Deep learning based timing calibration for PET.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Nuclear Medicine

Background:

• Neural networks are widely used across various fields.
• Traditional PET timing calibration algorithms like ISTA and ADMM have limitations.
• Optimization algorithms can be represented as networks, offering potential improvements.

Purpose of the Study:

• To introduce a structured deep network for PET timing calibration.
• To develop an ADMM-Net framework by reformulating the ADMM algorithm.
• To leverage the compatibility of the consistency condition method for enhanced calibration.

Main Methods:

• Reformulating the Alternating Direction Method of Multipliers (ADMM) algorithm into a deep network structure.
• Developing the ADMM-Net framework for PET timing calibration.
• Conducting Monte Carlo simulations using the GATE toolkit to evaluate performance.

Main Results:

• The ADMM-Net framework was successfully developed and applied to PET timing calibration.
• The proposed method demonstrated potential in overcoming shortcomings of traditional algorithms.
• Performance was validated through Monte Carlo simulations.

Conclusions:

• The ADMM-Net framework offers a promising approach for PET timing calibration.
• Deep learning reformulation of optimization algorithms can enhance medical imaging calibration.
• Further validation through simulations confirms the framework's potential.