Compartment Model Parameter Estimation From Sparse PET Data Using PINNs
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing compartment model parameter estimation methods face challenges with high noise sensitivity, instability at the voxel level, and inefficiency due to the need for full measurement data and long data acquisition times, particularly in medical imaging applications like PET, which can lead to inaccurate results from patient movement.
Innovation Solution
A parameter estimation method using physics-informed neural networks (PINNs) that estimates kinetic parameters from a small amount of measurement data, incorporating a physics-based model, with robustness to noise and flexibility in data acquisition, utilizing a neural network trained with measurement data and constrained by physical laws.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional parameter estimation methods (NLLS, IRWNLLS, NLRR, LLS, TLS, BF) are used, then parameter estimation can be performed, but the results become unstable at voxel level and ROI level when high noise is present
Solution Approach 1:
The patent replaces traditional mechanical/mathematical optimization methods (NLLS, IRWNLLS, NLRR, LLS, TLS, BF) with a neural network-based system. The neural network learns the mapping from measurement data to kinetic parameters through training, substituting the iterative mathematical optimization processes with a data-driven neural network model that provides stable predictions even in high noise conditions.
Solution Approach 2:
The patent creates a virtual copy of the physical measurement system through a neural network model. This neural network copy learns the relationship between input measurements and kinetic parameters during training, allowing it to reproduce accurate parameter estimates without being directly affected by noise in the original measurement data, thus improving both accuracy and stability.
2Measurement precision
If traditional methods require full measurement data for accurate parameter estimation, then estimation accuracy improves, but data acquisition time increases leading to inaccurate results due to object movement and reduced instrument efficiency
Solution Approach 1:
The patent applies partial action by using only a subset of the available measurement data for training the neural network. The network learns to extract meaningful parameters from incomplete data samples, enabling accurate parameter estimation without requiring the full measurement dataset, thus reducing acquisition time while maintaining accuracy.
Solution Approach 2:
The patent performs preliminary training of the neural network model using simulated or pre-acquired data to establish the mapping relationship between measurements and kinetic parameters. This preliminary action allows the network to make accurate predictions on new, incomplete measurement data without requiring extensive real-time data collection, thereby reducing the actual acquisition time needed during application.
3Measurement precision
If data acquisition time is extended to improve measurement accuracy, then parameter estimation improves, but patient movement causes inaccuracies and instrument efficiency decreases
Solution Approach 1:
The patent replaces time-consuming traditional parameter estimation methods with a neural network system that can process measurement data quickly. This substitution reduces the total time required for data acquisition, thereby minimizing the opportunity for patient movement to introduce errors and improving overall measurement accuracy under dynamic conditions.
Solution Approach 2:
The patent creates a neural network model that captures the essential relationships between measurements and kinetic parameters during training. This virtual model can then quickly infer parameters from incomplete or noisy data without requiring extended acquisition time, thus avoiding the harmful effects of patient movement that occur during prolonged scanning.
Data Source
AI summary
The present invention is a parameter estimation method for compartment model based on physics-informed neural networks. Starting from a physical model, the method extracts information from an AIF and a small amount of measurement data to obtain kinetic parameters, thereby greatly improving the scanning efficiency of a measuring instrument, and reducing occurrence of inaccurate estimation results due to patient movement. In addition, the present invention has the robustness to AIF noise and measurement data noise, and can flexibly arrange the time of data acquisition, reduce an error of inaccurate estimation caused by long time 10 acquisition and the patient movement, and improve the efficiency of data acquisition of the instrument. Experimental results show that the present invention is more stable and has less errors. Meanwhile, the present invention does not require the setup of training datasets, and is superior to an end-to-end supervised reconstruction method U-net network with fewer samples.