Scholarlink

Background <p>Breast cancer’s personalized management requires better risk stratification. Recent studies focus on differentiating the pathological complete response (pCR) from non-pCR, which lacks accuracy in prognostic prediction and therapy guidance for most non-pCR patients. We aimed to better stratify neoadjuvant chemotherapy (NAC) response and early identification of poor responders in the non-pCR population.</p> Methods <p>Pretreatment MRI scans were obtained retrospectively from breast cancer patients who had NAC followed by surgery (January 2021-October 2023). Pathological response to NAC was assessed using the Miller-Payne (MP) grading system, with grades 1–2 indicating poor response and grades 3–5 indicating good response. Logistic regression was used to identify variables associated with MP grading and to build predictive models based on the radiomics score, clinicopathological features, and their combination. Additionally, machine learning models were also trained. The models were assessed for discrimination, calibration, and decision-making ability. Shapley Additive Explanations (SHAP) analysis was specifically performed to interpret the final machine learning model.</p> Results <p>A total of 336 patients were included (mean age, 48.75 ± 9.52 years; training set, 235; test set, 101). Radiomics score (OR = 1.46, 95% CI: 1.09, 1.99; <i>P</i> = 0.013) and human epidermal growth factor receptor 2 (HER2) status (OR = 5.93, 95% CI: 2.58, 16.16; <i>P</i> &lt; 0.001) were independently associated with MP grades. The logistic regression, XGBoost, and decision tree combined models demonstrated enhanced discrimination performance, with area under the receiver operating characteristic curve (AUC) of 0.77 (95% CI: 0.67, 0.87), 0.74 (95% CI: 0.65, 0.84), and 0.71(95% CI: 0.59, 0.82), respectively.</p> Conclusions <p>The combined model integrating pretreatment MRI radiomics score and HER2 status effectively differentiated between MP grades 1–2 and 3–5 in breast cancer following NAC. The study improved response stratification, with a specific emphasis on early detection of poor NAC responders in order to provide precise prognostic guidance and influence treatment options for this patient population.</p> Trial registration <p>Not applicable.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Pretreatment MRI radiomics for predicting pathological Miller-Payne grading in breast cancer following neoadjuvant chemotherapy

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年1月16日

查看全文
  • Chengliu Bi,
  • Ao Chen,
  • Fengming Ran,
  • Zaoxiu Hu,
  • Shaomei Sun,
  • Ruolan Wang,
  • Xiaofeng Niu,
  • Lijuan Deng,
  • Depei Gao,
  • Qinqing Li,
  • Jun Yang

摘要

Background

Breast cancer’s personalized management requires better risk stratification. Recent studies focus on differentiating the pathological complete response (pCR) from non-pCR, which lacks accuracy in prognostic prediction and therapy guidance for most non-pCR patients. We aimed to better stratify neoadjuvant chemotherapy (NAC) response and early identification of poor responders in the non-pCR population.

Methods

Pretreatment MRI scans were obtained retrospectively from breast cancer patients who had NAC followed by surgery (January 2021-October 2023). Pathological response to NAC was assessed using the Miller-Payne (MP) grading system, with grades 1–2 indicating poor response and grades 3–5 indicating good response. Logistic regression was used to identify variables associated with MP grading and to build predictive models based on the radiomics score, clinicopathological features, and their combination. Additionally, machine learning models were also trained. The models were assessed for discrimination, calibration, and decision-making ability. Shapley Additive Explanations (SHAP) analysis was specifically performed to interpret the final machine learning model.

Results

A total of 336 patients were included (mean age, 48.75 ± 9.52 years; training set, 235; test set, 101). Radiomics score (OR = 1.46, 95% CI: 1.09, 1.99; P = 0.013) and human epidermal growth factor receptor 2 (HER2) status (OR = 5.93, 95% CI: 2.58, 16.16; P < 0.001) were independently associated with MP grades. The logistic regression, XGBoost, and decision tree combined models demonstrated enhanced discrimination performance, with area under the receiver operating characteristic curve (AUC) of 0.77 (95% CI: 0.67, 0.87), 0.74 (95% CI: 0.65, 0.84), and 0.71(95% CI: 0.59, 0.82), respectively.

Conclusions

The combined model integrating pretreatment MRI radiomics score and HER2 status effectively differentiated between MP grades 1–2 and 3–5 in breast cancer following NAC. The study improved response stratification, with a specific emphasis on early detection of poor NAC responders in order to provide precise prognostic guidance and influence treatment options for this patient population.

Trial registration

Not applicable.