The question at the heart of the study is deceptively simple: where does the diagnostic information associated with breast microcalcifications actually lie? Is it in the calcium deposits themselves, or in the tissue immediately surrounding them? The biological importance of this surrounding microenvironment is already known from histology and other research. What remained unclear was how much diagnostic information could be obtained from the immediate tissue surroundings in high-resolution 3D X-ray images, independently of the calcified core. According to the researchers, this is the first time, to their knowledge, that the contribution of the surrounding tissue has been isolated and directly compared with that of the calcified core in micrometer-scale 3D X-ray images.

Breast microcalcifications are well-established radiological findings in mammography and digital breast tomosynthesis, and their morphology and distribution play an important role in assessing suspicious breast lesions and determining whether further diagnostic investigation is warranted. Traditionally, image-based assessment has concentrated on the microcalcifications themselves, often examining their characteristics within clusters at mammographic resolution. Conventional clinical breast imaging, however, does not provide the spatial resolution required to investigate these structures at the micrometer scale, and this limitation has left the diagnostic value of the immediate tissue surroundings largely unexplored in three dimensions.

Ex vivo micro-computed tomography, or micro-CT, offers a way around that constraint. By scanning excised breast biopsy specimens, researchers can resolve individual microcalcifications and their immediate microanatomical context in 3D at a spatial resolution that is not achievable with clinical scanners. In this study, the team analysed 94 paraffin-embedded breast biopsy blocks using high-resolution micro-CT with a spatial resolution of approximately 8 micrometers. They segmented 3504 individual microcalcifications and extracted quantitative image measurements, known as radiomic features, from the resulting 3D images. These features were then used to train machine-learning classifiers to predict whether the corresponding breast lesions were histopathologically benign or malignant, providing an objective, computational test of how much discriminative information each image region contains.

Unlike previous approaches based on clusters of microcalcifications at mammographic resolution, the study investigated individual microcalcifications at the micrometer scale. The researchers designed three complementary analyses to separate the contribution of the calcified core from that of the immediately surrounding tissue, while controlling for preprocessing and segmentation effects. The first varied the image preprocessing window while keeping the segmentation mask constant, testing whether the amount of tissue visible around an unchanged calcification outline altered the extracted features. The second systematically expanded and contracted the microcalcification masks, dilating them to include adjacent tissue or eroding them toward the core.

The third analysis was particularly important: it extracted features exclusively from concentric shells inside and outside the calcified core, allowing the calcium deposit itself to be excluded from the analysis entirely. By training classifiers only on the surrounding tissue, the researchers could directly test how much diagnostic information remained without the calcium deposit itself. The results revealed a substantial improvement in classification performance when information from the microcalcifications’ surrounding tissue was incorporated, and the shell-based models delivered the most striking result of all.

Increasing the preprocessing window around an unchanged calcification mask improved the area under the receiver operating characteristic curve, or AUC, from 0.688 to 0.811. Similarly, moderate dilation of the segmentation masks to include adjacent tissue increased the AUC from approximately 0.689 to 0.813. Most notably, classifiers trained exclusively on features extracted from concentric tissue shells, with the calcified core excluded, maintained an AUC of approximately 0.81, comparable to models incorporating both the calcified core and the surrounding microenvironment. These findings indicate that the tissue surrounding individual microcalcifications is a dominant source of discriminative information in this experimental setting.

The study also highlights an important methodological consideration for the broader radiomics community: the size of the preprocessing window can substantially affect radiomic feature extraction and downstream classification performance, even when the segmentation mask remains unchanged. In other words, two analyses of the same calcification with the same outline can yield different feature values, and different classifier outcomes, depending solely on how much surrounding tissue the preprocessing step includes. This sensitivity means that preprocessing choices deserve explicit attention in any radiomics workflow that involves structures embedded in tissue.

By identifying the immediate tissue microenvironment as a particularly informative region, the research provides a rationale for developing radiomic biomarkers that explicitly characterise the interface between microcalcifications and adjacent breast tissue, rather than treating the interface as incidental background in feature extraction pipelines. The findings suggest that quantitative imaging approaches focusing exclusively on the mineralised deposits may overlook valuable diagnostic information, and that looking slightly beyond the calcified core, into its immediate surrounding tissue, can provide additional information about the underlying pathology. The researchers are careful, however, to frame the scope of these conclusions.

The study is part of fundamental research using ex vivo micro-CT imaging of breast biopsy specimens, an approach that enables detailed investigation of individual microcalcifications and their immediate tissue surroundings at a spatial resolution that is not achievable with conventional clinical breast imaging. The findings therefore do not imply that the same structures can currently be resolved, or that equivalent classification performance can be achieved, using clinical breast imaging. Instead, they provide new insights into the spatial origin of diagnostic imaging information and may guide future research into quantitative breast imaging. Further research will be needed to independently validate these findings and to investigate which tissue structures and cell types contribute to the diagnostic signal through detailed histological analysis, work that could help explain the biological basis of the imaging findings and inform the development of clinically applicable imaging biomarkers.

The research is a multidisciplinary collaboration bringing together expertise in medical image analysis, artificial intelligence, radiology and pathology from VUB’s Department of Electronics and Informatics (ETRO)-imec, VUB’s Laboratory of Experimental Pathology, and the Departments of Pathology and Radiology at UZ Brussel. The scientific publication, titled Micro-CT Radiomics Reveals the Microcalcification-Surrounding Tissue Interface as the Diagnostic Epicenter of Breast Microcalcifications, appears in Breast Cancer Research with the DOI 10.1186/s13058-026-02362-0. The authors are Redona Brahimetaj, Jan Cornelis, Guillaume Courtoy, Wim Waelput, Inneke Willekens, Johan De Mey, Ramses Forsyth and Bart Jansen. By pinpointing where the diagnostic signal resides in micrometer-scale 3D images, the study offers a compelling rethinking of one of radiology’s most familiar findings: in this experimental setting, the information that matters most may lie not in the calcium, but in the tissue that surrounds it.

Subject of Research: Micro-CT radiomics of breast microcalcifications and their surrounding tissue in benign versus malignant breast lesions

Article Title: Beyond the calcification: VUB-imec researchers identify the tissue surrounding individual breast microcalcifications as a key source of diagnostic information in 3D micro-CT images

Article References: Beyond the calcification: VUB-imec researchers identify the tissue surrounding individual breast microcalcifications as a key source of diagnostic information in 3D micro-CT images. Original publication (Bibliographic details unavailable.)

Image Credits: AI Generated

DOI: Not provided

Keywords: breast cancer, microcalcifications, micro-CT, radiomics, machine learning, diagnostic imaging, tissue microenvironment, AUC, image segmentation, Breast Cancer Research, VUB, UZ Brussel