251P - The changes of multiparametric magnetic resonance imaging radiomics and Ki67 after two cycles of neoadjuvant therapy builds an early prediction model of pathological complete response in breast cancer

Background

Neoadjuvant therapy (NAT) has emerged as a pivotal treatment modality for early-stage breast cancer, offering patients achieving pathological complete response (pCR) markedly improved prognoses. Therefore, early prediction of NAT response and optimization of therapeutic strategies are crucial for improving pCR rates. Our purpose was to develop a predictive model of pCR for the early stage of NAT based on the changes of multiparametric magnetic resonance imaging (mpMRI) radiomic features and Ki67 after two cycles of NAT.

Methods

A total of 150 breast cancer patients between January 2021 and August 2023 were retrospectively analyzed, which was divided into training set (n=106) and validation set (n=44). The mpMRI images and Ki67 were collected at baseline and after two cycles of NAT. Radiomic features were extracted, and selected by intraclass correlation coefficient test, analysis of variance, and Pearson correlation analysis. The baseline-radiomic model (pre-radiomic), 2^nd-cycle -radiomic model (2^nd-radiomic) and delta-radiomic model were constructed using random forest algorithm. Area under the receiver operating characteristic (ROC) curve was employed to evaluate efficacy of different models. The fusion model was constructed by combining the radiomic features of the optimal model with delta-ki67 and other clinicopathological factors. Delong test was used to compare the statistical differences between the models.

Results

The delta-radiomic model demonstrated favorable predictive performance in the validation set (AUC=0.810(0.679-0.940)), outperforming both the pre-radiomic model (AUC=0.613(0.441-0.784)) and 2^nd-radiomic model (AUC=0.710(0.542-0.878)). The fusion model significantly enhancing prediction efficiency (AUC=0.877(0.775-0.979). Delong test indicated significant differences in diagnostic performance between the fusion model and other models (P < 0.05).

Conclusions

The fusion model based on changes of mpMRI radiomic features and Ki67 after two cycles of NAT demonstrates promising potential for early pCR prediction in breast cancer.

Clinical trial identification

Editorial acknowledgement

Legal entity responsible for the study

The authors.

Funding

Has not received any funding.

Disclosure

All authors have declared no conflicts of interest.