554 Background: Homologous recombination deficiency (HRD) in breast cancer is an actionable target, with treatment efficacy potentially linked to timely detection. Notably, BRCA1 and BRCA2 are key genes implicated in HRD, and their pathogenic germline mutations are crucial criteria for the use of poly (ADP-ribose) polymerase inhibitors (PARPi). Although germline BRCA1/2 mutations are detectable in only a small fraction (1-5%) of breast cancers, recent whole-genome sequencing (WGS) studies have revealed that up to 22% of breast cancers also exhibit HRD-like genomic characteristics, suggesting a broader potential for HRD-targeted therapy. This has spurred the development of genomic-feature-based HRD identification methods. However, the invasive and costly nature of these methods poses a challenge. This has prompted our investigation into non-invasive image biomarkers for HRD identification. Methods: Invasive breast cancer patients were recruited from March 2021 to August 2022. Eligibility criteria included availability of fresh-frozen tumor tissue for WGS and suitability for dynamic contrast-enhanced (DCE) MRI to facilitate radiomics analysis. The association between HRD and radiomic features as well as clinicopathologic factors was investigated through rigorous genomic and statistical analysis. Results: This study encompassed 145 patients aged 20 to 69 years, of whom 16.6% (24 out of 145) were genomically identified as HRD. All patients harboring pathogenic germline mutations in BRCA1 (n=3) or BRCA2 (n=2), along with concomitant somatic LOH, exhibited HRD. A comprehensive analysis of 214 radiomic features revealed that tumor sphericity, informational measure of correlation 1 (IMC1), size-zone non-uniformity normalized (SZNN), and small area emphasis (SAE) derived from subtracted early dynamic T1-weighted imaging were significantly correlated with HRD. Moreover, these radiomic features demonstrated substantial diagnostic value, as evidenced by their ROC-AUC and PR-AUC performance against baseline models: Sphericity (ROC: 0.62, p=0.055; PR: 0.22, p=0.053), SAE (ROC: 0.63, p=0.052; PR: 0.32, p=0.007), SZNN (ROC: 0.63, p=0.04; PR: 0.30, p=0.007), and IMC1 (ROC: 0.62, p=0.067; PR: 0.29, p=0.02). Additionally, our findings suggest that these radiomic features are more directly related to HRD than to the triple-negative phenotype, which is commonly linked to HRD. Conclusions: We found that four radiomic features from MRI have significant predictive value for identifying HRD in breast cancer. Given their fair predictability and immediate availability, these features are ideally suited for a screening approach, potentially identifying more patients eligible for targeted treatment. This finding represents a significant advancement in the fields of radiology and precision oncology, opening new avenues for patient-specific treatment strategies.