79 3 Towards Automatization of Organoid Analysis: A Deep Learning Approach to Localize and Quantify Organoids Images framework based on convolutional neural network that outputs a mask channel consisting of a pixel-wise prediction of organoids’ location. As training this framework from scratch is a non-trivial task, we employ and adjust a pre-trained model18. It is relevant to mention that the computation time for each uploaded CZI file varies and depends on the number of stacks, images sizes, and the number of found organoids in each image stack. With this in mind, an optional step is offered for the user to provide an email address so that a notification will be sent to the email address when the computation is completed. Each of the segmented organoids will get an unique number (ID). * https://organelx.hpc.rug.nl/organoid/ Upon completing this step, an automated analysis step begins for grouping and assigning identical organoids of different stacks. Organoids of different stacks are called identical when their mask center-point is within a certain distance and the Interaction of Union19 value of their bounding boxes is within a certain range. Once all internal computations are finished, the results will be presented on the page to which the user is redirected, or an electronic notification is sent, if an email address has been provided. The number of the successfully detected and segmented organoids for all stack images will be given. The user can interactively perform the following operations: • Select a stack image and preview the organoids that are detected by MASKRCNN for that specific stack image. The corresponding number of detected organoids for this stack will be specified. • Select multiple organoids of interest. As a result, the area of each selected organoid will be shown. Additionally, the minimum, maximum, average, and total areas will be marked. The area of each organoid depends on the unit scale given by the CZI file. A radio button is provided to determine whether to select identical organoids of all stack or to only consider organoids of the chosen stack. • Extract and download a summary zip file containing: 1. PNG images of all stacks including the selected and unselected organoids. 2. CSV file of the identical organoids represented by unique ID and corresponding name. 3. Excel file, in which each sheet includes information of the selected organoids (ID, bounding box coordinates, mask border, and the area) for each stack image.
RkJQdWJsaXNoZXIy MTk4NDMw