Phase-contrast breast CT wins over radiologists in blinded reader study

Grating-interferometry breast CT, an experimental imaging technique that adds phase-contrast information to conventional X-ray attenuation images, has passed an important perceptual test. In a blinded reader study published in European Radiology Breast, five experienced breast radiologists consistently rated fused images, which combine attenuation and phase information, higher than attenuation-only images across every category they were asked to judge, including image sharpness, tumour delineation, noise and overall image quality.

The work was approved by the responsible ethics review board under EKNZ 2015-132, and all specimens were obtained from participants who provided written informed consent. Each sample was imaged at multiple dose levels ranging from 22 to 176 mGy mean absorbed breast dose, with the dose estimates referring to a full breast of 9-centimetre diameter.

Grating interferometry works by placing a set of micro-structured gratings in the X-ray beam. An absorption grating improves the spatial coherence of the beam, a phase grating generates a fine periodic interference pattern, and an analyser grating in front of the detector allows variations in that pattern to be measured even though detector pixels are too large to resolve the fringes directly. This Talbot-Lau interferometer configuration enables the simultaneous measurement of attenuation, phase-contrast and dark-field signals, the latter arising from small-angle scattering at microscopic structures that conventional breast CT cannot detect.

The specific system used in the study operated at a design energy of 46 keV with a tungsten-anode X-ray source at 70 kVp and a photon-counting detector with 100-micrometre pixels. A full CT scan, including reference scans and 1200 projections of the sample, took about five minutes. The field of view covered approximately one centimetre in height per scan, so taller specimens required multiple scans that were stitched together vertically. Because the samples had to be returned for clinical processing within a two-hour window, the measurement time for each specimen was strictly limited.

For the reader study, the radiologists, who were board-certified with breast imaging experience ranging from two years to more than twenty years, reviewed blinded datasets in random order on certified clinical screens. They saw only one image at a time and rated it on a 5-point Likert scale, without any side-by-side comparison. The images they judged were attenuation-only reconstructions and fused images created by combining a low-pass filtered attenuation image with a high-pass filtered phase-contrast image, exploiting the complementary noise properties of the two channels.

When fused images were compared with attenuation-only images acquired with the same photon statistics, meaning the same dose at the detector, the fused images were significantly preferred in all evaluated categories, with corrected p-values below 0.05 and a sample size of 90 paired ratings. However, the analyser grating absorbs half of the photon counts, so an attenuation image of equal quality could in principle have been acquired at half the sample dose without gratings. When the researchers instead paired fused images with attenuation images acquired at double the mean absorbed breast dose, the preference for fused images persisted as a trend but no longer reached statistical significance in any category, with p-values above 0.05 across the board and a reduced sample size of 45.

Because the samples were scanned at several dose levels, the team could also ask whether higher doses produced better perceived quality. Interestingly, for none of the investigated categories did they observe a trend towards better rankings with increasing dose. An ordinal regression analysis of ratings over image dose found no significant impact, with p-values of 0.527 for attenuation-only images and 0.073 for fused images across all rating categories. The authors suggest that the limited sample size may have masked potential dose-related trends, and they note that the dose levels studied were chosen because earlier work with the same system showed quantitative benefits of the fused images above approximately 16 mGy.

Agreement among the five readers was assessed with Krippendorff’s alpha and Kendall’s W. Agreement levels were similar between image types, but ratings of fused images were more consistent in the overall image quality category, and inter-reader agreement was lowest for tumour delineation. In an illustrative example from one sample, all five readers gave the fused image the highest score for tumour delineation, while three readers gave the attenuation-only image, acquired at double the mean absorbed breast dose, a rating of four points. The authors report that this study provides the first clinical reader-based evidence supporting grating-interferometry computed tomography, although they emphasise that the evaluation concerned perceived image quality rather than diagnostic accuracy.

Several limitations temper the promise. The doses studied, from 22 to 176 mGy, are well above clinically applied screening doses, which are typically on the order of a few mGy. The authors note that with ideal gratings the dose limit for their setup could be pushed down towards roughly 5 mGy, and a recent design study has aimed to identify parameters for a next-generation setup delivering better image quality at the same dose. Scanning times of roughly five minutes per CT are also too long for clinical practice, and a new prototype with imaging times below 30 seconds is currently being assessed. The dark-field signal, which the technique can also provide, holds promise for visualising microcalcifications, but not all samples in the current dataset included microcalcifications, so no conclusive evaluation of that contrast channel was possible.

The study’s conclusion is carefully worded: within the tested dose range, readers preferred fused images combining attenuation and phase information, but when accounting for the dose penalty imposed by the analyser grating, that preference did not reach statistical significance, an outcome the authors attribute partly to limited statistical power from the small sample size. They argue that the findings underscore the importance of continued system development, particularly in grating fabrication and dose efficiency, to fully leverage the diagnostic potential of grating-interferometry breast CT at clinically acceptable exposure levels. With continued advancements aimed at reducing the absorbed dose to the breast, the team writes, the technology is well-positioned to play a significant role in future in vivo diagnostic practices.

Subject of Research: Blinded reader evaluation of perceived image quality in grating-interferometry breast CT of fresh tumourectomy specimens

Article Title: Assessing perceived image quality in grating interferometry breast CT: a blinded reader study with scans of fresh tumourectomies

Article References: Kaser, S., Rawlik, M., Pereira, A., Spindler, S., Etter, D., Zdora, M.-C., Iori, G., Stauber, M., Marcon, M., Frauenfelder, T., Kroschke, J., Weber, J., Ochs, K., Kubik-Huch, R. A., Niemann, T., Leo, C., Rizzo, S. M. R., Schiaffino, S., Dugac, G., & Stampanoni, M. (2026). Assessing perceived image quality in grating interferometry breast CT: a blinded reader study with scans of fresh tumourectomies. European Radiology Breast, 1(1), Article 4. https://doi.org/10.1007/s44502-026-00005-w

Image Credits: AI Generated

DOI: 10.1007/s44502-026-00005-w

Keywords: grating interferometry, breast CT, phase contrast, dark-field imaging, blinded reader study, image quality, tumourectomy specimens, radiologist ratings, radiation dose, tumour delineation, Talbot-Lau interferometer, breast imaging