Modular training resources for bioimage analysis

Modular training resources for bioimage analysis Christian Tischer4*, Antonio Politi9*, Tim-Oliver Buchholz1, Elnaz Fazeli2, Nicola Gritti3, Aliaksandr Halavatyi4, Sebastián González-Tirado5, Julian Hennies4, Toby Hodges6, Arif Khan4, Dominik Kutra4, Stefania Marcotti7, Bugra Oezdemir8, Felix Schneider4, Martin Schorb4, Anniek Stokkermans10, Yi Sun4, Nima Vakili4 1Facility for Advanced Imaging and Microscopy, Friedrich Miescher Institute for Biomedical Research, Fabrikstrasse 24, 4058 Basel, Switzerland 2Biomedicum Imaging Unit, Faculty of Medicine and HiLIFE, University of Helsinki, Finland 3European Molecular Biology Laboratory, Carrer del Dr. Aiguader 88, 08003 Barcelona, Spain 4European Molecular Biology Laboratory, Meyerhofstraße 1, 69117 Heidelberg, Germany 5Institute for Computational Biomedicine, Heidelberg University, Im Neuenheimer Feld 130.3, 69120 Heidelberg, Germany 6The Carpentries, c/o Community Initiatives, 1000 Broadway, Suite #480, Oakland, CA 94607, USA 7Randall Centre for Cell and Molecular Biophysics, King’s College London, London, UK 8Euro-BioImaging Bio-Hub, EMBL, Meyerhofstraße 1, 69117 Heidelberg, Germany 9Facility for Light Microscopy, Max Planck Institute of Multidisciplinary Sciences, Am Fassberg 11, 37077 Göttingen 10Hubrecht Institute, Uppsalalaan 8, 3584 CT Utrecht, the Netherlands * corresponding to [email protected] and [email protected] Abstract Modern microscopy enables us to measure structural and dynamical properties of many biological processes and is therefore an indispensable research tool. However, the amount and complexity of the produced imaging data is steadily increasing. Thus, handling the data as well as reproducibly and automatically extracting accurate scientific information requires dedicated “bioimage analysis” expertise. To facilitate the dissemination of this ubiquitously required expertise we developed an open-access bioimage analysis training resource. The resource is designed to help trainers to design and run courses on bioimage analysis for life scientists. The material is modular where each module covers one concise topic and provides corresponding activities. The activities can be executed using various popular software packages (e.g. ImageJ, Python). The material is hosted on a public software repository allowing the bioimaging community to readily contribute new training modules or improve existing modules. Within the last three years, the material has been used by several trainers in numerous courses and continuously improved. 1 Graphical abstract Introduction Microscopy is a key technology driving biological research and a large and diverse software ecosystem exists to analyse microscopy images. Software solutions can be open-source or commercial, and vary from general to specific use-cases, ease of use, levels of documentation, and accessibility (Haase et al., 2022). Despite these available tools, the complexity and size of the corresponding imaging data can make it challenging to efficiently extract bio-physically meaningful information from the acquired images in a quantitative and reliable/reproducible manner. As a consequence, the bio-computational discipline “bioimage analysis” has emerged (GloBIAS, 2024, NEUBIAS, 2024). Scientists performing bioimage analysis have a diverse background and need to acquire a theoretical foundation on image formation as well as image processing and analysis algorithms. Furthermore, scientists must acquire the necessary skills to effectively and adequately use various software tools for image analysis tasks. During the last decade, online resources for self-study have emerged in the form of videos, slides, interactive books, and Massive Open Online Courses (MOOCs) (e.g., (Bankhead, 2024, Haase, 2020, Image.sc, 2019, Seitz et al., 2019, Jones et al., 2022)). In addition, numerous in-person or online bioimage analysis courses have been delivered and sometimes recorded (e.g., (NEUBIAS, 2022, ZIDAS, 2024)). Live online or in-person courses are very valuable as they provide a platform for networking, interacting with expert trainers, and addressing specific 2 questions with respect to learners’ needs. We found that an open-access resource allowing bioimage analysis trainers to efficiently design and deliver such courses to various target audiences would be helpful and is currently missing (Haase et al., 2024, Sivagurunathan et al., 2024). Such a resource should help and simplify the much-needed dissemination of bioimage analysis skills. To address those needs we started in early 2019 to create an open-access online resource for bioimage analysis training accessible via https://neubias.github.io/training-resources/ (Tischer et al., 2024). Since then, this resource has been used in at least 25 courses with in total nearly 600 participants (Supplemental Table 1). With this publication we aim to further spread the use of this material and also kindly invite the bioimaging community to contribute to any aspect of this