The authors searched IEEE Xplore, PubMed/MEDLINE and Scopus under a PRISMA 2020 protocol for peer-reviewed primary studies published in English between January 2020 and December 2025. Eligibility required clustering to be applied directly to breast images and sufficient methodological detail to appraise the method, dataset and assessment metrics. The search identified 1,552 candidate records, of which 11 met the criteria.

One reviewer performed screening and deduplication in Rayyan and checked each candidate citation against its source record. QUADAS-2, adapted for unsupervised segmentation, was used to assess risk of bias, while the results were synthesized narratively because of differences among the included studies. Mammography was the dominant modality, appearing in six studies.

Histopathology and dynamic contrast-enhanced MRI were each examined in two studies, and ultrasound was represented by one. The algorithmic mix included fuzzy C-means or a variant in five studies, K-means or a variant in four, and other partitioning methods in two. Five studies used a metaheuristic optimizer, and three performed clustering in a learned feature space.

The authors therefore characterized the field as having evolved from centroid-based partitioning toward optimization-enhanced methods and deep-feature clustering. These counts describe the design of the small reviewed literature rather than demonstrating that one algorithmic approach performs better than another. Accuracy, sensitivity and specificity were the dominant metrics across the studies, indicating that most evaluations compared clustering results against reference labels associated with supervised or otherwise reference-based procedures.

Only two studies reported per-image computational cost, and two reported internal cluster-validity indices. No study reported a silhouette coefficient, inter-observer agreement or prospective validation. The review’s evidence does not supply a comparative ranking of algorithms or establish clinical effectiveness; it instead shows that assessment remained concentrated on agreement with supervised labels while additional measures of unsupervised performance were uncommon.

Several features limit the conclusions that can be drawn from the available evidence. The authors identified uneven reporting, reliance on a small number of publicly available datasets and the absence of prospective validation in demographically diverse populations as barriers to clinical translation. The review was not registered, and the eligible literature comprised only 11 studies, with mammography accounting for six and ultrasound for one.

The findings therefore characterize a limited and heterogeneous body of work, not the full potential of image clustering in breast cancer detection and diagnosis.

Subject of Research: Systematic review of image-clustering methods in breast cancer imaging from 2020 to 2025

Article Title: Image clustering for breast cancer detection and diagnosis: a systematic review of algorithms, imaging modalities, feature representations and reported performance (2020–2025)

Article References: Odili, J., & John, O. (2026). Image clustering for breast cancer detection and diagnosis: a systematic review of algorithms, imaging modalities, feature representations and reported performance (2020–2025). BMC Medical Imaging. https://doi.org/10.1186/s12880-026-02929-5

Image Credits: AI Generated

DOI: 10.1186/s12880-026-02929-5

Keywords: breast cancer, image clustering, mammography, histopathology, magnetic resonance imaging, ultrasound, unsupervised learning, fuzzy C-means, K-means, clinical validation, Image, clustering

The review examines publicly available mammography datasets and their influence on the development of machine-learning and deep-learning systems for classification, lesion detection and segmentation. It covers conventional feature-based methods, convolutional neural networks, transfer learning, object detectors, segmentation networks, Vision Transformers, multimodal systems, explainable AI and privacy-preserving approaches. A structured narrative search covered studies published from 2016 through 2025, with preference given to research using public mammography collections. The authors also assessed each dataset qualitatively for scale, modality, annotation availability and label richness; those suitability ratings were not experimental performance scores.

Public collections have expanded markedly over three decades, but their capabilities remain uneven. MIAS, released in 1994, contains 322 film mammograms from 161 patients with image-level labels and has become a simple benchmark for conventional algorithms and lightweight convolutional networks. DDSM and its curated version CBIS-DDSM provide larger collections with pathology information, lesion regions and, in the curated dataset, segmentation masks. INbreast contains 410 high-quality full-field digital mammograms from 115 patients with pixel-level masks, making it valuable for lesion detection and segmentation research but too small for training large models from scratch. These older resources also rely heavily on digitized film or geographically limited populations.

More recent repositories offer scale, richer labels and, in some cases, clinical context. VinDr-Mammo contains 5,000 Vietnamese full-field digital examinations and 20,000 images with BI-RADS assessments and bounding boxes. CMMD includes 3,712 images from 1,775 Chinese patients, histopathology labels and molecular-subtype information, although it lacks region-of-interest annotations. OMI-DB provides a very large UK collection with lesion annotations, clinical metadata and longitudinal screening information, but access is restricted. DMID, derived from an Indian population, includes 510 full-field digital images with lesion and clinical metadata and remains comparatively limited and evolving. Together, these resources have enabled more ambitious research, but no single public collection combines large scale, broad demographics, standardized annotation, comprehensive clinical metadata and open access.

Annotation depth determines which questions an AI system can learn to answer. Whole-image labels support image classification but provide no clue about where a suspicious finding lies. Bounding boxes identify approximate lesion locations for detectors such as YOLO and Faster R-CNN, while pixel-level masks delineate boundaries for segmentation models such as U-Net and Mask R-CNN. Multi-level annotations can support systems that jointly classify, locate and outline lesions. Standardized BI-RADS descriptors, covering findings such as lesion shape, margins, calcifications and breast density, can add consistent clinical semantics. Yet annotation formats and labeling practices differ across collections, and inter-observer variability introduces label noise. Expert-drawn pixel masks are also labor-intensive, limiting the scale of precisely annotated data.

The technical trajectory has moved from manually engineered texture, shape and statistical features to end-to-end deep networks. Convolutional neural networks learn hierarchical image representations and remain the dominant approach, while transfer learning reuses pretrained weights to cope with smaller labeled collections. YOLO detectors favor speed and real-time inference, whereas Faster R-CNN generally provides stronger localization at greater computational cost. U-Net variants produce detailed lesion boundaries but depend on scarce pixel-level annotations. Vision Transformers use self-attention to capture long-range spatial relationships, and hybrid CNN–Transformer models combine local texture with global context. The review notes, however, that transformers usually require more training data than CNNs. Ensembles and multi-stage systems can integrate several tasks, but their computational complexity may complicate routine clinical use.

High benchmark results do not resolve the central generalization problem. The reviewed literature contains many reports of accuracy above 90%, and some studies describe values above 98% on benchmark collections. Such figures concern the datasets, preprocessing and evaluation protocols used in those studies and do not establish equivalent performance in hospitals with different equipment, acquisition protocols, patient populations or annotation practices. Models evaluated on an independent dataset can perform less well, particularly when breast density, disease distribution, image characteristics or preprocessing differ. Transfer learning, domain adaptation and ensemble methods may reduce these shifts, but the review says they do not eliminate dataset bias, class imbalance and limited interpretability.

Geographic concentration compounds the problem. Much of the research literature relies on North American and European collections, particularly DDSM, CBIS-DDSM, MIAS and INbreast. Public data from Africa, South America and parts of Asia remain scarce, including in India despite its substantial breast-cancer burden. DMID is an important step toward regional representation, but its institutional and geographic coverage is not yet sufficient. Private hospital datasets may contain richer clinical information and greater patient diversity, yet restricted access prevents reproduction and standardized comparison. This imbalance makes it difficult to determine whether a model trained in one population will retain useful sensitivity and specificity in another, or whether domain differences could increase false-positive or false-negative predictions.

