Prediction of prognosis in brain metastasis with artificial-intelligence-driven methods for whole brain radiotherapy

Purpose: Inferentially, 24%–45% of cancer patients develop brain metastases in their course. Individual survival estimation for these patients is crucial to identify the subset that may not benefit from whole-brain irradiation (WBI) due to a short survival time. This study aimed to identify variables and evaluate an artificial intelligence algorithm to determine which patients would benefit from WBI. Materials and Methods: The data of 345 patients with brain metastasis who were treated with 30 Gy in 10 fractions of WBI were retrospectively analyzed. In this cohort, a total of 15 clinical / laboratory factors are evaluated with 15 models of machine learning algorithms using Python 2.3, Pycaret library. Results: The Gradient Boosting Regressor was found to be the most accurate model, with a 0.68 R2 an R² value of 0.68, and a mean absolute error (MAE) of 12.90.The prediction error for the gradient Boosting Regressor was calculated as R2: 0.841. When the importance of features was investigated, time from diagnosis to metastasis was found to be the most important predictive variable for survival. Conclusion: The results of this study enable us to identify patients who may have an early death and provide a consequential decision guide in terms of whole-brain radiotherapy or additional labor-intensive techniques.