participants in this study were recruited as part of the STratifying Resilience and Depression Longitudinally (STRADL) study (2015-2019) which re-contacted participants from the Generation Scotland: Scottish Family Health Study (GS) via post for further assessment of mental health, specifically depression. Full details of the STRADL cohort and GS protocol are published elsewhere [32][33][34][35][36][37] . The current sample included a range of 566-798 participants depending on the biomarker/imaging modality investigated (Table 1).…”