Comparison of radiomics-based machine-learning classifiers for pretreatment prediction of pathologic complete response to neoadjuvant therapy in breast cancer
Author Block: X. Li, C. Li, L. Jiang, M. Chen; Beijing/CN
Purpose: In recent years, machine learning (ML) classifiers have been used to establish high-performance predictive models for pathological complete response (pCR) in breast cancer after neoadjuvant therapy (NAT). However, few studies have compared the effectiveness of different ML classifiers. This study investigated the ability of radiomics models based on pre- and post-contrast first-phase T1WI to predict breast cancer pCR after NAT and compared the performance of different ML classifiers.
Methods or Background: In this retrospective study, 300 patients from the Duke-Breast-Cancer-MRI dataset who underwent NAT were included, including pCR (n=76) and non-pCR (n=224) cases. These patients were randomly divided into training and validation groups at a ratio of 8:2. Radiomics features were extracted from pre- and post-contrast first-phase T1WI images of each patient. The radiomics model was built using features selected through the Spearman correlation analysis and the LASSO algorithm after normalisation. Seven ML classifiers were used to assess the predictive performance of the radiomics models.
Results or Findings: Out of the seven classifiers used, the LightGBM classifier performed best in predicting breast cancer pCR, with an AUC of 0.813 in the validation group (accuracy 78.3%, sensitivity 46.7%, specificity 100.0%). During subgroup analysis, RF achieved the highest AUC in pCR prediction in luminal breast cancers (0.859, accuracy 85.9%, sensitivity 68.8%, and specificity 83.3%), and DT yielded the highest AUC in pCR prediction in triple negative (TN) breast cancers (0.909, accuracy 88.2%, sensitivity 81.8%, and specificity 100%).
Conclusion: Overall, the LightGBM-based radiomics model demonstrated superior performance in predicting breast cancer pCR, while RF and DT displayed promising results in predicting pCR for luminal and TN breast cancers, respectively, during subgroup analysis.
Limitations: Our study included different NAT treatment regimens, and subgroup analysis based on treatment regimens was not performed.
Funding for this study: No funding was received for this study.
Has your study been approved by an ethics committee? Not applicable
Ethics committee - additional information: Data obtained from the Cancer Imaging Archive (TCIA, https://www.cancerimagingarchive.net/) did not require ethical approval; informed consent was waived since the TCIA dataset de-identified patient information.