The study demonstrated that the combined model yielded improved risk assessment for both interval and long-term breast cancers.
Recent advancements in mammography-based risk models have enabled the estimation of both short-term and long-term breast cancer risks. However, the potential benefit of combining these models to enhance risk assessment remains unexplored. A new study recently published in Radiology aimed to investigate whether combining a diagnostic artificial intelligence (AI) system for lesion detection and a mammographic texture model could improve breast cancer risk assessment.
The study involved a retrospective analysis of Danish women who underwent consecutive breast cancer screening from November 2012 to December 2015, with at least 5 years of follow-up data. Short-term risk was evaluated using a commercially available AI system that assigned a probability score for cancer detection. A separate mammographic texture model, trained on different data, assessed long-term cancer risk associated with specific textures. The area under the receiver operating characteristic curve (AUC) was used to evaluate individual and combined model performances for predicting future interval and long-term cancers in women with negative screening results.
The Danish screening cohort comprised 119,650 women, with 320 developing interval cancers and 1401 developing long-term cancers. The combination model, integrating AI lesion detection and mammographic texture assessment, demonstrated a higher AUC (0.73) for predicting interval and long-term cancers combined, compared to either the diagnostic AI system (AUC 0.70) or the texture model (AUC 0.66) alone. The top 10% of women identified by the combination model accounted for a significant proportion of interval cancers (44.1%) and long-term cancers (33.7%).
RELATED: New Research Shows Consequences of Delayed Diagnosis in Mammography Screening
The study demonstrated that the combined model yielded improved risk assessment for both interval and long-term breast cancers. The AUC of 0.73 was higher than that of previously established risk models. Notably, the combination model accurately identified women at high risk, representing a substantial proportion of interval and long-term cancers. The study's limitations include the need for external validation in diverse populations and the relatively short follow-up period.
"Recent advances in AI-based breast cancer risk prediction models suggest more personalized breast cancer screening is within reach," wrote Clare B. Poynton, M.D., Ph.D., and Priscilla J. Slanetz, M.D., M.P.H, in an accompanying editorial. They noted, however, that existing AI models still have substantial limitations, especially when it comes to identifying those with high-risk as well as in populations who are diverse in age, race, and ethnicity.
"Image-based AI risk prediction clearly has the potential to individualize screening protocols by helping determine when a woman should start screening, the optimal frequency for screening, and possibly even when screening could be delayed or avoided altogether," they wrote. "However, further study of multiple risk algorithms in large diverse imaging data sets will be critical to compare varied approaches and, ultimately, to ensure that resources are appropriately directed toward populations who will benefit most from a more personalized approach to early breast cancer detection."
Can CT-Based Deep Learning Bolster Prognostic Assessments of Ground-Glass Nodules?
June 19th 2025Emerging research shows that a multiple time-series deep learning model assessment of CT images provides 20 percent higher sensitivity than a delta radiomic model and 56 percent higher sensitivity than a clinical model for prognostic evaluation of ground-glass nodules.
Can Contrast-Enhanced Mammography be a Viable Screening Alternative to Breast MRI?
June 17th 2025While the addition of contrast-enhanced mammography (CEM) to digital breast tomosynthesis (DBT) led to over a 13 percent increase in false positive cases, researchers also noted over double the cancer yield per 1,000 women in comparison to DBT alone.
Multinational Study Reaffirms Value of Adjunctive AI for Prostate MRI
June 16th 2025The use of adjunctive AI in biparametric prostate MRI exams led to 3.3 percent and 3.4 percent increases in the AUC and specificity, respectively, for clinically significant prostate cancer (csPCa) in a 360-person cohort drawn from 53 facilities.