APD Detector Crystal Peak Tracking via SOFM Neural Network
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Avalanche-photodiodes (APD) detectors in PET scanners experience significant gain drifting due to thermal variations, causing crystal peak shifting in position profiles, which existing technologies have not effectively addressed.
Innovation Solution
A neural network-based self-organizing feature map (SOFM) is used to build crystal lookup tables on APD detectors, coupled with photomultiplier tubes, to compensate for thermal changes by updating weight vectors and constructing a lookup table, implemented on a field programmable gate array (FPGA) device for real-time tracking.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If APD detectors are used in PET scanners, then detection sensitivity is improved, but gain stability deteriorates due to thermal variations causing crystal peak shifting
Solution Approach 1:
The patent implements a feedback mechanism where the SOFM neural network continuously monitors crystal peak positions and dynamically adjusts the lookup table in real-time based on detected shifts. This closed-loop feedback system compensates for thermal drift by automatically updating the mapping between detector signals and crystal positions, thereby maintaining measurement precision despite gain variations.
Solution Approach 2:
The patent changes the parameter representation by using a self-organizing neural network to dynamically adjust the lookup table parameters. Instead of fixed calibration parameters, the system adaptively modifies the mapping parameters based on real-time thermal conditions, allowing the detector to maintain accuracy across varying temperature conditions.
2Stability of the object's composition
If real-time crystal peak tracking is implemented, then tolerance to thermal changes is improved, but device complexity increases due to neural network implementation
Solution Approach 1:
The patent replaces complex mechanical or electronic adjustment mechanisms with a software-based neural network solution. Instead of physically adjusting detector components to compensate for thermal drift, the system uses computational algorithms running on the FPGA to dynamically update the lookup table, thereby reducing mechanical complexity while achieving thermal compensation.
Solution Approach 2:
The SOFM neural network serves multiple functions: it performs real-time crystal peak tracking, dynamically updates the lookup table, and adapts to various thermal conditions. This multi-functional approach consolidates what would otherwise require separate calibration and correction systems into a single versatile computational module.
3Measurement precision
If SOFM neural network is used for crystal peak tracking, then pixel identification accuracy is improved, but training time and computational resources increase
Solution Approach 1:
The patent performs preliminary training of the SOFM neural network offline before clinical use. During this pre-training phase, the network learns the optimal mapping between detector signals and crystal positions using calibration data. This preliminary action allows the system to achieve high pixel identification accuracy without requiring extensive training time during actual PET scanning operations.
Solution Approach 2:
The patent implements dynamic adaptation where the pre-trained SOFM network continues to learn and refine its parameters during operation. The system balances between utilizing the pre-trained model for immediate accuracy and allowing continuous learning to adapt to specific scanner conditions, thereby optimizing the trade-off between training time and accuracy.
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 solution significantly increases the tolerance of APD detectors to thermal changes, achieving high pixel identification accuracy and real-time tracking of crystal peaks, reducing training time and addressing dead neuron and stability issues, while maintaining accuracy comparable to prior art.
Implementation Method 1
the neural network based self-organizing feature map (SOFM), and increases the tolerance of APD detectors to the thermal changes
Data Source
AI summary
The present invention provides a method of real-time crystal peak tracking for avalanche-photodiode (APD) detectors on positron emission tomography (PET) scanners that satisfies the need to compensate for the significant gain drifting due to thermal variations in APD detectors on PET scanners.