CA-UNet Segmentation Makes a Good Ischemic Stroke Risk Prediction.

Area of Science:

• Medical Imaging
• Artificial Intelligence in Medicine
• Cardiovascular Disease

Background:

• Stroke remains a leading cause of death and disability globally.
• Early detection and treatment are crucial for managing ischemic stroke.
• Limited data and privacy concerns hinder automated stroke diagnosis research.

Purpose of the Study:

• To develop an automated 3D carotid artery segmentation model using Computed Tomography Angiography (CTA) images.
• To propose an ischemic stroke risk prediction model integrating 3D CTA data, electronic medical records, and medical history.
• To enhance the accuracy and reliability of stroke diagnosis and risk assessment.

Main Methods:

• Developed CA-UNet, a 3D segmentation model for carotid artery extraction from CTA scans.
• Investigated optimal down-sampling strategies for carotid segmentation.
• Designed a multi-scale loss function to preserve detailed features during segmentation.
• Integrated the segmentation model with patient data for ischemic stroke risk prediction.

Main Results:

• The CA-UNet model demonstrated effective automated extraction of carotid arteries.
• The multi-scale loss function successfully addressed feature loss during down-sampling.
• The combined prediction model showed efficacy in assessing ischemic stroke risk.
• Validation tests confirmed the reliability of both the segmentation and prediction models.

Conclusions:

• The proposed CA-UNet model offers a reliable method for automated carotid artery segmentation.
• The integrated risk prediction model shows promise for improving patient care and medical professional decision-making.
• This approach contributes to advancing intelligent diagnosis systems for stroke prevention.