Attention-Based Deep Learning In Overall Survival Prediction for Patients with Brain Metastases Using MRI: A Multicenter Retrospective Study
Abstract
Purpose
To develop and validate an attention-based deep learning (DL) model to predict overall survival (OS) in patient with brain metastases (BM) using pre-treatment magnetic resonance image (MRI).
Methods
A total of 216 BM patients treated by radiotherapy or neurosurgical resection were retrospectively enrolled from two hospitals, which were divided into a training (N=128), an internal validation (N=56), and an external validation (N=32) cohort. Three feature aggregation methods and seven radiomics models were developed to predict OS for BM patients. A DL model incorporated cross-attention mechanism was built for predicting OS using pre-treatment MR images. The prognostic accuracy of the DL model was compared with radiomics model and graded prognostic assessment (GPA). A nomogram was developed by integrating GPA, radiomics score (Rad-score) and deep-learning score (DL_RiskScore). Model performance was assessed by concordance index (C-index).
Results
The ave-lesion radiomics achieved highest C-index of 0.685, 0.672 and 0.671 in training, internal validation and external validation cohorts. The DL model achieved a much higher C-index than radiomics model and GPA model (0.806 vs. 0.684 vs. 0.620 in training cohort, 0.755 vs. 0.673 vs. 0.628 in internal validation cohort, and 0.733 vs. 0.684 vs. 0.579 in external validation cohort). Nomogram integrating DL_RiskScore, Rad-score, and GPA achieved C-index values of 0.809, 0.764 and 0.733 in training, internal validation and external validation cohorts.
Conclusion
DL model based on pre-treatment MRI can predict survival outcome of patients with BM, which may help in risk stratification and guide treatment decision-making and follow-up management.