The review therefore calls for a comprehensive, geographically diverse mammography benchmark assembled across institutions and healthcare settings. Such a resource would need consistent BI-RADS labels, lesion localization, segmentation where feasible, pathology and follow-up information, and clearly documented data splits and evaluation metrics. Privacy-preserving collaboration through federated learning could allow institutions to contribute to model development without centralizing sensitive patient data. Prospective, multicenter validation would also be needed to test whether benchmark performance survives contact with routine clinical workflows. The central message is that AI-assisted mammography may become more capable through transformers, multimodal learning and explainable systems, but trustworthy deployment depends on data that represent the people and conditions the software is intended to serve.

Subject of Research: How dataset quality, demographic bias and annotation standards affect AI-based mammography for breast cancer detection

Article Title: Mammography datasets and artificial intelligence for breast cancer detection: challenges, bias, and future directions

Article References: Mohideen, A. K., Prakash, P. N. S., & Priyadharshini, M. K. (2026). Mammography datasets and artificial intelligence for breast cancer detection: challenges, bias, and future directions. Frontiers in Artificial Intelligence, 9, Article 1933770. https://doi.org/10.3389/frai.2026.1933770

Image Credits: AI Generated

DOI: 10.3389/frai.2026.1933770

Keywords: mammography, breast cancer, artificial intelligence, medical imaging, dataset bias, deep learning, Vision Transformers, annotation standards, external validation, federated learning, datasets, artificial

Performance in the training set was strong: the area under the curve was 0.92, with a 95% confidence interval of 0.86–0.96, while sensitivity reached 87.3% at 85.7% specificity. The independent validation set yielded an area under the curve of 0.95, with a 95% confidence interval of 0.91–0.97. At the fixed cutoff, sensitivity was 93.2% and specificity was 85%.

The investigators also performed internal validation through bootstrapping and Monte Carlo cross-validation, reporting that the assay’s discriminatory power remained robust under statistical resampling. These subgroup observations were derived from the retrospective study population, and the supplied results do not provide separate performance estimates for each subgroup. The study therefore supports the feasibility of BF-09 as a candidate blood-based marker, but it does not establish performance in a prospective screening population or show that use of the assay improves clinical outcomes.

Mammography is the standard-of-care screening method for breast cancer, although the paper identifies tissue density and overdiagnosis of benign conditions among its limitations and notes growing interest in alternative blood-based tests. Against that background, the findings offer preliminary support for BF-09 as a possible alternative screening assay rather than a validated replacement for established screening. Additional prospective evaluation would be needed to determine how the test performs when used in the intended screening setting.

The authors characterized BF-09 as having high potential as an alternative breast cancer screening test, based on the reported retrospective diagnostic performance and internal statistical validation.

Subject of Research: Retrospective evaluation of serum BF-09 for breast cancer detection.

Article Title: Diagnostic potential of serum BF-09 biomarker (protein SGF29) for early detection of breast cancer: high performance for early-stage cancer and major histological or molecular subtypes

Article References: Chavany, C., Dea, J., Valle, R. P. C., González, R. H., & Jendoubi, M. (2026). Diagnostic potential of serum BF-09 biomarker (protein SGF29) for early detection of breast cancer: high performance for early-stage cancer and major histological or molecular subtypes. Journal of Cancer Research and Clinical Oncology. https://doi.org/10.1007/s00432-026-06634-2

Image Credits: AI Generated

DOI: 10.1007/s00432-026-06634-2

Keywords: breast cancer, BF-09, SGF29, blood biomarker, ELISA, diagnostic accuracy, cancer screening, retrospective study, early detection, mammography, Diagnostic, potential

The core insight behind the work is that more computation does not always mean better diagnosis. Existing deep learning and optimization-assisted mammographic classification frameworks often suffer from high-dimensional feature redundancy, increased computational complexity, and limited interpretability. When a model relies on thousands of learned features, many of them turn out to be irrelevant noise that inflates memory requirements, slows training, and increases the risk of overfitting. Subramaniam’s framework, designated SFS-FS-BC, instead begins with just 30 handcrafted radiomic descriptors extracted from lesion regions and lets an optimization algorithm decide which of them actually matter for distinguishing benign from malignant tissue.

The choice of optimization engine is what gives the study its distinctive flavor. Stochastic fractal search is a population-based metaheuristic inspired by the growth and diffusion patterns of natural fractals, the branching structures seen in everything from river deltas to lung tissue. In the algorithm, each candidate solution, called a particle, represents a possible subset of features encoded as continuous values between zero and one. During the diffusion phase, particles perform Gaussian random walks through the search space using two complementary strategies: one directed toward the best solution found so far, which accelerates exploitation of promising regions, and one centered on each particle’s current position, which preserves the diversity needed for global exploration.

After diffusion, a ranking-based update mechanism refines the population. Particles are sorted by fitness, measured as the classification accuracy achieved by their feature subset under cross-validation, and weaker solutions are improved through stochastic interactions with stronger ones. Once the optimization converges, the continuous particle vectors are converted into binary feature masks using a fixed threshold of 0.5: values at or above the threshold select a feature, values below discard it. The result is a compact, optimized feature subset that feeds directly into the k-nearest neighbor classifier, which labels a test sample by majority vote among its five closest neighbors measured with Euclidean distance.

The experimental design reflects careful attention to a problem that plagues many machine learning studies in medicine: data leakage. The framework employed a patient-wise stratified 5-fold cross-validation protocol in which all images from the same patient were kept within a single fold, preventing the model from learning one patient’s imaging signature and then being tested on it again. Normalization parameters were computed exclusively from training subsets, and the feature selection optimization itself was run independently within each training fold. All reported metrics represent means and standard deviations across 30 independent experimental runs, and pairwise Wilcoxon signed-rank tests confirmed that the improvements over baseline methods were statistically significant at p < 0.01.

The numbers tell a compelling story. On CBIS-DDSM, a curated subset of the Digital Database for Screening Mammography containing 891 benign and 675 malignant lesions, the framework achieved 97.3 percent accuracy, 97.1 percent precision, 97.5 percent recall, an F1-score of 97.3 percent, and an area under the ROC curve of 0.991. On INbreast, a high-quality full-field digital mammography dataset with 220 benign and 190 malignant cases, performance climbed to 98.7 percent accuracy and 0.995 AUC. Perhaps most telling is the comparison with the baseline: a k-nearest neighbor classifier using all 30 features without selection managed only 87.6 percent accuracy on CBIS-DDSM, meaning the fractal optimization alone accounted for nearly a ten-point gain.

Feature reduction proved equally important. The stochastic fractal search compressed the original 30-dimensional feature space to an average of 12.67 optimized features, and an analysis of selection frequencies across 30 runs revealed which descriptors the algorithm consistently favored. Texture-based measures such as contrast, entropy, homogeneity, and correlation appeared again and again, confirming that tissue heterogeneity patterns are among the strongest signals of malignancy. Morphological descriptors like circularity and eccentricity were also repeatedly selected, underscoring the clinical observation that malignant lesions tend to have irregular, poorly defined shapes compared with benign masses.

