Multi-modal deep learning for predicting functional outcomes in intracerebral hemorrhage using 3D CT and clinical...

Area of Science:

• Medical Imaging and Artificial Intelligence
• Neurology and Clinical Outcomes Research
• Health Informatics and Predictive Modeling

Background:

• Intracerebral hemorrhage (ICH) poses significant challenges in predicting patient functional recovery.
• Accurate early prediction of outcomes is crucial for effective clinical management and personalized treatment strategies.
• Current prediction methods often rely on limited data, necessitating more comprehensive approaches.

Purpose of the Study:

• To develop and validate a novel multimodal deep learning framework for predicting 90-day functional outcomes in ICH patients.
• To integrate 3D computed tomography (CT) imaging and early clinical text data for enhanced predictive accuracy.
• To assess the model's performance against single-modality approaches and ensure interpretability.

Main Methods:

• A retrospective study of 508 ICH patients from two medical centers.
• A multimodal deep learning model integrating 3D CT features (3D CNN) and clinical text semantics (BioClinicalBERT) via a dual-stream attention mechanism.
• Functional outcomes assessed using the modified Rankin Scale (mRS); model interpretability explored using Grad-CAM.

Main Results:

• The multimodal model achieved high accuracy (0.867) and AUC (0.899) in internal validation.
• On an external test set, the model demonstrated robust performance with an accuracy of 0.821 and AUC of 0.846.
• Performance significantly outperformed the image-only model (internal: accuracy 0.696, AUC 0.718; external: accuracy 0.675, AUC 0.737).

Conclusions:

• The proposed framework effectively integrates 3D CT imaging and clinical text for accurate prediction of long-term functional outcomes in ICH patients.
• The model exhibits robust performance and interpretability, positioning it as a valuable early clinical risk stratification tool.
• This approach holds potential for advancing personalized precision medicine in ICH patient care.