Sparse Sampling Model for Volumetric Medical Image Segmentation

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Current methods for automated segmentation of volumetric medical images are computationally intensive and time-consuming, especially when dealing with large 3D images containing billions of voxels, and often require extensive training datasets and specialized hardware.

Innovation Solution

A computer-implemented method using sparse sampling to reduce the data processed, where a sparse sampling model defines sampling points in the volumetric medical image, allowing for efficient classification of labels and generation of segmentation masks, enabling fast and accurate organ identification with adjustable granularity.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If full segmentation of all organs is performed using conventional methods, then segmentation accuracy and completeness are improved, but computation time and computational resources increase significantly

Engineering Contradiction:
Improvesegmentation accuracyVSAvoidcomputation time
Core Design Contradiction:
Measurement precisionVSLoss of time

Solution Approach 1:

The patent extracts only the necessary information from the volumetric medical image by sampling voxels at specific locations rather than processing all voxels. The sampling model selects representative voxels that capture essential anatomical features, enabling accurate organ identification without processing the entire image volume, thus reducing computation time while maintaining segmentation accuracy.

Inventive Principle:
Principle #2Taking out (Extraction)

Solution Approach 2:

The patent segments the volumetric medical image into discrete sampling points distributed according to a sampling model. By dividing the continuous 3D space into discrete representative points, the system can process only the essential information needed for organ identification, reducing the computational burden from processing all billions of voxels to processing a manageable number of sampled points.

Inventive Principle:
Principle #1Segmentation

2Reliability

If conventional segmentation methods are used, then comprehensive organ identification is achieved, but the method requires large training datasets and specialized hardware

Engineering Contradiction:
Improveorgan identification reliabilityVSAvoidhardware requirements
Core Design Contradiction:
ReliabilityVSDevice complexity

Solution Approach 1:

The patent extracts only the essential information needed for reliable organ identification by sampling voxels at strategically selected locations. This extraction approach allows the system to achieve reliable organ identification using a reduced set of representative samples, thereby reducing the size of required training datasets and the computational hardware needed, while maintaining high identification reliability.

Inventive Principle:
Principle #2Taking out (Extraction)

3Manufacturing precision

If dense sampling of all voxels is performed, then segmentation detail and granularity are improved, but processing speed decreases

Engineering Contradiction:
Improvesegmentation detailVSAvoidprocessing speed
Core Design Contradiction:
Manufacturing precisionVSProductivity

Solution Approach 1:

The patent implements dynamic sampling where the sampling density can be adjusted based on the specific requirements of the analysis. The sampling model can be configured to use coarse sampling for fast processing or finer sampling for more detailed analysis, allowing the system to adaptively balance between processing speed and segmentation detail according to the application needs.

Inventive Principle:
Principle #15Dynamics

Solution Approach 2:

The patent applies local quality by sampling voxels with different densities in different regions of the image. The sampling model can concentrate sampling points in regions where detailed information is crucial while using sparser sampling in homogeneous regions, thereby maintaining segmentation detail where needed while improving overall processing speed through reduced sampling in less critical areas.

Inventive Principle:
Principle #3Local quality

Data Source

PatentEP4425427A1Method for automated processing of volumetric medical images
Publication Date: 2024.09.04 SIEMENS HEALTHINEERS AG
  • EP4425427A1 patent drawingFigure 1~2
  • EP4425427A1 patent drawingFigure 3~4
  • EP4425427A1 patent drawingFigure 5~6

AI summary

A method, device and system for automated processing of volumetric medical images (MI). In one aspect, the method comprises: receiving (301) a volumetric medical image (MI), the volumetric medical image (MI) comprising at least one organ or portion thereof, providing (302) a sparse sampling model (SM) for sparse sampling the volumetric medical image (MI), the sparse sampling model (SM) defining a number N of sampling points (400a-400c) distributed in the volumetric medical image (MI) and defining locations and distances of the distributed sampling points (400a-400c), sampling (303) voxels (306-308) from the volumetric medical image (MI) using the provided sparse sampling model (SM) for obtaining N sparse sampling descriptors (D), classifying (304) labels (L) for query points (701) in the volumetric medical image (MI) by applying a trained classifier to the obtained sparse sampling descriptors (D), and providing (305) a segmentation mask (MAP, 601, 602) for the volumetric medical image (MI) using the classified labels (L).