Learning SPECT detector angular response function with neural network for accelerating Monte-Carlo simulations.

Area of Science:

• Medical Imaging
• Computational Physics
• Artificial Intelligence

Background:

• Accurate simulation of single photon emission computed tomography (SPECT) imaging is crucial for quantitative analysis.
• Traditional simulation methods can be computationally intensive, limiting their application.
• Developing efficient simulation techniques is essential for advancing SPECT technology.

Purpose of the Study:

• To propose and evaluate a novel method for accelerating SPECT simulations.
• To leverage artificial neural networks (ANNs) for modeling the angular response function (ARF) of SPECT collimator-detector systems.
• To reduce computational load and data requirements for SPECT simulations.

Main Methods:

• An artificial neural network (ANN) was trained to learn the angular response function (ARF) of a collimator-detector system.
• The ANN was trained using data from a complete simulation of the SPECT head, including collimator, crystal, and digitization.
• Particle tracking in the simulation was replaced by a plane, with photon energy and direction serving as ANN inputs.

Main Results:

• The ANN-based method demonstrated comparable simulation efficiency to traditional histogram-based ARF methods.
• The proposed approach required less training data and showed reduced dependency on training data statistics.
• The implementation of the method is available within the GATE simulation platform.

Conclusions:

• The ANN-based method offers a computationally efficient alternative for SPECT simulations.
• This approach effectively models the ARF, leading to faster and less data-intensive simulations.
• The method holds promise for improving the speed and accessibility of SPECT imaging research and development.