169MO - Development and validation of a magnetic resonance imaging radiomics-based signature to predict axillary lymph node metastasis and disease-free survival in patients with breast cancer: A multicenter cohort study

Background

Preoperative non-invasive tools to accurately predict the axillary lymph node (ALN) status and disease-free survival (DFS) in early breast cancer are lacking. Therefore, we have developed and validated contrast–enhanced multiparametric magnetic resonance imaging (MRI) radiomic-based signatures for preoperative identification of ALN metastasis and assessment of individual disease-free survival (DFS) in early breast cancer.

Methods

In this multicentre, retrospective, cohort study, we included early stage breast cancer patients from four hospitals in China, which were divided randomly (7:3) into the development and validation cohorts. Radiomic features were extracted from preoperative MR imaging. LASSO and random forest algorithm were applied to select key radiomic features from the development cohort. Two nomograms incorporating radiomic and clinical signatures were developed to predict ALN status and individual DFS, respectively, based on multivariable logistic or radiomics signature penalized Cox regression models.

Results

218,729 MRI images from 1,214 individuals were included. The ALN radiomic nomogram accurately predicted ALN metastasis in the development cohort (Area Under Curve, [AUC] = 0.92), validation cohort (AUC = 0.90), and entire cohort (AUC = 0.91). The DFS radiomic nomogram could discriminate high- from low risk patients in the development, validation, and entire cohort (HR for all 0.04, P < 0.001). The radiomic nomogram was strongly correlated with 3-year DFS in the development cohort (AUC = 0.89), validation cohort (AUC = 0.90) and entire cohort (AUC = 0.89). Subgroup analysis showed that these nomograms had excellent and highly-generalized predictive ability for ALN metastasis and DFS.

Conclusions

This study used the largest database to date to describe the application of MRI-based artificial intelligence in patients with breast cancer, presenting novel individualized clinical-decision radiomic nomograms that could precisely predict ALN metastasis and DFS.

Clinical trial identification

NCT04003558; ChiCTR1900024020.

Editorial acknowledgement

Legal entity responsible for the study

HRY.

Funding

National Major Science and Technology Project of China, Medical artificial intelligence project of Sun Yat-Sen Memorial Hospital, National Natural Science Foundation of China, Natural Science Foundation of Guangdong Province, Guangzhou Science and Technology Major Program, Sun Yat-Sen University Clinical Research 5010 Program, Sun Yat-Sen Clinical Research Cultivating Program, Guangdong Science and Technology Department, Tencent Charity Foundation.

Disclosure

All authors have declared no conflicts of interest.