Training End-to-End Unrolled Iterative Neural Networks for SPECT Image Reconstruction

TL;DR

This paper introduces a memory-efficient Julia implementation of a forward-backward projector for SPECT image reconstruction, enabling effective end-to-end neural network training that improves image quality over traditional methods.

Contribution

The authors present an open-source Julia projector supporting exact adjoint backpropagation, significantly reducing memory usage and enabling superior end-to-end training of unrolled iterative neural networks for SPECT reconstruction.

Findings

End-to-end training with the Julia projector yields the best reconstruction quality.

Trade-off observed between computational cost and accuracy among training methods.

End-to-end training produces higher quality images than sequential training and OSEM.

Abstract

Training end-to-end unrolled iterative neural networks for SPECT image reconstruction requires a memory-efficient forward-backward projector for efficient backpropagation. This paper describes an open-source, high performance Julia implementation of a SPECT forward-backward projector that supports memory-efficient backpropagation with an exact adjoint. Our Julia projector uses only ~5% of the memory of an existing Matlab-based projector. We compare unrolling a CNN-regularized expectation-maximization (EM) algorithm with end-to-end training using our Julia projector with other training methods such as gradient truncation (ignoring gradients involving the projector) and sequential training, using XCAT phantoms and virtual patient (VP) phantoms generated from SIMIND Monte Carlo (MC) simulations. Simulation results with two different radionuclides (90Y and 177Lu) show that: 1) For 177Lu XCAT phantoms and 90Y VP phantoms, training unrolled EM algorithm in end-to-end fashion with our Julia projector yields the best reconstruction quality compared to other training methods and OSEM, both qualitatively and quantitatively. For VP phantoms with 177Lu radionuclide, the reconstructed images using end-to-end training are in higher quality than using sequential training and OSEM, but are comparable with using gradient truncation. We also find there exists a trade-off between computational cost and reconstruction accuracy for different training methods. End-to-end training has the highest accuracy because the correct gradient is used in backpropagation; sequential training yields worse reconstruction accuracy, but is significantly faster and uses much less memory.