Computer Vision News - September 2020

RSIP Vision Tools 4 MedIO is amedical images I/OPython package aimed at deep learning developers and people in the medical images industry. It is the first open source package that RSIP Vision has published. Currently, people use many different packages to load medical images into their deep learning architectures . This solution unifies the input and output engines of itk, nibabel, and pydicom (and dicom-numpy) in a simple comprehensive interface. Jonathan Daniel is the RSIP Vision engineer who developed MedIO . He explains: “With MedIO, developers can do everything they could do before, but just in a single consistent interface . Each package has its own conventions and metadata and we take care of all these details and provide a uniform solution for the developers to make their life easier. They can focus on the deep learning architectures, rather than the loading and saving of the data.” MedIO saves time and effort because you do not have to write the same code and procedures every time. It provides a consistent solution, meaning you do not have to check how itk conventions differ from nibabel, for example. You can write and read whatever you want and save in any format. You do not need to translate the conventions for RGB images, 2D images, or 3D images. The package takes care of any special cases that appear – including bugs, for example, which are not currently treated in the individual packages. “One example is saving dicom images ,” Jonathan tells us. “When you save a 3d dicom file, because of its metadata structure, you can only save 24 out of 48 orientations. Orientations are the way in which the numbers are ordered in the memory of the computer. For 2D images, there are 8 possible orientations. For 3D images, there are 48. If you use itk and you do not use one of the 24 allowed orientations, it results in a flipped image. The array would be okay, but its metadata would not fit what you save and cause a MedIO - Python package for medical images Jonathan Daniel - RSIP Vision