A Semi-Supervised Learning Framework to Leverage Proxy Information for Stroke MRI Analysis.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Neurology

Background:

• Acute ischemic stroke (AIS) treatment is time-sensitive, requiring rapid intervention to prevent brain damage.
• Thrombolytic therapy, like tissue-plasminogen activator (tPA), has a narrow therapeutic window.
• DWI-FLAIR mismatch on MRI indicates potential benefit from thrombolysis, but its assessment has high inter-reader variability.

Purpose of the Study:

• To develop an automated method for detecting DWI-FLAIR mismatch in AIS patients.
• To compare the performance of the automated deep learning model against expert neuroradiologists.
• To investigate the utility of semi-supervised learning (SSL) with clinical proxy labels for model training.

Main Methods:

• A deep learning model was trained using MRI data to classify tissue clock.
• Time clock was used as a weak proxy label to supplement training in an SSL framework.
• The model was evaluated on an external, unseen dataset.

Main Results:

• The automated deep learning framework achieved a Receiver Operating Characteristic Area Under the Curve (ROC-AUC) of 74.30%.
• The study demonstrated improved detection of DWI-FLAIR mismatch compared to previous methods.
• Incorporating clinical proxy information into SSL enhanced model optimization.

Conclusions:

• Automated detection of DWI-FLAIR mismatch is feasible and effective for AIS patients.
• Deep learning with SSL can overcome inter-reader variability in MRI interpretation.
• This approach has the potential to improve the selection of patients for thrombolytic therapy.