Related Experiment Video
Updated: Sep 12, 2025
14:08
Automated Midline Shift and Intracranial Pressure Estimation based on Brain CT Images
Published on: April 13, 2013
42.8K
A Deep Learning Model to Detect Acute MCA Occlusion on High-Resolution Noncontrast Head CT
David A Fussell1, Jasmine L Lopez2, Peter D Chang2
1From the Department of Radiological Sciences, University of California, Irvine, Irvine, California fusselld@hs.uci.edu.
AJNR. American Journal of Neuroradiology
|August 8, 2025
Summary
A deep learning model accurately detects acute middle cerebral artery (MCA) occlusion using non-contrast CT (NCCT) scans. This tool can help identify stroke patients for thrombectomy, especially in resource-limited settings.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Neurology
Background:
- Acute ischemic stroke diagnosis relies heavily on advanced imaging techniques.
- Identifying middle cerebral artery (MCA) occlusions is critical for timely treatment decisions.
- Computed Tomography Angiography (CTA) is a standard but resource-intensive method for detecting occlusions.
Purpose of the Study:
- To evaluate the feasibility and accuracy of a deep learning (DL) model for identifying acute MCA occlusions.
- To determine if high-resolution non-contrast CT (NCCT) data can be effectively used by a DL model for this purpose.
- To assess the potential of DL in improving stroke triage and treatment accessibility.
Main Methods:
- A 3D deep learning (DL) model was trained on 4,648 retrospective NCCT scans (1.0 mm slice thickness or less).
- Ground truth for MCA thrombus was established using same-day CTA.
- The model performed per-voxel thrombus segmentation to estimate the likelihood of acute MCA occlusion.
- An independent test set of 1,011 NCCT scans was used for validation.
Main Results:
- The DL model achieved an AUROC of 0.952 for detecting M1 segment MCA occlusions.
- Accuracy for M1 segment occlusion detection was 93.6%, with 90.9% sensitivity and 93.6% specificity.
- Performance slightly decreased for M2 segment occlusions, with an AUROC of 0.884 and 93.2% accuracy.
- The model demonstrated high accuracy in identifying acute MCA occlusions from NCCT data.
Conclusions:
- A deep learning model can accurately detect acute MCA occlusions using high-resolution NCCT imaging.
- The model's performance approaches that of CTA, offering a potentially faster and more accessible diagnostic tool.
- This DL tool could significantly aid stroke triage by identifying thrombectomy candidates using NCCT alone, particularly in resource-constrained environments.