resource. In the following sections we will describe the rationale, content, implementation details and usage instructions. Training resource design and organization We aimed for a resource to facilitate delivering in-person or online live courses. The material should include activities to illustrate use cases. Given the large number of image analysis software, we wanted to separate an activity from a particular software implementation. Finally, the material should allow bioimage analysts to taylor courses for different target audiences. This led to the following design constraints: open-access online hosting of the material, modularity, interactivity, usability within a live context, and easy extension by the community. Design principles We found that existing online material is primarily conceived as self-study resources and does not yet fulfil all our prerequisites. While a simple collection of slides could be used within a live teaching context, they are rather static, often lack integrated activities, and are hard to develop/maintain as a community. Pre-recorded videos are inappropriate for a live, interactive setting and also lack integrated activities. MOOCs provide comprehensive material often enriched with videos, exercises, and assessments that are best suited for self-paced study (Seitz et al., 2019, Jones et al., 2022). The interactive book by P. Bankhead (Bankhead, 2024) contains comprehensive explanations, example images, and activities, representing a good reference for bioimage analysis concepts. While all the aforementioned material are great resources, they impose a pre-defined order of topics and lack modularity. Modifications or additions of new concepts may therefore require major revisions; in addition, in our view, the existing resources are not designed to support agile contributions by a community of bioimage analysts. 3 We decided to base our training resource on the didactically excellent training material of The Carpentries (https://carpentries.org/) (Backhaus et al., 2024, Brown et al., 2023). A Carpentries expert has been a critical contributor to the initial development of the bioimage analysis training resource. Among other things, lessons by The Carpentries are aimed at novices as well as experts and gradually build knowledge, avoiding too much information at once. A lesson is composed of several so-called episodes, lasting for 20 to 60 minutes with clearly stated learning objectives. Each episode includes activities suited for live demonstration/live-coding, so to explain every step of the process, and formative assessments to identify misconceptions. The Carpentries provide a course on image-processing with python that can be used in a live context with a variety of example images and activities (Meysenburg et al., 2023). Our material differs from The Carpentries' primarily in its increased modularity and specific application to bioimaging data. Unlike The Carpentries, which delivers predefined lessons, we designed small, independent training modules (reminiscent to Carpentries episodes) that can be flexibly assembled into various courses — differing in length, target audience, and software packages covered (Fig. 1). To manage and optimize cognitive load (Sweller, 1988), each training “module” focuses on just one or very few new concepts that should be taught within 30-45 min. This inherently restricts the complexity or number of concepts included within a single module. A 'concept' in this context refers to a generic, commonly used principle or method in bioimage analysis that is agnostic to specific software implementations, such as 'object shape measurement'. A module is designed to provide all the necessary information to effectively motivate, explain (e.g., through figures, concept maps), and apply (e.g., using example images and code) its content. Figure 1: Training resource design: Hierarchy of the bioimage analysis training course. A course is designed by assembling several existing modules. Each module explains a concept/method and provides activities. An activity has general instructions, an example image, and different software implementations. 4 Module structure The above design principles, refined via repeated teaching of different modules, led to the subdivision of each module into nine distinct sections, as detailed in Figure 2 and Table 1. After the Prerequisites section, the following four sections (Fig. 2A-C) serve to teach the concept in a lecture style presentation, where the Concept map and a Figure are used as a visual support for the instructor to explain the concept to the learners. Some additional detailed explanations could be added after the main figure. The main section of a module is Activities that demonstrates different applications of the concepts that are taught in the module. Each activity is described in general software-agnostic terms and has linked example data that can be readily used. The instructor and learner can then select instructions for specific software packages (Fig. 2D, E). This separation of general concepts from specific implementations was a major design decision for this resource. This makes it possible to continuously add new software implementations of an activity, without needing to change much of the generic teaching content. In addition, this design also increases the didactic quality of the material, because it forces the trainer and students to disentangle universally applicable concepts and methods such as, e.g. “connected component analysis” and “object shape measurements”, from specific implementations such as ImageJ’s Particle Analyzer (Schneider et al., 2012) or scikit-image’s regionprops (van der Walt et al., 2014). 