Neuroimaging-aided differential diagnosis of the depressive state

Highlights

• No imaging-guided differential diagnosis has yet been established in psychiatry.

• We measured fronto-temporal NIRS signals induced by a brief verbal fluency task.

• We examined the accuracy of single-subject classification among major disorders.

• NIRS distinguished unipolar depression from schizophrenia and bipolar depression.

• Our results suggest a potential role for NIRS in guiding personalised care.

Abstract

A serious problem in psychiatric practice is the lack of specific, objective biomarker-based assessments to guide diagnosis and treatment. The use of such biomarkers could assist clinicians in establishing differential diagnosis, which may improve specific individualised treatment. This multi-site study sought to develop a clinically suitable neuroimaging-guided diagnostic support system for differential diagnosis at the single-subject level among multiple psychiatric disorders with depressive symptoms using near-infrared spectroscopy, which is a compact and portable neuroimaging method. We conducted a multi-site, case–control replication study using two cohorts, which included seven hospitals in Japan. The study included 673 patients (women/men: 315/358) with psychiatric disorders (major depressive disorder, bipolar disorder, or schizophrenia) who manifested depressive symptoms, and 1007 healthy volunteers (530/477). We measured the accuracy of the single-subject classification in differential diagnosis among major psychiatric disorders, based on spatiotemporal characteristics of fronto-temporal cortical haemodynamic response patterns induced by a brief (< 3 min) verbal fluency task. Data from the initial site were used to determine an optimal threshold, based on receiver-operator characteristics analysis, and to generate the simplest and most significant algorithm, which was validated using data from the remaining six sites. The frontal haemodynamic patterns detected by the near-infrared spectroscopy method accurately distinguished between patients with major depressive disorder (74.6%) and those with the two other disorders (85.5%; bipolar disorder or schizophrenia) that presented with depressive symptoms. These results suggest that neuroimaging-guided differential diagnosis of major psychiatric disorders developed using the near-infrared spectroscopy method can be a promising biomarker that should aid in personalised care in real clinical settings. Potential confounding effects of clinical (e.g., age, sex) and systemic (e.g., autonomic nervous system indices) variables on brain signals will need to be clarified to improve classification accuracy.

Introduction

Among non-communicable diseases, neuropsychiatric conditions, including depression, contribute most significantly to overall disability-adjusted life years (DALYs), surpassing both cardiovascular disease and cancer (Mathers and Loncar, 2006, Prince et al., 2007). Therefore, early and accurate diagnosis and treatment are critical in psychiatric disorders, for which the development of specific biomarkers is of special importance. Currently, however, the diagnostic process in psychiatry is mainly based on patients' reports of symptoms, observed behaviours and disease course. Overcoming the limitations of relying on clinical interviews alone for the diagnosis of psychiatric disorders has been a great challenge.

To complicate this issue further, the manifestation of only a major depressive episode hampers the reliable differentiation of major depressive disorder (MDD) from bipolar disorder (BP) or schizophrenia (SZ) based on the Diagnostic and Statistical Manual of Mental Disorders (DSM) criteria alone (Zimmermann et al., 2009). Although many clinical symptoms are common to various psychiatric disorders, depressive symptoms are particularly ubiquitous in the disease process or clinical staging of various psychiatric disorders (Hafner et al., 2005). For instance, differentiation between BP presenting with depressive symptoms and unipolar MDD is a topical issue (Akiskal et al., 1995). Indeed, most patients with BP with depressive symptoms are initially diagnosed with and treated for MDD (Akiskal et al., 1995, Goldberg et al., 2001). Therefore, biomarkers that can facilitate early and accurate differentiation of BP with depressive symptoms from MDD are necessary.

