Elizabeth S. McDonald

Rationale: Routine clinical imaging captures anatomy under a limited set of physiologic and acquisition conditions, leaving segmentation models with sparse examples of how lesions and surrounding tissues deform with compression, respiration, or vascular pulsation. Standard image augmentation expands training datasets but does not explicitly model the mechanics underlying these changes. Patient-specific biomechanical simulation offers an approach to generate anatomically consistent images and corresponding segmentation masks under conditions absent from the original scan, without additional imaging or training a generative model. We hypothesized that incorporating this mechanistically grounded variability into training improves segmentation across imaging modalities. Methods: We evaluate across three segmentation tasks: breast cancer on 3D digital breast tomosynthesis, lung tumors on chest CT, and carotid plaque on ultrasound. From each routine scan, patient-specific biomechanical models generate acquisitions absent from the original study—a range of breast compressions, the respiratory cycle, and arterial wall distension across the cardiac cycle. The resulting displacement fields deform both images and masks. Segmentation models were trained with and without the generated data across a range of standard image augmentation techniques, evaluated via cross-validation and held-out test sets. Realism was assessed through blinded reader studies, and generated respiratory motion was validated against real 4D CT. Results: Training on the generated acquisitions improved segmentation across modalities, increasing Dice scores by 6.6% for breast (p<0.05), 10.1% for lung (p0.05) relative to the routine scans alone. Generated respiratory motion closely matched real gated 4D CT, with Dice of 0.94 ± 0.02 for lung volume overlap. Blinded readers discriminated generated from real images at near chance (AUC: breast 0.61, lung 0.41, carotid 0.46). Conclusions: Mechanistic simulation generates imaging acquisitions that clinical systems do not supply, without training data or additional scanning, and improves AI-based disease detection across modalities.

Background: A subset of biopsy‐confirmed ductal carcinoma in situ (DCIS) cases can be upstaged to invasive breast cancer at surgery. Preoperative identification of upstaging risk is essential for treatment planning. Purpose: To develop clinical and breast MRI radiomics models to preoperatively predict upstaging in DCIS. Study Type: Retrospective and prospective. Population: Three hundred forty‐five women (median age, 59 years [IQR, 49–67]) with 362 DCIS lesions diagnosed via core‐needle biopsy (training: n = 216, internal testing: n = 53, external testing: n = 93). Field Strength/Sequences: 1.5 T 3D T1‐weighted spoiled gradient‐echo MRI: pre‐contrast non‐fat‐suppressed and dynamic contrast‐enhanced (DCE) fat‐suppressed sequences. Assessments: Clinical and radiomics features were extracted from radiologist‐segmented lesions on DCE breast MRI. Seven machine learning algorithms were evaluated across clinical‐only (demographic and clinicopathologic variables), radiomics‐only (whole‐lesion/peritumoral), and combined models, using nested cross‐validation and internal and external testing. Feature importance was assessed using Shapley additive explanation (SHAP). Statistical Tests: Mann–Whitney, Chi‐square, DeLong tests; area under the receiver operating characteristic curve (AUC), sensitivity, negative predictive value (NPV). Results: The clinical model achieved AUC 0.70 (95% CI: 0.51–0.89), with 36% sensitivity and 84% NPV. A multilayer perceptron integrating clinical and peritumoral radiomics non‐significantly increased AUC to 0.75 (95% CI: 0.58–0.92; ΔAUC = +0.05; False Discovery Rate p‐ adj = 0.65), improving sensitivity to 82% (Δ = +46%) and NPV to 92% (Δ = +8%). On external testing, AUC was 0.63 (95% CI: 0.50–0.76) with no significant change over the clinical baseline (ΔAUC = −0.03; FDR p ‐adj = 0.857) but demonstrated superior sensitivity (68%; Δ = +30%) and NPV (84%; Δ = +3%). SHAP identified peritumoral center‐of‐mass shift as the dominant predictor. Data Conclusion: Clinical‐peritumoral modeling did not significantly improve discrimination over the clinical baseline but showed higher NPV and sensitivity, suggesting the tumor microenvironment may provide complementary information for upstaging risk stratification. Level of Evidence: 3. Technical Efficacy: Stage 2.

