Phase Retrieval in Grating-Based X-ray Imaging: A Comparison of Maximum Likelihood and Fourier-Based Methods in Low-count Regimes

Sammanfattning

X-ray imaging as used in clinical practice today relies solely on the attenuationsignal. Phase-contrast X-ray imaging has emerged as an promising technologyoffering additional information beyond attenuation. Phase-contrast X-rayimaging uses the phase shift the X-ray exhibits when interacting with anobject as its source of contrast. To enable this, the phase shift needs to bereconstructed from the measured intensity on the detector. Advances in high-resolution detectors introduce a need for phase-retrieval algorithms capable ofoperating in low-count, high-noise regimes. In this work, through numericalsimulation of X-ray wave propagation, we evaluate maximum likelihoodestimation (MLE) with and without a regularization term alongside a Fourier-based method. We find that MLE outperforms the Fourier-based methodacross the full range of photon counts evaluated. This work is limited toa simplified geometry and a single sample type; nevertheless, the resultsmotivate further investigation of MLE-based phase-retrieval in medical X-rayimaging