Deep learning models trained on video, equipment logs and clinical data from 222 real coronary angiograms recognized catheterization lab workflow phases with over 90 percent accuracy.