Radiomics Features for Predicting Lung Adenocarcinoma Invasiveness

A two-center retrospective study analyzes 260 patients with subcentimeter subsolid lung nodules. The dataset contains a set of ten selected radiomic features used to train machine learning models for predicting invasive adenocarcinoma, with an IAC prevalence of 44.6%. It was authored by Wenfeng Feng and includes results from internal and external validation cohorts.

Use Cases

• Train a logistic regression classifier using the ten selected radiomic features to predict invasive adenocarcinoma status.
• Apply SHAP analysis to interpret the global importance of features like wavelet_HLL_glszm_LowGrayLevelZoneEmphasis and original_shape_Flatness.
• Validate model generalizability by testing performance across the 83-patient external cohort from Hospital 2.
• Conduct decision curve analysis to compare the clinical utility of the radiomics-LR model against empirical strategies.

Strengths

• Includes data from 260 patients across two hospitals, providing multi-center validation.
• Model development used a set of ten radiomic features selected via mRMR and LASSO regression.
• The best-performing logistic regression model achieved an AUC of 0.842 in internal validation.

Limitations

• The dataset is small in scale at 17.3 KB, containing summary results rather than raw patient-level data.
• Sample size is limited to 260 patients, which may restrict statistical power for some analyses.
• Data is retrospective and sourced from only two centers, potentially limiting generalizability.

Provenance

Source

figshare, authored by Wenfeng Feng.

Collection Method

Retrospective study using preoperative CT scans from two hospitals; radiomic features extracted via uAI Research Portal.

Time Range

null

Freshness

Last updated March 26, —.

Geography

Data from patients at Hospital 1 and Hospital 2 (specific locations not provided).