Deep Belief Network for Dual-Tracer PET Signal Separation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current dual-tracer PET imaging methods struggle to effectively separate signals from tracers labeled with the same isotope, leading to reduced distinguishability and inefficiencies in clinical applications, particularly in reducing scan time and improving diagnostic accuracy.
Innovation Solution
A deep belief network (DBN) based method is employed to separate dual-tracer PET signals by training a neural network using time activity curves from mixed and single-tracer PET images, enabling the extraction of individual tracer distributions from a mixture of dual-tracer PET images labeled with the same isotope.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If dual tracers labeled with the same isotope are used, then diagnostic information coverage is improved, but signal distinguishability deteriorates
Solution Approach 1:
The patent introduces an isotope different from the primary isotope used in both tracers. A third tracer labeled with this different isotope is injected to provide reference information. This intermediary tracer serves as a mediator that helps distinguish the signals of the two tracers labeled with the same isotope by providing a comparative reference, enabling the separation algorithm to differentiate between the overlapping signals of the primary dual-tracer pair.
2Ease of manufacture
If two tracers are injected to provide non-mixed single tracer information, then separation modeling is facilitated, but scanning time is lengthened
Solution Approach 1:
The patent combines multiple tracer injections into a single mixed injection. Instead of injecting tracers separately to obtain non-mixed single tracer information, the method injects a mixture of multiple tracers (including the primary dual-tracer pair and an intermediary tracer) simultaneously. This merged approach simplifies the scanning process, reduces scanning time, and still enables separation modeling through the deep learning algorithm that processes the mixed signal data.
Solution Approach 2:
The patent replaces the mechanical/physical approach of separate tracer injections with a computational approach. Instead of using physical separation of tracer injections to obtain reference data, the method uses a deep learning algorithm to computationally separate and analyze the mixed tracer signals. This substitution of mechanical injection protocols with computational signal processing reduces scanning time while maintaining the ability to perform accurate separation modeling.
3Productivity
If conventional separation algorithms are used for dual-tracer PET, then scan time is reduced, but effectiveness deteriorates for same-isotope tracers
Solution Approach 1:
The patent changes the fundamental parameters of the separation approach by transitioning from traditional mathematical models based on half-life differences to a deep learning algorithm. The neural network is trained to recognize and separate tracer signals based on learned temporal and spatial patterns rather than relying on fixed physical parameters like half-life. This parameter change enables effective separation of same-isotope tracers while maintaining fast scan times, overcoming the limitations of conventional algorithms.
Solution Approach 2:
The patent replaces conventional mathematical separation algorithms with a deep learning-based computational system. The neural network learns complex non-linear relationships in the tracer kinetics data and performs separation through trained computational models rather than traditional iterative mathematical optimization. This substitution enables the system to handle the challenging case of same-isotope tracers while maintaining high productivity and scan time efficiency.
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 DBN method achieves effective separation of dual-tracer PET signals, improving diagnostic accuracy and reducing scan time by learning a point-to-point mapping relationship between mixed and single-tracer images, demonstrating robust performance across various tracer combinations.
Implementation Method 1
With the help of information exaction ability of deep learning, two individual PET volumetric images can be separated from a mixture of dual-tracer PET images.
Data Source
AI summary
The present invention discloses a DBN based separation method of a mixture of dual-tracer single-acquisition PET signals labelled with the same isotope. It predicts the two separate PET signals by establishing a complex mapping relationship between the dynamic mixed concentration distribution of the same isotope-labeled dual-tracer pairs and the two single radiotracer concentration images. Based on the compartment models and the Monte Carlo simulation, the present invention selects three sets of the same radionuclide-labeled tracer pairs as the objects and simulates the entire PET process from injection to scanning to generate enough training sets and testing sets. When inputting the testing sets into the constructed universal deep belief network trained by the training sets, the prediction results show that the two individual PET signals can been reconstructed well, which verifies the effectiveness of using the deep belief network to separate the dual-tracer PET signals labelled with the same isotope.