Performance of clinical breast cancer risk prediction models versus a mammography-based artificial intelligence risk model

JNCI Journal of the National Cancer Institute · 2026-03-24

BACKGROUND: Artificial intelligence (AI) -based, mammography breast cancer (bc) risk prediction models show improved discriminatory accuracy relative to clinical risk models. However, data on their calibration are limited. This study compared model performance of three clinical bc risk models-Gail, Tyrer-Cuzick (TC) v8, and Breast Cancer Surveillance Consortium (BCSC) v3 - to the MIRAI AI-risk model. METHODS: Digital mammograms were ascertained from a screening mammography cohort of 12,308 women within the Mayo Clinic Biobank with 250 incident BCs (176 invasive) within five years. We predicted five-year bc risk, estimated discriminatory accuracy (concordance [C]-index) and calibration (observed to expected ratio [O/E]) of both overall and invasive bc, and compared estimates using bootstrapping approaches. RESULTS: MIRAI demonstrated similar or improved discriminatory accuracy of overall bc (C-index = 0.71, 95% confidence interval [CI]=0.68-0.74) and invasive bc (C-index = 0.71, 95%CI = 0.67-0.75) compared to clinical models (Overall bc: C-index = 0.59-0.68, Invasive bc: C-index = 0.60-0.68). MIRAI's calibration for risk of overall bc (O/E = 0.96, 95%CI = 0.85-1.08) was improved compared to Gail (O/E = 1.22, 95%CI = 1.07-1.38) and BCSC (O/E = 1.38, 95%CI = 1.22-1.56) but similar to TC with volumetric percent density and polygenic risk score (O/E = 0.99, 95%CI = 0.87-1.13). However, for low-risk women (approximately 50%), MIRAI overestimated risk of overall bc. MIRAI also overestimated risk of invasive bc across the risk spectrum (O/E = 0.68, 95%CI = 0.58-0.78), while clinical models had good calibration (O/E = 0.86-0.99). CONCLUSION: MIRAI demonstrated stronger discriminatory accuracy than clinical models for five-year overall and invasive bc risk prediction but overestimated risk for both bc endpoints. AI-based risk models should consider discriminatory accuracy and calibration for invasive cancer before implementation.

Transformer-based Fusion of Longitudinal Multimodal Radiomic Features from Chest Radiography and CT in COVID-19

Radiology Artificial Intelligence · 2026-03-18

A transformer-based framework integrating longitudinal multimodal medical data from chest radiographs and CT images achieved robust performance in clinical outcome prediction in patients with COVID-19.

Abstract 2319: Variation in urinary estrogen levels according to breast parenchymal texture in postmenopausal women

Cancer Research · 2026-04-03

Abstract Background: Estrogen levels and breast parenchymal texture are both risk factors for breast cancer, but the relationship between the two is incompletely understood. This study evaluated associations between urinary estrogens, their metabolites, and parenchymal texture features in postmenopausal women to clarify how hormonal pathways contribute to radiomic breast tissue characteristics. Methods: Urinary concentrations of estradiol, estrone, and 13 estrogen metabolites were quantified using liquid chromatography/tandem mass spectrometry (LC-MS/MS) and standardized to urinary creatinine levels (pmol/mg creatinine) among 294 postmenopausal women undergoing screening mammography at the University of North Carolina between 2020 and 2022. Women who were using menopausal hormones, oral contraceptives, or chemoprevention, or who had breast implants were excluded. An automated radiomic pipeline was used to quantify 344 parenchymal texture features from bilateral mammograms. Features were harmonized using ComBat to reduce batch effects. Principal components analysis and unsupervised clustering of features were used to define texture groups. Multinomial regression was used to assess associations between texture groups and estrogen levels, with and without adjustment for age and body mass index (BMI). Results: Participants had a median age of 64 years (IQR: 59-71), a median BMI of 28 kg/m2 (IQR: 24-33), and a median total estrogen level of 20.9 pmol/mg creatinine. Participants clustered into three groups (Group 1: n=120, Group 2: n=154, Group 3: n=21). In the unadjusted analysis, Group 1 had lower levels of all parent estrogens and metabolites compared with Group 2, including lower levels of estrone (β =-0.046, p=0.04), 2-hydroxyestrone (β=-0.064, p=0.03), 16-epiestriol (β=-0.177, p = 0.03), and 17-epiestriol (β=-0.449, p=0.02). Differences in some, but not all, metabolites were attenuated after adjustment. In contrast, there was no consistent difference in estrogens between Group 3 and Group 2 in unadjusted or adjusted analyses. Conclusion: For some postmenopausal women, estrogen metabolism, particularly through the 2- and 16-hydroxylation pathways, was associated with distinct mammographic parenchymal texture profiles. Identification of texture patterns linked to specific hormonal profiles may help explain the biological factors that shape breast composition and inform strategies for prevention of hormonally driven breast cancers. Citation Format: Onyedikachi Adike, Kajita Yukie, Xianming Tan, Gretchen Gierach, Cherie Kuzmiak, Despina Kontos, Eric A. Cohen, Walter C. Mankowski, Sarah J. Nyante. Variation in urinary estrogen levels according to breast parenchymal texture in postmenopausal women [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2026; Part 1 (Regular Abstracts); 2026 Apr 17-22; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2026;86(7 Suppl):Abstract nr 2319.

