Research on the deep learning model based on the combination of intratumoral and peritumoral dynamic contrast-enhanced MRI for predicting axillary lymph node metastasis in breast cancer
10.3969/j.issn.1002-1671.2024.06.011
- VernacularTitle:瘤内联合瘤周动态对比增强MRI乳腺癌腋窝淋巴结转移深度学习预测模型研究
- Author:
Yijun GUO
1
;
Rui YIN
;
Junqi HAN
;
Zhaoxiang DOU
;
Jingjing CHEN
;
Peifang LIU
;
Hong LU
;
Wenjuan MA
Author Information
1. 天津市肿瘤医院 肿瘤研究所 天津医科大学肿瘤医院乳腺影像诊断科 国家恶性肿瘤临床医学研究中心天津市恶性肿瘤临床医学研究中心 乳腺癌防治教育部重点实验室 天津市肿瘤防治重点实验室,天津 300060
- Keywords:
breast cancer;
deep learning;
magnetic resonance imaging;
intratumoral;
peritumoral
- From:
Journal of Practical Radiology
2024;40(6):907-912
- CountryChina
- Language:Chinese
-
Abstract:
Objective To explore the value of deep learning models in predicting axillary lymph node(ALN)metastasis of breast cancer based on intratumoral and peritumoral dynamic contrast-enhanced magnetic resonance imaging(DCE-MRI).Methods A retrospective analysis was conducted on cases from Tianjin Medical University Cancer Hospital and Laoshan Branch of Affiliated Hospital of Qingdao University,involving a total of 850 lesions in 850 patients.The region of interest within the tumor was delineated at the largest area of the lesion on the first enhancement images and automatically expanded by 3 mm and 6 mm in a conformal fashion.Deep learning prediction models based on ResNet50 were developed via intratumoral,peritumoral,and intratumoral combined peritumoral models,respectively,and a comprehensive prediction model was developed by integrating semantic features of imaging reports.Cases from Tianjin Medical University Cancer Hospital were randomly divided into training and test cohorts in a 7∶3 ratio,while cases from Laoshan Branch of Affiliated Hospital of Qingdao University served as the external validation cohort.The area under the curve(AUC),accuracy,sensitivity,specificity,F1-score,and Brier-score were calculated,respectively.Results The model incorporating intratumoral,peritumoral(3 mm),and semantic features demonstrated the highest performance,with AUC of 0.801[95％confidence interval(CI)0.765-0.845],0.781(95％CI 0.745-0.817),and 0.752(95％CI 0.700-0.793)in the training cohort,test cohort,and external validation cohort,respectively,and there was no significant difference in AUC between combined model and intratumoral/peritumoral model,respectively,but it demonstrated the higher sensitivity and F1-score,and the lower Brier-score.Conclusion Incorporating peritumoral images into the conventional model based on intratumoral images enhanced the predictive ability of ALN metastasis in breast cancer.
