Multi-ROI Deep Learning Model for Opportunistic Osteoporosis Screening
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Osteoporosis is often under-diagnosed and under-treated due to low awareness and limited accessibility of Dual-energy X-ray Absorptiometry (DXA) examinations, which are the current standard for measuring bone mineral density (BMD), and existing opportunistic screening methods are not effective in utilizing readily available medical images for osteoporosis screening.
Innovation Solution
A method and system that utilize a plain film chest X-ray (CXR) to extract regions of interest (ROIs) and apply a multi-ROI model, involving a backbone network to generate individual and joint feature vectors, which are then decoded to provide BMD scores, allowing for opportunistic screening of osteoporosis without additional cost or time.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If Dual-energy X-ray Absorptiometry (DXA) examination is used to measure BMD, then measurement precision is improved, but device complexity and accessibility are worsened
Solution Approach 1:
The patent uses plain film CXR images (a simple, widely available copy) to substitute for DXA examinations (the complex gold standard). The deep learning model creates a computational copy of the BMD measurement process by training on paired CXR-DXA datasets, enabling the network to predict DXA-equivalent BMD values from routine chest X-rays without requiring actual DXA hardware
Solution Approach 2:
The patent replaces the mechanical DXA measurement system with an information-processing system based on deep learning. Instead of using physical DXA scanners to measure bone density, the system uses a neural network that processes CXR images and predicts BMD values, substituting computational mechanics for physical measurement mechanics
2Ease of operation
If opportunistic screening using plain film CXR is implemented, then accessibility and cost are improved, but measurement precision is worsened
Solution Approach 1:
The patent transforms the CXR image parameters by processing them through multiple convolutional layers that extract hierarchical features. The model changes the parameter representation from raw pixel values to meaningful bone structure features, and finally to predicted BMD values, effectively bridging the precision gap between simple CXR and complex DXA
Solution Approach 2:
The patent combines multiple CXR images (frontal and lateral views) and processes them through a composite deep learning architecture that integrates feature extraction, fusion, and prediction modules. This composite approach synthesizes information from multiple sources to achieve DXA-level precision
3Measurement precision
If multiple ROIs are extracted and processed individually, then measurement precision is improved, but device complexity is worsened
Solution Approach 1:
The patent segments the CXR image into multiple anatomical ROIs (vertebrae, ribs, clavicles, scapulae) and processes each region independently through separate feature extraction pathways. This segmentation allows the model to capture bone density information from multiple skeletal sites, improving overall BMD prediction precision while maintaining manageable complexity through modular processing
Data Source
AI summary
A method of opportunistic screening of osteoporosis includes obtaining a plain film chest X-ray (CXR); extracting regions of interest (ROIs) from the plain film CXR; and providing individual bone mineral density (BMD) scores corresponding to the extracted ROIs and a joint BMD corresponding to the plain film CXR based on a multi-ROI model by performing: inputting the extracted ROIs into a backbone network to generate individual feature vectors, each individual feature vector corresponding to one of the extracted ROIs; concatenating the individual feature vectors into a joint feature vector; individually decoding the individual feature vectors by a shared fully connected (FC) layer to generate the individual BMDs, each individual BMD corresponding to one of the individual feature vectors; and decoding the joint feature vector by a separate FC layer to generate the joint BMD.