An explainable streaming early identification model for early neurological deterioration based on coordinated fusion of ECG waveforms and vital signs - Takeaways - MDSpire

An explainable streaming early identification model for early neurological deterioration based on coordinated fusion of ECG waveforms and vital signs

  • By

  • Yuyan Zhang

  • Shihan Yao

  • Bo Wen

  • Jinjie Liu

  • May 7, 2026

  • 0 min

Share

  • 1

    DSF-Net is a dual-stream multimodal system designed for continuous postoperative monitoring of stroke patients to predict Early Neurological Deterioration (END).

  • 2

    The model combines high-frequency physiological signals processed by a 1D-CNN with clinical metrics analyzed by a multilayer perceptron (MLP).

  • 3

    DSF-Net achieves an AUC of 0.9996 and an F1-score of 0.9841, significantly improving END detection compared to traditional methods.

  • 4

    The model utilizes cost-sensitive learning and dynamic threshold optimization to address class imbalance and enhance predictive accuracy.

  • 5

    Integrated Gradients analysis demonstrates DSF-Net's ability to identify subtle waveform changes, providing interpretable early warnings for clinical use.

Original Source(s)

An explainable streaming early identification model for early neurological deterioration based on coordinated fusion of ECG waveforms and vital signs

  • by Yuyan Zhang, Shihan Yao, Bo Wen, Jinjie Liu

  • May 7, 2026

Related Content

Linking atrial fibrillation to cerebral small vessel disease: a cross-sectional study with predictive analytics

Top 10 States With Rising Demand for Physicians

These 10 states report physician shortages, projected physician supply gaps, or physician-to-population ratios below national benchmarks.

Case Report: Neuronal intranuclear inclusion disease mimicking recurrent stroke in the setting of intracranial stenosis