AI BMD Model Training via Chest X-Ray Positioning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The limited availability and high cost of dual-energy X-ray absorptiometry (DXA) equipment restrict the amount of training data for AI algorithms, making them less accurate for bone mineral density (BMD) measurements, which is a concern given the increasing prevalence of osteoporosis and its silent progression.
Innovation Solution
A BMD model training method using chest X-ray images and lumbar vertebra BMD values measured by DXA to train AI models, focusing on specific bone positions like the 1st lumbar vertebra and 12th thoracic vertebra, which bear the body's weight and are more susceptible to osteoporosis, thereby increasing the volume of training data and improving accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If dual-energy X-ray absorptiometry (DXA) is used for BMD measurement, then measurement precision is improved, but device complexity and cost increase, limiting widespread availability
Solution Approach 1:
The patent creates a virtual copy of the DXA measurement capability through AI. The model is trained on paired data consisting of chest X-ray images and corresponding DXA BMD measurements, enabling the system to replicate DXA's measurement precision using ordinary chest X-ray equipment. This copying approach allows widespread deployment without requiring actual DXA devices.
Solution Approach 2:
The patent replaces the mechanical DXA measurement system with an AI-based image analysis system. Instead of using dual-energy X-ray absorptiometry hardware to directly measure BMD, the system uses a trained neural network that processes chest X-ray images to predict BMD values, substituting complex mechanical measurement with computational analysis.
2Measurement precision
If dual-energy X-ray absorptiometry (DXA) is used for BMD measurement, then measurement precision is improved, but the amount of training data is reduced due to limited equipment availability
Solution Approach 1:
The patent uses DXA BMD measurements as reference labels to train an AI model that operates on chest X-ray images. By copying the measurement capability through supervised learning rather than requiring actual DXA devices for each training sample, the system can utilize large volumes of chest X-ray data paired with DXA-derived BMD values for model training.
Solution Approach 2:
The patent introduces an intermediary approach where DXA measurements serve as ground truth labels for training, but the actual prediction is performed using chest X-ray image analysis. This intermediary training strategy allows the system to leverage the precision of DXA for model development while using more accessible chest X-ray data for actual training.
3Quantity of substance
If chest X-ray images are used instead of DXA for BMD measurement, then equipment availability and training data quantity increase, but measurement precision deteriorates
Solution Approach 1:
The patent creates a computational model that copies DXA's measurement precision through training on paired chest X-ray and DXA BMD data. The AI model learns to map chest X-ray image features to accurate BMD values, effectively copying the precision of DXA while using the more accessible chest X-ray imaging modality.
Solution Approach 2:
The patent transforms the relationship between chest X-ray images and BMD measurement through learned parameters in the AI model. By training the neural network on large datasets with DXA-verified BMD values, the model adjusts its internal parameters to achieve precision comparable to DXA, despite using different input imaging data.
Data Source
AI summary
A method of training a bone mineral density (BMD) model, including a training data retrieval step, a positioning step, and a model training step. The training data retrieval step retrieves one or more chest X-ray images with the BMD of a person. The positioning step locates and retrieves specific bone positions in the chest X-ray image. The model training step trains the AI model based on the bone positions in the chest X-ray image and the BMD. The specific bone position includes the first segment of the lumbar vertebrae. The BMD is the lumbar BMD value measured by dual-energy X-ray absorptiometry (DXA).