Generative Network for Lung Fissure Localization in CT Imaging

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Current automated lung lobe segmentation methods face challenges in accurately segmenting lungs into lobar regions due to variations in fissure completeness and vessel crossings, particularly due to the scarcity of annotated training data and the complexity of anatomical variations in CT images.

Innovation Solution

The use of deep learning techniques, specifically a machine-learnt generative network for fissure localization, which generates labeled imaging data to aid in lobar segmentation, and an image-to-image network for fissure localization, reduces the need for extensive manual annotation and improves segmentation accuracy by inferring incomplete fissures.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If manual annotation of fissures is used for training data, then training data accuracy is improved, but annotation time and labor cost increase significantly

Engineering Contradiction:

Improvefissure annotation accuracyVSAvoidannotation time

Core Design Contradiction:

Measurement precisionVSLoss of time

Solution Approach 1:

The system performs preliminary fissure detection and localization using automated algorithms before final annotation. The generative network pre-identifies fissure locations and characteristics, creating a preliminary annotation that experts then verify and refine, rather than annotating from scratch

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

A generative adversarial network acts as an intermediary between raw imaging data and final annotations. The network generates synthetic fissure annotations that serve as training data, mediating the gap between limited manual annotations and the need for extensive training data

Inventive Principle:

Principle #24Intermediary (Mediator)

2Measurement precision

If deep learning with generative networks is used, then segmentation accuracy is improved, but computational complexity increases

Engineering Contradiction:

Improvelobar segmentation accuracyVSAvoidcomputational system complexity

Core Design Contradiction:

Measurement precisionVSDevice complexity

Solution Approach 1:

The computational task is segmented into distinct stages: a generative network first localizes fissures and generates preliminary annotations, then a separate segmentation network uses these annotations to perform lobar segmentation. This divides the complex task into manageable components that can be trained and optimized independently

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The generative network performs preliminary processing to generate fissure annotations and localization information before the main segmentation network processes the data. This preliminary action simplifies the input to the segmentation network, reducing its computational burden while improving overall accuracy

Inventive Principle:

Principle #10Preliminary action

Data Source

PatentUS10607114B2Trained generative network for lung segmentation in medical imaging

Publication Date: 2020.03.31 SIEMENS HEALTHINEERS AG

AI summary

A generative network is used for lung lobe segmentation or lung fissure localization, or for training a machine network for lobar segmentation or localization. For segmentation, deep learning is used to better deal with a sparse sampling of training data. To increase the amount of training data available, an image-to-image or generative network localizes fissures in at least some of the samples. The deep-learnt network, fissure localization, or other segmentation may benefit from generative localization of fissures.