A Machine Learning Approach for Classifying Ischemic Stroke Onset Time From Imaging.

Area of Science:

• Neurology
• Radiology
• Artificial Intelligence in Medicine

Background:

• Current stroke treatment decisions rely on clinical history for time since stroke (TSS) determination.
• Patients with unknown or unwitnessed TSS are often excluded from time-sensitive treatments like thrombolysis.
• Accurate TSS estimation is crucial for optimizing acute stroke care.

Purpose of the Study:

• To develop and validate a machine learning (ML) approach for classifying TSS using routinely acquired magnetic resonance (MR) imaging sequences.
• To enhance ML model performance by incorporating deep learning-extracted features from MR perfusion-weighted images.
• To provide an imaging-based decision support tool for guiding acute stroke treatment.

Main Methods:

• Extracted imaging features from standard MR sequences.
• Trained ML models for TSS classification.
• Developed a deep learning model to extract hidden representations from MR perfusion-weighted images.
• Integrated deep features with conventional MR features for improved classification.

Main Results:

• The best ML classifier achieved an area under the curve (AUC) of 0.765.
• Achieved a sensitivity of 0.788 and a negative predictive value (NPV) of 0.609.
• Demonstrated superior performance compared to existing methods.
• Validated model robustness across variations in imaging parameters and acquisition years.

Conclusions:

• ML analysis of MR imaging can accurately classify time since stroke (TSS).
• Incorporating deep learning features significantly improves classification performance.
• This imaging-based approach offers a promising tool to aid clinical decisions for stroke treatment, particularly for patients with unknown onset times.