Convergence behavior further distinguished the approach from its competitors. The optimization typically stabilized within roughly 35 to 40 iterations, despite a maximum limit of 100, thanks to a convergence-based early stopping mechanism that halts the search when the best fitness fails to improve beyond a tolerance of 10 to the minus 6 for ten consecutive iterations. In head-to-head comparisons, particle swarm optimization required nearly 65 iterations and genetic algorithms about 80 to approach comparable fitness values. An ablation study confirmed that every component contributes: removing fractal diffusion, ranking-based updates, or Z-score normalization each degraded performance, while the full pipeline reached 98.47 percent accuracy on the combined datasets.

The study is candid about its limits. The framework depends on expert-provided lesion annotations and segmentation masks rather than performing fully automated lesion localization, and it was evaluated only on public datasets without an independent external test set, so the results cannot yet be read as proof of clinical generalizability. Runtime comparisons with deep learning methods were conducted on CPU hardware and should be interpreted cautiously, since many competing architectures were originally implemented on GPU-accelerated systems. The work also addresses only binary benign-malignant classification, not the full clinical workflow of detection, segmentation, and risk assessment.

Even with those caveats, the implications are significant. Because the selected features correspond to interpretable morphological, intensity, and texture characteristics that radiologists already assess, the framework offers a transparency that black-box deep learning models often lack. Its low computational footprint makes it a plausible candidate for resource-constrained clinical environments where GPU infrastructure is unavailable, potentially supporting screening workflows and second-opinion analysis. The author points toward multi-institutional validation, automated segmentation integration, and hybrid radiomics-deep learning architectures as the next steps. For now, the study stands as an elegant demonstration that sometimes the smartest way forward in medical AI is not a bigger model, but a better-chosen handful of features, found with a little help from the mathematics of fractals.

Subject of Research: Fractal optimization-driven feature selection and k-nearest neighbor classification of benign and malignant breast lesions in mammographic images

Article Title: Breast cancer classification in mammographic images using fractal diffusion feature selection and k nearest neighbor

Article References: Subramaniam, K. (2026). Breast cancer classification in mammographic images using fractal diffusion feature selection and k nearest neighbor. Discover Artificial Intelligence, 6(1), Article 1236. https://doi.org/10.1007/s44163-026-02284-y

Image Credits: AI Generated

DOI: 10.1007/s44163-026-02284-y

Keywords: breast cancer, mammography, stochastic fractal search, feature selection, k-nearest neighbor, computer-aided diagnosis, radiomics, machine learning, CBIS-DDSM, INbreast, metaheuristic optimization, medical imaging

The research, led by Aika Hayashi of Yokkaichi Municipal Hospital and colleagues at Suzuka University of Medical Science in Japan, focused on the newer of two mammography-guided biopsy approaches. Conventional stereotactic-guided vacuum-assisted biopsy triangulates the lesion from two angled two-dimensional images, while digital breast tomosynthesis-guided biopsy reconstructs the target in three dimensions from multiple tomogram layers. Performed with the patient lying face down, the tomosynthesis-guided technique allows more accurate coordinate determination and shorter procedure times than its stereotactic predecessor. What it has lacked, the authors note, is any clear criterion for choosing the optimal compression direction based on where the calcifications sit within the breast and what the biopsy device itself can physically do.

To fill that gap, the team built a variable proprietary breast phantom embedded with simulated calcifications in all four breast quadrants — the upper-outer, upper-inner, lower-outer and lower-inner regions that radiologists use to map findings. They then compressed and imaged the phantom from eight different directions, spanning the standard craniocaudal, mediolateral, lateromedial and oblique projections used in clinical practice. For every combination of calcification location and compression direction, the researchers measured the distance from the needle tip to the support platform, the distance from the tissue sampling port to the center of the target, and the spatial relationship between the needle assembly and the compression paddle. They also recorded whether the device displayed warning messages signaling that tissue sampling would be difficult.

The statistical analysis delivered an unambiguous result: across every breast region, the choice of compression direction was significantly associated with whether an alert occurred, with p values below 0.001. In practical terms, the geometry of the approach — not chance — determines whether the biopsy proceeds cleanly. Compression direction also dictates the trajectory the biopsy needle must follow through the compressed breast, and certain trajectories are inherently riskier than others.

Two failure modes emerged from the phantom experiments, and they sit at opposite ends of the same geometric line. When the breast is compressed along the target’s diagonal from the opposite quadrant — that is, from 180 degrees across from the calcification — the needle trajectory lengthens and the risk of target deviation rises, along with the possibility of damage to the detector. Push too far in that direction and the needle may simply miss the calcifications it was aimed at, or the hardware itself may pay the price.

The mirror-image problem is just as troublesome. Compressing from the opposite direction on the same diagonal — the approach that brings the compression paddle closest to the calcification — can push the tissue sampling port above the skin level. In that configuration, the vacuum mechanism may fail to aspirate tissue, and the port itself risks bruising the patient’s skin. A biopsy designed to be gentle and precise can thus be undermined by the very maneuver intended to bring the target within reach.

These findings give device-generated warnings a new interpretation. In prone tomosynthesis-guided biopsy, the authors conclude, the alerts the machine displays serve as a proxy for technical difficulty rather than as predictors of clinical outcome. An alert does not tell the radiologist whether the biopsy will ultimately succeed in diagnostic terms; it signals that the physical setup has pushed the device toward the edge of its operating envelope. Reading the warning as a prompt to reconsider the compression direction, rather than as a verdict on the procedure, is the practical takeaway.

The study’s recommendation is correspondingly concrete: avoid the two compression directions most likely to produce errors — the 180-degree opposite approach across the target’s diagonal, and the direction immediately adjacent to the calcification area. Steering around those two angles, the authors argue, helps clinicians manage the physical constraints of the biopsy system while preserving reliable diagnostic capability. Because the phantom placed calcifications in all four quadrants and tested eight compression directions, the guidance applies across the full map of the breast rather than to a single favored region.

As with any phantom study, the results come from a laboratory model rather than living tissue, and the authors note the work involved no human or animal subjects. Real breasts vary in density, elasticity and size in ways a phantom cannot fully reproduce. Nevertheless, the systematic geometry the study documents — the diagonal trade-off between needle trajectory length and sampling-port exposure — reflects the fixed mechanics of the biopsy device itself, which behaves the same way whether the medium is gel or glandular tissue. That makes the findings directly actionable for the radiographers and radiologists who set up these procedures every day.

The broader significance lies in how small procedural choices ripple into diagnostic quality. Vacuum-assisted biopsy has become the standard alternative to surgical excision for non-palpable microcalcification lesions precisely because it is minimally invasive yet highly precise. A technique that depends on millimeter-scale targeting cannot afford avoidable geometric errors, and this study shows that the compression paddle’s angle is one of those controllable variables. By turning an intuitive choice into an evidence-based rule — never approach from directly across the diagonal, never from immediately beside the calcification — the Japanese team has given breast biopsy teams a simple decision aid grounded in measured data rather than habit.

