Brain metastases detection and segmentation on magnetic resonance images is laborious, error-prone and often irreproducible for radiologists and radiation oncologist. We present a specific deep slice-crossed network with local weighted loss to automatically detect and segment brain metastases on contrast-enhanced T1WI images. The results demonstrated the good performance, high robustness and generalizability of the model. In addition, compared with radiologists, the model showed higher sensitivity and increased efficiency in identifying and segmenting brain metastases. The results jointly suggested that the proposed model is a promising tool to assist the workflow in the clinical practice.