Scholarlink

Purpose <p>This study aimed to enhance the differential diagnosis of primary malignant brain tumors and lung cancer brain metastases using a radiomics approach. Leveraging MRI-derived imaging features and machine learning, this study sought to improve diagnostic accuracy and robustness while addressing the limitations of subjective radiological assessments.</p> Methods <p>Radiomics features were extracted from magnetic resonance imaging (MRI) scans of patients with primary malignant brain tumors and lung cancer brain metastases. An extreme gradient boosting (XGBoost) classification model was trained and evaluated using metrics such as precision, recall, F1-score, receiver operating characteristic (ROC) curve, area under the curve (AUC), and confusion matrix. The robustness of the model was validated through 5-fold cross-validation.</p> Results <p>The XGBoost model demonstrated a robust performance in classifying primary brain tumors and lung cancer brain metastases, with ROC-AUC values ranging from 0.93 to 0.98. Confusion matrices revealed true positive rates of 84.62–97.00% and true negative rates of 82.71–95.65%, with low fold-specific variability. Cross-validation metrics showed median accuracy, precision, recall, and F1-Scores above 0.90, with narrow interquartile ranges.</p> Conclusion <p>The XGBoost model exhibited high accuracy, precision, and discriminative ability in classifying primary malignant brain tumors and lung cancer brain metastases, with consistent performance across cross-validation folds.</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

Identifying Primary Brain Tumors and Lung Cancer Brain Metastases by Training XGBoost Models Based on Radiomics Features from Brain MRI Data

  • 期刊论文

  • 发表日期：

    2025年6月9日

查看全文
  • Qi Liu,
  • Haikuan Liu,
  • Jianyu Xu,
  • Wencheng Shao,
  • Yanling Bai

摘要

Purpose

This study aimed to enhance the differential diagnosis of primary malignant brain tumors and lung cancer brain metastases using a radiomics approach. Leveraging MRI-derived imaging features and machine learning, this study sought to improve diagnostic accuracy and robustness while addressing the limitations of subjective radiological assessments.

Methods

Radiomics features were extracted from magnetic resonance imaging (MRI) scans of patients with primary malignant brain tumors and lung cancer brain metastases. An extreme gradient boosting (XGBoost) classification model was trained and evaluated using metrics such as precision, recall, F1-score, receiver operating characteristic (ROC) curve, area under the curve (AUC), and confusion matrix. The robustness of the model was validated through 5-fold cross-validation.

Results

The XGBoost model demonstrated a robust performance in classifying primary brain tumors and lung cancer brain metastases, with ROC-AUC values ranging from 0.93 to 0.98. Confusion matrices revealed true positive rates of 84.62–97.00% and true negative rates of 82.71–95.65%, with low fold-specific variability. Cross-validation metrics showed median accuracy, precision, recall, and F1-Scores above 0.90, with narrow interquartile ranges.

Conclusion

The XGBoost model exhibited high accuracy, precision, and discriminative ability in classifying primary malignant brain tumors and lung cancer brain metastases, with consistent performance across cross-validation folds.