Chest X-ray Abnormality Detection via Linear Structure Segmentation

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Current methods for detecting abnormalities in chest X-ray images, particularly when anatomical structures overlap, struggle to accurately recognize lesions due to overlapping anatomical structures, limiting the ability to determine various abnormal states independently of specific anatomical structures.

Innovation Solution

A method using machine learning to detect linear structures in chest X-ray images, calculating indicators for abnormal states, and comparing these indicators with reference values to determine if the structures are in an abnormal state, allowing for the display of affected areas and details on a unified framework.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Extent of automation

If machine learning is used to detect lesions in chest X-ray images, then automation and productivity are improved, but measurement precision deteriorates when anatomical structures overlap with lesions

Engineering Contradiction:

Improveautomation of lesion detectionVSAvoidlesion recognition accuracy

Core Design Contradiction:

Extent of automationVSMeasurement precision

Solution Approach 1:

The patent segments the chest X-ray image into multiple anatomical structure components (ribs, spine, clavicles, heart shadow, lung fields) and processes each segment separately to identify lesions, avoiding the interference of overlapping structures

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The patent introduces an intermediary processing step that generates synthetic training data with artificially added lesions, which serves as a bridge to train the machine learning model to recognize lesions even when they overlap with anatomical structures

Inventive Principle:

Principle #24Intermediary (Mediator)

2Measurement precision

If techniques focus on specific anatomical structures like rib spacing to detect abnormalities, then measurement precision for those structures is improved, but adaptability to detect various abnormal states deteriorates

Engineering Contradiction:

Improveabnormality detection accuracyVSAvoidcapability to determine various abnormal states

Core Design Contradiction:

Measurement precisionVSAdaptability or versatility

Solution Approach 1:

The patent creates a universal machine learning model that can detect multiple types of abnormalities (pneumonia, tuberculosis, lung cancer, rib fractures, spinal abnormalities) across different anatomical structures simultaneously, rather than creating separate techniques for each structure or disease type

Inventive Principle:

Principle #6Universality (Multi-functionality)

Data Source

PatentUS11475568B2Method for controlling display of abnormality in chest x-ray image, storage medium, abnormality display control apparatus, and server apparatus

Publication Date: 2022.10.18 PANASONIC HOLDINGS CORP

AI summary

A method for controlling display of an abnormality includes obtaining a target chest X-ray image, detecting a structure including a linear structure formed of a first linear area that has been drawn by projecting anatomical structures whose X-ray transmittances are different from each other or a second linear area drawn by projecting an anatomical structure including a wall of a trachea, a wall of a bronchus, or a hair line, calculating an indicator for determining the abnormal state from the structure, comparing the indicator with a reference indicator, and determining whether the structure is in the abnormal state, and displaying, if it is determined that the structure is in the abnormal state, an image of an area of the target chest X-ray image including the structure determined to be in the abnormal state and details of the abnormal state.