CT and PET Radiomics for Predicting Benign vs. Malignant Lung Lesions

A dataset of 384 patients with pulmonary space-occupying lesions used to develop a multimodal machine learning model. The data includes computed tomography radiomics, positron emission tomography metabolic parameters, and clinical variables, with malignant lesions confirmed by pathology. The dataset was created by Xue Liu and last updated on 2026-04-21.

Use Cases

• Training a classifier to differentiate benign from malignant lung lesions based on CT radiomic features.
• Validating the predictive power of PET metabolic parameters like ΔSUVmax and TLG_d for tumor characterization.
• Developing an interpretable clinical nomogram using key features identified by SHAP analysis.
• Comparing the performance of integrated models (e.g., XGBoost, SVM) against single-modality models for medical diagnosis.

Strengths

• Includes data from 384 patients, split into a training set (n=268) and an independent temporal validation set (n=116).
• Model performance was rigorously evaluated, with the best integrated XGBoost model achieving an AUC of 0.967 on the validation set.
• All malignant lesions were pathologically confirmed, and 82% of benign lesions were confirmed by pathology, with the remainder followed clinically for at least 12 months.

Limitations

• Row count is unknown, which may limit suitability assessment.
• Column-level documentation is absent; field semantics must be inferred after download.
• The dataset is very small at 22.1 KB, indicating limited scope, likely containing model results or summary statistics rather than raw patient data.

Provenance

Source

figshare, author Xue Liu.

Collection Method

Data was collected from 384 patients who underwent dual-time-point ¹⁸F-FDG PET/CT examinations.

Freshness

Last updated 2026-04-21 05:25:24; freshness should be verified.