Can AI have an impact in predicting breast cancer recurrence for women undergoing surgery for ductal carcinoma in situ (DCIS)?
For a new multicenter retrospective study, recently published in the American Journal of Roentgenology, researchers assessed the use of an AI software (Lunit INSIGHT for Mammography, version 1.1.7., Lunit) in comparison to clinical models for predicting second breast cancers in 1,740 patients who had breast-conserving surgery (BCS) for DCIS. The clinical risk models included the Van Nuys Prognostic Index (VNPI) and the Memorial Sloan Kettering Cancer Center (MSKCC) nomogram, according to the study.
The researchers found that women with an AI score > 73.5 percent had a cumulative incidence rate (CIR) of ipsilateral recurrence of 4.13 percent at five years in comparison to 0.86 percent for those with an AI score < 73.5. Similarly, those who met or exceeded the 73.5 percent AI score threshold had a higher CIR at 10 years (7.26 percent) for ipsilateral recurrence in contrast to women with an AI score below 73.5 percent (3.72 percent).1
There was also no significant difference between the AI software, VNPI and MSKCC nomogram with respect to the AUC for predicting ipsilateral recurrence (70 percent vs. 73 percent vs. 63 percent respectively), according to the study authors.1
“These findings support a role of the AI score on preoperative mammography in predicting the risk of ipsilateral recurrence after BCS for DCIS, potentially aiding proactive treatment and surveillance,” noted lead study author Jung Hyun Yoon, M.D., Ph.D., who is affiliated with the Department of Radiology and the Research Institute of Radiological Science in the College of Medicine at Yonsei University in Seoul, Korea, and colleagues.
Three Key Takeaways
• AI risk stratification may identify higher-risk patients after BCS for DCIS. An AI score > 73.5% on preoperative mammography was associated with significantly higher cumulative incidence rates of ipsilateral recurrence at five years (4.13 percent vs. 0.86 percent) and 10 years (7.26 percent vs. 3.72 percent), suggesting potential value in guiding surveillance or adjuvant treatment decisions.
• AI performance was comparable to established clinical models. The AI model demonstrated similar discriminative ability (AUC 70 percent) to the Van Nuys Prognostic Index (73 percent) and outperformed the MSKCC nomogram (63 percent), indicating that AI-based imaging biomarkers may provide risk prediction on par with traditional clinicopathologic tools.
• Objective AI assessment may offer advantages over subjective imaging features. Unlike prior studies linking calcification morphology and breast density to recurrence risk, this study found no such associations, highlighting the potential consistency and objectivity of AI-derived risk scoring in breast cancer recurrence prediction.
While previous literature has identified fine linear branching calcifications and breast density as being correlated with second breast cancers after DCIS treatment, the authors of the current study did not find associations between these factors and DCIS recurrence.1-3
“The reason for these conflicting results is unclear although may relate to the subjective variable nature of assessments of mammographic abnormalities, thereby highlighting a relative strength of objective AI,” posited Yoon and colleagues.
(Editor’s note: For related content, see “Study: Radiomic Mammography Features Lead to Upstaging of DCIS in More than 14 Percent of Patients,” “Can AI Assessment of Microcalcifications on Mammography Improve Differentiation of DCIS and Invasive Ductal Carcinoma?” and “Breast MRI-Based Radiomics May Reduce Overtreatment of DCIS.”)
In regard to study limitations, the authors acknowledged the retrospective nature of the research, use of a single AI software, a small number of second breast cancers and a Korean cohort, which may preclude broader extrapolation of the study findings to broader populations.
