Using Deep Learning and B-Splines to Model Blood Vessel Lumen from 3D Images.

Area of Science:

• Medical Imaging
• Computational Geometry
• Machine Learning

Background:

• Accurate geometric modeling of blood vessel lumen is essential for diagnosing and monitoring vascular diseases.
• Existing methods often assume simplified vessel cross-sections (circular/elliptical), limiting accuracy for complex geometries.
• Image segmentation can introduce spatial discretization errors, affecting lumen boundary localization.

Purpose of the Study:

• To develop a novel method for accurate geometric modeling of blood vessel lumen from 3D images.
• To overcome limitations of existing methods by avoiding assumptions of circular/elliptical cross-sections and image segmentation.
• To enable faster and more accurate vessel quantification for clinical applications and research.

Main Methods:

• Utilized parametric B-splines combined with image formation system equations to model lumen boundaries.
• Employed a feedforward neural network to efficiently identify model parameters from cross-section images.
• Avoided traditional image segmentation to maintain localization accuracy.

Main Results:

• The B-spline method accurately localized highly curved lumen boundaries without segmentation.
• The feedforward neural network identified model parameters significantly faster than a least-squares fitting algorithm.
• Demonstrated successful application in modeling lower extremity artery-vein complexes (MRI) and coronary arteries (CTA).

Conclusions:

• The proposed method offers a robust and efficient approach for accurate blood vessel lumen modeling.
• This technique enhances diagnostic capabilities for vascular diseases and supports applications like blood-flow simulation.
• The method can also automate image dataset annotation for training machine learning algorithms in medical imaging.