A Dynamic Machine Learning Model to Predict Angiographic Vasospasm After Aneurysmal Subarachnoid Hemorrhage

Area of Science:

• Neuroscience
• Medical Imaging
• Machine Learning

Background:

• Aneurysmal subarachnoid hemorrhage (aSAH) poses significant risks, including the development of angiographic vasospasm (AV).
• Early and accurate prediction of AV is crucial for timely intervention and improved patient outcomes.
• Current prediction methods may lack the precision and dynamism required for complex clinical scenarios.

Purpose of the Study:

• To develop a precise, dynamic machine learning model for predicting AV in aSAH patients.
• To leverage daily transcranial Doppler ultrasound (TCD) data for enhanced predictive capabilities.
• To compare the performance of a dynamic LSTM model against static models.

Main Methods:

• Retrospective review of 424 aSAH patients.
• Collection of demographic, clinical, radiographic, and quantitative data (e.g., MAP, CPP, serum sodium, TCD parameters).
• Development and comparison of three models: Baseline Model (BM), Standard TCD Model (SM), and Long Short-Term Memory (LSTM) machine learning model.

Main Results:

• The LSTM model demonstrated the highest precision (0.571) and accuracy (0.776) in predicting AV.
• The LSTM model maintained high precision (0.488) and accuracy (0.803) for predicting AV within 5 days.
• Removing non-TCD data improved LSTM model precision to 0.824, highlighting the value of TCD data.

Conclusions:

• Longitudinal TCD data can power dynamic machine learning models for AV prediction.
• The developed LSTM model offers superior precision compared to static TCD measurements.
• This dynamic approach holds promise for improving AV prediction in aSAH management.