Scholarlink

Purpose <p>Radiomics enables decoding of phenotypic heterogeneity in malignant tissues. This study evaluated the capability of radiomics for interpreting thoracic malignancies and developing a machine-learning–based virtual biopsy model.</p> Methods <p>This study included 300 patients with thoracic lesions who underwent CT-guided biopsy. Lesions were segmented, and 106 radiomics features were extracted after preprocessing. Feature harmonization was performed using the ComBat algorithm, followed by feature selection using ANOVA and the ReliefF algorithm. Features were normalized, and class imbalance was managed using SMOTE. The dataset was split into training (80%) and test (20%) sets, with five-fold cross-validation applied to the training data. Binary classifications were conducted to distinguish tumoral subtypes—adenocarcinoma (ADC), squamous cell carcinoma (SCC), small cell lung cancer (SCLC), lymphoma, and mesenchymal tumors—from normal tissue. Modeling was conducted using logistic regression (LR), support vector machine (SVM), and random forest (RF) classifiers, with evaluation based on several metrics across three feature types: unharmonized features, ComBat-harmonized features, and the best ComBat-harmonized features selected by ANOVA.</p> Results <p>Using the best ComBat-harmonized features، AUCs for ADC, SCC, SCLC, lymphoma, and mesenchymal tumors versus normal tissue were 0.58, 0.70, 0.62, 0.70, 0.77, and 0.91, respectively. The combined clinical-radiomics model demonstrated higher performance, achieving AUCs of 0.83, 0.76, 0.94, 0.95, 0.98, and 0.98.</p> Conclusion <p>The results add strength to the body of evidence that virtual biopsy can serve as an intelligent bedside assistant alongside image-guided biopsy, potentially decreasing the inherent risks of biopsy procedures.</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

Development of a Radiomics-Based Machine Learning Model Utilizing CT-Guided Biopsy Images: Virtual Biopsy of Thoracic Cancers

  • 期刊论文

  • 发表日期：

    2026年2月18日

查看全文
  • Maryam Cheraghi,
  • Mehran Malekshoar,
  • Fatemeh Pakniyat,
  • Seyed Mahmoud Reza Aghamiri,
  • Sougand Setareh,
  • Mehrdad Bakhshayesh Karam,
  • Hamid Abdollahi

摘要

Purpose

Radiomics enables decoding of phenotypic heterogeneity in malignant tissues. This study evaluated the capability of radiomics for interpreting thoracic malignancies and developing a machine-learning–based virtual biopsy model.

Methods

This study included 300 patients with thoracic lesions who underwent CT-guided biopsy. Lesions were segmented, and 106 radiomics features were extracted after preprocessing. Feature harmonization was performed using the ComBat algorithm, followed by feature selection using ANOVA and the ReliefF algorithm. Features were normalized, and class imbalance was managed using SMOTE. The dataset was split into training (80%) and test (20%) sets, with five-fold cross-validation applied to the training data. Binary classifications were conducted to distinguish tumoral subtypes—adenocarcinoma (ADC), squamous cell carcinoma (SCC), small cell lung cancer (SCLC), lymphoma, and mesenchymal tumors—from normal tissue. Modeling was conducted using logistic regression (LR), support vector machine (SVM), and random forest (RF) classifiers, with evaluation based on several metrics across three feature types: unharmonized features, ComBat-harmonized features, and the best ComBat-harmonized features selected by ANOVA.

Results

Using the best ComBat-harmonized features، AUCs for ADC, SCC, SCLC, lymphoma, and mesenchymal tumors versus normal tissue were 0.58, 0.70, 0.62, 0.70, 0.77, and 0.91, respectively. The combined clinical-radiomics model demonstrated higher performance, achieving AUCs of 0.83, 0.76, 0.94, 0.95, 0.98, and 0.98.

Conclusion

The results add strength to the body of evidence that virtual biopsy can serve as an intelligent bedside assistant alongside image-guided biopsy, potentially decreasing the inherent risks of biopsy procedures.