DualDyConvNet: Dual-Stream Dynamic Convolution Network via Parameter-Efficient Fine-Tuning for Predicting Motor...

Area of Science:

• Neuroscience
• Biomedical Engineering
• Rehabilitation Medicine

Background:

• Stroke significantly impacts global health, affecting motor function recovery.
• Predicting early motor recovery potential is crucial for personalized stroke rehabilitation.
• Current methods for predicting motor prognosis have limited performance.

Purpose of the Study:

• To propose a novel DualDyConvNet framework for predicting motor recovery in subacute stroke patients.
• To utilize resting-state electroencephalogram (EEG) data for motor prognosis prediction.
• To enhance the accuracy and generalizability of motor recovery prediction models.

Main Methods:

• Developed a dual-stream dynamic convolution network (DualDyConvNet) with channel and spatial streams.
• Employed resting-state EEG data from subacute stroke patients (SMC and KIST datasets).
• Quantified motor recovery using the Fugl-Meyer Assessment of the upper limb.

Main Results:

• Achieved low root mean squared errors (RMSE) on internal validation: 0.070 ± 0.045 (SMC) and 0.223 ± 0.148 (KIST).
• Demonstrated superior performance compared to existing models.
• Showcased strong generalization through cross-dataset validation with and without Euclidean-space alignment (EA).

Conclusions:

• The DualDyConvNet framework effectively predicts motor function prognosis in stroke patients.
• This prediction capability can facilitate early, personalized rehabilitation planning.
• Improved prognostication has the potential to enhance the quality of life for stroke survivors.