e-Poster Display Session (ID 87) Poster Display

7P - Machine learning intratumoral and axillary lymph node magnetic resonance imaging radiomics for predicting axillary lymph node metastasis in patients with early-stage invasive breast cancer (RBC-01 Study) (ID 991)

Presentation Number
Lecture Time
09:00 - 09:00
  • Yujie Tan (Guangzhou, China)
On-Demand e-Poster Display, Virtual Meeting, Virtual Meeting, Singapore
09:00 - 20:00



In current clinical practice, the routine approaches of axillary lymph node (ALN) status evaluation through sentinel lymph node biopsy (SLNB) is unsatisfied with high false-negative rate and brings significant complications. We aimed to develop a preoperative magnetic resonance imaging radiomic-based signature for predicting ALN metastasis (ALNM) in a non-invasive way.


A total of 1,090 early-stage invasive breast cancer patients from 4 institutions were enrolled in this multicenter, retrospective, diagnositc study. Radiomic signature for ALNM prediction were constructed by machine learning in 803 patients from Sun Yat-sen Memorial Hospital and Sun Yat-sen University Cancer Center (Training cohort). The clinical-radiomic siganture was constructed by combining radiomic signature and significant clinic-pathological risk factors and was validated in patients from prospective phase III trials [NCT01503905] (Internal validation cohort, n=106), and Shunde Hospital and Tungwah Hospital (External validation cohort, n=181). This study is registered with ClinicalTrials.gov (NCT04003558) and Chinese Clinical Trail Registry (ChiCTR1900024020).


The radiomic signature for predicting ALNM consisted of intratumoral and ALN features showed AUCs of 0.91, 0.88, and 0.85 in the training, internal validation and external validation cohorts. The clinical-radiomic signature achieved the highest AUCs of 0.93, 0.91, and 0.91 in the training, internal validation and external validation cohorts, which successfully discriminate high- from low risk relapse patients (HR 0.12, 95% CI 0.03–0.53; P<0.001) and was similar to the performance in ALNM and non-ALNM (HR 0.28, 95% CI 0.09–0.87; P=0.002). In additon, the clinical-radiomic signature also performed well in the subgroup of N1, N2, N3 status (AUCs of 0.89, 0.90, 0.97).


This study developed a clinical-radiomic signature incorporated the intratumoral and ALN radiomic features and clinical risk factors, which could serve as a non-invasive tool to evaluate ALN status for guiding surgery plans of early-stage breast cancer patients.

Legal entity responsible for the study

The authors.


Has not received any funding.


All authors have declared no conflicts of interest.