Frontiers in psychiatry · 2026
Multiplex functional connectome graph transformer for cognitive vulnerability in dialysis-treated chronic kidney disease.
СтатьяНазвание и аннотация: Перевод в очереди
Обработка статьи · этапы, ошибки и управление
Загрузка состояния…
Ошибки и ограничения
История · было → стало
Состояние обновляется каждые 5 секунд, пока блок открыт. Процент показывается только там, где известен объём работы. Пауза останавливает продолжение этапа; текущий запрос к источнику может завершиться.
Zhai Junjie, Pang Jiachang, Lei Xiangdong, Yan Kexu, Zhou Jingren
Аннотация
OBJECTIVE: Chronic kidney disease (CKD) requiring maintenance dialysis is accompanied by cognitive vulnerability and distributed brain-network dysfunction. This study evaluated whether complementary resting-state functional magnetic resonance imaging (rs-fMRI) connectome views could support subject-level characterization of CKD-related neurocognitive brain dysfunction.
METHODS: We analyzed a de-identified rs-fMRI dataset of 100 participants, including 52 patients with CKD receiving maintenance dialysis and 48 normal controls (NCs). The CKD group had lower Mini-Mental State Examination and Montreal Cognitive Assessment scores. AAL116 regional time series were represented as three 116 × 116 graph views: Pearson correlation, sparse representation, and Granger causality mapping. We developed MTGAF, a multiplex graph transformer combining modality-specific graph encoding, graph-transformer refinement, disease-aware readout, cross-modal attention, and adaptive fusion. Performance was evaluated with participant-level stratified 5-fold cross-validation and compared with BrainNetTF, BrainGNN, SVM, Random Forest, single-modality variants, and additional component-ablation variants. Robustness was assessed using pooled confusion matrices, bootstrap confidence intervals, calibration summaries, paired comparisons, and ROI importance analysis.
RESULTS: MTGAF achieved accuracy 0.8800 ± 0.0678, macro-F1 0.8796 ± 0.0680, and AUC 0.9297 ± 0.0437. The pooled confusion matrix contained 42 true negatives, 46 true positives, 6 false positives, and 6 false negatives. Bootstrap 95% CIs were 0.8100-0.9400 for accuracy, 0.8095-0.9396 for macro-F1, and 0.8696-0.9770 for pooled AUC. The full PC-SR-GCM model achieved higher accuracy and macro-F1 than single-modality variants. ROI importance highlighted cingulate, temporoparietal, sensorimotor, visual-association, frontostriatal, posterior cingulate, and temporal-association systems.
CONCLUSIONS: Multiplex modeling of complementary connectome information provides signals for characterizing CKD-related neurocognitive brain-network dysfunction. MTGAF is a research-stage imaging-AI framework for individualized brain-risk characterization in dialysis-treated CKD and should be further evaluated in larger multi-center cohorts.
METHODS: We analyzed a de-identified rs-fMRI dataset of 100 participants, including 52 patients with CKD receiving maintenance dialysis and 48 normal controls (NCs). The CKD group had lower Mini-Mental State Examination and Montreal Cognitive Assessment scores. AAL116 regional time series were represented as three 116 × 116 graph views: Pearson correlation, sparse representation, and Granger causality mapping. We developed MTGAF, a multiplex graph transformer combining modality-specific graph encoding, graph-transformer refinement, disease-aware readout, cross-modal attention, and adaptive fusion. Performance was evaluated with participant-level stratified 5-fold cross-validation and compared with BrainNetTF, BrainGNN, SVM, Random Forest, single-modality variants, and additional component-ablation variants. Robustness was assessed using pooled confusion matrices, bootstrap confidence intervals, calibration summaries, paired comparisons, and ROI importance analysis.
RESULTS: MTGAF achieved accuracy 0.8800 ± 0.0678, macro-F1 0.8796 ± 0.0680, and AUC 0.9297 ± 0.0437. The pooled confusion matrix contained 42 true negatives, 46 true positives, 6 false positives, and 6 false negatives. Bootstrap 95% CIs were 0.8100-0.9400 for accuracy, 0.8095-0.9396 for macro-F1, and 0.8696-0.9770 for pooled AUC. The full PC-SR-GCM model achieved higher accuracy and macro-F1 than single-modality variants. ROI importance highlighted cingulate, temporoparietal, sensorimotor, visual-association, frontostriatal, posterior cingulate, and temporal-association systems.
CONCLUSIONS: Multiplex modeling of complementary connectome information provides signals for characterizing CKD-related neurocognitive brain-network dysfunction. MTGAF is a research-stage imaging-AI framework for individualized brain-risk characterization in dialysis-treated CKD and should be further evaluated in larger multi-center cohorts.