Multi-Phase 3D Liver Lesion Segmentation From CT or MRI

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing methods for 3D lesion segmentation in CT or MRI imaging primarily rely on single-phase or phase-specific neural networks, which are confined to 2D applications, or combine separate phase-specific models post-processing, lacking an efficient method for enhanced 3D segmentation directly from multi-phase imaging datasets.

Innovation Solution

A computer-implemented method using a trained function that encodes and aggregates feature information from multiple phases of 3D CT imaging datasets through an encoder-decoder architecture with a spatial aggregation module, generating 3D segmentation data by aligning and combining 2D slices across phases.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If phase-specific neural networks are used for 3D lesion segmentation, then segmentation can be performed on specific imaging phases, but the method is confined to 2D applications and lacks efficient 3D segmentation capability

Engineering Contradiction:
Improvelesion segmentation accuracyVSAvoid3D segmentation capability
Core Design Contradiction:
Measurement precisionVSAdaptability or versatility

Solution Approach 1:

The patent transitions from 2D phase-specific segmentation to 3D multi-phase segmentation by processing volumetric data across multiple imaging phases simultaneously. The system analyzes lesions in three-dimensional space while incorporating temporal information from different contrast phases, enabling comprehensive 3D segmentation that leverages both spatial and temporal dimensions of the imaging data.

Inventive Principle:
Principle #17Another dimension (Dimensionality change)

Solution Approach 2:

The patent creates a unified segmentation framework that can handle multiple imaging phases and 3D volumetric data within a single system. This multi-functional approach allows the same methodology to process different phases (arterial, portal venous, delayed) and generate consistent 3D segmentation results across various lesion types and imaging scenarios.

Inventive Principle:
Principle #6Universality (Multi-functionality)

2Loss of information

If separate phase-specific models are combined in post-processing, then multi-phase information can be utilized, but the combination process is complex and less efficient

Engineering Contradiction:
Improvemulti-phase information utilizationVSAvoidmodel combination process
Core Design Contradiction:
Loss of informationVSDevice complexity

Solution Approach 1:

The patent merges multiple phase-specific processing streams into a unified 3D segmentation framework. Instead of separately processing each phase and then combining results, the system integrates arterial, portal venous, and delayed phase information within a single coherent processing pipeline that operates on 3D volumetric data, simplifying the overall process while preserving multi-phase information.

Inventive Principle:
Principle #5Merging (Combining)

Solution Approach 2:

The patent performs preliminary alignment and registration of multi-phase 3D volumetric data before segmentation. By pre-processing the data to establish consistent spatial relationships across phases, the system eliminates the need for complex post-processing combination steps, as the segmentation is performed on already-integrated multi-phase information.

Inventive Principle:
Principle #10Preliminary action

Data Source

PatentEP4625316A1A system and a method for three-dimensional lesion segmentation using multi-phase computed tomography or magnetic resonance imaging data
Publication Date: 2025.10.01 SIEMENS HEALTHINEERS AG
  • EP4625316A1 patent drawingFigure 1~2
  • EP4625316A1 patent drawingFigure 3~4
  • EP4625316A1 patent drawingFigure 5

AI summary

Computer-implemented methods and computer systems are provided for locating a liver lesion in the input data. In an example, a computer implemented method of the current disclosure comprises receiving, by a first computer interface, input data; applying a trained function to the input data; generating, by the trained function, output data, wherein the output data is associated with the input data; providing, via a second computer interface, the output data, wherein the input data comprises a multi-phase three-dimensional, 3D, computed tomography, CT, imaging dataset of a liver lesion associated with a patient, and wherein the output data comprises a three-dimensional, 3D, segmentation data for locating the liver lesion in the input data.