Objective: Allostatic load (AL) is an emerging framework quantifying physiological dysregulation from cumulative exposure to stressors over an individual’s life. We evaluated the relationship between AL and malignant breast biopsy results in women across sociodemographic subgroups. Methods: Women 18 years and older who underwent at least one image-guided breast biopsy at a safety-net hospital between April 2016 and December 2023 were identified. Consistent with prior studies, AL was defined as a score based on eight values: resting pulse rate, systolic blood pressure, diastolic blood pressure, body-mass index, serum albumin, creatinine, estimated glomerular filtration rate, and white blood cell count. Multivariable logistic regression models were used to assess the association between AL and malignant breast biopsy results among different sociodemographic groups. Effect modification by racial/ethnic category was then examined using stratified models. Results: 668 participants had data available for all eight biomarkers. Most participants were Hispanic (n = 376, 54.5%), reported English as their primary language (n = 423, 63.3%), and had Medicaid for insurance (n = 300, 44.9%). In stratified analyses, the association between AL and malignant breast biopsy result was significant in White, non-Hispanic patients. In White, non-Hispanic patients, after adjusting for age, every one-unit increase in AL was associated with a 46% increased likelihood of malignant breast biopsy result (OR 1.46, P = 0.03). Conclusion: AL was associated with malignant breast biopsy results in specific sociodemographic subgroups, potentially representing a modifiable risk factor for breast cancer prevention. Further research with large, diverse populations is required to evaluate additional factors impacting breast cancer outcomes across sociodemographic groups.

Purpose: Excess weight is a key modifiable risk factor for breast cancer. Glucagon-like peptide-1 receptor agonists (GLP-1) promote weight loss and improve metabolic health, but their effect on breast cancer risk remains unclear. Methods: We conducted a retrospective cohort study from January 1, 2022, to June 30, 2025, using electronic health records. We identified 217,025 unique patients who underwent breast imaging, restricting to women aged 45 to 80 years with a BMI ≥ 25 and a documented imaging outcome (n = 111,646; median age 61). The primary outcome was breast cancer detection. GLP-1 use was defined as a first prescription prior to the exam date and assessed in relation to race, ethnicity, age, and type 2 diabetes. To address potential confounding between these covariates and GLP-1 exposure, we performed one-to-one, case-control matching using propensity scores based on age, race, ethnicity, highest BMI, breast density, and history of type 2 diabetes. The propensity matching to the GLP-1 use group was performed using the greedy nearest neighbor approach. Results: GLP-1 exposure was associated with a lower incidence of breast cancer (OR 0.649, 95% CI 0.569-0.741; p < 0.0001). In the matched logistic regression (30,528 observations; 600 cancer cases), GLP-1 exposure was associated with a lower breast cancer incidence (OR 0.695, 95% CI 0.590–0.819; p < 0.0001). Conclusion: In this large observational study of women undergoing breast imaging at a major academic center and affiliated sites, GLP-1 treatment was associated with a lower incidence of breast cancer, independent of age, race, ethnicity, BMI, breast density, and diabetes. Findings support the need for prospective trials investigating GLP-1 agonists for breast cancer prevention.

Randomized controlled trials (RCTs) have confirmed the mortality benefits of screening mammography and are the gold standard for evaluating new diagnostic tests and medical interventions. Reliable and rigorous execution of RCTs can be complex and require high levels of funding and prolonged time from technology implementation to trial completion. Breast radiology and artificial intelligence technologies are being developed and are receiving regulatory approval faster than RCTs may be effectively conducted. Regulatory approval can lead to societal recommendations, legislation, clinical availability, purchasing, and patient care, even in the absence of reliable evidence indicating increased efficacy. Marketing and patient demand and physician perceptions influence technology use. There is a wide chasm between using data provided by RCTs and site-specific care decisions. Can we find a middle ground? What types of data can influence guidelines in the absence of rigorous clinical trials? This article explores the benefits and drawbacks of RCTs and describes professional initiatives, such as widespread use of method of detection, that could allow the breast imaging community to more quickly evaluate and integrate new technologies in a reasoned and evidence-based fashion.

This shows the number of samples with available data for analysis.

This shows the association of TK1 with early breast cancer relapse segregated by molecular subtype.

These are the publicly available microarray datasets combined for this analysis.

This shows the trimeric complex correlation with PAM50 ROR.