Validation of Deep-Learning Models for Autosegmentation of Brain Metastases

Targeted radiotherapy planning for treating brain metastases is labor-intensive and time-consuming. nnU-Net and MedNeXt are deep-learning frameworks widely explored in automated brain tumor segmentation. This study aimed to validate four pre-trained deep learning models based on nnU-Net and MedNeXt for brain metastases segmentation by using an external cohort of 243 patients with pre-treatment T1-weighted contrast-enhanced 3D brain MRI. Models trained on combined data achieved better performance metrics, while nnU-Net models slightly outperformed MedNeXt models. Further validation with follow-up images is ongoing to evaluate advanced performance.