In addition, depressive symptoms that fulfil the operational diagnostic criteria for a depressive episode/major depression can also occur at any stage of SZ and can contribute substantially to its associated morbidity and even mortality (an der Heiden et al., 2005). The differentiation of SZ from MDD, especially in the early stages, is also important because patients with SZ also exhibit non-psychotic and non-specific prodromal symptoms (e.g., depressive or negative symptoms and cognitive deficits) for several years before the onset of full-blown psychosis (McGorry et al., 2008). Therefore, the availability of clinically useful and cost-effective biomarkers for the differential diagnosis of major psychiatric disorders would likely enhance patient management, improve treatment/therapeutic response and lead to targeted therapies tailored to the individual (Holsboer, 2008). Despite their potential, to date, no such biomarkers have been established.

Functional imaging studies are one source of potential biomarkers (Gur et al., 2007, Phillips and Vieta, 2007); these studies have previously elucidated subtle brain abnormalities in patients with major psychiatric disorders relative to healthy control (HC) individuals and have been applied to the differential diagnosis of psychiatric disorders (e.g., to differentiate MDD from SZ, Barch et al., 2003 or BP, Almeida et al., 2009). However, some functional neuroimaging techniques are limited by the fact that, during the procedure, the individuals need to be placed in an uncomfortable or unnatural setting (e.g., lying in a supine position in a narrow gantry with the head fixed during the entire examination), for accurate measurement.

In contrast, multi-channel near-infrared spectroscopy (NIRS) using near-infrared light provides a completely non-invasive measurement of the spatiotemporal characteristics of brain function in ordinary clinical settings and allows patients to be comfortably seated in a well-lit room; therefore, it is considered a method for ‘real-world neuroimaging’. Additionally, NIRS has relatively low maintenance costs and does not involve ionising radiation or objectionable noise; thus, it can be repeated as needed even for patients with psychiatric disorders. The utility and limitations of NIRS have been discussed extensively in previous reports (Ferrari and Quaresima, 2012, Obrig and Villringer, 2003, Strangman et al., 2002a). NIRS allows the measurement of haemoglobin concentration changes (1) only in the cortical surface area located immediately beneath the probes, but not in deeper brain structures, and (2) with limited spatial resolution, although it has a high temporal resolution. In NIRS, typical cortical activation represents not only decreased concentration of deoxy-haemoglobin ([deoxy-Hb]), which is considered the main source of blood oxygenation level-dependent (BOLD) contrast increase in functional magnetic resonance imaging (fMRI), but also a relatively larger increase in oxy-haemoglobin concentration ([oxy-Hb]) (Fig. 1).

The verbal fluency task (VFT) is a cognitive task that is used as a neuropsychological test or a neuroimaging task. The VFT elicits different abnormalities relevant to each diagnostic group of major psychiatric disorders (Curtis et al., 2001, Zanelli et al., 2010). We previously developed a very brief (< 3 min) VFT and used it to investigate the differential fronto-temporal haemodynamic pattern between MDD and SZ (Suto et al., 2004) or MDD and BP (Kameyama et al., 2006), as well as the relationship between NIRS signals and functional impairment in SZ (Takizawa et al., 2008). We also found functional NIRS abnormalities in individuals at ultra-high risk for SZ and patients with first-episode psychosis (Koike et al., 2011). However, the clinical applicability of NIRS to the differential diagnosis of individuals remains uncertain. In this study, we extended our translational approach to replicate our previous findings (Kameyama et al., 2006, Suto et al., 2004) in a seven-site collaborative study using a large, fully independent sample set, and to evaluate the application of NIRS to psychiatric differential diagnosis in natural clinical settings.

Specifically, we used NIRS with wide coverage of the prefrontal and temporal cortices to investigate whether the frontal and temporal brain haemodynamic responses induced by cognitive activation could serve as biomarkers of underlying major psychiatric disorders with depression. To validate the reproducibility and generalisability of the results, we applied an algorithm developed using the data generated at the initial site to the test data derived from the remaining 6 sites. We hypothesised that the spatiotemporal characteristics of the haemodynamic responses detected by NIRS would not only differentiate patients with psychiatric disorders from HCs with acceptable sensitivity and specificity, but would also differentiate correctly and with a high concordance rate patients with MDD from patients with bipolar disorder and schizophrenia who present with depressive symptoms.