# Developing a predictive model for neoadjuvant therapy in HER2 overexpression breast cancer using multi-parameter MRI radiomics: two-center retrospective study

## TL;DR

This study develops an MRI-based radiomics model to predict the effectiveness of neoadjuvant therapy in HER2-positive breast cancer patients.

## Contribution

The novel contribution is a multi-parameter MRI radiomics nomogram model that outperforms individual clinical and imaging models in predicting treatment response.

## Key findings

- The nomogram model achieved high AUC (0.894 in training, 0.878 in testing) for predicting pathological complete response.
- Seven key radiomics features were selected from 3375 extracted features to build the predictive model.
- Calibration and decision curve analyses confirmed the model's strong consistency and clinical utility.

## Abstract

To explore an MRI-based radiomics model for predicting the efficacy of neoadjuvant therapy (NAT) for breast cancer with HER2 overexpression.

A total of 133 patients with HER2 positive breast cancer who underwent neoadjuvant therapy were retrospectively enrolled and divided into pathological complete response (PCR) and non-PCR groups. The patients from two centers were split into a training group (n=68) and a test group (n=65). MRI sequences (fs-T2WI, DWI, DCE-MRI) were used to outline regions of interest (ROI). Optimal features were selected using f-classif function and LASSO regression, and a multi-parameter MRI radiomics score (Rad-score) was constructed via logistic regression. Clinical independent predictors were identified to build a clinical model, and a nomogram was developed by combining Rad-score with these predictors. Model performance was evaluated using AUC, DeLong test, calibration curves, and decision curve analysis (DCA).

In this study, multivariate analysis identified key predictive clinical factors for pCR, including Ki-67 increment index and tumor morphology. Additionally, a total of 3375 radiomics features were extracted, and 7 key features were selected for model construction. Compared with the image group model and clinical model, the nomogram model based on imaging group had the best predictive performance (training group AUC: 0.894, sensitivity 83.72%, specificity 84.00%, test group AUC: 0.878, sensitivity 88.64%, specificity 71.43%). The calibration and decision curve analyses confirmed its strong consistency and clinical utility compared to individual models.

The nomogram model based on multi-parameter MRI has a steady performance in predicting the efficacy of NAT in breast cancer patients with HER2 overexpression, which provides important guidance for clinical treatment and decision-making.