Subject of Research: Optimal breast compression direction for tomosynthesis-guided vacuum-assisted biopsy of calcifications

Article Title: Assessment of breast compression direction based on calcification location in prone digital breast tomosynthesis-guided vacuum-assisted breast biopsy: a phantom study

Article References: Hayashi, A., Inagaki, Y., Matsuura, K., & Muto, H. (2026). Assessment of breast compression direction based on calcification location in prone digital breast tomosynthesis-guided vacuum-assisted breast biopsy: a phantom study. BMC Medical Imaging. https://doi.org/10.1186/s12880-026-02776-4

Image Credits: AI Generated

DOI: 10.1186/s12880-026-02776-4

Keywords: breast biopsy, vacuum-assisted biopsy, digital breast tomosynthesis, microcalcifications, mammography, phantom study, prone position, breast compression, breast cancer diagnostics, interventional radiology, medical imaging, BMC Medical Imaging

The study, led by Fatema Habbash of Arabian Gulf University and colleagues, used multistage and systematic sampling to recruit participants and conducted face-to-face interviews with a structured questionnaire. Among the 302 women surveyed, 181, or 59.9%, had undergone mammography at least once. But only 59, or 19.6%, had been screened within the recommended interval. That distinction matters enormously. A single mammogram can detect an existing tumour, but regular screening at fixed intervals is what allows radiologists to compare images over time and identify subtle changes that signal early malignancy. The low regular-screening rate suggests that many women view mammography as a one-time event rather than an ongoing preventive habit.

The barriers the researchers identified are striking in their ordinariness. The most common reason for not screening, cited by 67.9% of women, was simply believing mammography was unnecessary. Close behind, at 64.4%, was the absence of a physician recommendation. Insufficient information accounted for 42.4% of non-uptake. Fear of pain and fear of a cancer diagnosis each affected 30.9% of women, while fear of treatment or mastectomy affected 27.6%. Being asymptomatic was cited by 23.4%, embarrassment by 23.0%, concerns about radiation by 15.6%, and lack of time by 10.0%. Taken together, these numbers paint a picture of a population that largely does not see itself as at risk and has not been told otherwise by the medical professionals it trusts.

That last point deserves particular attention. The belief that mammography is unnecessary unless symptoms appear is a well-documented misconception in many populations, and it is especially consequential because early breast cancer is frequently asymptomatic. The whole purpose of screening is to detect disease before it announces itself. When nearly a quarter of women cite being asymptomatic as a reason not to screen, it signals a fundamental misunderstanding of what screening is for. Similarly, the finding that two-thirds of women lacked a physician recommendation highlights the decisive role that primary care providers play in translating public health guidelines into individual action.

When the researchers analysed which factors were actually associated with screening uptake, the results were unambiguous. Screening was significantly linked to age, family history of breast cancer, previous breast-related problems, and receiving a screening invitation from a healthcare provider. In multivariable analysis, which adjusts for the influence of other variables, receiving a formal screening invitation within the previous two years was strongly associated with higher uptake, with an adjusted odds ratio of 5.23 and a 95% confidence interval of 2.55 to 10.72. In practical terms, women who received a formal invitation were more than five times as likely to have been screened as those who did not. That is one of the strongest effects a public health intervention can produce, and it points to a remarkably simple lever for change.

The other significant predictors reinforce the clinical logic of existing screening programmes. Women aged 50 and above were more likely to undergo screening, consistent with the fact that breast cancer risk rises with age and that most international guidelines prioritise this group. Women with a family history of breast cancer were also more likely to screen, reflecting heightened perceived susceptibility. And those with previous breast-related problems were more likely to have had a mammogram, likely because such problems prompt clinical follow-up. What emerges is a pattern in which women who already perceive themselves as at risk, or who have been personally prompted by the health system, take action, while everyone else falls through the cracks.

The implications for Bahrain and the wider Gulf region are significant. The Gulf Cooperation Council countries share cultural, demographic, and healthcare-system characteristics, and the barriers identified here—misconceptions about necessity, fear of diagnosis, embarrassment, and reliance on physician prompting—are likely to resonate across the region. The study’s authors argue that culturally tailored health education, systematic screening invitations, and proactive physician recommendations may improve participation. The strength of the invitation effect suggests that a structured invitation system, in which every eligible woman receives a formal prompt at recommended intervals, could dramatically raise regular screening rates without requiring new technology or infrastructure.

The researchers also point toward the future of screening outreach. They call for future research to assess technology- and AI-supported outreach strategies, especially for younger, asymptomatic, and under-screened women. The logic is compelling: the women least likely to screen are those who feel well and see no reason to visit a clinic, and reaching them may require digital reminders, mobile health applications, or AI-driven risk stratification that identifies who needs prompting most. In a country with high smartphone penetration and a well-developed primary healthcare network, the technical foundations for such outreach already exist.

What makes this study particularly valuable is its methodological rigour. By randomly selecting participants across all four governorates using multistage and systematic sampling, the researchers produced a sample that reflects the broader population of Bahraini women in the target age range. The face-to-face interview approach minimised literacy barriers and ensured consistent data collection. The multivariable analysis distinguished between factors that merely correlate with screening and those that independently drive it. The finding that a formal invitation independently predicts uptake, even after adjusting for age, family history, and breast-related problems, gives policymakers a clear, evidence-based target.

For a condition that is the most common cancer among women worldwide, the gap between what screening can achieve and what regular screening rates actually are remains one of public health’s most stubborn challenges. This study shows that in Bahrain, the solution may not lie in convincing women one by one, but in building a system that reaches out to them automatically. A formal invitation, delivered at the right interval, backed by culturally attuned education and proactive primary care physicians, could transform mammography from an occasional event into a routine preventive habit. The evidence suggests that the simplest intervention—asking women to come in for screening—may be the most powerful one available.

Subject of Research: Public Health

Article Title: Determinants and barriers to mammography screening uptake among Bahraini women: a cross-sectional study in primary healthcare settings

Article References: Habbash, F., AlMohri, M., Waleed, A., Hamdan, H., Salman, K., Obaid, S., Abdulla, Z., Almarabheh, A., & Salah, A. B. (2026). Determinants and barriers to mammography screening uptake among Bahraini women: a cross-sectional study in primary healthcare settings. BMC Public Health. https://doi.org/10.1186/s12889-026-29859-9

Image Credits: AI Generated

DOI: 10.1186/s12889-026-29859-9

Keywords: mammography, breast cancer screening, Bahrain, public health, preventive medicine, primary healthcare, cancer epidemiology, Determinants, barriers, screening, uptake, Bahraini

The clinical rationale for AI augmentation is stark. Perceptual errors account for an estimated 60 to 80 percent of missed diagnoses in retrospective audits, and fatigue during high-volume reporting can reduce radiologist performance. Inter-reader variability among experienced radiologists, with Cohen’s kappa around 0.4 to 0.6, affects recall and biopsy thresholds. Meanwhile, mammographic screening sensitivity falls to 48 percent in dense breasts, and false-positive rates of 10 to 12 percent per round generate unnecessary procedures and patient anxiety. Five-year relative survival approaches 99 percent for localized disease but drops to 28 percent for distant disease, making early detection central to improving outcomes.

