Related Experiment Video
Updated: Apr 16, 2026
09:59
A Magnetic Resonance Imaging Protocol for Stroke Onset Time Estimation in Permanent Cerebral Ischemia
Published on: September 16, 2017
14.7K
Excellent agreement between automated deep learning-based and manual diffusion-weighted imaging infarct volume
Yuki Sakamoto1, Junya Aoki2, Yuji Nishi1
1Department of Neurology, Nippon Medical School, Tokyo, Japan.
Journal of the Neurological Sciences
|April 14, 2026
Summary
Deep learning automated infarct volume measurement closely matches manual assessments in acute ischemic stroke patients. This reliable method shows similar prognostic value, supporting its use in hyperacute stroke management.
Area of Science:
- Neurology
- Medical Imaging
- Artificial Intelligence
Background:
- Diffusion-weighted imaging (DWI) infarct volume and growth are key biomarkers in acute ischemic stroke patients undergoing mechanical thrombectomy (MT).
- Manual measurement of these biomarkers is time-consuming and resource-intensive.
- The performance and prognostic value of deep learning (DL)-based automated segmentation for hyperacute DWI have not been fully validated.
Purpose of the Study:
- To evaluate the agreement between manual and DL-based automated infarct volume measurements on DWI in patients with acute ischemic stroke treated with MT.
- To assess the prognostic value of automated infarct volume measurements for predicting functional outcomes.
- To determine the feasibility and reliability of DL-based automated segmentation in hyperacute stroke management.
Main Methods:
- Retrospective analysis of 371 acute ischemic stroke patients treated with MT (2014-2019) with admission and 24-hour follow-up DWI scans.
- Infarct volumes measured manually by neurologists and automatically using DL software.
- Agreement assessed using Pearson's correlation, Bland-Altman analysis, and concordance correlation coefficient (CCC); prognostic ability compared using C-statistics.
Main Results:
- Manual and automated infarct volumes showed very strong correlation (r=0.94-0.97) with minimal bias.
- Concordance correlation coefficients (CCC) for agreement were high (0.935-0.971), comparable to human inter/intra-rater reliability.
- Predictive abilities for good functional outcome were similar for manual and automated measurements of infarct volume and growth.
Conclusions:
- DL-based automated infarct volume measurement demonstrates substantial agreement with manual assessment in acute ischemic stroke.
- The automated method shows comparable prognostic performance to manual measurements.
- DL-based automated segmentation is feasible and reliable for hyperacute stroke management.
