Multi-Channel ResNet for Bone Fracture Risk Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for analyzing ultrasonic radio-frequency signals from quantitative ultrasound (QUS) devices are inadequate for comprehensively and accurately extracting characteristics related to osteoporotic fracture risk, leading to incomplete bone quality assessment.
Innovation Solution
A model training method using a multi-channel residual neural network (ResNet) is proposed, which includes determining an initial multi-channel ResNet model, obtaining a training dataset with multi-channel bone radio-frequency data and corresponding fracture evaluation labels, and training the model to extract features and predict fracture risk.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional parameter-based analysis is used for ultrasonic radio-frequency signals, then the analysis process is simple, but a large amount of bone quality information is lost and measurement precision is insufficient
Solution Approach 1:
The patent divides the ultrasonic radio-frequency signal analysis into multiple independent channels (e.g., different frequency bands or time windows), allowing each channel to extract specific features independently. This segmentation enables comprehensive capture of bone quality information across different signal dimensions while maintaining manageable complexity through modular processing.
Solution Approach 2:
The patent transforms the traditional single-parameter analysis into multi-dimensional feature space by applying deep learning models that process signals across multiple channels and layers. This dimensional expansion captures complex patterns in the radio-frequency signals that single-parameter methods miss, significantly improving bone quality assessment accuracy.
2Measurement precision
If comprehensive feature extraction from ultrasonic radio-frequency signals is implemented, then fracture risk prediction accuracy is improved, but the difficulty of detecting and measuring key variables increases
Solution Approach 1:
The patent employs self-supervised learning or pre-training approaches where the model automatically learns relevant features from the ultrasonic radio-frequency signals without requiring manual annotation of key variables. The model identifies and extracts meaningful patterns independently, reducing the need for expert knowledge in variable selection while improving prediction accuracy.
Solution Approach 2:
The patent replaces manual feature engineering and expert-based variable selection with automated deep learning models. The neural network automatically discovers and extracts relevant features from raw signals, substituting the mechanical process of manual analysis with an intelligent system that handles the complexity of key variable identification.
3Reliability
If multi-channel residual neural network model is trained with comprehensive bone radio-frequency data, then the model's fracture risk prediction capability is enhanced, but training time and computational resources increase
Solution Approach 1:
The patent implements data preprocessing steps before model training, including signal normalization, noise filtering, and feature standardization. By preparing the data in advance, the model training process becomes more efficient and converges faster, reducing overall training time while maintaining prediction reliability.
Solution Approach 2:
The patent employs continuous training strategies where the model is trained incrementally on batches of data rather than requiring complete retraining. This allows the model to learn from comprehensive datasets efficiently, maintaining high prediction reliability while reducing total training time through continuous, incremental learning.
Applied Scientific Principles
This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.
Function Achieved in This Case
The method enables the comprehensive and accurate extraction of ultrasonic radio-frequency signal characteristics, resulting in a robust multi-channel ResNet model that effectively predicts fracture risk, thereby reducing the incidence of osteoporotic fractures and alleviating patient pain.
Implementation Method 1
Quantitative ultrasound (QUS) is a bone mineral density measurement technique. Its working principle is to detect bone quality by using the different propagation speeds and attenuations of ultrasound in different bone components.
Data Source
AI summary
A model training method comprises: determining an initial multi-channel residual neural network model; acquiring a multi-channel residual neural network model training data set, the multi-channel residual neural network model training data set comprises multi-channel bone radio-frequency data obtained by a QUS device, and a fracture evaluation value label corresponding to the multi-channel bone radio-frequency data; and training the initial multi-channel residual neural network model by taking the multi-channel bone radio-frequency data as an input and taking the fracture evaluation value label as an output, so as to obtain a multi-channel residual neural network model. Features of an ultrasonic radio-frequency signal can be extracted, and a multi-channel residual neural network model is obtained by training, which model can be applied to the field of fracture risk prediction, such that an effective prevention measure is taken in a timely manner.