Breast Imaging Reporting and Data System
BI-RADS is the standardized vocabulary and reporting framework breast radiologists use to describe findings and assign a final assessment with a recommended next step. Maintained by the American College of Radiology, it spans the breast across mammography (including tomosynthesis), ultrasound, MRI, and contrast-enhanced mammography, so a report means the same thing to anyone who reads it. Its hallmark is a numbered assessment scale that ties each imaging conclusion to a management recommendation. It is the dominant breast-imaging lexicon worldwide and, in the United States, is woven into federal mammography reporting requirements.
The first major overhaul in over a decade and the largest expansion yet (the illustrated guide was rebranded from 'Atlas' to 'Manual'). Breast-density reporting became a mandatory standalone section across all modalities, tomosynthesis was formally folded into how masses are characterized, and contrast-enhanced mammography was elevated to a full core modality. The lexicons were modernized — ultrasound gained a 'non-mass' concept, several ambiguous MRI terms were retired and new descriptors added (with recognition of abbreviated MRI protocols), and categories 0 and 6 were clarified rather than renumbered. It matters because reporting now better matches how breast imaging is practiced today.
The reference standard until v2025. It delivered a comprehensive update across mammography, ultrasound, and MRI, sharpened breast-density language toward a composition/masking emphasis, and strengthened management and outcome-monitoring guidance.
A major expansion that added dedicated lexicons for breast ultrasound and breast MRI for the first time, moving BI-RADS beyond mammography, and introduced the 4A/4B/4C subdivision to better stratify biopsy risk.
Broadened the illustrated atlas with feature illustrations and added practice-auditing guidance and sample reports to support quality assurance.
A refinement update that improved descriptor clarity and consistency over the original.
The original, mammography-only release that established standardized terminology, a structured report format, and the numbered assessment-category concept to cut interpretive variability.
At its core, BI-RADS sorts every study into a small set of numbered final assessment categories, conventionally 0 through 6, where each number bundles a level of suspicion with the action it implies — from 'more imaging needed' through 'benign' and 'probably benign' to escalating degrees of cancer concern and, finally, already-proven cancer. The scale is ordered so a higher number broadly signals greater concern or a more definitive next step, and the 'suspicious' tier is further split into low/intermediate/high sub-levels. Separately, the system standardizes how findings are described (shape, margin, density, calcifications, enhancement, breast density). Consult the official ACR BI-RADS v2025 Manual for the exact criteria, definitions, and management language.
These are our plain-language summaries. For the exact criteria, thresholds, and management rules, see the official source.
In 1847 breast MRI exams, a DINOv2-based medical slice transformer triaged cases without suspicious findings (BI-RADS ≥4) at 97.5% sensitivity. Specificity was 19% for contrast-enhanced (T1sub+T2w) and 17% for non-contrast-enhanced (DWI1500+T2w) MRI. External validation AUC 0.77.
Thirteen Takeaways on BI-RADS v2025 Changes for Mammography, Ultrasound and Breast MRI Reporting
Thirteen Takeaways on BI-RADS v2025 Changes for Mammography, Ultrasound and Breast MRI Reporting
<![CDATA[Thirteen Takeaways on BI-RADS v2025 Changes for Mammography, Ultrasound and Breast MRI Reporting]]>
BI-RADS v2025 brings mammographic density alignment with Mammography Quality Standards Act, nonmass lesions on ultrasound, revised MRI enhancing lesion terminology, and first dedicated contrast-enhanced mammography section, harmonizing lexicon across modalities.
LLM for breast ultrasound report quality control: 80% accuracy for margins (vs 64% manual), 74% for echo patterns (56%), 50 reports in 13 min (vs 213 min), suggesting time-saving automation.
For BI-RADS 4 breast lesions, Kaiser score AUC 0.901 (95% CI 0.856-0.941) outperformed ADC alone (AUC 0.733) and combination (0.890). ADC had high false-positive rate (66.1%) in benign cellular lesions. DCIS and mucinous carcinoma were often false negatives.
Objective and subjective evaluation of a mammogram-reporting AI found BI-RADS agreement 58.1%, mass 76.7%, calcification 81.4%, ROUGE-L F1 0.672, but some cases with high objective scores were rated low subjectively, flagged as over/underestimation.
An AI-assisted decision-support model using patient-level variables reduced categorization disagreement by 26% for breast density and background parenchymal enhancement on contrast-enhanced mammography, achieving AUC 0.75, precision 0.72, and recall 0.69, with greatest benefit i…
Combining cfDNA tumor mutation burden and gene mutation score with age and BI-RADS classification yielded an AUC of 0.923, 88.1% sensitivity, 89.1% specificity for predicting malignancy in 212 patients with BI-RADS 4 breast nodules.
In a 5-year review of BI-RADS 4C and 5 lesions, radiologist-determined radiology-pathology (rad-path) discordance carried a 47.3% upgrade to malignancy versus 9.4% for concordant benign biopsies, supporting its role in guiding surgical excision.
A 2.5D ABUS deep learning model with BI-RADS shape and margin auxiliary tasks achieved an external test AUC of 0.91 (95% CI 0.83-0.96) for breast lesion classification, outperforming 3D Swin Transformer (0.76) and other benchmarks.