Table S7 from Phase I/II Study of AXL-Specific Antibody–Drug Conjugate Enapotamab Vedotin in Patients with Advanced Solid Tumors

2025-11-26

<p>Loading values of radiomic features for principal component 1 comprising radiomics signatures at baseline, cycle 2, and longitudinal; the top 5 contributors to each radiomics signature highlighted with colored arrows</p>

Figure 1 from Phase I/II Study of AXL-Specific Antibody–Drug Conjugate Enapotamab Vedotin in Patients with Advanced Solid Tumors

2025-11-26

<p>Best change from baseline in sum of target lesions (investigator assessment) in the dose-expansion phase for the NSCLC cohorts (<b>A</b>, <b>C</b>, and <b>E</b>) and baseline tumor H-score levels for AXL positivity by investigator-assessed confirmed best overall response (<b>B</b>, <b>D</b>, and <b>F</b>). <b>A</b> and <b>B,</b> Cohort 1 (Q3W): NSCLC with sensitizing <i>EGFR</i> mutations and/or mutations targeted by third-generation TKIs (<i>n</i> = 22). <b>C</b> and <b>D,</b> Cohort 2 (2.2 mg/kg Q3W): NSCLC without activating <i>EGFR</i> mutations or <i>ALK</i> rearrangements (<i>n</i> = 55). <b>E</b> and <b>F,</b> Cohort 8 (3Q4W): NSCLC without activating <i>EGFR</i> mutations or <i>ALK</i> rearrangements (<i>n</i> = 26). NE, not evaluable; PD, progressive disease; PR, partial response.</p>

Novel 4D radiomics applied to dynamic FES PET images to improve prediction of breast cancer response to ER-targeted therapy

European Journal of Nuclear Medicine and Molecular Imaging · 2025-11-23

PURPOSE: F] fluoroestradiol (FES) is an FDA-approved tracer that measures functional estrogen receptor (ER) expression and can estimate the likelihood of response to ER-targeted therapy. In this exploratory analysis, we tested a novel radiomics based analysis of dynamic volumetric FES PET images to predict outcomes in patients with metastatic ER positive breast cancer treated with endocrine therapy. METHODS: We utilized the Rad-Fit method, previously tested in an FDG PET data set, to identify and characterize intratumor subregions of heterogeneous time-activity through an unsupervised clustering approach. A scaled silhouette score was implemented to determine the optimal number of intratumor subregions on a per-tumor basis. Summary statistics of sum of squared error (SSE) and distance between sub regions as well as the total number of intratumor subregions were used to build prognostic models of overall survival (OS) and progression free survival (PFS). We employed Kaplan-Meyer analysis to determine model performance. RESULTS: The radiomic phenotype differentiated between a high and low risk group for progression free survival (C = 0.67, p = 0.025) in the single tumor scenario. Radiomic features of subregion distance classified a high and low risk group for OS in a single tumor (C = 0.67, p = 0.008) and average tumor (C = 0.65, p = 0.017) scenario. CONCLUSIONS: In this exploratory study, 4D radiomic features extracted from dynamic FES PET images can improve the prediction of outcomes in metastatic ER positive breast cancer. Metrics of tumor subregion distance and radiomic phenotype appear to perform as the best radiomic predictors for risk stratification of OS and PFS respectively by potentially reflecting characteristics of the overall tumor heterogeneity in FES PET images. CLINICAL TRIAL NUMBER: not applicable.