The methodological landscape covered by the review spans classical machine learning and radiomics, convolutional neural networks, YOLO-family detectors, vision transformers and hybrid CNN-Transformer architectures, U-Net variants, Mask R-CNN, SAM, and MedSAM across mammography, tomosynthesis, MRI, ultrasound, contrast-enhanced mammography, and emerging photoacoustic imaging. Convolutional neural networks have dominated breast imaging AI since the ImageNet deep-learning breakthrough, exploiting spatial structure through hierarchical local feature extraction. Single-stage YOLO detection frameworks formulate lesion localization as a direct regression problem, enabling fast inference suitable for real-time detection workflows. Vision transformers apply global self-attention to sequences of image patches, motivating hybrid CNN-Transformer approaches for medical image analysis.

Representative CNN studies show impressive retrospective benchmarks. A deep-learning mammography system evaluated on independent UK and US screening cohorts achieved an AUC of 0.871 in a reader study compared with a mean reader AUC of 0.750, with absolute reductions in false positives of 1.2 percent in the UK and 5.7 percent in the US, and absolute reductions in false negatives of 2.7 and 9.4 percent respectively. A simulated UK double-reading workflow maintained non-inferior performance while reducing second-reader workload by 88 percent. The open-source Mirai model, developed at Massachusetts General Hospital and evaluated on held-out cohorts from Sweden and Taiwan, obtained C-indices of 0.76 to 0.81 for one- to five-year breast cancer risk prediction, outperforming established clinical risk models where direct comparison was available.

YOLO-family models have shown promising retrospective evidence for fast detection and localization. One early YOLO-based CAD system using 600 original DDSM mammograms and 2,400 augmented images achieved 99.7 percent mass-detection accuracy and 97.0 percent benign/malignant classification accuracy under five-fold cross-validation, although the review cautions that such very high accuracies on single benchmarks or heavily augmented data carry substantial overfitting risk and do not equate to clinical sensitivity or specificity in prospective screening workflows. A YOLOv5-CAD framework using transfer learning across CBIS-DDSM, VinDr-Mammo, INbreast, MIAS, and a private cohort achieved mAP 0.843 with 14.8 millisecond inference time when transferring from CBIS-DDSM to INbreast, with external MIAS and private cohort mAP values of 0.756 and 0.816 respectively.

Segmentation architectures have evolved from standard U-Net through attention mechanisms, transformer integration, and foundation model-based promptable segmentation. The original U-Net, introduced in 2015 with its symmetric encoder-decoder structure and skip connections, achieved Dice of 0.92 on histological cell segmentation and has been forked over 10,000 times on GitHub. Attention U-Net applied to the BUSI ultrasound dataset achieved Dice of 0.87 and Hausdorff distance of 5.8 millimeters for breast lesion boundary delineation. The nnU-Net framework applied to 100 three-dimensional fat-suppressed breast DCE-MRI scans achieved Dice of 0.968 for whole breast and 0.877 for fibroglandular tissue. MedSAM, fine-tuned on 1.5 million medical image-mask pairs spanning 10 imaging modalities, achieved Dice of 0.88 on the breast imaging subset. Mammo-SAM, a breast-specific adaptation of the Segment Anything Model, improved breast mass segmentation over unadapted SAM on CBIS-DDSM and INbreast, demonstrating that foundation segmentation models require domain adaptation before clinical use in mammography.

Explainable AI has become essential because diagnostic models must support radiologist trust, error analysis, and regulatory transparency under the EU AI Act and FDA guidance. The review highlights Grad-CAM and Grad-CAM++ saliency heatmaps, bounding-box and lesion-localization outputs, segmentation masks, attention visualization, and case-based interpretability as the explainability approaches most relevant to breast imaging. Grad-CAM generates class-discriminative heatmaps by weighting convolutional feature maps using gradient information, helping radiologists assess whether model attention overlaps with suspicious masses, calcifications, or enhancement regions. Case-based interpretable models support predictions by comparison with visually similar mammographic mass cases, a clinically intuitive approach because radiologists often reason by comparison with prior cases. The review emphasizes that localization maps and segmentation masks should not be interpreted as full mechanistic explanations, and attention weights should be interpreted cautiously because they do not always provide faithful explanations of model causality.

Multimodal fusion, radiogenomics, and federated learning represent emerging directions. A machine-learning radiogenomic analysis using 922 patients and 529 DCE-MRI features demonstrated moderate associations between MRI-derived quantitative features and molecular markers including ER, PR, HER2, Ki-67, and molecular subtype. Federated learning addresses the privacy challenge of building generalizable models across institutions without centralizing sensitive patient data, sharing model updates rather than raw imaging data. Public datasets including CBIS-DDSM, INbreast, VinDr-Mammo, BUSI, Duke Breast Cancer MRI, and EMBED, with approximately 3.4 million mammograms and linked demographic metadata, enable reproducible benchmarking and fairness auditing, while open-source systems such as Mirai, nnU-Net, and MedSAM allow local validation.

Recent prospective and implementation evidence strengthens the clinical-translation component. The final MASAI randomized trial, including more than 105,000 women, showed that AI-supported screening achieved a non-inferior interval cancer rate, higher sensitivity, unchanged specificity, and 12 percent interval-cancer reduction versus standard double reading. The AI-STREAM prospective multicenter cohort in South Korea, evaluating 24,543 women, reported a 13.8 percent increase in cancer detection without a significant recall-rate increase. The AITIC paired noninferiority trial showed that partially autonomous AI triage reduced radiologist workload by 63.6 percent while increasing cancer detection by 15.2 percent, although recall rates increased in the overall cohort and did not meet the prespecified noninferiority criterion. Differences in recall burden probably reflect threshold setting, whether AI was used as triage or decision support, and local workflow design.

Despite these advances, persistent challenges include data heterogeneity, class imbalance, domain shift, bias, incomplete calibration, and limited prospective validation. The review found that only 38 of the 284 evidence-map entries were verified from full text, 48 were based on abstracts, and 198 were based on DOI or source pages, and the authors emphasize that the supplementary data sheet should be interpreted as a provenance-tagged evidence map rather than a formal study-level risk-of-bias assessment. The most frequent methodological concerns were retrospective single-institution development, internal-only train-test splits with possible patient or lesion-level leakage, limited independent external validation, scanner and vendor-related domain shift, and incomplete reporting of calibration, subgroup performance, and failure cases. The authors conclude that broader clinical translation requires robust external and prospective validation, improved calibration, assessment of generalizability and bias, and integration into clinical workflows, with future research prioritizing prospective multicenter trials, lesion-level evaluation, subgroup fairness analysis, uncertainty quantification, and calibration across diverse populations and imaging systems.

