Development of fully automated deep-learning-based approach for prediction of sentinel lymph node metastasis in
Taixia Liu1,2,3, Guojia Zhao4, Wei Wei5
1Cheeloo College of Medicine, Shandong University, Jinan, Shandong, China.
Purpose:
This study aimed to develop a novel predicting model based on deep learning (DL) to predict sentinel lymph node (SLN) metastasis in breast cancer (BC) patients using ultrasound (US) imaging.
Methods:
A retrospective cohort consisting of 692 female BC patients from two hospitals was analyzed, with data collected from January 2020 to October 2023. Patients from Hospital A were randomly allocated to training (n = 405) and internal validation (n = 174) sets (7:3 ratio), with Hospital B patients (n = 113) serving as the external test set. A post-fusion model integrating the DeepLabV3, U-Net, and U-Net++ segmentation algorithms, respectively, was utilized to automatically delineate regions of interest (ROIs). Furthermore, three convolutional neural networks (CNNs)-ResNet50, ResNet101, and DenseNet121, respectively-were employed to analyze the cropped regions and concurrently construct a predictive model. A composite model that incorporates the DL signature (DL Sig) alongside clinical factors was developed by utilizing logistic regression (LR). A database to compare human and machine performance was created to evaluate the model's effectiveness. A nomogram was ultimately constructed to forecast the occurrence of SLN metastasis. The evaluation of model performance involved the utilization of receiver operating characteristic (ROC) curves, calibration curves, and decision curve analysis (DCA), respectively.
Results:
The post-fusion model demonstrated a robust correlation with manual delineation, yielding Dice coefficients of 0.893 and 0.855 in the internal validation and external test sets, respectively. The ResNet50 model, recognized as the most effective base model, demonstrated an area under the curve (AUC) of 0.773 (95% CI: 0.706-0.840) and an accuracy of 68% in the internal validation set (VS). In the external test set (TS), it achieved 0.765 AUC (95% CI: 0.674-0.856) with accuracy of 74%. The integrated model, which combined the DL Sig with clinical factors, exhibited the most effective performance in forecasting SLN metastasis, achieving 0.763 AUC (95% CI: 0.671-0.855) with accuracy of 69% in the TS. The DCA demonstrated notable clinical utility in the integrated model, surpassing the performance of both senior and junior radiologists.
Conclusion:
Our novel predictive model exhibited superior performance compared to both senior and junior radiologists in predicting SLN metastasis. Its capability for automatic segmentation and prediction highlights its potential for clinical applications.
