Subject Motion Measuring Apparatus Using Trained Model for Error Reduction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Magnetic resonance imaging (MRI) systems face challenges in accurately measuring subject motion due to tracking errors caused by disturbances like camera vibrations and skin movements, leading to artifacts in image quality.
Innovation Solution
A subject motion measuring apparatus and method using a trained model to reduce tracking errors from motion information, employing machine learning to output corrected motion data with reduced errors, applicable to various medical imaging modalities including MRI, CT, and PET systems.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If optical tracking system is used to measure subject motion, then motion information can be obtained, but tracking errors occur due to disturbances such as camera vibration and skin movement
Solution Approach 1:
A deep learning-based error correction model is introduced as an intermediary between the optical tracking system and the motion correction processing. This model receives motion information containing tracking errors and outputs corrected motion information by learning the relationship between tracking errors and actual subject motion from training data, thereby eliminating the harmful effect of tracking errors without affecting the core tracking function
Solution Approach 2:
The system implements feedback by using the corrected motion information to adjust the gradient magnetic field in real-time during MRI scanning. The deep learning model continuously processes motion data and provides feedback-corrected motion vectors that guide the gradient field adjustment, creating a closed-loop system that adapts to actual subject motion while compensating for tracking disturbances
2Reliability
If smoothing processing is applied to reduce tracking errors, then noise is reduced, but actual subject motion is also distorted
Solution Approach 1:
Instead of applying fixed smoothing parameters that blur all high-frequency components, the system uses a deep learning model to dynamically change the correction parameters based on the characteristics of the input motion data. The model learns from training data to identify which frequency components correspond to noise and which correspond to actual motion, applying selective correction that preserves motion accuracy while removing noise
Solution Approach 2:
The patent replaces traditional mechanical smoothing filters (which uniformly blur signal) with a neural network-based correction system. The deep learning model processes motion information through multiple layers of neural networks that have learned to distinguish between noise and actual motion patterns, substituting the blunt instrument of conventional smoothing with an intelligent system that can selectively correct errors while preserving motion fidelity
3Measurement precision
If tracking error reduction is performed, then measurement accuracy improves, but processing time increases
Solution Approach 1:
The deep learning correction model is trained in advance using extensive training data that captures various types of tracking errors and corresponding corrected motion information. This preliminary training allows the model to store correction knowledge in its weights, enabling it to quickly process new motion data during actual MRI scanning without requiring real-time complex calculations, thus reducing processing time while maintaining high accuracy
Data Source
AI summary
A subject motion measuring apparatus includes at least one memory storing a program, and at least one processor which, by executing the program, causes the subject motion measuring apparatus to measure motion of a subject and output motion information related to the motion of the subject, reduce, from the motion information, an error caused by a disturbance other than the motion of the subject by using a trained model, and output the motion information in which the error is reduced. The trained model has functions of receiving a data set including the motion information with a plurality of degrees of freedom and outputting the motion information with the plurality of degrees of freedom in which the error is reduced.