Subject of Research: Artificial intelligence in breast cancer imaging

Article Title: Artificial intelligence in breast cancer imaging: a systematic review of detection, segmentation, explainability, and clinical translation

Article References: Khan, I. I., Shah, S. A. H., Tsipourakis, A., Malavolta, M., Aleesha, A., Shah, S. B. H., Hajiani, M., Buccoliero, A., Panagiotopoulos, K., Shah, S. T. H., & Deriu, M. A. (2026). Artificial intelligence in breast cancer imaging: a systematic review of detection, segmentation, explainability, and clinical translation. Frontiers in Imaging, 5, Article 1910013. https://doi.org/10.3389/fimag.2026.1910013

Image Credits: AI Generated

DOI: 10.3389/fimag.2026.1910013

Keywords: artificial intelligence, breast cancer, mammography, deep learning, medical imaging, image segmentation, explainable AI, clinical translation, systematic review, radiomics, federated learning, vision transformers

The research, conducted across eight Providence Health and Services clinics in California, examined two critical questions in the rapidly expanding field of population-scale genetic testing: which outreach method best reconnects patients with testing opportunities they declined or deferred, and whether requiring patients to travel to a different building for sample collection suppresses participation. The answers carry significant implications for health systems attempting to scale genetic services beyond specialty clinics.

The scale of unmet need is substantial. Despite established national guidelines from the National Comprehensive Cancer Network, fewer than 20 percent of eligible US patients with a personal or family history of breast or ovarian cancer undergo genetic testing. That gap translates to an estimated 1.2 to 1.3 million women in the United States who qualify for testing but have not received it. Among women with a family history of breast or ovarian cancer but no prior diagnosis, only about 3.6 percent have completed testing. These shortfalls are compounded by well-documented disparities tied to race, age, education, and insurance coverage—disparities that the Centers for Disease Control and Prevention has called a public health imperative to address before precision medicine widens existing health inequities.

Providence’s population-scale cancer risk assessment program offered an ideal testing ground. The program uses the Ambry CARE platform to deliver electronic risk assessments aligned with NCCN guidelines to patients before their mammogram appointments. Patients flagged as high risk for hereditary cancer are informed of their eligibility for a multigene pan-cancer hereditary panel and offered genetic counseling. At the mammogram encounter itself, mammography technicians hand eligible patients informational materials and confirm their decision about testing. For the outreach experiment, 7,112 patients who had not formally agreed to testing at their visit were randomized to receive one of three follow-up contacts: a mailed brochure, an email, or an SMS text message, each reminding them of their high-risk status and inviting them to call and pursue testing.

The results were sobering. Fewer than 3 percent of patients who received any outreach went on to have a genetic test ordered, and the rate did not differ meaningfully by method. Brochures yielded a 3.2 percent order rate, while email and SMS each produced 2.7 percent—differences that vanished statistically. The one factor that did correlate with uptake was timing: patients who received outreach closer to their mammogram appointment were somewhat more likely to order a test, averaging 98 days between appointment and outreach compared with 107 days for those who did not order. Younger patients, those aged 45 and under, also responded at higher rates of roughly 4.4 to 4.5 percent compared with less than 2.5 percent for older groups.

The workflow analysis told a far more compelling story. Drawing on 16,965 high-risk patients seen between 2023 and 2024, the researchers compared clinics where genetic testing was offered in the same building—and sometimes the same room—as the mammogram against clinics that required patients to travel to a different building for sample collection. The overall order rate in this high-risk population was 11.7 percent, but the same-building workflow achieved 13 percent versus 9 percent for the different-building approach. After adjusting for age, language, race and ethnicity, and insurance payor using modified Poisson regression, patients in the different-building group had a 28 percent lower rate of test orders, with an incidence rate ratio of 0.72 and a 95 percent confidence interval of 0.65 to 0.80.

The stratified analysis revealed something more troubling than a simple convenience effect. In clinics requiring patients to go elsewhere for testing, disparities by patient characteristics widened dramatically. Not a single patient who completed their risk assessment in a language other than English had a test ordered at a different-building clinic—a 0 percent order rate among 35 patients—while the same-building clinics showed no language-based difference. Black patients at different-building clinics had an order rate of just 2.3 percent and Asian patients 6.5 percent, compared with rates within a percentage point or two of White and Hispanic patients at same-building clinics. Younger patients aged 18 to 39 also fared worse when sent elsewhere, ordering at 11 percent versus 15 percent for the 40-to-45 group, a reversal of the pattern seen at same-building sites.

These findings align with prior research showing that point-of-care testing models can double participation compared with referral-based approaches. Studies by Wang and colleagues and Brabender and colleagues documented genetic testing completion rates of 35 percent and 50.1 percent respectively in high-risk populations offered same-day testing—figures well above the 13 percent seen in Providence’s same-building group, though the researchers note that differences in patient demographics, provider capacity, and the fact that their different-building patients did not need to schedule a separate appointment may have attenuated the gap. The Providence order rates instead track more closely with the roughly 15 percent observed in other Southern California populations and national survey data.

The modest outreach results also contrast instructively with the Geno4ME population-scale whole genome sequencing program, which achieved 7.5 percent enrollment using similar outreach methods. The researchers suggest several explanations: Geno4ME was free of charge, whereas clinical genetic testing involves insurance billing and potential out-of-pocket costs; Geno4ME used multiple outreach attempts rather than a single contact; and qualitative research has shown that repetition helps patients reconsider testing they initially declined. Notably, the finding that no single outreach channel outperformed the others suggests health systems can simply choose whichever method is cheapest and most feasible—email being the obvious candidate—without sacrificing effectiveness.

The study’s limitations warrant consideration. The workflow variable served as a proxy for convenience but could not capture unmeasured differences in how clinics implemented the program, such as staff capacity to discuss testing during appointments. With only eight clinics, facility-level variation could not be fully modeled, and the analysis relied on electronic health records that might have missed testing done at other health systems, potentially inflating the denominator. Personal cancer history, which could strongly influence a patient’s decision, was not visible in the data. Still, the core message is difficult to escape: outreach alone, however delivered, moves only a small fraction of eligible patients, while embedding testing directly into the care patients are already receiving—same building, same day—substantially raises participation and, critically, does so more equitably across language and racial groups. As health systems invest in scaling genetic services beyond specialty clinics, the evidence suggests that dollars spent on provider capacity and point-of-care access will yield far greater returns than any brochure, email, or text message ever could.

Subject of Research: Effectiveness of outreach methods and testing workflows on genetic testing participation among high-risk mammography patients

Article Title: Evaluating the effectiveness of outreach efforts and testing workflow on genetic testing participation in a high-risk population

Article References: Dickey, L., Bryce, E., Rinaldi, J. B., Emery, K., Childers, K., Brown, S., Gordon, O., & Vartanian, K. (2026). Evaluating the effectiveness of outreach efforts and testing workflow on genetic testing participation in a high-risk population. Frontiers in Cancer Control and Society, 4, Article 1765262. https://doi.org/10.3389/fcacs.2026.1765262

Image Credits: AI Generated

DOI: 10.3389/fcacs.2026.1765262

Keywords: genetic testing, hereditary cancer, outreach methods, point-of-care testing, health equity, mammography, precision medicine, NCCN guidelines, health disparities, patient participation, clinical workflow, cancer prevention

The study, published in BMC Medical Imaging, analyzed 173 invasive breast carcinomas treated at a single Turkish center between 2018 and 2023. Researchers Şahinde Atlanoğlu and Figen Aslan from Kütahya Health Sciences University systematically re-examined mammographic and ultrasonographic images using a standardized reporting lexicon, looking for patterns that might predict which molecular subtype a tumor belonged to before any tissue sample was taken.

