Objective:Neoadjuvant therapy(NAT)has become the standard treatment option for patients with locally advanced breast cancer.How to non-invasively screen out patients with pathological complete response(pCR)after NAT h...Objective:Neoadjuvant therapy(NAT)has become the standard treatment option for patients with locally advanced breast cancer.How to non-invasively screen out patients with pathological complete response(pCR)after NAT has become an urgent world-wide clinical problem.Our work aims to the assessment of neoadjuvant treatment response in breast cancer patients for higher accuracy prediction using innovative artificial intelligence system.Methods:In this study,we retrospectively collected longitudinal(pre-NAT and post-NAT)multi-parametric magnetic resonance imaging(MRI)and clinicopathologic data of a total of 1,315 breast cancer patients(clinical stageⅠ-Ⅲ)who had undergone NAT followed by standard surgery and treated across 5 independent medical centers from January 2010 to January 2023.We used radiomics,3D convolutional neural network technology and clinical data statistical analysis methods to extract and screen multimodal features,and then developed and validated a Clinical-Radiomics-Deep-Learning(CRDL)model to predict patients'pCR outcomes based on multimodal fusion features.Results:We use the area under the receiver operating characteristic curve(AUC)in the primary cohort(PC)and3 external validation cohorts(VC_(1-3))to evaluate the model performance.The results showed that the AUC in the PC composed of 2 medical centers was 0.947[95%confidence interval(95%CI):0.931-0.960],and the AUC values in VC_(1-3)were 0.857(95%CI:0.810-0.901),0.883(95%CI:0.841-0.918)and 0.904(95%CI:0.860-0.941),respectively.Conclusions:The CRDL model demonstrated high accuracy and robustness in predicting pCR to NAT using multimodal fusion data.This study provides a strong foundation for non-invasive assessment of pCR status in breast cancer patients following NAT and offers critical insights to guide clinical decision-making in post-NAT treatment planning.展开更多
基金supported by the Non-profit Central Research Institute Fund of Chinese Academy of Medical Sciences(No.2023-JKCS-23)the Special Research Fund for Central Universities,Peking Union Medical College[No.2022-I2M-C&T-A-014,CAMS Innovation Fund for Medical Sciences(CIFMS)]。
文摘Objective:Neoadjuvant therapy(NAT)has become the standard treatment option for patients with locally advanced breast cancer.How to non-invasively screen out patients with pathological complete response(pCR)after NAT has become an urgent world-wide clinical problem.Our work aims to the assessment of neoadjuvant treatment response in breast cancer patients for higher accuracy prediction using innovative artificial intelligence system.Methods:In this study,we retrospectively collected longitudinal(pre-NAT and post-NAT)multi-parametric magnetic resonance imaging(MRI)and clinicopathologic data of a total of 1,315 breast cancer patients(clinical stageⅠ-Ⅲ)who had undergone NAT followed by standard surgery and treated across 5 independent medical centers from January 2010 to January 2023.We used radiomics,3D convolutional neural network technology and clinical data statistical analysis methods to extract and screen multimodal features,and then developed and validated a Clinical-Radiomics-Deep-Learning(CRDL)model to predict patients'pCR outcomes based on multimodal fusion features.Results:We use the area under the receiver operating characteristic curve(AUC)in the primary cohort(PC)and3 external validation cohorts(VC_(1-3))to evaluate the model performance.The results showed that the AUC in the PC composed of 2 medical centers was 0.947[95%confidence interval(95%CI):0.931-0.960],and the AUC values in VC_(1-3)were 0.857(95%CI:0.810-0.901),0.883(95%CI:0.841-0.918)and 0.904(95%CI:0.860-0.941),respectively.Conclusions:The CRDL model demonstrated high accuracy and robustness in predicting pCR to NAT using multimodal fusion data.This study provides a strong foundation for non-invasive assessment of pCR status in breast cancer patients following NAT and offers critical insights to guide clinical decision-making in post-NAT treatment planning.
