Deep learning ultrasound radiomics nomogram for predicting Ki-67 expression in invasive breast cancer
LU Lili
LI Lin
DU Huan
ZHANG Panpan
ZHU Yinhua
JIA Xiaohan
LI Yang
Abstract：Objective To explore the value of a deep learning-based ultrasound radiomics nomogram in predicting Ki-67 expression levels in invasive breast cancer.Methods A retrospective single-center study was conducted,collecting complete preoperative clinical data and ultrasound images from 465 patients with pathologically confirmed invasive breast cancer at the First Affiliated Hospital of Bengbu Medical University from January to December 2024.Image acquisition was performed using Mindray Resona 7 and Samsung HS60 color Doppler ultrasound systems.Based on immunohistochemical results,patients were divided into high and low Ki-67 expression groups and randomly assigned to training(n=326)and validation(n=139)cohorts at a 7:3 ratio.ITK-SNAP software was used to segment tumors from the largest 2D ultrasound cross-sectional images,with interobserver consistency of ROI delineation assessed by ICC.Pyradiomics was employed to extract radiomics features from tumor tissues,and four deep learning networks were pretrained to construct clinical,ultrasound radiomics,fusion,and combined nomogram models.Diagnostic performance and clinical utility were evaluated using ROC curves,calibration curves,and decision curve analysis.Results Nineteen optimal ultrasound radiomics features and the DenseNet121 deep learning model showed the best performance(P<0.05).In the training cohort,the AUCs for the clinical model,ultrasound radiomics model,deep learning model,fusion model,and nomogram were 0.79(95%CI:0.74-0.84),0.85(95%CI:0.81-0.90),0.87(95%CI:0.83-0.91),0.94(95%CI:0.91-0.97),and 0.95(95%CI:0.93-0.98),respectively.In the validation cohort,the corresponding AUCs were 0.76(95%CI:0.68-0.84),0.78(95%CI:0.70-0.85),0.81(95%CI:0.74-0.88),0.91(95%CI:0.86-0.96),and 0.93(95%CI:0.89-0.98).Conclusion The deep learning-based ultrasound radiomics nomogram can effectively predict Ki-67 expression in invasive breast cancer.
Keywords：deep learningradiomicsbreast cancerKi-67nomogram
Publication Date：2025-11-20
Online Publishing Date：2025-12-16(First online date of this platform, not the publication date of the document)
Pages：8( 1325-1332 )