Brief reportPredicting obsessive–compulsive disorder severity combining neuroimaging and machine learning methods
Introduction
Several analytical approaches have been used to classify or discriminate patients with psychiatric disorders from healthy controls through brain magnetic resonance imaging (MRI) as an attempt to understand the neurobiological underpinnings of mental disorders (Pereira et al., 2009).
Structural MRI has been previously used to assess the feasibility of differentiating individual subjects with obsessive–compulsive disorder (OCD) from healthy controls. The overall classification accuracy to discriminate a group of OCD patients (not used for training the classifier) from control subjects was 76.6% (Soriano-Mas et al., 2007). More recently, using a functional MRI protocol involving the processing of fear and disgust stimuli in OCD, a multivariate pattern recognition approach revealed that the orbitofrontal cortex (OFC) and caudate nucleus encoded diagnostic information that differentiated OCD patients from healthy controls with an accuracy of 100% (Weygandt et al., 2012). However, most of patients included in previous studies have been exposed to medications, which most probably have influenced the results (Hoexter et al., 2012).
Given the limitations in the diagnostic constructs currently under use in psychiatry (Insel et al., 2010), it seems mandatory to evaluate if additional characteristics correlate with brain imaging patterns beyond the presence of a specific disorder. One of the such characteristics could be the global symptom severity that varies across sufferers from a given disorder. The combination of computational neuroanatomy based on neuroimaging and machine learning methods is a suitable approach to address this issue (Sato et al., 2012a, Sato et al., 2012b) and may allow the identification of predictive neurobiological markers of future outcomes in individual subjects (de Almeida and Phillips, 2013).
Numerous neuroimaging studies have reported associations between OCD severity and gray matter (GM) volume abnormalities in several brain regions (Radua and Mataix-Cols, 2009, Rotge et al., 2010). Moreover, OCD severity is also a consistent variable for predicting outcome in OCD (Denys et al., 2003, Mataix-Cols et al., 1999).
Herein, based on structural MRI data from a sample described previously (Hoexter et al., 2012), we combine support vector regression to evaluate whether specific GM volumes encompassing cortical–subcortical loops contain relevant information to predict OCD severity in a sample of treatment-naïve patients. This approach has not been applied previously and may provide objective measures of the severity of the disorder, and address specific dimensions that may characterize subgroups of patients.
The enrolling of only treatment-naive patients in such a study design is essential to rule out the potential influence of previous treatments on brain morphometry.
Section snippets
Subjects
Thirty-seven adult OCD patients participated in this study. Patients were recruited from the Obsessive–Compulsive Spectrum Disorders Program at the Institute of Psychiatry, University of São Paulo Medical School, Brazil. All subjects provided written informed consent to participate in this study, approved by the local Ethics Committee (Hoexter et al., 2009).
Participants were treatment-naïve, aged between 18 and 65 years, had OCD as the primary diagnosis (DSM-IV) and presented a Yale-Brown
Results
The mean age±standard deviation (SD) of our sample was 31.9±10.1 years. Twenty-two of 37 patients were female (59%) and 36 were right-handed (97%). Clinical characteristics, Y-BOCS and DY-BOCS scores of the sample are presented in Table 1.
Fig. 1 shows a scatter-plot of predicted and observed values and the relevance (contribution) of each brain region for DY-BOCS and Y-BOCS scores. Of note, the regions that contained the most discriminative information were the left medial OFC and the left
Discussion
Herein we have applied a SVR analysis to predict OCD severity based on individual structural MRI scans obtained from treatment-naïve patients. The enrollment of only treatment-naive patients was essential to rule out the potential influence of previous treatments on brain morphometry. While previous MRI studies in OCD have used classification models to discriminate patients from controls (categorical outcome) (Soriano-Mas et al., 2007, Weygandt et al., 2012), a methodological advantage of
Role of funding source
This study received financial support in the form of Grants provided by the Fundação de Amparo à Pesquisa do Estado de São Paulo (FAPESP, Foundation for the Support of Research in the State of Sao Paulo) to Dr. Miguel (2005/55628-8), to Dr. Shavitt (06/61459-7) and to Dr. Diniz (06/50273-0). Dr Hoexter was supported by a Ph.D. Grand from FAPESP (2005/04206-6) and by a doctorate “sandwich” scholarship from the Coordenação de Aperfeiçoamento de Pessoal de Nível Superior (CAPES, Agency for Support
Conflict of interest
Dr. Hoexter, Dr. Miguel, Dr. Diniz, Dr. Shavitt, Dr. Busatto and Dr. Sato have declared no conflict of interest.
Acknowledgments
The authors thank Fabio L.S. Duran, Marcelo C. Batistuzzo and Antonio C. Lopes for previous contribution. We also wish to thank all patients who participated in this investigation.
References (24)
- et al.
Obsessive–compulsive symptom dimensions correlate to specific gray matter volumes in treatment-naive patients
Journal of Psychiatric Research
(2012) - et al.
Clinical features associated to refractory obsessive–compulsive disorder
Journal of Affective Disorders
(2006) - et al.
Cortical surface-based analysis.II: inflation, flattening, and a surface-based coordinate system
NeuroImage
(1999) - et al.
Machine learning classifiers and fMRI: a tutorial overview
NeuroImage
(2009) - et al.
Can neuroimaging be used as a support to diagnosis of borderline personality disorder? An approach based on computational neuroanatomy and machine learning
Journal of Psychiatric Research
(2012) - et al.
Functional neuroimaging and the neuroanatomy of obsessive–compulsive disorder
Psychiatric Clinics of North America
(2000) - et al.
Identifying patients with obsessive–compulsive disorder using whole-brain anatomy
NeuroImage
(2007) - et al.
fMRI pattern recognition in obsessive–compulsive disorder
NeuroImage
(2012) - et al.
Distinguishing between unipolar depression and bipolar depression: current and future clinical and neuroimaging perspectives
Biological Psychiatry
(2013) - et al.
Methods for analysis of skewed data distributions in psychiatric clinical studies: working with many zero values
American Journal of Psychiatry
(2004)