A deep learning and radiomics based Alberta stroke program early CT score method on CTA to evaluate acute ischemic...

Area of Science:

• Medical Imaging
• Artificial Intelligence in Medicine
• Neurology

Background:

• Alberta Stroke Program Early CT Score (ASPECTS) is crucial for evaluating early ischemic changes in acute ischemic stroke.
• Physician-based ASPECTS assessment faces challenges with significant inter-observer variance.
• Accurate ASPECTS scoring guides critical treatment decisions and prognostic judgments.

Purpose of the Study:

• To develop an automated method combining deep learning and radiomics to improve ASPECTS accuracy and consistency.
• To reduce inter-observer variability in ASPECTS scoring for acute ischemic stroke.
• To assist physicians in more accurate and comprehensive stroke detection.

Main Methods:

• Utilized an improved encoding-decoding deep convolutional neural network for precise brain region segmentation.
• Employed Pyradiomics to extract and select radiomic features significantly associated with cerebral infarction.
• Trained machine learning classifiers using selected radiomic features to identify infarction in each brain region.

Main Results:

• Achieved a Dice coefficient of 0.79 for brain region segmentation.
• Identified three reliable radiomic features for cerebral infarction detection.
• The automated ASPECTS method demonstrated strong agreement with physician scores (intraclass correlation coefficient = 0.86).
• Achieved a high prediction performance (AUC = 0.95) for stroke detection.

Conclusions:

• Deep learning offers superior brain region segmentation compared to traditional methods.
• The radiomics-based classifier shows potential for assisting clinicians in stroke detection.
• This automated approach can significantly improve the consistency and accuracy of ASPECTS scoring.