Extending the field of view in modulation-based X-ray phase microtomography

Recent advances in propagation-based phase-contrast imaging, such as hierarchical imaging, have enabled the visualization of internal structures in large biological specimens and material samples. However, modulation-based techniques, which provide quantitative electron density information, face challenges when imaging larger objects due to stringent beam stability requirements and detector distortions. Extending the field of view of these methods is crucial for obtaining comparable quantitative results across beamlines and adapting to the smaller beam profiles of fourth-generation synchrotron sources. We introduce a novel image processing technique combining an eigenflat optimization with deformable image registration to address the challenges and enable quantitative high-resolution scans of centimeter-sized objects with multiple-micrometer resolution. We demonstrate the potential of the method by obtaining an electron density map of a rat brain sample 15 mm in diameter despite the limited horizontal field of view of 6 mm of the beamline. This showcases the technique’s ability to significantly widen the range of applications of modulation-based techniques in both biological and materials science research.

💡 Research Summary

The paper addresses a fundamental limitation of modulation‑based X‑ray phase‑contrast micro‑tomography: the inability to image objects larger than the available field of view (FOV) while preserving quantitative electron‑density information. Conventional flat‑field correction using eigenflats (principal‑component‑derived flat‑field images) requires a sample‑free region to fit the eigenflat weights. When the object completely fills the detector, as is the case for centimeter‑scale specimens, this approach fails, and time‑varying beam profiles further degrade image quality by leaving residual modulator patterns in the reconstructed phase images.

To overcome these challenges, the authors develop a comprehensive processing pipeline that combines eigenflat optimization with deformable image registration (DIR). The acquisition strategy divides the specimen into multiple overlapping sub‑scans (typically left, middle, and right). Each sub‑scan is performed with the same rotation axis but with a horizontal offset of the detector and source, creating a modest overlap (≈10 % of the single‑scan FOV) between adjacent datasets.

For the outer scans (left and right), the presence of sample‑free margins permits the standard eigenflat method: a set of flat‑field images is decomposed by principal component analysis, and a linear combination of the resulting eigenflats is fitted to the sample‑free region to generate a synthetic reference flat for each projection. The middle scan, however, lacks any sample‑free area, so a direct eigenflat fit is impossible. The authors therefore first apply DIR to align the middle scan with the outer scans. A free‑form deformation (FFD) model based on B‑splines is employed; the control‑point grid is optimized to maximize a similarity metric (e.g., mutual information) between the differential phase images of the overlapping regions. The outer scans are deformed to match the middle scan, while the middle scan remains fixed, which simplifies subsequent calculations.

With the deformed outer scans in hand, the authors formulate a dynamic flat‑field correction for the middle scan. They iteratively adjust the eigenflat weights for the middle scan, each time generating a synthetic flat, retrieving differential phase images (using Unified Modulated Pattern Analysis, UMPA), and computing the difference image between the middle and the corresponding outer region. Assuming the DIR has already removed geometric mismatches, any residual structure in the difference image should be due solely to the modulator pattern or noise. The total variation (TV) of this difference image is used as a cost function; minimizing TV drives the synthetic flat toward eliminating the modulator pattern. The optimization proceeds until the TV falls below a predefined threshold, yielding the optimal eigenflat weights for the middle scan.

After flat‑field correction, the three sets of differential phase images (x‑ and y‑directions) are blended across the overlap zones using a distance‑weighted sum, ensuring a smooth transition. The stitched differential phase projections are then integrated via a 2‑D Fourier‑based integration method, producing final phase projection images. A filtered back‑projection (FBP) algorithm reconstructs the three‑dimensional electron‑density volume.

The method is demonstrated on a rat brain specimen 15 mm in diameter, imaged at the P05 beamline of PETRA III (Hereon, DESY). The beamline provides a horizontal FOV of only 6 mm, far smaller than the specimen. Using Talbot‑array and sandpaper modulators, the authors acquire three overlapping scans and apply their pipeline. The resulting electron‑density map covers the full lateral extent of the brain with a spatial resolution of about 4 µm, and the residual modulator artifacts are dramatically reduced compared with conventional mean‑flat correction. Quantitative consistency across the stitched volume is confirmed by comparing δ/µ ratios with reference measurements, showing improved reproducibility.

In discussion, the authors note that the approach relaxes the stringent beam‑stability requirements that have limited modulation‑based techniques on fourth‑generation synchrotrons, where source size and angular divergence produce very narrow FOVs. Because the eigenflat weights are determined adaptively for each sub‑scan, the method tolerates temporal beam fluctuations. Moreover, the DIR component compensates for spatially varying detector distortions that cannot be modeled by simple rigid translations. The pipeline is largely automated and can be extended to any number of sub‑scans, making it suitable for larger specimens or for laboratory‑scale X‑ray sources with limited coherence.

Future work suggested includes real‑time implementation of the DIR and eigenflat optimization, incorporation of more sophisticated regularization terms (e.g., sparsity or edge‑preserving priors) in the TV cost, and validation on heterogeneous materials where the single‑material assumption (fixed δ/µ ratio) is invalid. The authors conclude that their combined eigenflat‑DIR strategy opens the door for quantitative, high‑resolution phase‑contrast tomography of centimeter‑scale biological and material samples, substantially widening the applicability of modulation‑based imaging techniques.