Multi-Center Facial Image Features for Non-Invasive Lung Cancer Risk Assessment

124 facial image features were extracted from multi-center patients with benign pulmonary nodules and lung cancer using a TFDA-1 Digital Tongue and Face Diagnosis Instrument. The dataset was used to train and validate four machine learning models, with the best-performing XGBoost model achieving an AUC of 0.900 on an internal test set and 0.906 on an external validation set. The research was authored by Yulin Shi and last updated on figshare in May 2026.

Use Cases

• Training machine learning models for lung cancer risk assessment based on facial image features.
• Validating the performance of auxiliary diagnostic models using an independent external cohort.
• Analyzing the clinical utility of facial imaging for early cancer screening via calibration and decision curves.
• Investigating feature importance for model interpretability using the SHAP method.

Strengths

• Model performance was validated on an independent external cohort, demonstrating generalization.
• The XGBoost model achieved an AUC of 0.906 in external validation.
• 124 facial image features were extracted for analysis.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• The dataset is a 22.2 KB DOCX file, suggesting it contains summary results or a paper, not the raw image data.

Provenance

Source

figshare

Collection Method

Facial images were collected via a TFDA-1 Digital Tongue and Face Diagnosis Instrument from multi-center patients.

Freshness

Last updated 2026-05-15 04:27:51; freshness should be verified.