A Multiparametric MRI-based Radiomics Model for Stratifying Postoperative Recurrence in Luminal B Breast Cancer

Kepei Xu, Meiqi Hua, Ting Mai, Xiaojing Ren, Xiaozheng Fang, Chunjie Wang, Min Ge, Hua Qian, Maosheng Xu, Ruixin ZhangJournal of Imaging Informatics in Medicine 2024 · DOI

TL;DR

This study introduces a novel MRI-based radiomics model to predict the risk of recurrence in Luminal B breast cancer patients after surgery. By analyzing features from dynamic contrast-enhanced MRI (DCE-MRI) and combining them with clinical data, the model offers improved accuracy in stratifying patients into high and low recurrence risk groups, potentially guiding more personalized treatment decisions.

What Problem Does This Paper Address?

Luminal B breast cancer is a common and aggressive subtype with a higher risk of recurrence. Accurately predicting this recurrence risk is crucial for tailoring adjuvant therapy, as over-treatment can lead to unnecessary side effects and costs, while under-treatment can result in poorer outcomes. Traditional methods rely on histopathological factors, which may not fully capture the complexity of tumor heterogeneity. While radiomics has shown promise in predicting recurrence for breast cancer in general, few studies have specifically focused on the Luminal B subtype, which has distinct biological characteristics. This paper aims to fill this gap by developing a multiparametric MRI radiomics model specifically for Luminal B breast cancer.

Methods

The researchers employed a multi-step approach to develop their predictive model. First, they extracted a large number of radiomics features (over 5000) from four MRI sequences (T2WI, DWI, ADC map, and DCE-MRI). Feature selection was a critical step, involving univariate Cox risk regression to identify initial candidates, followed by Least Absolute Shrinkage and Selection Operator (LASSO) regression for further reduction and to prevent overfitting. Finally, multivariate Cox regression was used to identify the most significant features for predicting recurrence.

A key aspect of their methodology was the development of a "risk score" based on two selected radiomics features from DCE-MRI. This score was then integrated with clinical and pathological characteristics (specifically lymph node status (pN) and endocrine therapy) to create a combined predictive model, a radiomics nomogram. The performance of both the radiomics risk score alone and the combined model was evaluated using Area Under the Curve (AUC) from Receiver Operating Characteristic (ROC) curves and calibration curves to assess predictive accuracy and consistency.

Datasets

The study utilized a retrospective dataset of 244 patients diagnosed with Luminal B breast cancer who underwent mastectomy or breast-conserving surgery. These patients were collected from two institutions. The dataset was divided into a training set (n=171) and a testing set (n=73) in a 7:3 ratio. Out of the 244 patients, 35 experienced disease recurrence, while 209 did not. No new benchmarks were introduced; the study used existing patient data.

Key Insights and Ideas

  • DCE-MRI Radiomics for Luminal B Specificity: The study highlights that specific radiomics features from DCE-MRI (GLSZM and GLRLM) are particularly valuable for stratifying recurrence risk in the Luminal B subtype, suggesting that this subtype's unique characteristics are captured by dynamic contrast enhancement patterns.
  • The Power of Combination: While the radiomics risk score alone showed good predictive power, integrating it with key clinical factors like lymph node status (pN) and endocrine therapy significantly boosted the model's accuracy, underscoring the importance of a multimodal approach.
  • Single-Sequence Advantage: The researchers propose that using a single MRI sequence (DCE-MRI in this case) for radiomics might be more practical and robust due to the lack of standardized MRI protocols across different institutions. This could facilitate wider clinical adoption.
  • Personalized Treatment Potential: The developed model has the potential to move beyond traditional risk assessment, offering a more nuanced understanding of individual patient risk, which can inform more precise and personalized treatment strategies.

Key Results

The study demonstrated promising results for both the radiomics risk score and the combined model:

  • Radiomics Risk Score:
    • 3-year recurrence risk: AUCs of 0.860 (training) and 0.816 (testing).
  • 5-year recurrence risk: AUCs of 0.868 (training) and 0.714 (testing).
  • Combined Model (Risk Score + pN + Endocrine Therapy):
    • 3-year recurrence risk: AUCs of 0.857 (training) and 0.943 (testing).
  • 5-year recurrence risk: AUCs of 0.912 (training) and 0.945 (testing).

The combined model showed excellent discriminative power and good calibration, indicating a strong agreement between predicted and actual recurrence risk. Kaplan-Meier survival curves also showed significant differences in recurrence-free survival between high- and low-risk groups identified by the model.

Limitations and Future Work

The authors acknowledge several limitations:

  • Retrospective Nature: The study's retrospective design may introduce selection bias, potentially limiting the generalizability of the findings.
  • Sample Size: The relatively small sample size (244 patients) might limit the statistical power of the study.
  • Generalizability Across Scanners: While data from two institutions were used, the equipment parameters were similar. Future studies should include multiple MRI protocols and external validation across different MRI scanners to ensure the model's robustness and compatibility.

The authors suggest that future research should focus on larger, multi-center studies with diverse patient populations and MRI protocols to further validate and refine this radiomics model for clinical application.

Summary updated

Abstract

This study aims to develop an MRI-based radiomics model to assess the likelihood of recurrence in luminal B breast cancer. The study analyzed medical images and clinical data from 244 patients with luminal B breast cancer. Of 244 patients, 35 had experienced recurrence and 209 had not. The patients were randomly divided into the training set (51.5 ± 12.5 years old; n = 171) and the test set (51.7 ± 11.3 years old; n = 73) in a ratio of 7:3. The study employed univariate and multivariate Cox regression along with the least absolute shrinkage and selection operator (LASSO) regression methods to select radiomics features and calculate a risk score. A combined model was constructed by integrating the risk score with the clinical and pathological characteristics. The study identified two radiomics features (GLSZM and GLRLM) from DCE-MRI that were used to calculate a risk score. The AUCs were 0.860 and 0.868 in the training set and 0.816 and 0.714 in the testing set for 3- and 5-year recurrence risk, respectively. The combined model incorporating the risk score, pN, and endocrine therapy showed improved predictive power, with AUCs of 0.857 and 0.912 in the training set and 0.943 and 0.945 in the testing set for 3- and 5-year recurrence risk, respectively. The calibration curve of the combined model showed good consistency between predicted and measured values. Our study developed an MRI-based radiomics model that integrates clinical and radiomics features to assess the likelihood of recurrence in luminal B breast cancer. The model shows promise for improving clinical risk stratification and treatment decision-making.
BibTeX
@article{xu2024multiparametric,
  author = {Kepei Xu and Meiqi Hua and Ting Mai and Xiaojing Ren and Xiaozheng Fang and Chunjie Wang and Min Ge and Hua Qian and Maosheng Xu and Ruixin Zhang},
  title = {{A Multiparametric MRI-based Radiomics Model for Stratifying Postoperative Recurrence in Luminal B Breast Cancer}},
  journal = {Journal of Imaging Informatics in Medicine},
  year = {2024},
  doi = {10.1007/s10278-023-00923-9},
}
Figure