Kannan, Vignesh

Abstract Current genomic assays provide prognostic information and guide chemotherapy decisions in early-stage hormone receptor–positive/human epidermal growth factor receptor 2–negative (HR+/HER2−) breast cancer, but limited tissue sampling may incompletely capture tumor heterogeneity. We developed and validated an image-based artificial intelligence (AI) model integrating dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) and clinicopathologic features to predict recurrence risk in early-stage HR+/HER2 − breast cancer. In this retrospective multi-institutional study, the model was trained using pre-treatment imaging and clinicopathologic data from 522 patients and independently validated in 550 additional patients treated between 2010 and 2023. By analyzing spatial tumor characteristics, the model generated individualized recurrence risk scores that outperformed clinicopathologic features alone. In the validation cohort, median age was 53 years (range, 24–90), 15.6% of patients self-identified as African American, 89.7% had clinical T1–T2 disease, and 34.4% had 1–3 positive lymph nodes. Five-year recurrence-free survival was 93.8% (95% CI, 90.0–96.2%) among patients classified as low risk compared with 82.6% (95% CI, 74.8–88.1%) among those classified as high risk, corresponding to an adjusted hazard ratio of 2.28. Prognostic performance was maintained across clinically relevant subgroups, including age and nodal status. These findings support the potential of a noninvasive image-based AI approach for real-time prognostication and treatment personalization in early-stage HR+/HER2 − breast cancer.

560 Background: Patients (pts) with early stage hormone receptor-positive, human epidermal growth factor receptor 2-negative (HR+/HER2-) breast cancer (BC) remain at both short and long term risk of recurrence. While clinico-pathological and gene-based biomarkers provide insight into early and late recurrence risk, additional prognostic tools are needed. We previously developed (Saha et al. Clin Cancer Res , 2025) a new breast MRI image-based artificial intelligence prognostic tool (TumorSight Risk) composed of both radiographic and clinical information. Here we sought to validate this tool in an independent cohort. Methods: In this multi-center, blinded, prospective/retrospective validation study, we sought to validate the 5-year recurrence-free survival (RFS) according to pre-defined risk scores (0-100) and categories (“high” or “low”) via pre-specified cutoff (≥47) for pts with HR+/HER2- BC. Kaplan-Meier and Cox regression analyses were then used to assess the unadjusted and adjusted association between risk category/score and the 5-year RFS. Results: 2,129 pts with HR+/HER2- BC diagnosed between January 1, 2010, and December 31, 2024, who underwent breast MRI prior to surgery (no neoadjuvant therapy) were included from four institutions. Median age was 57 years, 16% were Black, 15% had grade 3 disease, 93% were clinical T1 or T2, and 85%/15% were clinical N0/N1. At a median follow up of 6.24 years, the unadjusted hazard ratio (HR) for RFS comparing low (n=1808) and high risk (n=321) pts was 2.71 (95% CI = 1.95 – 3.76, p = 3x10 -9 ). The estimated 5 year RFS rate according to risk score (low and high) was 95.5% (95% CI: 94.63 – 96.3%) and 89.6% (95% CI = 86.4 – 92.6%) while the 10 year RFS estimates were 88.9% (95% CI = 86.5 - 90.9%) and 69.6% (95% CI = 59.1 – 77.9%). Univariate subgroup analysis demonstrated that the risk score was prognostic by age (< 50 vs. ≥50 years), pathologic nodal status, grade, and surgery type, regardless of chemo-, endocrine, or radiation therapy administration. The observed continuous risk score was strongly associated with risk of recurrence, with a 22% increase in relative risk per 10-unit change in risk score (0-100), translating to a 2.4% (95% CI = 1.2 – 4.5%) and 25.5% (95% CI = 15.4 – 40.4%) risk of recurrence at 5-years for pts at the lowest (0-9) versus highest (90-100) risk score. The prognostic association between risk score and RFS remained after adjusting for pathologic T- and N-stage, grade, age, race, and treatment (p = 0.002). Conclusions: TumorSight Risk is a new image-based biomarker for HR+/HER2- BC that provides prognostic information independent of standard clinico-pathological parameters. Given the ready availability of breast MRI in most pts with BC, TumorSight Risk may be a convenient tool for estimating risk for recurrence.