Breast cancer poses a significant global health challenge, especially among women, with high rates of recurrence and distant spread despite early interventions. The timely identification of metastasis risk and accurate prediction of treatment strategies are critical for improving prognosis. However, the complex heterogeneity of breast tumors presents challenges in precise prognosis prediction. Therefore, the development of innovative methods for tumor segmentation and prognosis assessment is essential.

The research conducted is a multicenter study that enrolled 1,199 non-metastatic breast cancer patients from four independent centers. Our study leverages the advancements in artificial intelligence (AI) to address this challenge. This study is the first successful application of MRI-based multimodal prediction system to precisely identify the risk of postoperative recurrence in breast cancer patients.

Inclusion Criteria:

• Histologically confirmed stage I-III invasive BC
• Age ≥ 18 years
• The patient having undergone surgery
• The existence of MRI scans

Exclusion Criteria:

• Lacked pathological results
• Had other, simultaneous malignancies
• Had MR imaging issues were excluded

In the study's initial phase, 1199 patients were randomly allocated at a ratio of 8:2 to training and testing datasets for automatic tumor segmentation. Subsequently, we randomly assigned 569 patients from Sun Yat-sen Memorial Hospital of Sun Yat-sen University (SYSMH; Guangzhou, China) at a ratio of 3:1 to training (n = 456) and internal-validation (n = 113) cohorts for DFS prediction. The remaining patients were divided into two independent external-validation cohorts: 432 from Sun Yat-sen University Cancer Center (SYSUCC; Guangzhou, China) into external testing cohort 1, and 198 from Dongguan Tungwah Hospital (DTH; Dongguan, China) and Shunde Hospital of Southern Medical University (SDHSMU; Guangzhou, China) into external testing cohort 2.