Computationally efficient collimator-detector response compensation in high energy SPECT using 1D convolutions and...

Area of Science:

• Medical Imaging Physics
• Computational Imaging
• Nuclear Medicine

Background:

• Accurate collimator-detector response (CDR) modeling is crucial for quantitative single photon emission computed tomography (SPECT) reconstruction, impacting resolution and accuracy.
• Traditional 2D convolution-based CDR modeling can be computationally intensive, especially with septal penetration and scatter, creating bottlenecks in image reconstruction.
• Existing methods struggle with computational efficiency, limiting their application in complex quantitative imaging scenarios.

Purpose of the Study:

• To develop and validate a computationally efficient 1D convolution and rotation-based CDR model for SPECT.
• To reduce reconstruction times while maintaining the accuracy of traditional 2D convolution methods.
• To provide an open-source tool (SPECTPSFToolbox) for broader research and development in SPECT image reconstruction.

Main Methods:

• Formulated a 1D convolution and rotation-based CDR model.
• Fitted the model to Monte Carlo (MC) point source data for isotopes 177Lu, 131I, and 225Ac.
• Compared computational times and reconstruction accuracy against a traditional 2D convolution model using MC, phantom, and patient data.

Main Results:

• The 1D/rotation-based CDR model achieved a two-fold computational speed-up compared to the 2D model on GPUs for typical SPECT matrix sizes.
• Reconstructions using the 1D/rotation model showed negligible differences (around 1%) in count and contrast quantification compared to the 2D model.
• The SPECTPSFToolbox was released on GitHub to facilitate open-source development and application of these CDR models.

Conclusions:

• The proposed 1D convolution and rotation technique significantly accelerates SPECT reconstruction times.
• This method maintains high fidelity with traditional 2D convolution-based CDR modeling, ensuring quantitative accuracy.
• The open-source toolbox enables wider adoption and further development of advanced CDR modeling in SPECT imaging.