Building a computer vision machine learning model is hard at the best of times. The challenges increase dramatically when it comes to building a medical AI model, especially one that needs to use images from radiology modalities. In this talk, Encord will take a closer look at some of the most innovative use cases for AI in radiology, the inherent challenges that come from creating training data from DICOM and NIfTI files, how to overcome these challenges, and best practice for building a computer vision model for radiology diagnostics.