Objective. This study investigates the use of list-mode (LM) maximum a posteriori (MAP) expectation maximization (EM) incorporating prior information predicted by a convolutional neural network for image reconstruction in fast neutron (FN)-based proton therapy range verification. Approach. A conditional generative adversarial network (pix2pix) was trained on progressively noisier data, where detector resolution effects were introduced gradually to simulate realistic conditions. FN data were generated using Monte Carlo simulations of an 85 MeV proton pencil beam in a computed tomography-based lung cancer patient model, with range shifts emulating weight gain and loss. The network was trained to estimate the expected two-dimensional ground truth FN production distribution from simple back-projection images. Performance was evaluated using mean squared error, structural similarity index (SSIM), and the correlation between shifts in predicted distributions and true range shifts. Main results. Our results show that pix2pix performs well on noise-free data but suffers from significant degradation when detector resolution effects are introduced. Among the LM-MAP-EM approaches tested, incorporating a mean prior estimate into the reconstruction process improved performance, with LM-MAP-EM using a mean prior estimate outperforming naïve LM maximum likelihood EM (LM-MLEM) and conventional LM-MAP-EM with a smoothing quadratic energy function in terms of SSIM. Significance. Findings suggest that deep learning techniques can enhance iterative reconstruction for range verification in proton therapy. However, the effectiveness of the model is highly dependent on data quality, limiting its robustness in high-noise scenarios.

This study aims to investigate the capability of U-Nets in improving image reconstruction accuracy for proton range verification within the framework of the NOVO (Next generation imaging for real-time dose verification enabling adaptive proton therapy) project. NOVO aims to enhance the accuracy of proton range verification by imaging the distribution of prompt gamma-rays (PGs) and fast neutrons (FNs) produced by nuclear interactions within tissue. In this work, focus lies on FNs, leaving PGs as future work. A dataset consisting of Monte Carlo-based simple back-projection and ground truth images of FN production distributions in a homogeneous water phantom was utilized. Various U-Net models were trained to predict FN distributions, and a set of range landmark (RL) metrics were computed for evaluation. Linear regression models were established to correlate shifts in mean RL with true range shift magnitudes. Our findings demonstrate a strong linear correlation between the shifts in mean RL in U-Net predictions and the true range shift magnitudes. Multiple RL metrics, including weighted average, inflection point, edge, and peak, were explored. This study highlights the potential utility of U-Nets in enhancing image reconstruction accuracy for proton range verification. The observed correlations between RL shifts and true range shifts provide evidence of the ability of U-Nets to accurately predict images containing range information. Future research will focus on generating more realistic training data incorporating more clinically relevant phantoms, including tissue heterogeneities.