Related Experiment Video
Updated: Dec 31, 2025
10:25
Brain Infarct Segmentation and Registration on MRI or CT for Lesion-symptom Mapping
Published on: September 25, 2019
49.2K
Automated segmentation of acute stroke lesions using a data-driven anomaly detection on diffusion weighted MRI
Sanaz Nazari-Farsani1, Mikko Nyman2, Tomi Karjalainen1
1Turku PET Centre and Turku University Hospital, University of Turku, Finland.
Journal of Neuroscience Methods
|January 7, 2020
Summary
A new automated method accurately detects acute ischemic stroke (AIS) lesions on 3D MRI scans. This fast, low-computation approach aids clinical diagnosis and improves patient outcomes.
Area of Science:
- Medical Imaging
- Neurology
- Artificial Intelligence
Background:
- Accurate delineation of acute ischemic stroke (AIS) lesions is critical for improving patient outcomes.
- Existing lesion segmentation methods often suffer from low sensitivity, high computational demands, and interpretability challenges.
Purpose of the Study:
- To develop and validate a fully automated method for localizing and segmenting AIS lesions using multimodal 3D MRI.
- To create a classifier for distinguishing stroke from non-stroke cases based on segmented lesions.
Main Methods:
- A fully automated method was developed using the Crawford-Howell t-test to compare 192 multimodal 3D MRI scans (106 stroke, 86 healthy).
- The method localizes and segments AIS lesions, followed by a classifier to discriminate stroke/non-stroke images.
Main Results:
- The automated method achieved a mean Dice similarity coefficient (DSC) of 0.50 ± 0.21 for lesion segmentation.
- Stroke classification yielded a mean accuracy of 73%, with 84% sensitivity and 69% specificity.
Conclusions:
- The developed method is a significant improvement over existing techniques, offering comparable performance with reduced computational requirements.
- The approach is fast, straightforward, requires minimal computation, and can be integrated into clinical diagnostic workflows, showing good agreement with expert assessments.
