Related Experiment Video
Updated: Dec 16, 2025
04:48
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
3.2K
Automated stroke lesion segmentation in non-contrast CT scans using dense multi-path contextual generative
Hulin Kuang1, Bijoy K Menon1, Wu Qiu1
1Department of Clinical Neurosciences, University of Calgary, Calgary, Alberta, T2N 2T9 Canada.
Physics in Medicine and Biology
|July 1, 2020
Summary
This study introduces an automated method using a 2D dense multi-path contextual generative adversarial network (MPC-GAN) to segment stroke lesions in CT scans. The MPC-GAN accurately identifies ischemic and hemorrhagic lesions, aiding in acute ischemic stroke patient prognosis.
Area of Science:
- Radiology
- Artificial Intelligence
- Medical Imaging
Background:
- Accurate stroke lesion volume measurement is crucial for assessing prognosis in acute ischemic stroke (AIS) patients.
- Current segmentation methods can be time-consuming and may lack precision.
- Automated segmentation of both ischemic and hemorrhagic lesions from non-contrast CT (NCCT) is needed.
Purpose of the Study:
- To develop and validate an automated segmentation method for ischemic and hemorrhagic lesions in AIS patients using NCCT scans.
- To improve the accuracy and efficiency of lesion volume assessment for patient prognosis.
Main Methods:
- Proposed a 2D dense multi-path contextual generative adversarial network (MPC-GAN).
- The generator utilized a dense multi-path 2D U-Net, regularized by a discriminator network.
- Incorporated contextual information, including bilateral intensity difference, distance map, and lesion location probability.
Main Results:
- Achieved a Dice coefficient (DC) of 70.6% for ischemic infarct segmentation and 76.5% for hemorrhage segmentation.
- Outperformed several benchmark segmentation methods.
- Demonstrated strong correlation between MPC-GAN segmented lesion volumes and manual measurements (Pearson correlation coefficients of 0.926 and 0.927).
Conclusions:
- The proposed MPC-GAN method accurately segments ischemic infarcts and hemorrhages from NCCT volumes in AIS patients.
- This automated approach shows potential for improving prognostic assessment in acute ischemic stroke care.
- The method offers a reliable tool for lesion volume quantification in clinical practice.
