Value of Deep Learning Ultrasound Radiomics Nomogram to Assess Invasive Metastasis in Invasive Breast Cancer
10.3969/j.issn.1005-5185.2024.08.009
- VernacularTitle:基于深度学习超声组学列线图评估浸润性乳腺癌侵袭转移的价值
- Author:
Songhua LI
1
;
Chaojun WU
;
Dayou WEI
;
Shaofeng LI
;
Youshi LUO
;
Yan LIN
;
Linyong WU
Author Information
1. 茂名市人民医院超声医学科,广东 茂名 525011
- Keywords:
Breast neoplasms;
Deep learning;
Radiomics;
Ultrasonography;
Nomograms
- From:
Chinese Journal of Medical Imaging
2024;32(8):803-808
- CountryChina
- Language:Chinese
-
Abstract:
Purpose To explore the value of deep learning ultrasound radiomics nomogram in assessing the biological characteristics of invasive metastases in invasive breast cancer.Materials and Methods A retrospective collection of ultrasound imaging data from 180 pathologically confirmed invasive breast cancer between January 2021 to December 2022 in Maoming People's Hospital was conducted,with pathological reports indicating the status of lymph node metastasis(LNM),lymphovascular space invasion(LVSI)or perineural invasion(PNI),according to the LNM/LVSI/PNI status,the three indexes were divided into the training cohort and the verification cohort by 8∶2.Based on Pyradiomics and ResNet50 deep learning extractor,1 316 radiomic features and 2 048 deep learning features were extracted,respectively.The random forest machine learning algorithm was employed to develop evaluation models,and the model scores were calculated.The deep learning radiomics nomograms were developed based on the radiomic and deep learning model scores.The receiver operating characteristic curve was used to assess the performance of the models.The Delong test was applied to analyze the performance differences between different models.Results In the evaluation of LNM,LVSI and PNI status,the area under the curve of all the nomogram in the cohorts demonstrated moderate or above assessment performance(≥0.73),with accuracies all greater than 0.70.Specifically,in the LNM evaluation,the area under the curve of the training cohort was 0.97,the accuracy was 0.93,the sensitivity was 0.88 and the specificity was 0.96.Through the Delong test,the assessment performance of the nomograms was superior to the radiomics models(LNM,Z=2.04,P=0.04;LVSI,Z=2.80,P=0.01;PNI,Z=3.52,P<0.01),and was superior to or similar to the deep learning models(LNM,Z=4.52,P<0.01;LVSI,Z=1.86,P=0.06;PNI,Z=0.31,P=0.76)in the training cohort.Conclusion The deep learning radiomics nomogram can effectively evaluate the biological characteristics of invasion and metastasis in invasive breast cancer.The nomogram improves the assessment performance by integrating the radiomic and deep learning feature information.
