Patient Body Representation via ML Landmark Alignment
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current medical imaging technologies face challenges in accurately representing the human body for targeted imaging, particularly when radiation is involved, as existing methods often rely on crude body representations that are not specific to the patient's position or anatomy, leading to inefficient radiation exposure and imaging quality.
Innovation Solution
A computer-implemented method is developed to train a machine learning system using medical imaging data sets from various modalities (CT, MRI, X-ray) by estimating landmarks, aligning data, sampling points, and adjusting body representations to create an accurate, n-dimensional vector model of the patient's body, allowing for precise radiation targeting and reduced exposure.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If crude body representations are used for targeted imaging, then device complexity is reduced, but measurement precision of body shape and size deteriorates
Solution Approach 1:
The patent segments the body representation problem into multiple components: landmark detection, pose estimation, and body shape/size parameters. By dividing the complex task into manageable segments, the system achieves high measurement precision without requiring an overly complex monolithic system.
Solution Approach 2:
The system performs preliminary actions by detecting landmarks and estimating pose before determining body shape and size. This preliminary processing prepares the data in advance, enabling accurate measurements without needing complex real-time computation during the actual imaging process.
2Measurement precision
If accurate body representations are determined using multiple imaging modalities, then measurement precision improves, but loss of time increases due to processing multiple data sets
Solution Approach 1:
The patent applies preliminary action by pre-aligning and pre-processing imaging data from multiple modalities before the actual body representation determination. Landmarks are detected and poses are estimated in advance, so that when multiple data sets need to be processed, the time-consuming alignment and preparation steps have already been completed, reducing overall processing time.
3Object-generated harmful factors
If radiation is targeted accurately using body representation, then object-generated harmful factors are reduced, but measurement precision of body position must be improved
Solution Approach 1:
The patent replaces direct mechanical/radiation-based measurement systems with a computational approach using machine learning models. Instead of relying on radiation patterns to determine body position, the system uses trained neural networks to predict body shape and size from imaging data, enabling more precise positioning with reduced radiation exposure.
4Productivity
If quick determination of body shape and size is achieved, then productivity increases, but measurement precision may deteriorate
Solution Approach 1:
The system performs preliminary actions by pre-training machine learning models on large datasets and pre-aligning imaging data. During actual use, the pre-trained models can quickly determine body shape and size without needing complex real-time computations, achieving both high speed and high accuracy simultaneously.
Solution Approach 2:
The patent uses copying by training machine learning models on copies of imaging data from multiple modalities and patients. The models learn from these synthetic copies and examples, enabling them to quickly and accurately determine body representations for new patients without requiring complex real-time analysis of original data.
Data Source
AI summary
For training a machine learning system for representing a patient body a plurality of stored medical imaging data sets each representing at least a part of a respective patient are obtained. A first one of the plurality of stored medical imaging data sets represents a different part of the patient body than a second one of the plurality of stored medical imaging data sets. A plurality of landmarks in the stored medical imaging data sets are estimated, and each of the stored medical imaging data sets are aligned to a predefined pose using the plurality of landmarks. A plurality of points in the aligned medical imaging data sets are sampled, and the machine learning system is trained based on at least the plurality of points. The learned parameters of the machine learning system are then stored and used in a method for inferring a body representation.