Real-time interventional MRI（I-MRI）would be very helpful to track the position of the surgical device during MR-guided neurosurgery.This would improve the patient outcome.Specially,in deep brain stimulation（DBS）,the visualization of the surgical procedure in real-time using I-MRI could improve the accuracy of the electrode placement.However,the reconstruction of interventional images using high undersampling rate and fast reconstruction speed for real-time imaging pose a great challenge.We proposed a feature-based convolutional neural network（FbCNN）for reconstructing for reconstructing interventional images from golden-angle radially sampled data,based on recent advances in deep learning（DL）.The reconstruction is composed of two steps: 1)reconstruction of the interventional feature;2)feature refinement and postprocessing.We showed that the interventional feature can be reconstructed with a cascade CNN,using an undersampling of only 5 radially sampled spokes.We obtained a final interventional image from a refined feature and a fully sampled reference image.FbCNN was compared with traditional reconstruction techniques and recent DL-based method,and it was shown that only FbCNN could reconstruct the interventional feature and the final interventional image.Furthermore,FbCNN demonstrated that had the potential in real-time I-MRI applications with a reconstruction time of ～500ms per frame and an acceleration factor of ～80.