AI post-intervention operational and functional outcomes prediction in ischemic stroke patients using MRIs

Area of Science:

• Artificial Intelligence in Medicine
• Medical Imaging Analysis
• Stroke Outcome Prediction

Background:

• Predicting short-term (length of stay) and long-term (90-day Modified Rankin Scale) outcomes for acute ischemic stroke (AIS) patients is clinically challenging.
• Current machine learning models primarily use clinical features, with limited exploration of advanced deep learning for integrating imaging biomarkers.
• Whole 2.5D image fusion techniques using deep learning for stroke prognosis remain underexplored.

Purpose of the Study:

• To develop and evaluate a novel autoencoder-based method for integrating 2.5D diffusion weighted imaging (DWI) with clinical features.
• To refine the prediction of operational and functional outcomes in acute ischemic stroke patients.
• To explore the potential of deep learning for whole image fusion in stroke prognosis.

Main Methods:

• Utilized autoencoders for integrating 2.5D DWI data with patient clinical features.
• Developed a novel deep learning approach for whole 2.5D image fusion.
• Evaluated the method on a comprehensive dataset of AIS patients.

Main Results:

• The autoencoder-based method demonstrated comparable performance to traditional convolutional neural network (CNN) fusion methods and clinical data alone.
• Achieved an AUC of 0.817 for predicting length of stay > 8 days.
• Achieved an AUC of 0.754 for predicting 90-day mRS > 2.

Conclusions:

• The novel integration of imaging and clinical data offers computational and operational advantages for stroke prognosis.
• This approach has the potential to enhance personalized patient management and healthcare operational decision-making.
• Further validation is required before widespread clinical adoption.