Abstract PO3-07-07: Predictive Modeling for Identifying Breast Cancer Patients Eligible for Axillary Lymph Node Dissection Exemption Following Neoadjuvant Therapy: A Longitudinal MRI-based Radiomics and Deep Learning Features Analysis

Abstract

Abstract Objective: Neoadjuvant therapy (NAC) has emerged as a pivotal treatment modality for breast cancer. However, accurately identifying patients who can safely avoid axillary lymph node dissection (ALND) following NAC remains challenging. In this study, our aim was to develop a predictive model using longitudinal MRI-based radiomics and deep learning features to identify breast cancer patients suitable for exemption from ALND. Methods: A total of 140 patients with cN1-2 breast cancer who underwent NAC were included in this study between January 2021 and October 2022. MRI images were collected before and after two cycles of NAC. The dataset was randomly divided into training and validation sets using a 7:3 ratio. Logistic regression (LR), K-nearest neighbors (KNN), LightGBM, and multilayer perceptron (MLP) machine learning models were utilized to predict axillary lymph node pathological complete response (ypN0) in patients following NAC. Finally, a prediction model with a single modality feature was trained by integrating all the extracted data. Results: Among the 140 patients included, 55 achieved ypN0 following NAC, while 85 did not achieve ypN0. Modeling the features extracted from pre- and post-chemotherapy evaluations revealed that the LR model achieved the highest area under the curve (AUC) values, reaching 0.842 and 0.841, respectively. Moreover, by integrating the evaluation of pre- and post-chemotherapy images and the net change of features between the two time points, the novel developed Amalgamation model demonstrated the highest AUC value of 0.958. The Amalgamation model exhibited an accuracy of 0.9, sensitivity of 0.833, and specificity of 1. Conclusion: Our study developed a predictive model using MRI images obtained before and after two cycles of NAC to identify breast cancer patients with ypN0 at an early stage. This model has the potential to avoid unnecessary ALND, significantly reducing complications. Keywords: breast cancer, neoadjuvant chemotherapy, axillary lymph node dissection, MRI, radiomics, deep learning, predictive modeling Figure 1: The ROC curves of Amalgamation model, Delta-model, Two-cycle reassessment-model and Baseline-model based on Logistic regression. Table 1: Performances of combining different machine learning models for predicting axillary lymph node pathological complete response to neoadjuvant therapy in pre- and post-neoadjuvant therapy MRI images. Citation Format: Yushuai Yu, Jialu Yi, Ruiliang Chen, Kaiyan Huang, Jie Zhang, Chuangui Song. Predictive Modeling for Identifying Breast Cancer Patients Eligible for Axillary Lymph Node Dissection Exemption Following Neoadjuvant Therapy: A Longitudinal MRI-based Radiomics and Deep Learning Features Analysis [abstract]. In: Proceedings of the 2023 San Antonio Breast Cancer Symposium; 2023 Dec 5-9; San Antonio, TX. Philadelphia (PA): AACR; Cancer Res 2024;84(9 Suppl):Abstract nr PO3-07-07.