Method, Device and Computer Program for Mapping a Three-Dimensional Branched Tubular Structure into at Least One Two-Dimensional Image Plane

Simple SummaryContent extracted from patent full text and abstract with AI.

This invention describes a computer-implemented method, device, and computer program that map complex three-dimensional (3D) branched tubular structures—such as blood vessels or bronchial trees—in medical imaging data into one or more two-dimensional (2D) image planes. The process intelligently selects and transforms subsets of 3D image data (voxels) based on segmentation and optimizes this transformation to produce a flattened, easy-to-analyze 2D image. The approach handles highly branched and complex anatomical networks, such as brain vasculature, where previous flattening techniques often fail. The resulting 2D images can be used for visualization, analysis, or as input for machine learning applications.

Use CasesContent extracted from patent full text and abstract with AI.

• Medical diagnostics: Visualizing complex vascular networks, such as brain or liver vasculature, in a more interpretable 2D format for easier and faster diagnosis.
• Surgical planning: Allowing surgeons to review and plan interventions by seeing detailed 2D representations of 3D anatomic pathways.
• Teaching and communication: Providing textbook-like, standardized 2D images of complex patient anatomy for medical education and interdisciplinary discussions.
• Automated disease detection: Supplying flattened 2D images as inputs to machine learning models (neural networks, etc.) for tasks such as detecting vessel occlusions, stenosis, or abnormal branching patterns.
• Data reduction for AI: Reducing the size and complexity of input data for computer vision models, enabling more efficient machine learning workflows with less computational resource demand.
• Comparative studies: Facilitating side-by-side comparison of anatomical variations or disease progression using consistent 2D maps.

BenefitsContent extracted from patent full text and abstract with AI.

• Enables effective visualization of highly complex 3D branched structures that traditional methods struggle to flatten.
• Reduces the manual effort and time needed to extract and interpret relevant structures from volumetric scans.
• Provides images that closely resemble standard anatomical representations, aiding comprehension and clinical communication.
• Optimizes for smooth, symmetric, and high-contrast output images, improving readability and diagnostic accuracy.
• Can be used to generate training data or inputs for machine learning, streamlining development of automated analysis tools.
• Generic enough to be modularly applied to various types of branched tubular structures (vessels, airways, nerve tracts, etc.).
• May decrease the amount of data and computational resources needed for downstream image analysis through data reduction.

Technical Classifications (CPCs)

Inventors & Applicants

Patent Abstract

Computer implemented method for mapping a three-dimensional branched tubular structure (73) depicted in a three-dimensional image data set (33) into at least one twodimensional image plane (50), comprising the steps:- Selecting a first group (39) of voxels (40) based on segmentation data (34),- Determining a respective transformation vector (41) for each voxel (40) of the first group (39) using a given model (37) that specifies the respective position of at least part of the tubular structure (73) in the or a respective image plane (50),- Selecting a second group (42) of voxels (43) in the image data set (33), especially based on the image data set (33) and/or the segmentation data (34), and determining a respective transformation vector (48) for each voxel (43) of the second group (42), wherein the selection of the second group (42) of voxels (43) and/or the determination of the transformation vectors (48) for the second group (42) of voxels (43) is performed by optimizing a cost function (45, 47) that depends on the transformation vectors (41) for the first group (39) of voxels (40), and- Generating an output image (49) for the at least one image plane (50).