Related Experiment Video
Updated: Oct 18, 2025
14:08
Automated Midline Shift and Intracranial Pressure Estimation based on Brain CT Images
Published on: April 13, 2013
43.0K
A Robust Deep Learning Segmentation Method for Hematoma Volumetric Detection in Intracerebral Hemorrhage
Nannan Yu1, He Yu1, Haonan Li2
1Department of Artificial Intelligence, School of Electrical Engineering and Automation, Jiangsu Normal University, Xuzhou, China (N.Y., H.Y.).
Stroke
|October 4, 2021
Summary
A novel deep learning model, DR-UNet, accurately segments hematomas on CT scans for intracerebral hemorrhage (ICH) patients. This method provides efficient and reliable hematoma volume analysis, comparable to expert clinicians.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Neurosurgery
Background:
- Hematoma volume (HV) is crucial for diagnosing intracerebral hemorrhage (ICH) severity and guiding treatment.
- Accurate and rapid HV measurement is essential for clinical decision-making.
Purpose of the Study:
- To develop a robust deep learning segmentation method for fast and accurate HV analysis using computed tomography (CT).
- To evaluate the performance of the developed model against existing methods and expert clinicians.
Main Methods:
- A novel dimension reduction UNet (DR-UNet) model was developed for CT image segmentation and HV measurement.
- The model was trained and validated on retrospective (12,568 slices) and prospective (1,257 slices) ICH patient datasets.
- DR-UNet's performance was compared with UNet, fuzzy clustering, active contour methods, and the Coniglobus formula.
Main Results:
- DR-UNet achieved expert-level performance in segmenting hematomas on independent test datasets (Dice scores of 0.861±0.139 and 0.874±0.130).
- HV measurements from DR-UNet strongly correlated with manual segmentation (R²=0.9979).
- DR-UNet demonstrated superior robustness in segmenting irregularly shaped and extra-axial hematomas compared to UNet.
Conclusions:
- DR-UNet accurately segments hematomas and quantifies HV from CT scans in ICH patients.
- The model offers improved accuracy and efficiency over existing methods, performing comparably to expert clinicians.
- DR-UNet's robust performance supports its potential for integration into clinical deep learning systems for various applications.
