Related Experiment Video
Updated: Aug 16, 2026
14:08
Automated Midline Shift and Intracranial Pressure Estimation based on Brain CT Images
Published on: April 13, 2013
Assessing CT-based volumetric analysis via deep learning for idiopathic normal pressure hydrocephalus
Meera Srikrishna1,2, Woosung Seo3, Anna Zettergren4
1Wallenberg Centre for Molecular and Translational Medicine, University of Gothenburg, Gothenburg 40530, Sweden.
Brain Communications
|August 15, 2026
Summary
Deep learning models can now accurately segment brain ventricles in CT scans for diagnosing normal pressure hydrocephalus (NPH). This automated approach shows high accuracy, comparable to expert assessments, and has significant clinical potential.
Area of Science:
- Neuroimaging
- Medical Diagnostics
- Artificial Intelligence
Background:
- Idiopathic normal pressure hydrocephalus (iNPH) diagnosis often relies on manual assessment of brain imaging.
- Current methods for evaluating ventriculomegaly in iNPH using computed tomography (CT) can be subjective and time-consuming.
- Deep learning has shown promise in medical image analysis, including intracranial tissue segmentation.
Purpose of the Study:
- To enhance the segmentation of ventricular cerebrospinal fluid (VCSF) in brain CT scans using a deep learning model.
- To assess the performance of automated CT-based volumetrics for diagnosing iNPH.
- To evaluate the clinical utility of deep learning-derived measures in hydrocephalus assessment.
Main Methods:
- A two-stage deep learning approach was developed, initially using a 2D U-Net model trained on healthy controls' MRI-VCSF labels and then refined with iNPH patient CT-VCSF labels.
- The model was trained on a large dataset including CT scans from healthy controls and iNPH patients, with external validation from multiple international clinical sites.
- Three CT-based volumetric measures (CTVMs) were derived for iNPH assessment.
Main Results:
- The automated VCSF segmentation demonstrated strong volumetric correlation with manual measurements in iNPH patients (ρ = 0.91, P < 0.001).
- CTVMs achieved high accuracy in differentiating iNPH patients from controls, with AUCs of 0.97 and 0.99 in external and internal validation datasets, respectively.
- The deep learning-derived measures performed comparably to gold-standard neuroradiology assessments, even with intraventricular shunt catheters present.
Conclusions:
- Deep learning-based automated CT volumetrics offer a robust method for quantifying morphological features in hydrocephalus.
- These automated measures show potential to improve the diagnostic work-up and treatment monitoring of iNPH patients.
- Given the wider availability of CT compared to MRI, this approach has considerable clinical impact for hydrocephalus management.
