PET Image Reconstruction via Quantum Annealing Optimization
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current image reconstruction methods in PET systems, such as ML-EM and OS-EM, require numerous iterations for convergence and lack a guaranteed mathematical convergence criterion, leading to inefficiencies and potential divergence of pixel values.
Innovation Solution
An image reconstruction method utilizing a quantum computer or pseudo-quantum computer to solve a combinatorial optimization problem based on an objective function, incorporating detection data to perform PET image reconstruction, thereby reducing reconstruction time and eliminating imprecision in convergence criteria.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If ML-EM is used for image reconstruction, then pixel value convergence can be achieved, but a large number of iterations are required
Solution Approach 1:
The patent replaces the classical ML-EM iterative optimization algorithm with a quantum annealing approach. The combinatorial optimization problem is mapped to an Ising model Hamiltonian and solved using quantum annealing, which utilizes quantum mechanical effects (tunneling, superposition) to find the global minimum of the objective function. This substitution of the optimization mechanism enables significantly faster convergence without requiring the numerous iterations needed by classical ML-EM methods.
Solution Approach 2:
The patent transforms the image reconstruction problem by changing the mathematical formulation from a statistical optimization problem to a combinatorial optimization problem expressed as an Ising model. This parameter transformation allows the use of quantum annealing algorithms that can solve combinatorial problems more efficiently than classical iterative methods, thereby reducing the time required for reconstruction while maintaining convergence reliability.
2Productivity
If OS-EM is used to accelerate calculation speed, then reconstruction time is reduced, but pixel values diverge after certain iterations
Solution Approach 1:
The patent replaces the classical OS-EM algorithm with quantum annealing to solve the combinatorial optimization problem. Quantum annealing uses quantum mechanical phenomena to explore the solution space more efficiently, finding the global minimum without getting trapped in local minima or diverging oscillations that plague classical accelerated methods like OS-EM. This ensures both speed and convergence reliability.
Solution Approach 2:
The quantum annealing process inherently includes feedback mechanisms through the annealing schedule and energy landscape evolution. The system continuously adjusts the Hamiltonian parameters during annealing, providing feedback that guides the system toward the optimal solution while preventing divergence. This feedback mechanism ensures convergence without requiring manual intervention to prevent oscillations.
3Reliability
If classical iterative algorithms are used, then mathematical convergence criteria can be established, but the operation lacks specific logical criteria for convergence
Solution Approach 1:
The patent substitutes classical iterative optimization with quantum annealing, which provides a different mathematical foundation for convergence. Quantum annealing converges to the global minimum of the Ising model Hamiltonian through quantum mechanical effects, providing a more robust and easier-to-verify convergence criterion. The annealing process naturally terminates when the quantum system reaches the ground state, providing a clear and simple convergence indicator.
Data Source
AI summary
An image reconstruction method of an embodiment is to perform PET image reconstruction based on an objective function that solves a combinatorial optimization problem. The image reconstruction method of the embodiment includes a step of obtaining detection data and a step of performing PET image reconstruction by solving a combinatorial optimization problem based on the objective function into which the detection data is incorporated.