Using Machine Learning and CT Radiomics to Predict Brain Metastasis in Patients With NSCLC

By: Chris Schimpf, BS
Posted: Thursday, February 1, 2024

Cancer Imaging has published the results of a Chinese study exploring the use of machine learning to predict brain metastasis in patients with non–small cell lung cancer (NSCLC). For the study, Tong Tong, MD, of Fudan University, Shanghai, and colleagues developed a deep learning–based segmentation and CT radiomics–based ensemble learning model. The researchers found that by fusing CT radiomics and clinical features, the model improved prediction of brain metastasis and proved effective in stratifying patients into high- and low-risk groups. In addition, they reported the model demonstrated prognostic value in predicting both brain metastasis–free survival and overall survival among patients with NSCLC.

“Accurately predicting the risk of brain metastasis is a critical aspect of personalized treatment planning for advanced NSCLC patients to improve treatment outcomes,” the investigators stated. “This approach could have significant implications for the early detection and treatment of this challenging complication.”

A total of 602 patients with stage IIIA to IVB NSCLC were included in the retrospective, two-center study, 309 of whom were diagnosed with brain metastasis, and 293 of whom were not. A three-dimensional deep residual U-Net network was used to segment patients’ lung tumors; then, 1,106 radiomics features were computed to decode the imaging phenotypes of primary lung cancer. The investigators employed an ensemble learning algorithm of the extreme gradient boosting (XGBoost) classifier to train and build a prediction model by fusing radiomics features and clinical features. The risk score generated by the model (through Kaplan-Meier survival analysis) achieved a significant prognostic value for brain metastasis–free survival and overall survival in the two cohorts (P < .05).

The researchers noted that the study’s results supported additional research into optimal machine learning models to combine different phenotype features and further improve the prediction of brain metastasis among patients with NSCLC.

Disclosure: The study authors reported no conflicts of interest.

Cancer Imaging

---