Neighbor2Inverse: Self-Supervised Denoising for Low-Dose ROI Phase Contrast CT

Abstract— Propagation-based X-ray phase-contrast imaging (PBI) enables high-contrast visualization of lung structures and holds strong clinical potential. However, safe translation requires substantial radiation dose reduction, which inevitably increases image noise. Supervised Convolutional Neural Network-based denoising can restore image quality but depends on paired low- and high-dose datasets, which are rarely available in practice. Self-supervised methods avoid this limitation, yet most are not well adapted to the inverse problem of PBI computed tomography (CT).

We introduce Neighbor2Inverse, a self-supervised denoising framework for low-dose PBI CT. Building on the Neighbor2Neighbor principle, each noisy projection is subsampled into two variants that preserve structural information but contain independent noise realizations. These are reconstructed separately, and the resulting pairs are used to train a denoising network directly in the reconstruction domain. We systematically investigate multiple framework variants and benchmark against established denoising methods.

In region-of-interest PBI CT experiments, Neighbor2Inverse achieves superior noise suppression while maintaining fine structural details, as confirmed by visual assessment, increased contrast-to-noise ratio, enhanced spatial resolution, and improved composite image quality metrics.

Code and data are publicly available at https://github.com/J-3TO/Neighbor2Inverse.

Under submission.