A new AJR article guides radiologists on adopting the BI-RADS 2025 lexicon in clinical breast imaging reports, covering updated terminology and structured reporting recommendations.
ACR BI-RADS v2025 adds new descriptors like "lobulated" and "non-mass lesion," renames MRI kinetics to "early phase," and separates assessment from management. It formally integrates contrast-enhanced mammography and supports surgical excision as definitive treatment under a rev…
Older age and larger tumor size independently predicted malignancy in 104 women with BI-RADS 4/5 breast lesions (aOR 1.13/yr, 1.10/mm). BI-RADS 5 was not significant after adjustment, but the small subgroup (n=11) limited power.
A nomogram combining DBT-based intralesional and perilesional radiomics with patient age and BI-RADS 4 subclassification achieved an AUC of 0.942 in classifying benign vs malignant breast masses in the validation set, with SHAP analysis showing feature contributions.
AI-based biopsy sparing for BI-RADS 4A+ breast lesions can be translated into clinically actionable decision-making, suggests a new study in J Ultrasound Med.
Combined BreastViT+GPT-4o achieved top radiologist-rated scores for structural integrity (8.03), terminology (8.02), accuracy (7.31) and completeness (7.61) on breast US, significantly outperforming standalone models (P<0.05) in external validation (49 cases).
BI-RADS v2025 now recommends routine audits of category 3 assessments (malignancy likelihood ≤2%), a new auditing framework for staging breast MRI, and prospective recording of initial cancer detection method to link outcomes to screening.
MRI-based breast texture analysis may assist in diagnosing BI-RADS category 4 lesions, potentially reducing unnecessary biopsies. (Performance metrics and study details are not reported in the source snippet.)
A new AuntMinnie podcast episode examines the potential for artificial intelligence to transform BI-RADS classification in breast imaging. The discussion, hosted by women's imaging experts, explores whether AI might become a standard component of future reporting.
AJR. American journal of roentgenology
In 112 mammograms read by 10 residents, AI changed the BI-RADS category in 9.7% of assessments and altered diagnostic confidence in 19.2%, with upgrades outnumbering downgrades. Sensitivity for BI-RADS 4+ rose from 72.0% to 82.3% against an expert-consensus standard, though this…
A logistic regression model combining mammographic descriptors, age, and DCE-MRI enhancement classified BI-RADS 4/5 microcalcifications with 90.9% accuracy (F1 0.889) in preliminary internal testing of 53 biopsy-proven cases. Findings require multicenter validation before practi…
Dense breast categorization rose from 38.4% to 39.1% (p=0.01) after Breast Imaging Reporting and Data System (BI-RADS) 5th edition, with negligible practical effect size (Phi=0.01), in a single-institution study of 132,326 mammograms.
A practical review synthesizes how preoperative breast MRI—detecting additional ipsilateral disease in ~20% of newly diagnosed breast cancers—should guide surgical planning under BI-RADS v2025, including new 'Additional Close Findings' category and key oncoplastic thresholds.
ACR BI-RADS v2025 — the first year-versioned edition — drops 'focus' and 'developing asymmetry,' adds ultrasound 'non-mass,' standardizes Category 4 subdivisions across modalities, and reframes mammographic density around masking risk rather than volume.
A retrospective study (n=390) found a wavelet-spectrum multi-channel deep learning model could spare 26% of biopsies in BI-RADS 4a+ breast lesions while keeping missed malignancy at 1.8% — a ~2% accuracy gain over B-mode alone. Not yet externally validated.
In 216 BI-RADS 4 breast lesions, a logistic regression model with 6 imaging/elastography features best predicted malignancy — random forest appeared superior (AUC 1.00) but showed clear overfitting. LR model AUC in validation not reported in source. Single-center, n=212.
BI-RADS v2025 breast ultrasound updates refine glandular tissue descriptors, introduce structured non-mass lesion characterization, and standardize lymph node reporting, with emphasis on dense-breast screening in Asian populations.
BI-RADS is effectively the universal lexicon for breast imaging, with little real competition — it is the template that later 'RADS' systems (PI-RADS, LI-RADS, and others) were modeled on. In the United States its use is reinforced by federal mammography regulation, which cements its dominance over ad hoc reporting. Where alternatives exist they tend to be complementary rather than rival: automated/volumetric breast-density software, for example, maps its output back onto BI-RADS density classes rather than replacing the framework.
BI-RADS underpins large national audit datasets, and its predictive value rises steeply across the suspicious sub-tiers — reported positive predictive values climb from roughly a third for 4A to the high-80s–90s percent for 4B/4C and category 5 — which validates the stratification while exposing that sub-categorizing category 4 is subjective and reader-dependent. Inter-reader reproducibility is generally substantial but imperfect, with breast-density assignment in particular showing meaningful variability between editions and experience levels. Historical analyses credit BI-RADS with materially reducing reporting ambiguity and enabling outcomes monitoring since the 1990s. Recurring criticisms — interpretive variability in the probably-benign and suspicious zones, and density-assessment subjectivity — are explicitly among the issues the 2025 6th edition set out to address.
RadPigeon is an independent radiology news digest and is not affiliated with or endorsed by American College of Radiology (ACR). “BI-RADS” is a trademark of its owner and is named here only to refer to the system. Always consult the official source for the exact, current criteria.
