3D Convolutional Neural Networks for Medical Image Segmentation

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Current medical image segmentation technologies, particularly in clinical settings, face challenges in accurately segmenting multiple overlapping organs and are time-consuming, often requiring human expertise, which can lead to inconsistencies and increased healthcare costs.

Innovation Solution

A segmentation neural network system utilizing a sequence of encoder and decoder blocks with residual connections, including 2D and 3D convolutional layers, and a sigmoid activation function in the final layer to generate precise segmentation outputs for multiple organs, capable of processing medical images from various modalities like CT, MRI, and PET scans.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Productivity

If automated segmentation systems are used, then productivity is improved, but manufacturing precision deteriorates

Engineering Contradiction:
Improvesegmentation speedVSAvoidsegmentation accuracy
Core Design Contradiction:
ProductivityVSManufacturing precision

Solution Approach 1:

The neural network is divided into multiple encoder blocks with sequential processing stages. Each encoder block independently processes specific spatial resolutions and feature types, with the first encoder block handling high-resolution spatial features and subsequent blocks handling lower-resolution contextual features. This segmented architecture enables parallel processing of different feature types while maintaining overall segmentation accuracy.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The system processes medical images in multiple spatial dimensions simultaneously through the encoder blocks. The first encoder block operates at high spatial resolution to capture detailed anatomical boundaries, while subsequent encoder blocks operate at progressively lower resolutions to capture broader contextual relationships. This multi-dimensional processing approach maintains precision while improving computational efficiency and productivity.

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

2Manufacturing precision

If multiple encoder blocks are used to improve precision, then device complexity increases

Engineering Contradiction:
Improvesegmentation accuracyVSAvoidnetwork architecture complexity
Core Design Contradiction:
Manufacturing precisionVSDevice complexity

Solution Approach 1:

Each encoder block in the sequence is designed with a universal architecture that can handle multiple types of features simultaneously. The encoder blocks use identical structural components (convolutional layers, batch normalization, ReLU activations) that serve multiple functions: feature extraction, spatial relationship detection, and contextual understanding. This universal design reduces overall system complexity while maintaining high segmentation precision through the sequential arrangement of these multi-functional blocks.

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

3Loss of time

If automated segmentation is implemented, then loss of time is reduced, but reliability may worsen due to lack of human expertise

Engineering Contradiction:
Improvemanual segmentation timeVSAvoidconsistency of results
Core Design Contradiction:
Loss of timeVSReliability

Solution Approach 1:

The neural network performs self-correction and self-verification through its multi-block encoder architecture. Each encoder block independently processes the input data and generates feature representations that are progressively refined through the sequence. The first encoder block establishes accurate spatial boundaries, while subsequent blocks automatically adjust and refine these boundaries based on contextual information from lower resolutions. This self-service mechanism ensures consistent, reliable results without requiring human intervention, thereby reducing time loss while maintaining high reliability.

Inventive Principle:
Principle #25Self-service

Data Source

PatentEP3830791B13-d convolutional neural networks for organ segmentation in medical images for radiotherapy planning
Publication Date: 2023.11.01 GOOGLE LLC
  • EP3830791B1 patent drawingFigure 1
  • EP3830791B1 patent drawingFigure 2A
  • EP3830791B1 patent drawingFigure 2B

AI summary

Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for segmenting a medical image. In one aspect, a method comprises: receiving a medical image that is captured using a medical imaging modality and that depicts a region of tissue in a body; and processing the medical image using a segmentation neural network to generate a segmentation output, wherein the segmentation neural network comprises a sequence of multiple encoder blocks, wherein: each encoder block is a residual neural network block comprising one or more two-dimensional convolutional neural network layers, one or more three-dimensional convolutional neural network layers, or both, and each encoder block is configured to process a respective encoder block input to generate a respective encoder block output wherein a spatial resolution of the encoder block output is lower than a spatial resolution of the encoder block input.