What they discovered was unexpected. Two ultrasound features — a circumscribed, smooth-appearing margin and increased blood flow detected on Doppler imaging — were independently associated with the triple-negative phenotype, which lacks estrogen receptors, progesterone receptors, and excess HER2 protein. This subtype is typically the hardest to treat and carries a poorer prognosis than hormone-responsive forms of the disease.

The counterintuitive association between a benign-looking margin and an aggressive cancer challenges conventional radiological wisdom. Most teaching emphasizes that cancers with irregular, spiculated borders are more likely to be high-grade and biologically aggressive. Yet the triple-negative tumors in this cohort frequently presented with well-circumscribed edges, mimicking the appearance of fibroadenomas or other benign lesions.

To ensure their findings were not simply artifacts of subjective interpretation, the researchers had a second blinded radiologist independently re-read 50 of the lesions. Inter-reader agreement was nearly perfect for circumscribed margins and Doppler vascularity, with kappa statistics of 0.90 and 0.96 respectively. This reproducibility is critical, because previous studies linking imaging features to molecular subtypes have often been criticized for inconsistent descriptor use across different observers and institutions.

Not every imaging feature held up to scrutiny. The echo pattern of the lesion — how its internal texture appears relative to surrounding tissue — showed only slight agreement between readers, with a kappa of just 0.19. Isoechogenicity, meaning the lesion appeared similar in brightness to the surrounding breast tissue, was initially associated with triple-negative status but lost its statistical significance once the researchers controlled for margin characteristics in their multivariable model.

The team used Firth penalized logistic regression, a statistical method particularly suited to small sample sizes and rare outcomes, restricting their analysis to variables available before biopsy. After adjusting for confounders, circumscribed margin carried an odds ratio of 3.57 and Doppler vascularity an odds ratio of 3.07 for predicting the triple-negative phenotype. The optimism-corrected area under the curve was 0.71, indicating moderate discriminative ability.

Beyond subtype prediction, the study also examined axillary ultrasound performance for detecting lymph node metastases. The technique proved highly specific at 97.3 percent but only moderately sensitive at 75.8 percent, meaning a normal-appearing axilla on ultrasound does not reliably rule out nodal spread. This has implications for surgical planning, where axillary ultrasound results often influence whether a patient undergoes sentinel node biopsy or more extensive dissection.

Mammographic features, by contrast, showed weaker and less consistent associations. Margin characteristics on mammography differed between subtypes before statistical adjustment but lost significance afterward. Only 52 of the 173 lesions, roughly 30 percent, had mammographic images available for analysis, limiting the statistical power of that component of the study.

The authors are careful to frame their findings as exploratory. The single-center retrospective design, the modest sample size, and the exclusion of patients who received neoadjuvant therapy all limit generalizability. Larger, prospective, multi-center studies with independent image reading will be needed before these imaging signatures can be incorporated into clinical decision-making. Still, the idea that a smooth-bordered, well-vascularized mass on ultrasound might warrant heightened suspicion for triple-negative biology — rather than reassurance — represents a provocative shift in how radiologists might approach the everyday work of breast cancer detection.

The biological explanation for why triple-negative tumors can appear deceptively well-circumscribed on ultrasound remains an active area of investigation. Triple-negative carcinomas are defined by the absence of estrogen receptors, progesterone receptors, and HER2 overexpression, and they disproportionately affect younger women and carriers of BRCA1 mutations. Histologically, many of these tumors grow with a pushing rather than infiltrative border, sometimes accompanied by a prominent lymphocyte infiltrate at the tumor edge. That pushing growth pattern can translate into a smooth, rounded contour on sonography even while the tumor itself behaves aggressively, offering a plausible bridge between the imaging observation and the underlying pathology. The association with Doppler vascularity fits this picture as well, since rapidly proliferating tumors typically recruit new blood vessels, a process that can manifest as increased internal blood flow on color Doppler examination.

The methodological choices in this study deserve attention because they address common weaknesses in the imaging-subtype literature. By re-reading every examination under the fifth edition of the BI-RADS lexicon with a radiologist blinded to pathology results, the authors reduced the risk of circular reasoning, in which readers unconsciously assign descriptors they believe correlate with aggressive biology. The use of Benjamini-Hochberg adjustment across multiple comparisons guards against the false positives that arise when dozens of imaging features are tested simultaneously. Firth penalized logistic regression, meanwhile, is designed for situations where outcome events are sparse and conventional maximum-likelihood estimates can be biased or fail to converge, which is a realistic concern when only about a third of a modest cohort carries the phenotype of interest. Bootstrap internal validation, which produced the optimism-corrected area under the curve of 0.71, provides a more honest estimate of model performance than apparent statistics calculated on the same data used to fit the model.

Even so, an area under the curve of 0.71 falls well short of the discrimination needed for imaging features alone to drive clinical decisions. In practical terms, a circumscribed margin or increased vascularity shifts the probability of triple-negative biology upward but leaves substantial uncertainty. The clinical value of such associations, if confirmed, would more likely lie in combination with other pre-biopsy information, such as patient age, family history, and lesion location, rather than as standalone predictors. It is also worth noting that the analysis set consisted of 173 invasive carcinomas in women, with the lesion rather than the patient as the unit of analysis, and that surrogate subtyping by immunohistochemistry is itself an approximation of the gene-expression-based classifications used in contemporary oncology.

The axillary ultrasound findings carry independent practical weight. A specificity of 97.3 percent means that when ultrasound shows a clearly abnormal node, the likelihood of true metastatic involvement is high, which can prompt needle biopsy and alter surgical planning before the operating room. The sensitivity of 75.8 percent, however, reinforces long-standing guidance that a normal-appearing axilla cannot substitute for nodal staging, since roughly one in four metastatic nodal basins will escape sonographic detection. This asymmetry between ruling in and ruling out disease mirrors the performance of axillary ultrasound reported across broader literature and underscores why sentinel node biopsy remains the standard for surgical staging in most patients.

The mammographic component of the study illustrates how data availability shapes conclusions. With mammograms available for only 52 of the 173 lesions, roughly 30 percent of the analysis set, the power to detect subtype differences was inherently limited, and the loss of significance for mammographic margin after adjustment may reflect either a genuinely weaker association or simply insufficient numbers. In many screening and diagnostic settings mammography is routine, but in this retrospective cohort its absence for most patients likely reflects referral patterns, patient age, or institutional practice, a reminder that single-center imaging studies inherit the biases of the populations they serve.

The reproducibility findings may ultimately prove as consequential as the subtype associations themselves. A kappa of 0.19 for echo pattern, indicating only slight agreement between two blinded readers at the same institution, raises questions about how reliably this descriptor can be used in any predictive model, whether in this study or elsewhere. By contrast, the near-perfect agreement for circumscribed margin and Doppler vascularity suggests these are features radiologists can identify consistently, a prerequisite for any descriptor hoping to survive translation from retrospective research into prospective clinical use. The authors’ decision to report both the promising and the fragile findings in equal detail reflects a growing emphasis in the imaging literature on transparent reporting of measurement reliability alongside associations.

