Related Experiment Video
Updated: Aug 6, 2026
Machine Learning Algorithms for Early Detection of Bone Metastases in an Experimental Rat Model
Published on: August 16, 2020
Machine Learning-Based Prediction of Brain Metastasis at Initial Diagnosis in Small-Cell Lung Cancer: Model
Yu-Long He1, Qin-Ling Jiang1, Yong Zhai2
1Department of Oncology, Nanxishan Hospital of the Guangxi Zhuang Autonomous Region (The Second People's Hospital of Guangxi Zhuang Autonomous Region), Guilin, Guangxi, China.
Background:
Brain metastasis (BM) in small cell lung cancer (SCLC) is typically associated with poor survival rates and quality of life, making the timely identification of patients with a high likelihood of BM at diagnosis crucial.
Aims:
To develop and validate a machine learning (ML) prediction model for BM in the overall SCLC population and provide an interpretable and clinically accessible risk assessment tool.
Methods And Results:
Univariate and multivariate logistic regression analyses were performed to identify BM-associated factors. Eight ML algorithms were applied to build the model. The model performance was quantified using the area under the curve (AUC), area under the precision-recall curve (AUPRC), and Matthews correlation coefficient (MCC). SHapley Additive exPlanations (SHAP) was used to interpret the best-performing model. A web calculator was developed to facilitate individualized BM risk estimation. Multivariate logistic regression revealed that age, T stage, tumor size, bone metastasis, lung metastasis, and distant lymph node metastasis were independently associated with BM. Extreme Gradient Boosting (XGB) achieved the best discrimination in the validation cohort, with an AUC of 0.8762, AUPRC of 0.9025, accuracy of 0.7974, precision of 0.8009, recall of 0.7974, specificity of 0.8516, MCC of 0.5983, F1-score of 0.7968, and Brier score of 0.1377. Cross-validation demonstrated similarly strong performance. SHAP analysis identified age, tumor size, T stage, distant lymph node metastasis, bone metastasis, and lung metastasis as the strongest contributors to BM risk. A web-based risk calculator was developed to facilitate exploratory risk stratification and individualized BM risk estimation.
Conclusion:
We created and internally validated an interpretable ML model to identify brain metastasis at diagnosis in SCLC, with XGB showing the best performance. Its web-based tool may assist in identifying patients with a higher likelihood of brain metastasis at diagnosis and provide supplementary risk stratification information for exploratory clinical assessment. Further prospective validation is required before routine clinical implementation of this model.
