Structural covariance is described as coordinated variation in brain morphological features, such as cortical thickness and volume, among brain structures functionally or anatomically interconnected to one another. Structural covariance networks, based on graph theory, have been studied in mental disorders. This analysis can help in understanding the brain mechanisms of schizophrenia and bipolar disorder. We investigated cortical thickness-based individualized structural covariance networks in patients with schizophrenia and bipolar disorder. T1-weighted magnetic resonance images were obtained from 39 patients with schizophrenia, 37 patients with bipolar disorder type I, and 32 healthy controls, and cortical thickness was analyzed via a surface-based morphometry analysis. The structural covariance of cortical thickness was calculated at the individual level, and covariance networks were analyzed based on graph theoretical indices: strength, clustering coefficient (CC), path length (PL) and efficiency. At the global level, both patient groups showed decreased strength, CC and efficiency, and increased PL, compared to healthy controls. In bipolar disorder, we found intermediate network measures among the groups. At the nodal level, schizophrenia patients showed decreased CCs in the left suborbital sulcus and the right superior frontal sulcus, compared to bipolar disorder patients. In addition, patient groups showed decreased CCs in the right insular cortex and the left superior occipital gyrus. Global-level network indices, including strength, CCs and efficiency, positively correlated, while PL negatively correlated, with the positive symptoms of the Positive and Negative Syndrome Scale for patients with schizophrenia. The nodal-level CC of the right insular cortex positively correlated with the positive symptoms of schizophrenia, while that of the left superior occipital gyrus positively correlated with the Young Mania Rating Scale scores for bipolar disorder. Altered cortical structural networks were revealed in patients, and particularly, the prefrontal regions were more altered in schizophrenia. Furthermore, altered cortical structural networks in both patient groups correlated with core pathological symptoms, indicating that the insular cortex is more vulnerable in schizophrenia, and the superior occipital gyrus is more vulnerable in bipolar disorder. Our individualized structural covariance network indices might be promising biomarkers for the evaluation of patients with schizophrenia and bipolar disorder.

Mild cognitive impairment (MCI) is an intermediary stage between typical cognitive aging and dementia. Identifying reliable biomarkers for early detection of MCI is crucial for slowing disease progression. This study explored multimodal biomarkers associated with amyloid-positive MCI and assessed wearable EEG/ERP and MRI features using machine learning. This study included 70 patients with MCI with amyloid-positive positron emission tomography (PET) scans and 140 cognitively normal (CN) individuals with negative PET scans. Data were collected using prefrontal wearable EEG/ERP and MRI devices. EEG data were obtained in three paradigms: resting-state, sensory-evoked potentials, and selective attention tasks. Machine learning techniques were applied to classify MCI and CN groups using these biomarkers. EEG/ERP analysis suggested differences in frontal alpha asymmetry, power ratio, and ERP components between the MCI and CN groups. MRI revealed atrophy of the hippocampus and accumbens along with cortical thinning in specific brain regions. Combining EEG/ERP and MRI features in a machine learning model yielded the highest classification accuracy of 80.90%. Prefrontal wearable EEG/ERP and MRI offered complementary candidate biomarkers of MCI in community-dwelling individuals using a machine learning approach. These findings support further validation of EEG/ERP and MRI biomarkers in larger and externally validated cohorts.

The neuropathology of mood disorders, including the diagnostic transition from major depressive disorder (MDD) to bipolar disorder (BD), is poorly understood. This study investigated resting-state electroencephalography (EEG) activity in patients with MDD and those whose diagnosis changed from MDD to BD. Among sixty-eight enrolled patients with MDD, the diagnosis of 17 patients converted to BD during the study period. We applied machine learning techniques to differentiate the two groups using sensor- and source-level EEG features. At the sensor level, patients with BD showed higher theta band power at the AF3 channel and low-alpha band power at the FC5 channel compared to patients with MDD. At the source level, patients with BD showed higher theta band activity in the right anterior cingulate and low-alpha band activity in the left parahippocampal gyrus. These four EEG features were selected for discriminating between BD and MDD with the best classification performance showing an accuracy of 80.88%, a sensitivity of 76.47%, and a specificity of 82.35%. Our findings revealed distinct theta and low-alpha band activities in patients with BD and MDD. These differences could potentially serve as candidate neuromarkers for the diagnosis and diagnostic transition between the two distinct mood disorders.