For clinicians interpreting these results, the most defensible takeaway is a heightened index of suspicion rather than a change in biopsy thresholds. A smooth-bordered, hypervascular mass in an appropriate clinical context may merit the same urgency as an obviously malignant-appearing lesion, particularly in younger patients or those with risk factors for triple-negative disease. But the authors themselves emphasize that these associations emerged from an exploratory, single-center cohort and require confirmation in larger, prospective, independently read studies before they can inform guidelines or everyday practice. Until such validation arrives, the study stands as a careful, methodologically disciplined challenge to the assumption that benign-appearing imaging features reliably signal benign biology.

Subject of Research: Associations between breast imaging features and immunohistochemistry-defined molecular subtypes of breast cancer

Article Title: Associations of mammographic and ultrasonographic features with immunohistochemistry-defined surrogate subtypes of breast cancer: a single-center retrospective study

Article References: Atlanoğlu, Ş., & Aslan, F. (2026). Associations of mammographic and ultrasonographic features with immunohistochemistry-defined surrogate subtypes of breast cancer: a single-center retrospective study. BMC Medical Imaging. https://doi.org/10.1186/s12880-026-02953-5

Image Credits: AI Generated

DOI: 10.1186/s12880-026-02953-5

Keywords: breast cancer, ultrasonography, mammography, triple-negative, BI-RADS, immunohistochemistry, surrogate subtype, Doppler vascularity, radiology, cancer imaging, axillary ultrasound, tumor subtypes

The research team measured breast cancer screening compliance during calendar year 2025 using tools embedded in the EPIC electronic health record system. Patients were counted as compliant if they were women aged 40 to 74 who had received a mammogram within the previous two years and had no disqualifying characteristics, such as a prior bilateral mastectomy. The investigators then abstracted detailed patient data from the medical record and collected information about each physician, including gender, language facility, and years in practice. Using both bivariate comparisons and multivariable logistic regression, they teased apart how much of the screening gap could be explained by patient characteristics, clinical and utilization patterns, and physician attributes.

The headline numbers are striking. Of the 8,825 patients eligible for breast cancer screening at the practice, 1,024 — about 12 percent — were cared for by residents. Faculty physicians achieved a screening rate of 83 percent among their eligible patients, while residents reached only 66 percent, a difference that was highly statistically significant. In the unadjusted analysis, a resident’s patient had just 0.41 times the odds of being up to date on screening compared with a faculty patient — a 95 percent confidence interval of 0.35 to 0.47, meaning resident-panel patients were roughly 60 percent less likely to have a current mammogram.

But the raw gap tells only part of the story, and this is where the study becomes technically interesting. The resident and faculty panels were not comparable populations to begin with. Faculty panels included a higher percentage of patients who were white, spoke English, and carried commercial insurance, along with more medical comorbidities. The resident panel, by contrast, contained more patients with serious mental illness and had fewer primary care encounters overall. These differences matter enormously, because decades of research have shown that screening mammography rates vary systematically with race, ethnicity, socioeconomic status, language, and engagement with the health care system.

To account for this, the researchers built a series of progressively more rigorous statistical models. In the first model, adjusting only for patient sociodemographic characteristics, the odds ratio for a resident patient being screened improved from 0.41 to 0.50. When clinical and utilization variables were added in the second model, the odds ratio rose further to 0.59. And in the third model, which incorporated physician attributes such as gender, language facility, and years in practice, the odds ratio climbed to 0.73 — with a confidence interval of 0.54 to 0.98 that only barely excludes the null. In other words, most of the apparent screening deficit attributable to being a resident’s patient dissolved once the underlying differences between patient populations were taken into account, though a meaningful residual gap remained.

This pattern of attenuation carries an important message about how quality disparities in teaching practices should be interpreted. Earlier studies comparing resident and faculty care have often reported that trainees achieve poorer cancer screening rates, and some have framed this as a conflict between training and equity. A 2019 study in the same journal documented disparities in quality of primary care by resident and staff physicians, and a multi-center study published in 2025 confirmed differences in ambulatory care outcomes between trainees and faculty across institutions. But those studies, like this one, must grapple with the fact that resident panels are not random samples of the practice population. Residents often inherit panels of patients with the greatest social and medical complexity — patients who are harder to reach, less likely to attend appointments, and more likely to face logistical barriers to preventive care.

The setting of the new study makes its findings particularly notable. The authors deliberately chose to examine this question in an academic primary care practice with robust practice support, reasoning that if residents’ disadvantages in experience and continuity could be overcome anywhere, it would be in a high-functioning environment. Teaching practices like this one typically offer embedded care managers, screening outreach programs, and structured panel management — infrastructure that should, in theory, buffer less experienced clinicians from their own limitations. The fact that a gap persisted even under these favorable conditions suggests that the problem is not simply one of inadequate support, but may be woven into the structure of residency training itself.

One structural feature looms large: the annual turnover inherent in residency. Residents graduate and hand their panels to incoming trainees every year, creating what researchers have called the revolving door of resident continuity practice. Prior work has identified gaps in transitions of care during these year-end handoffs, and a 2009 commentary in JAMA described the academic year-end transfer of outpatients as an unaddressed patient safety issue. Patients who lose their trusted physician at a moment of transition may be less likely to follow through on screening recommendations, particularly if the handoff is poorly executed. The lower number of primary care encounters observed in the resident panel in this study may reflect both patient disengagement and the discontinuity that comes with annual rotations.

The study’s conclusion is measured but pointed: while rigorous statistical adjustment substantially improved the apparent performance of resident physicians, residency training programs may need to be redesigned to eliminate the remaining disparities in care. Possible redesigns discussed in the broader literature include longer continuous blocks in continuity clinic, structured year-end sign-out programs, and X-plus-Y scheduling models that cluster ambulatory time to protect continuity. Quality improvement efforts targeting specific screening gaps — such as a 2025 initiative addressing colorectal cancer screening disparities between faculty and residents — offer a template for how practices might close the mammography gap as well.

For patients, the practical takeaway is not to avoid resident physicians, whose care is supervised by attending faculty and whose patients often have the greatest need. Rather, the findings highlight where the health care system should direct its attention: toward the patients of trainees, who are disproportionately likely to face barriers to preventive care, and toward the design of the training programs that will produce the next generation of physicians. Breast cancer screening saves lives, and the US Preventive Services Task Force recommends biennial screening for women aged 40 to 74. Ensuring that every eligible woman — regardless of whether her physician is a first-year resident or a twenty-year veteran — receives that recommendation in time is a test of whether academic medicine can align its training mission with its equity obligations.

Subject of Research: Disparities in breast cancer screening rates between resident and faculty physicians in an academic primary care practice

Article Title: Understanding Low Breast Cancer Screening Rates by Resident Physicians

Article References: Weingart, S. N., Kufer, K. S., Short, M. I., Gillis, J., Sathi, K., Chawla, P., Schelling, K., Huang, K. T. L., & Tishler, J. (2026). Understanding Low Breast Cancer Screening Rates by Resident Physicians. Journal of General Internal Medicine. https://doi.org/10.1007/s11606-026-10899-1

Image Credits: AI Generated

DOI: 10.1007/s11606-026-10899-1

Keywords: breast cancer screening, mammography, resident physicians, primary care, health disparities, graduate medical education, residency training, quality of care, continuity of care, academic medicine, preventive care, logistic regression