Scholarlink

Background <p>Early identification of patients at risk of cancer therapy-related myocardial injury remains challenging. Conventional surveillance mainly relies on echocardiographic indices and circulating biomarkers, which often indicate injury after myocardial damage has already occurred. This study aimed to determine whether left ventricular myocardial radiomics features extracted from pretreatment chest computed tomography could predict myocardial injury after cancer therapy.</p> Methods <p>This single-center retrospective study included patients with malignant tumors who received anticancer therapy at Xuzhou Central Hospital between January 2020 and December 2025. A total of 129 patients with complete imaging and outcome data entered the radiomics modelling workflow. One patient with missing clinical covariates was excluded from baseline comparisons and logistic regression analyses, leaving 128 complete clinical cases. Left ventricular myocardial regions of interest were manually delineated on pretreatment chest computed tomography images. Radiomics features were screened using inter-observer reproducibility analysis, variance filtering, Spearman correlation analysis, and Elastic Net regularization. Clinical, radiomics, and combined models were developed and evaluated using receiver operating characteristic analysis, calibration assessment, the Hosmer-Lemeshow test, decision curve analysis, and five-fold cross-validation.</p> Results <p>In the complete clinical cohort, alcohol use and hypertension differed significantly between patients with and without myocardial injury. In univariable logistic regression, alcohol use, hypertension, absolute neutrophil count, diabetes, and NLR met the candidate threshold for multivariable analysis. No clinical variable remained independently significant in multivariable logistic regression, although alcohol use showed a borderline association. Five radiomics features were retained in the final model. In the test set, the areas under the curve of the clinical, radiomics, and combined models were 0.601, 0.867, and 0.805, respectively. The radiomics model showed significantly better discrimination than the clinical model, whereas its difference from the combined model was not statistically significant. The radiomics model achieved a sensitivity of 0.692 and a specificity of 0.962 at the optimal cutoff. Calibration should be interpreted cautiously because the Hosmer–Lemeshow <i>p</i> value was close to the conventional significance threshold.</p> Conclusions <p>A pretreatment chest computed tomography-based myocardial radiomics model showed preliminary potential for identifying patients at increased risk of myocardial injury after cancer therapy. This approach may support low-burden baseline cardiotoxicity risk assessment using imaging data already acquired during routine cancer care. Larger external cohorts are required for validation.</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

Clinical value of a CT radiomics model for predicting myocardial injury after cancer therapy

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年8月1日

查看全文
  • Fei Zhao,
  • Xiang Hu,
  • Heyao Xu,
  • Jingwen Liu,
  • Weijia Li,
  • Yuhang Wu,
  • Kangning Liu,
  • Cuixia Wen,
  • Chong Zhou,
  • Xiaojin Wu

摘要

Background

Early identification of patients at risk of cancer therapy-related myocardial injury remains challenging. Conventional surveillance mainly relies on echocardiographic indices and circulating biomarkers, which often indicate injury after myocardial damage has already occurred. This study aimed to determine whether left ventricular myocardial radiomics features extracted from pretreatment chest computed tomography could predict myocardial injury after cancer therapy.

Methods

This single-center retrospective study included patients with malignant tumors who received anticancer therapy at Xuzhou Central Hospital between January 2020 and December 2025. A total of 129 patients with complete imaging and outcome data entered the radiomics modelling workflow. One patient with missing clinical covariates was excluded from baseline comparisons and logistic regression analyses, leaving 128 complete clinical cases. Left ventricular myocardial regions of interest were manually delineated on pretreatment chest computed tomography images. Radiomics features were screened using inter-observer reproducibility analysis, variance filtering, Spearman correlation analysis, and Elastic Net regularization. Clinical, radiomics, and combined models were developed and evaluated using receiver operating characteristic analysis, calibration assessment, the Hosmer-Lemeshow test, decision curve analysis, and five-fold cross-validation.

Results

In the complete clinical cohort, alcohol use and hypertension differed significantly between patients with and without myocardial injury. In univariable logistic regression, alcohol use, hypertension, absolute neutrophil count, diabetes, and NLR met the candidate threshold for multivariable analysis. No clinical variable remained independently significant in multivariable logistic regression, although alcohol use showed a borderline association. Five radiomics features were retained in the final model. In the test set, the areas under the curve of the clinical, radiomics, and combined models were 0.601, 0.867, and 0.805, respectively. The radiomics model showed significantly better discrimination than the clinical model, whereas its difference from the combined model was not statistically significant. The radiomics model achieved a sensitivity of 0.692 and a specificity of 0.962 at the optimal cutoff. Calibration should be interpreted cautiously because the Hosmer–Lemeshow p value was close to the conventional significance threshold.

Conclusions

A pretreatment chest computed tomography-based myocardial radiomics model showed preliminary potential for identifying patients at increased risk of myocardial injury after cancer therapy. This approach may support low-burden baseline cardiotoxicity risk assessment using imaging data already acquired during routine cancer care. Larger external cohorts are required for validation.