A new network for sleep and breathing analysis after stroke
The MM-Net framework performs simultaneous sleep staging and respiratory-event detection.
Researchers developed MM-Net, a multimodal multi-task network. It performs sleep staging and sleep apnea assessment after ischemic stroke.
The system uses cross-fitting and ridge-based embedding alignment to resolve representation inconsistency. It fuses sleep-adapted EEG-EOG representations and engineered neural descriptors with learned cardiorespiratory features. These are combined through cross-attention and shared temporal modeling.
EEG-EOG representations are sleep-adapted. Cardiorespiratory features are learned. Sleep staging is the classification of sleep stages. Respiratory-event detection is the identification of breathing disturbances.
On the iSLEEPS dataset with 99 patients with subacute ischemic stroke, the model achieved 76.3% staging accuracy with a Cohen's = 0.656. It also reached a 0.825 respiratory-event AUC.
The model achieved 86.4% staging accuracy on Sleep-EDF. On ISRUC-I, it reached 80.3% staging accuracy and a 0.889 respiratory AUC. Without retraining, the iSLEEPS-trained model retained respiratory AUCs of 0.833 on ISRUC-I and 0.841 on UCD.
The findings support representation-consistent neural adaptation and joint physiological modeling across stroke, healthy-sleep, and sleep-disorder cohorts.
Key facts
- MM-Net achieved 76.3% staging accuracy on 99 subacute ischemic stroke patients.
- The model reached 86.4% staging accuracy on the Sleep-EDF dataset.
- The system performs simultaneous sleep staging and respiratory-event detection.