Scholarlink

<p>Breast cancer remains a global health issue, with radiotherapy being a critical component of its treatment. However, radiotherapy can lead to shoulder morbidity, adversely affecting patients' quality of life. This study investigates the predictive potential of radiomics-derived features, termed dosiomics, from dose map to predict shoulder impairment in breast cancer patients. We analyzed data from 50 patients undergoing regional lymph node irradiation (RNI), using dosiomics features alongside dose-volume histogram (DVH) parameters to train machine learning models. Three feature selection methods were applied: correlation matrix, feature importance, and recursive feature elimination (RFE). Eight classification models were trained and evaluated on both dosiomics and DVH feature sets. Our results show that dosiomics models consistently outperformed DVH models in predicting shoulder morbidity, achieving an AUC of 0.81 compared to 0.69 for the best performing DVH model. Additionally, dosiomics models demonstrated superior predictive ability across all feature selection methods. Our results indicate that dosiomics features could contribute to more accurate prediction models for shoulder complications in breast cancer radiotherapy, warranting further investigation in larger patients’ cohort.</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

Dosiomics versus DVH-based machine learning models for predicting shoulder impairment in breast cancer patients

  • 期刊论文

  • 发表日期：

    2026年4月8日

查看全文
  • Moghadaseh Khalghi Bizaki,
  • Mohammad Fasih,
  • Zahra Bgherpour,
  • Pedram Fadavi,
  • Majid Akhavan Hejazi Seyed,
  • Donya Davoodi,
  • Sepideh Soltani,
  • Mojtaba Safari,
  • Manijeh Beigi

摘要

Breast cancer remains a global health issue, with radiotherapy being a critical component of its treatment. However, radiotherapy can lead to shoulder morbidity, adversely affecting patients' quality of life. This study investigates the predictive potential of radiomics-derived features, termed dosiomics, from dose map to predict shoulder impairment in breast cancer patients. We analyzed data from 50 patients undergoing regional lymph node irradiation (RNI), using dosiomics features alongside dose-volume histogram (DVH) parameters to train machine learning models. Three feature selection methods were applied: correlation matrix, feature importance, and recursive feature elimination (RFE). Eight classification models were trained and evaluated on both dosiomics and DVH feature sets. Our results show that dosiomics models consistently outperformed DVH models in predicting shoulder morbidity, achieving an AUC of 0.81 compared to 0.69 for the best performing DVH model. Additionally, dosiomics models demonstrated superior predictive ability across all feature selection methods. Our results indicate that dosiomics features could contribute to more accurate prediction models for shoulder complications in breast cancer radiotherapy, warranting further investigation in larger patients’ cohort.