Machine learning-based radiomics for differentiating lung cancer subtypes in brain metastases using CE-T1WI

Area of Science:

• Radiology
• Oncology
• Artificial Intelligence

Background:

• Distinguishing between non-small cell lung cancer (NSCLC) and small cell lung cancer (SCLC) brain metastases (BMs) is crucial for treatment.
• Accurate differentiation can be challenging using conventional imaging.

Purpose of the Study:

• To develop and validate machine learning-based radiomic models.
• To discriminate between primary NSCLC and SCLC in patients with BMs using contrast-enhanced T1-weighted imaging (CE-T1WI).

Main Methods:

• Extracted 833 radiomic features from 260 patients (173 NSCLC, 87 SCLC) with BMs.
• Utilized multistep selection (LASSO) to identify 15 optimal features.
• Trained ten machine learning models, including LightGBM, RandomForest, and XGBoost, on CE-T1WI data.

Main Results:

• Top models (XGBoost, LightGBM, SVM, RandomForest) achieved high AUCs (0.855-0.963) in training.
• LightGBM demonstrated superior performance in the test cohort with an AUC of 0.853.
• Radiomic features significantly aided in differentiating NSCLC from SCLC.

Conclusions:

• Machine learning radiomics using CE-T1WI is effective for differentiating NSCLC and SCLC brain metastases.
• The LightGBM model shows potential as a supportive, non-invasive diagnostic tool.
• Further prospective validation is recommended for clinical application.