Deep Learning BMD Estimation from Plain Radiographs

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Current methods for estimating bone mineral density (BMD) using plain radiographs are inadequate due to their qualitative nature and limited accuracy, making it difficult to identify individuals at risk for osteoporosis, especially in areas where dual-energy X-ray absorptiometry (DXA) is underutilized.

Innovation Solution

A computer-implemented method and device that uses a deep neural network to estimate BMD by detecting anatomical landmarks in plain radiographs, extracting regions of interest, and performing BMD estimation, thereby providing a more accurate and quantitative assessment.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Extent of automation

If rule-based methods (H index, fractal analysis) are used to assess bone texture, then the analysis can be performed automatically, but the measurement precision and accuracy of BMD estimation remain insufficient

Engineering Contradiction:

Improveautomatic bone texture analysisVSAvoidBMD estimation accuracy

Core Design Contradiction:

Extent of automationVSMeasurement precision

Solution Approach 1:

The patent replaces rule-based mechanical analysis methods with deep learning neural networks that automatically learn complex patterns from bone texture images. The CNN model substitutes manual feature engineering and rule-based algorithms with data-driven learning, enabling both automation and high measurement precision simultaneously.

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

Solution Approach 2:

The patent transforms the approach by changing from traditional rule-based parameters (H index, fractal dimension) to deep learning-based feature representations. The model learns optimal feature parameters from training data, allowing flexible adaptation to different bone structures and improving measurement precision while maintaining automation.

Inventive Principle:

Principle #35Parameter changes

2Ease of operation

If qualitative visual assessment is used by physicians, then the method is simple and requires minimal equipment, but the accuracy is inadequate to support clinical decisions

Engineering Contradiction:

Improvesimplicity of assessmentVSAvoidBMD measurement accuracy

Core Design Contradiction:

Ease of operationVSMeasurement precision

Solution Approach 1:

The system enables self-service automated assessment where the deep learning model independently performs BMD estimation without requiring physician intervention. The model processes bone texture images autonomously, providing quantitative results that maintain simplicity while dramatically improving accuracy for clinical decision-making.

Inventive Principle:

Principle #25Self-service

Solution Approach 2:

The patent substitutes qualitative visual assessment with automated deep learning-based quantitative analysis. The neural network replaces human visual interpretation with computational models that provide objective, reproducible measurements, improving accuracy while maintaining ease of operation through automated processing.

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

3Measurement precision

If DXA is used for BMD screening, then measurement precision is high, but availability is limited and cannot be performed frequently

Engineering Contradiction:

ImproveBMD measurement accuracyVSAvoidavailability and frequency of screening

Core Design Contradiction:

Measurement precisionVSAdaptability or versatility

Solution Approach 1:

The patent creates a copying approach by using plain radiographs as substitutes for DXA scans. Instead of requiring expensive DXA equipment, the system uses readily available radiographs and applies deep learning to extract BMD information, providing high measurement precision through a more versatile and frequently performable imaging modality.

Inventive Principle:

Principle #26Copying

Solution Approach 2:

The patent employs a cost-effective approach by using standard radiographs that are already part of routine patient care. These readily available, low-cost images serve as the basis for BMD estimation, eliminating the need for expensive DXA scans while maintaining measurement precision through advanced computational analysis.

Inventive Principle:

Principle #27Cheap short-living objects (Disposable)

4Adaptability or versatility

If plain radiograph screening is implemented, then availability and frequency of screening improve, but measurement precision of BMD estimation deteriorates

Engineering Contradiction:

Improveavailability of screeningVSAvoidBMD estimation accuracy

Core Design Contradiction:

Adaptability or versatilityVSMeasurement precision

Solution Approach 1:

The patent applies deep learning technology to replace traditional rule-based analysis of plain radiographs. The neural network processes bone texture images to extract quantitative BMD measurements, transforming the limited precision of radiograph-based methods into high-accuracy measurements comparable to DXA, while maintaining the availability and frequency benefits of using plain radiographs.

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

Data Source

PatentUS11704796B2Estimating bone mineral density from plain radiograph by assessing bone texture with deep learning

Publication Date: 2023.07.18 PING AN TECH (SHENZHEN) CO LTD

AI summary

The present disclosure provides a computer-implemented method, a device, and a computer program product for radiographic bone mineral density (BMD) estimation. The method includes receiving a plain radiograph, detecting landmarks for a bone structure included in the plain radiograph, extracting an ROI from the plain radiograph based on the detected landmarks, estimating the BMD for the ROI extracted from the plain radiograph by using a deep neural network.