Semi-Automatic Tumor Segmentation with Interactive Feedback

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Current tools for quantitative analysis in medical imaging lack the speed, precision, and consistency required for wider clinical use, making manual delineation of tumor boundaries time-consuming and prone to variable results.

Innovation Solution

A system and method for volumetric segmentation in medical images that combines automated segmentation with a real-time user interface, allowing radiologists to interactively confirm candidate structures and refine segmentation results, thereby enhancing efficiency and accuracy.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If manual delineation is used to define tumor boundaries, then complete control and precision are achieved, but the process becomes time-consuming and produces variable results

Engineering Contradiction:
Improvetumor boundary delineation precisionVSAvoidsegmentation speed
Core Design Contradiction:
Measurement precisionVSProductivity

Solution Approach 1:

The system performs preliminary automated segmentation to generate initial tumor boundaries before user review. This preliminary action provides a starting point that reduces the time needed for manual delineation while maintaining precision, as users only need to adjust the automated results rather than draw boundaries from scratch.

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The system implements feedback by displaying the automated segmentation results to the user for review and adjustment. Users can interact with the system to correct boundaries, and the system incorporates these corrections to produce final segmentation. This feedback loop ensures precision while reducing time compared to purely manual methods.

Inventive Principle:
Principle #23Feedback

2Productivity

If fully automated segmentation methods are used, then speed is improved, but precision and consistency are insufficient for clinical use

Engineering Contradiction:
Improvesegmentation speedVSAvoidtumor boundary delineation precision
Core Design Contradiction:
ProductivityVSMeasurement precision

Solution Approach 1:

The system performs preliminary automated segmentation to generate initial tumor boundaries before user review. This preliminary action provides a starting point that reduces the time needed for manual delineation while maintaining precision, as users only need to adjust the automated results rather than draw boundaries from scratch.

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The system implements feedback by displaying the automated segmentation results to the user for review and adjustment. Users can interact with the system to correct boundaries, and the system incorporates these corrections to produce final segmentation. This feedback loop ensures precision while reducing time compared to purely manual methods.

Inventive Principle:
Principle #23Feedback

3Ease of operation

If manual delineation is performed by different users, then complete control is achieved, but variable results are produced due to variable performance

Engineering Contradiction:
Improveuser control over segmentationVSAvoidconsistency of segmentation results
Core Design Contradiction:
Ease of operationVSReliability

Solution Approach 1:

The system performs preliminary automated segmentation to generate initial tumor boundaries before user review. This preliminary action provides a starting point that reduces the time needed for manual delineation while maintaining precision, as users only need to adjust the automated results rather than draw boundaries from scratch.

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The system implements feedback by displaying the automated segmentation results to the user for review and adjustment. Users can interact with the system to correct boundaries, and the system incorporates these corrections to produce final segmentation. This feedback loop ensures precision while reducing time compared to purely manual methods.

Inventive Principle:
Principle #23Feedback

Data Source

PatentEP3877949B1Systems and methods for semi-automatic tumor segmentation
Publication Date: 2025.04.09 GENENTECH INC
  • EP3877949B1 patent drawingFigure 1
  • EP3877949B1 patent drawingFigure 2
  • EP3877949B1 patent drawingFigure 3

AI summary

A method for volumetric segmentation in a plurality of planar medical images includes, receiving, at an electronic processor, the plurality of planar medical images. A boundary of a candidate structure in the plurality of medical images is generated using a segmentation model. A first planar medical image from the plurality of planar medical images is displayed on a display. A user input is received using a user interface indicating a region in the first planar medical image. A first planar contour of the candidate structure is generated. The region is compared to the boundary. Responsive to the region being at least partially within the boundary, the first planar medical image is re-displayed on the display showing the first planar contour of the structure, and a finding record for the candidate structure including the boundary is generated.