Fig. 1Map of regions which showed a significant difference in grey and white matter in MDD patients and healthy controls: (a) grey matter regions demonstrating atrophy in MDD patients relative to healthy controls, (b) white matter regions demonstrating atrophy in MDD patients relative to healthy controls. Green indicates significant regions at P<0.001 uncorrected, and areas of red colour (threshold P<0.05) indicates the trend towards significance characterised by –log(p) values as shown in the colour bar.
Fig. 2(a) Schematic map showing white matter regions which contributed towards diagnostic classification of MDD, regions are presented at P<0.05 uncorrected. Blue indicates regions showing atrophy in MDD patients relative to controls and yellow indicates regions of greater volume in MDD patients compared with healthy controls. (b) Receiver operating characterstic (ROC) curve for the comparison between MDD and healthy participants, area under curve (AUC)=0.73, P=0.02. The x-axis is the false positive rate (1-specificity) and the y-axis is the true positive rate (sensitivity). (c) Graph illustrating change in classification rate with number of attributes selected. The x-axis indicates the number of attributes and the y-axis the classification rate. The highest classification rate was 81.4% based on 47 features, while the most stable pattern was observed with an accuracy of around 70% based on 50–70 features.
Med, Medication; HAMD, 17-item Hamilton Depression Rating Scale; MDD, major depressive disorder; TRD, treatment-resistant depression; pMDD, psychotic MDD; HV, healthy volunteers; GM, grey matter; WM, white matter; CSF, cerebrospinal fluid; SVM, support vector machines; PCA, principle component analysis; RFE, recursive feature elimination; LLE, locally linear embedding; VBM, voxel based morphometry; RVM, relevance vector machine; FBM, feature based morphometry; EM, expectation-maximisation dustering algorithm; KMeans, simple K means classification via clustering; TUD, treatment unresponsive patients.
Depression status of MDD patients: first-episode – Liu (2012), Qiu (2014), Serpa (2014); first-episode and recurrent – Costafreda (2009); recurrent: Mwangi (2012); not stated: Gong (2011), Kipli (2013).
↵+ Mwangi (2012): data were randomly divided into two sets (training set, testing set) of equal number of patients and controls (n=31). In patients, depression was considered to be treatment unresponsive. Minimum duration of illness was >3 months with antidepressant medication.
↵++ Kipli (2013), accuracy of 82.3% also obtained with other classifiers: information gain:-J48, information gain-RandomForest, SVM-K Means, SVM-RandomForest, ReliefF-RandomTree, all-naïve bayes.
↵+++ Combined parameters: Qiu (2014) integrated all the morphometric parameters (i.e. cortical thickness, volume, plial area, curvature area, sulcal depth and Jacobian metric distortion) of the left and right hemispheres within a single model to investigate the discriminative power of the resulting combination.