Schizophrenia (SZ) stands as a severe psychiatric disorder. This study applied diffusion tensor imaging (DTI) data in conjunction with graph neural networks to distinguish SZ patients from normal controls (NCs) and showcases the superior performance of a graph neural network integrating combined fractional anisotropy and fiber number brain network features, achieving an accuracy of 73.79% in distinguishing SZ patients from NCs. Beyond mere discrimination, our study delved deeper into the advantages of utilizing white matter brain network features for identifying SZ patients through interpretable model analysis and gene expression analysis. These analyses uncovered intricate interrelationships between brain imaging markers and genetic biomarkers, providing novel insights into the neuropathological basis of SZ. In summary, our findings underscore the potential of graph neural networks applied to multimodal DTI data for enhancing SZ detection through an integrated analysis of neuroimaging and genetic features.
Humans ; Schizophrenia/pathology* ; Diffusion Tensor Imaging/methods* ; Male ; Female ; Adult ; Brain/metabolism* ; Young Adult ; Middle Aged ; White Matter/pathology* ; Gene Expression ; Nerve Net/diagnostic imaging* ; Graph Neural Networks

ObjectiveTo develop and validate a predictive model to individually predict the risk of patients with stroke in the eICU Collaborative Research Database for early clinical identification and intervention. MethodIndividual patient data （200 859 cases） from a national multicenter cohort study （eICU database） were selected， and the patients with stroke in neurological diseases （9 037 cases） were selected for statistical analysis. The main outcome was hospital mortality. The Glasgow Coma scale （GCS） was used to divide all patients with stroke into stroke in meridian and stroke in viscera （GCS≤14 for stroke in viscera and GCS=15 for stroke in meridian）. The patients were then divided into a training set and a test set according to 7∶3， respectively， to evaluate the differences in hospital mortality between the two groups. The multivariate logistic regression was used to analyze the related factors affecting the prognosis of the two groups， and a predictive model was established. Receiver operator characteristic （ROC） curves were used to assess the discrimination of the predictive model. ResultThe predictive model based on 9 037 patients with stroke was established. The predictors of the stroke in meridian （4 475 cases） included pulmonary infection， mechanical ventilation， acute physiology， and chronic health status scoring system Ⅳ （APACHE Ⅳ） score. The predictors of the stroke in viscera （4 562 cases） included anticoagulation therapy （AT）， mechanical ventilation， acute physiology， and APACHE Ⅳ score. According to the predictors， the predictive models of the stroke in meridian and the stroke in viscera were constructed， respectively. The areas under the curve （AUC） of ROC of the training set and the test set of the predictive models of the stroke in meridian were 0.845 ［95% confidence interval （CI） （0.811， 0.879）］ and 0.807 ［95% CI （0.751， 0.863）］， respectively. The areas under the ROC curve of the training set and test set of the predictive models of the stroke in viscera were 0.799 ［95% CI （0.781， 0.817）］ and 0.805 ［95% CI （0.778， 0.832）］， respectively. The AUC of the predictive model of the training set and the test set were both above 0.7. ConclusionThe model established in this study can conveniently， directly， and accurately predict the hospital mortality risk of patients with stroke. Physicians and other healthcare professionals can use this predictive approach to provide early care planning and clinical interventions for patients with stroke during their hospital stay.