Computer Vision News - August 2021

3 Summary 11 InnerEye by Microsoft In theFig1. a comparison, taken fromthis paper, is shownwith themain results showing that the ML model reduces the time it takes for end-to-end image segmentation and annotation in radiotherapy. Both the time to draw segmentation contours on the image and the time to correct inaccuracies in the automated (or semi-automated) system are included in this comparison. Hospitals and healthcare providers could utilize InnerEye as a toolkit to create personalized products and services using the layers of machine learning. They can also deploy them in hospitals and clinics using Azure Machine Learning and/or Azure Stack Hub (of course in that case regulation needs to be in place, e.g. FDA clearance or in-house exemption controls ). I am sure you are now curious to see some of the code and ideas on using it. So, let’s go! Dataset creation To create a dataset for segmentation or classification tasks, the following methods need to be used, followed by uploading the dataset as an AzureML blob storage . Here an example about the segmentation dataset will be given. Figure 1: The dark colurs are the algorithm predictions which are compared to the ground truth annotations (lighter colors are the reader annotations). From left to right (columns displaying mid-axial and mid-coronal slices): first left scan is from the main dataset and the remaining two are from the external dataset. The differences between the datasets on patient anatomy and through-plane scan resolution are shown.