A radiomic model leveraging conventional and Hessian matrix-based radiomic features from DCE-MRI for predicting efficacy of neoadjuvant chemotherapy in patients with HER2-low breast cancer

TL;DR

A new model using MRI imaging and clinical data helps predict how well chemotherapy will work for patients with HER2-low breast cancer.

Contribution

A novel radiomic model combining conventional and Hessian matrix-based features improves neoadjuvant chemotherapy prediction in HER2-low breast cancer.

Findings

The radiomic model achieved an AUC of 0.84 in training and 0.74 in validation.

The integrated nomogram reached an AUC of 0.89 in training and 0.79 in validation.

SHAP analysis identified key features contributing to model predictions.

Abstract

This study aimed to develop a predictive model for assessing the efficacy of neoadjuvant chemotherapy (NAC) in patients with Human Epidermal Growth Factor Receptor 2 (HER2)-low breast cancer, integrating clinical factors and radiomic features.

We retrospectively analyzed data from patients with HER2-low breast cancer who underwent NAC. Radiomic features were extracted from pre-treatment imaging, including wavelet-based and Hessian matrix-based features. Various machine learning models were constructed using radiomic and clinicopathological features. The Shapley additive explanations (SHAP) analysis was used to assess feature contributions. Model performance was evaluated based on the area under the receiver operating characteristic curve (AUC), accuracy, sensitivity, specificity, and other metrics. Finally, a nomogram was established by combining the best-performing models to enhance predictive utility.

For the clinicopathological model, the Random Forest (RF) algorithm was determined as the most effective. In the radiomic model, which incorporated both conventional and Hessian matrix-based features, RF exhibited superior performance compared to other models, achieving an AUC of 0.84 in the training cohort and 0.74 in the validation cohort. Based on these findings, a nomogram was developed that integrated the best-performing RF models from both the clinicopathological and radiomic feature sets. This nomogram attained an AUC of 0.89 in the training cohort and 0.79 in the validation cohort. Decision curve analysis further validated that the nomogram provided significant clinical benefits over models using only clinical or radiomic features.

Our current study successfully constructed a predictive nomogram that offers a promising strategy for predicting NAC efficacy in patients with HER2-low breast cancer.

Figures