Figure 2 from Phase I/II Study of AXL-Specific Antibody–Drug Conjugate Enapotamab Vedotin in Patients with Advanced Solid Tumors

2025-11-26

<p>PK profile for EnaV and MMAE in the dose-escalation phase. Plasma/serum concentrations of (<b>A</b>) conjugated EnaV and (<b>B</b>) MMAE at Q3W. Dots indicate observed data; lines indicate mean data. LLOQ, lower limit of quantitation.</p>

Table S6 from Phase I/II Study of AXL-Specific Antibody–Drug Conjugate Enapotamab Vedotin in Patients with Advanced Solid Tumors

2025-11-26

<p>Coefficient p-values for logistic regression models of investigator confirmed ORR and DCR built from radiomic feature-derived signatur</p>

Table S5 from Phase I/II Study of AXL-Specific Antibody–Drug Conjugate Enapotamab Vedotin in Patients with Advanced Solid Tumors

2025-11-26

<p>Safety summary: TEAEs in ≥20% of patients in any cohort in dose-expansion phase</p>

Sex‐specific patterns of a machine learning‐derived Alzheimer's brain atrophy imaging signature in participants without diagnosed cognitive impairment: A multi‐cohort study

Alzheimer s & Dementia · 2025-12-01

Abstract Background We investigated sex differences in a machine learning‐derived imaging signature of AD brain atrophy (i.e., SPARE‐AD 5 ), in relation to age, genetic factors ( APOE ε4 allele), and multi‐organ biological age gap (BAG 2,3 ). Methods Data from the iSTAGING and MULTI consortia included 53,622 participants without diagnosed cognitive impairment (mean age: 61.8 ± 12.6 years; 54% women). The SPARE‐AD model uses a support vector machine with a linear kernel to distinguish between cognitively normal individuals and those with AD 5 . Generalized linear models assessed sex differences and nine BAG associations with SPARE‐AD, adjusting for age, sex, APOE ε4, and interactions, and analysis of covariance (ANCOVA) with Tukey's test to assess differences in SPARE‐AD scores between APOE ε4 allele carrier groups. Results Overall, SPARE‐AD increased with age (β = 0.018, p < 2e‐16). Women had higher SPARE‐AD scores than men (β = ‐0.393, p < 2e‐16). Women had higher SPARE‐AD scores at younger ages but lower values at older ages (β = 0.006, p < 2e‐16 for the age‐sex interaction term) when compared to males (Figure 1a). Furthermore, SPARE‐AD was positively associated with the number of APOE ε4 alleles (β = 0.018, p = 1.06e‐6). Non‐carriers and heterozygous carriers of the APOE ε4 allele exhibited lower SPARE‐AD scores compared to homozygous carriers in analyses of both combined sexes and in men alone; this pattern was not observed in women (Figure 1b‐d). Among the nine BAGs, the brain BAG was most strongly associated with SPARE‐AD in both sexes combined (β = 0.018, p = 1.09e‐302) (Figure 2a) and separately (women: β = 0.017, p = 5.08e‐128; men: β = 0.019, p = 4.24e‐175) (Figure 2b‐c). Other significant BAG associations were observed in men and not in women, including musculoskeletal (β = 0.004, p = 0.02), immune (β = 0.004, p = 0.02), and metabolic BAGs (β = 0.005, p = 0.02) (Figure 2b‐d). Conclusion SPARE‐AD scores increased with age and were higher in women at younger ages but lower than men at older ages, with a significant age*sex interaction, and were positively associated with the number of the APOE ε4 allele, particularly in men.