Related Experiment Video
Updated: Mar 6, 2026
14:08
Automated Midline Shift and Intracranial Pressure Estimation based on Brain CT Images
Published on: April 13, 2013
43.7K
PItcHPERFeCT: Primary Intracranial Hemorrhage Probability Estimation using Random Forests on CT
John Muschelli1, Elizabeth M Sweeney1, Natalie L Ullman2
1Department of Biostatistics, Bloomberg School of Public Health, Johns Hopkins University, Baltimore, MD, USA.
Neuroimage. Clinical
|March 10, 2017
Summary
This study developed an automated method for segmenting intracerebral hemorrhage (ICH) on CT scans using a random forest algorithm. The automated approach achieved high accuracy, comparable to manual segmentation, offering a faster and more consistent alternative for stroke assessment.
Area of Science:
- Medical imaging analysis
- Neurology
- Artificial intelligence in healthcare
Background:
- Intracerebral hemorrhage (ICH) is a severe type of stroke, often assessed using CT scans.
- Manual segmentation of ICH on CT scans is the gold standard but is time-consuming and prone to variability.
- Automated segmentation methods are needed to improve efficiency and consistency in ICH volume estimation.
Purpose of the Study:
- To develop and validate a fully automated segmentation approach for intracerebral hemorrhage (ICH) using CT scans.
- To compare the performance of a random forest algorithm against other machine learning models for ICH segmentation.
- To provide an accessible tool for automated ICH segmentation.
Main Methods:
- Utilized 112 baseline CT scans from the MISTIE trial patients.
- Developed four models: logistic regression, LASSO-penalized logistic regression, generalized additive model (GAM), and random forest classifier.
- Validated models using Dice Similarity Index (DSI) and volume correlation against expert manual segmentations.
Main Results:
- The random forest model achieved the highest median Dice Similarity Index (DSI) of 0.899, outperforming logistic, LASSO, and GAM models.
- Correlation between manual and random forest predicted hemorrhage volumes was high (0.93).
- The automated random forest approach demonstrated accurate and reliable ICH segmentation across diverse imaging centers.
Conclusions:
- A fully automated random forest segmentation approach provides accurate and efficient assessment of intracerebral hemorrhage.
- This method offers a viable alternative to manual segmentation, reducing time and inter-reader variability.
- An R package and Shiny application are available for implementing and testing the proposed ICH segmentation approach.
