Related Experiment Video
Updated: Jul 6, 2026
14:08
Automated Midline Shift and Intracranial Pressure Estimation based on Brain CT Images
Published on: April 13, 2013
42.6K
USING CONVOLUTIONAL NEURAL NETWORK-BASED SEGMENTATION FOR IMAGE-BASED COMPUTATIONAL FLUID DYNAMICS SIMULATIONS OF
Mostafa Rezaeitaleshmahalleh1, Zonghan Lyu1, Nan Mu1
1Dept. of Biomedical Engineering, Michigan Technological University, 1400 Townsend Drive Houghton, Michigan 49931, USA.
Journal of Mechanics in Medicine and Biology
|March 25, 2024
Summary
Artificial intelligence (AI) segmentation accelerates the creation of patient-specific computational fluid dynamics (CFD) models for intracranial aneurysms. AI segmentation shows high agreement with manual methods, making it a feasible approach for clinical translation.
Area of Science:
- Biomedical Engineering
- Medical Imaging
- Computational Science
Background:
- Image-based computational fluid dynamics (CFD) simulations offer patient-specific hemodynamic insights.
- Current manual segmentation for CFD model creation is time-consuming, limiting clinical application.
- Deep learning-based AI segmentation presents a potential solution to automate and accelerate this process.
Purpose of the Study:
- To evaluate the feasibility of using AI segmentation for rapid CFD model generation.
- To compare the accuracy of AI-segmented CFD models against manually segmented models.
- To assess the impact of AI segmentation on morphological and hemodynamic analysis of intracranial aneurysms.
Main Methods:
- Two AI segmentation methods (MIScnn, DeepMedic) were applied to 3D rotational angiography data of intracranial aneurysms.
- CFD models were generated using both AI and manual segmentation by two human users.
- Morphological and hemodynamic parameters were compared between AI-generated and manually-generated models using ICC, Bland-Altman plots, and PCC.
Main Results:
- AI segmentation demonstrated almost perfect agreement with manual segmentation for all morphological parameters.
- High agreement was observed for five out of eight hemodynamic parameters.
- Moderate agreement was found for the remaining three hemodynamic parameters, indicating areas for further refinement.
Conclusions:
- AI segmentation is a feasible method for accelerating the creation of image-based CFD models.
- The high agreement suggests AI segmentation can reliably capture key patient-specific hemodynamic information.
- Further development of AI segmentation techniques holds promise for routine clinical translation of CFD simulations.
