Radiomics-Based Classification of Tumor and Healthy Liver on Computed Tomography Images.

Area of Science:

• Radiology
• Medical Imaging
• Machine Learning

Background:

• Liver malignancies are a major cause of cancer mortality.
• Abdominal CT scan data is often underutilized in clinical practice.
• Radiomics offers a method to extract quantitative features from medical images.

Purpose of the Study:

• To investigate the use of machine learning and radiomics for differentiating malignant liver tumors from healthy liver tissue.
• To assess the performance of various machine learning classifiers in this differentiation task.

Main Methods:

• Extraction of 1686 radiomics features (first-order, second-order, higher-order, shape statistics) from contrast-enhanced CT images of 94 patients.
• Feature selection using variance threshold, Student's t-test, and lasso regression.
• Training and evaluation of six classifiers (random forest, SVM, naive Bayes, AdaBoost, XGBoost, logistic regression) with grid search hyperparameter tuning and 10-fold cross-validation.

Main Results:

• The area under the receiver operating curve (AUROC) ranged from 0.5929 to 0.9268.
• Naive Bayes classifier achieved the highest AUROC score of 0.9268.
• Radiomics features were successfully classified, indicating good performance.

Conclusions:

• Radiomics features extracted from CT images can effectively differentiate between tumor and non-tumor liver tissue.
• The developed radiomics signature shows potential as a prognostic biomarker for hepatic tumor screening.
• Machine learning application in radiomics can enhance the diagnostic utility of underused CT imaging data.