An arXiv preprint reports an 8.8-fold enhancement in the edge resolution of a laboratory X-ray detector when a bulk scintillator was paired with a metalens. The CT claims in the same work are projections from computer models, not direct experimental validation of a complete clinical system.
The study asks whether a metalens can route high-spatial-frequency information from a thick bulk scintillator while retaining X-ray absorption. The proof of concept used a silicon-nitride metalens fabricated on YAG:Ce. Scanning electron microscope images showed nanopillar features down to 50 nanometres, with a 14-fold aspect ratio. The broader work paired this laboratory device with X-ray images of inorganic and biological specimens and with a model-based CT comparison against a pixelated scintillator.
The laboratory test
To quantify the comparison, the researchers used a sharp-edge target and measured the line-spread function, a way of describing how sharply the detector responds to a sudden boundary. Its full width at half maximum was 202.2 micrometres for bulk-scintillator readout and 22.9 micrometres for metalens readout. The supplied analysis reports no confidence interval or repeatability estimate.
The sharper edge response was accompanied by finer-looking specimen images. Compared with bulk readout, the metalens-scintillator images qualitatively recovered more detail in a TEM grid and a needle pinhole. The same pattern appeared in iodine-stained images of an adult ant and a fruit fly. These were qualitative recoveries rather than a numerical clinical image-quality result, so they strengthen the laboratory demonstration without turning it into a clinical test.
The CT case rests on simulations
The CT analysis used detective quantum efficiency, or DQE, a detector measure tracked across different levels of spatial detail. A common stochastic computational pipeline was used for both the metalens and pixelated architectures, with differences limited to architecture-specific optical and detector parameters. In the simulations, the metalens-scintillator preserved DQE at spatial frequencies approaching 70 line pairs per millimetre. The pixelated reference had a Nyquist frequency of 0.97 line pairs per millimetre, the sampling limit used in that comparison.
Using the pixelated detector's DQE threshold as a benchmark, the metalens reached the same threshold at a 25-fold higher spatial frequency in the model. Its full simulated spatial-frequency span was 72 times larger and extended to 70 line pairs per millimetre. The authors also modeled detectability for a 3-mm task. Under fixed acquisition assumptions, the metalens was estimated to preserve that detectability at approximately fivefold lower incident fluence and corresponding modeled radiation dose. That dose figure belongs to the selected task and is not evidence that the same reduction would apply to finer details beyond the pixelated detector's Nyquist limit.
To make the bandwidth comparison more concrete, the study generated simulated X-ray projections of trabecular bone, lung tissue and breast microcalcifications. The modeled projections contained spatial-frequency content beyond the pixelated detector's Nyquist limit, while the pixelated images showed severe loss of fine detail. These were anatomical projections, not reconstructed CT volumes, and the analysis did not provide quantitative clinical detection statistics. They show what the model predicts the detector could preserve in selected structures, not whether patients would receive better diagnoses or lower-dose examinations.
A result still waiting for a full-system test
The modeling was checked against the laboratory edge experiment. Wave-optics calculations reproduced the direction of the sharpening, with predicted line-spread-function widths of 7.9 micrometres and 11.8 micrometres, compared with 22.8 micrometres experimentally. The match validates the trend, not an exact numerical result. The analysis notes that fabrication variation, optical aberrations or misalignment, and approximations in the modeling could contribute to the difference, and it reports no formal uncertainty analysis.
That distinction between a working laboratory demonstration and a projected CT architecture is central to reading the result. The laboratory evidence supports sharper readout from a bulk scintillator, but the CT dose, DQE bandwidth and anatomical findings remain model-based projections rather than direct experimental validation of a complete clinical system. Open questions include whether the full CT-relevant architecture can be fabricated and tested, how larger-area arrays and fabrication variation will affect performance, and how reconstruction will alter the result. Until those tests are done, the modeled gains should be treated as targets for validation.
The paper is an arXiv preprint rather than a journal publication. Its competing-interests statement says J.C., S.V., S.P., C.R.-C. and M.S. are pursuing patent protection for ideas in the work. The authors reported support from government, institutional, fellowship and startup-funding sources. The simulation and analysis code is publicly available, while supporting data can be requested from the corresponding authors. Taken together, the work offers direct laboratory evidence of sharper scintillator readout and CT projections that still require experimental and clinical-task validation.
Paper data and sources
Original title: Nanophotonic control of spatial information in scintillation detectors
Authors: Joshua Chen, Simo Pajovic, Seou Choi et al.
Journal/Repository: arXiv
Status: Preprint, not yet peer-reviewed
First online: 2026-08-26
DOI: Not available
Original paper · Full text