# Predicting relative efficacy of anthracyclines and taxanes in breast cancer neoadjuvant AC-T chemotherapy using longitudinal MRI radiomic model

## TL;DR

This study uses MRI radiomic models to predict which breast cancer patients respond better to anthracycline or taxane treatments during chemotherapy, helping personalize treatment plans.

## Contribution

A hybrid radiomic and clinicopathological model is developed to predict relative efficacy of AC and T treatments in breast cancer neoadjuvant chemotherapy.

## Key findings

- The Delta radiomic model achieved an AUC of 0.887 in the training set and 0.757 in the test set for predicting relative efficacy.
- The Delta+clinicopath hybrid model improved performance with AUCs of 0.887 and 0.772 in training and test sets, respectively.
- The hybrid model showed favorable calibration and clinical net benefit for stratifying patients.

## Abstract

Neoadjuvant chemotherapy (NAC) is a standard treatment strategy for breast cancer, with a commonly used regimen consisting of 4-cycle anthracycline and cyclophosphamide (AC) treatment followed sequentially by 4-cycle taxane (T) treatment. Variations in treatment efficacy are observed at different stages of AC-T regimen. Stratifying patients based on the efficacy variations could provide insights to prolong the cycle of AC or T treatment, potentially enhancing the overall efficacy of NAC. Therefore, this study aimed to evaluate the feasibility of developing magnetic resonance imaging (MRI) radiomic models for predicting the relative efficacy of AC versus T treatments.

This retrospective study included 190 breast cancer patients, who were randomly allocated into a training set (n=133) and a test set (n=57). All patients received NAC treatment consisting of four cycles of AC followed by four cycles of T. Breast MRI examinations were conducted before NAC (pre-NAC), before the fifth cycle (mid-NAC), and before surgery (post-NAC). Relative efficacy was defined by comparing tumor volume change rates between the AC and T treatment stages. Radiomic features were extracted from dynamic contrast-enhanced (DCE) and apparent diffusion coefficient (ADC) images based on the intratumoral and peritumoral regions at the pre-NAC and mid-NAC stages. Radiomic models were first developed, and hybrid models were then established by integrating radiomic and clinicopathological data to predict relative efficacy.

For radiomic models, the Delta model demonstrated effective discrimination of relative efficacy, achieving areas under the curve (AUCs) of 0.887 [95% confidence interval (CI): 0.816-0.930] in the training set and 0.757 (95% CI: 0.683-0.817) in the test set. For hybrid models, the Delta+clinicopath model showed improved performance, with AUCs of 0.887 (95% CI: 0.873-0.892) in the training set and 0.772 (95% CI: 0.744-0.786) in the test set. The Delta+clinicopath model also exhibited favorable calibration in both sets and provided a substantial clinical net benefit.

The hybrid model is a reliable and reproducible tool for predicting the relative efficacy between AC and T treatments in breast cancer NAC. The model could help to stratify patients for personalized adjustment of NAC regimens.