Coronary angiography represents the GOLD standard for diagnosing and treating obstructive coronary artery disease. Its analysis relies on cardiologist?s visual assessment which is subjective and leads to inter-observer variability and different treatment strategies. Objective methods have been proposed and implemented in catheterization laboratories. However, their input is manually chosen by the cardi-ologist. In this paper, we present a new coronary angiogram keyframe extraction, Deep AngioKey. It is based on feature extraction. The frame with the highest vessel structure was identified as keyframe. Feature extraction was done using a U-Net model trained for binary vessel segmentation. Our solution surpassed existing methods and reached an overall accuracy of 98.1% in keyframe identification with an average main frame distance of 1.63.