A Multimodal Ensemble Deep Learning Model for Functional Outcome Prognosis of Stroke Patients.

Area of Science:

• Neuroimaging
• Artificial Intelligence in Medicine
• Stroke Research

Background:

• Accurate prediction of functional outcomes in acute ischemic stroke (AIS) is critical for clinical decision-making.
• Current prediction methods often rely on single data modalities, limiting accuracy.
• Optimal resource utilization in stroke care necessitates precise outcome forecasting.

Purpose of the Study:

• To develop an ensemble deep learning model for predicting 90-day functional outcomes in AIS patients.
• To integrate multimodal data, including imaging and clinical information, for enhanced prediction accuracy.
• To compare the performance of the multimodal ensemble model against single-modality models.

Main Methods:

• Utilized data from the Korean Stroke Neuroimaging Initiative database.
• Constructed an ensemble model combining 3D CNNs for diffusion-weighted imaging and FLAIR, and a DNN for clinical data.
• Predicted 90-day functional independence using the modified Rankin Scale (mRS) score of 3-6.

Main Results:

• The ensemble model achieved an Area Under the Curve (AUC) of 0.830 (standard CV) and 0.779 (time-based CV).
• This multimodal model significantly outperformed individual models based on diffusion-weighted imaging (b-value 1,000 s/mm2, apparent diffusion coefficient map) and FLAIR.
• Performance was also superior to models using only clinical data.

Conclusions:

• Integrating multimodal imaging and clinical data significantly improves the prediction of 90-day functional outcomes in AIS patients.
• The developed ensemble deep learning model offers a superior approach compared to single-data-modality predictions.
• This finding supports the use of multimodal data integration for more accurate stroke outcome prediction.