Parameters are typically given in the order: input images, output images, other parameters. Next, it reads CLIJ_, CLIJ2_ or CLIJx_ followed by the specific method and, in brackets, the parameters passed over to this method. Thus, it is recommended to write this line initially, to start at a predefined empty state.Īnother typical step in CLIJ macros is to push image data to the GPU memory:Īll CLIJ methods start with the prefix Ext., a convention by classical ImageJ, indicating that we are calling a macro extension optionally installed to ImageJ. Hence, a macro is not executed until the very end, where GPU memory typically gets cleaned up. A macro under development unintentionally stops every now and then with error messages. It is not mandatory to write it at the beginning, however, it is recommended while elaborating a new ImageJ macro. This command is typically called by the end of a macro. The second line, in the example shown above, cleans up GPU memory. One can explore available GPU devices by using the menu Plugins> ImageJ on GPU (CLIJ2)> Macro tools> List available GPU devices. If only a part of the name is specified, such as nVendor or some, CLIJ will select a GPU which contains that part in the name. One can specify the name of the GPU in brackets, for example nVendor Awesome Intelligent. In the first line, the parameter cl_device can stay blank, imposing that CLIJ will select automatically an OpenCL device, namely the GPU. Four example time points of the dataset are shown in Fig. The protein-channel (C2), excited with 488 nm wavelength light, represents the distribution of the cytoplasmic Lamin B protein, which accumulates at the inner nuclear membrane (Lamin B receptor signal). The nuclei-channel (C1), excited with 561 nm wavelength light, consists of Histone H2B-mCherry signals within the nucleus. The dataset has a pixel size of 0.165 \(\mu \)m per pixel and a frame delay of 400 s. As a representative dataset for this domain, we process a two-channel time-lapse showing a HeLa cell with increasing signal intensity in one channel (Boni et al. The method of live-cell imaging, taken as long-term time-lapses, is important when studying dynamic biological processes. Fluorescent labeling techniques allow the study of certain structures and cell components, in particular to trace dynamic processes over time, such as changes in intensity and spatial distribution of fluorescent signals. We also give an insight into quality assurance methods, which help to ensure good scientific practice when modernizing BIA workflows and refactoring code.Ĭell membranes create functional compartments and maintain diverse content and activities. Accordingly, we show how to measure workflow performance. In terms of image processing, refactoring means restructuring an existing macro without changing measurement results, but rather improving processing speed. These commands are then assembled to refactor the pre-existing workflow. We then introduce ways to discover CLIJ commands as counterparts of classic ImageJ methods. To demonstrate the procedure, we translate a formerly published BIA workflow for examining signal intensity changes at the nuclear envelope, caused by cytoplasmic redistribution of a fluorescent protein (Miura, 2020). Our suggested approach neither requires a profound expertise in high performance computing, nor to learn a new programming language such as OpenCL. We present a guide for transforming state-of-the-art image processing workflows into GPU-accelerated workflows using the ImageJ Macro language. , 2020), enables biologists and bioimage analysts to speed up time-consuming analysis tasks by adding support for the Open Computing Language (OpenCL) for programming GPUs (Khronos-Group, 2020) in ImageJ. As an alternative to established acceleration techniques, such as ImageJ’s batch mode, we explore how GPUs can be exploited to accelerate classic image processing. Even though general machine learning and convolutional neural networks are not new approaches to image processing, their importance for life science is increasing.Īs their application is now at hand due to the rise of advanced computing hardware, namely graphics processing units (GPUs), a natural question is if GPUs can also be exploited for classic image processing in ImageJ (Schneider et al. Nowadays, image data scientists join forces with artificial intelligence researchers, incorporating more and more machine learning algorithms into BIA workflows. Run("Make Montage.Modern life science increasingly relies on microscopic imaging followed by quantitative bioimage analysis (BIA).
0 Comments
Leave a Reply. |
AuthorWrite something about yourself. No need to be fancy, just an overview. ArchivesCategories |