5 Figure 2: Structure of a training module: This figure shows screenshots of a module website and activity. (A) Knowledge required to follow a given module is provided via links to other modules (Prerequisites). Learning objectives and Motivation provide the why and what will be learned during the module. The (B) Concept Map and (C) Figure visuals provide and overview of the key concepts. Additional explanations may appear after the main figure (not shown here). In Activities (D) the activity instruction and its implementations provide the necessary information to showcase a concept using a specific software (e.g. python and scikit-image/napari). (E) Example output generated when executing an activity. (F) Formative assessments are provided at the end of a module to reinforce what has been learned. 6 Module section Explanation Prerequisites Contains links to other modules of our material, defining the content that learners should know before starting Learning objectives Defines the skills that learners will acquire Motivation Explains why it is important to acquire these skills Concept map A schematic drawing that defines the core new concepts that are taught Figure A figure illustrating the teaching content Explanations (optional) Additional details typically used to help self-study and add mathematical context Activities Interactive instructions for applying the concepts taught in various software platforms Assessment Questions and answers for testing the conceptual understanding Follow-up material Links to suitable follow-up training modules and extra-curricular material Table 1: Training module sections: A training module is composed of difference sections to introduce the concept, provide examples, and further reading. The subsequent “assessment” section (Fig. 2F) comprises short quizzes that allow the participants to test their understanding of the main concepts. For instance, a question and answer could be: ”Q: What is the typical data type of a label mask image? A: Typically, label mask images are of positive integer data types (unsigned integer 8 or 16 bit) with zero being the background label”. Finally, the Follow-up Material section provides links to additional modules that can be learned after the current module or external resources. In our experience, this predefined module structure significantly facilitates the addition of new material. While it still forces the contributor to carefully think about what they actually want to teach in the module, it frees one from thinking too much about how to structure and format the material. Resource content Over the last 5 years we created ~50 teaching modules (Supplemental table 2). Supplemental video 1-3 illustrate the content of the training material and how to teach a module. The current material covers content typically taught in introductory and intermediate bioimage analysis courses. Among others, the resource includes modules on image data handling and the understanding of the different data formats, image data inspection, image filtering and 7 thresholding, segmentation and measurements, as well as basic scripting, batch processing, and cloud based computing. We also included so-called workflows, where learners, starting from example images, apply a sequence of methods to solve an image analysis task and obtain a result table. Finding suitable example image data for teaching a certain concept is a major effort. The repository contains small example image data that have been carefully selected to fulfil this need. The activities are designed for different levels of understanding and complexity. 8 Figure 3: Module dependencies and courses: This figure shows screenshots of the resource home page. (A) Alphabetical list of modules. (B) Network of dependencies between modules. Each module is a node and the Prerequisites define the edges of the graph. By clicking on one of the nodes (arrow) the dependencies are highlighted (green). This information can be used to define a set of modules for a course. (C) Link to course suggestions that can 9 References Backhaus, A., Becker, E., Bellini Saibene, Y., Brooks-Kieffer, J., Companjen, B., Curty, R., Davey, R., Hertweck, K., Hodges, T., Kamvar, Z., Khan, B. J., Kohrt, F., Kuzak, M., McClatchy, S., Michonneau, F., Nenadic, A., Stevens, S., Takemon, Y., Trizna, M., Trusler, A., ven der Burg, S., Allegra, V., Wilson, G. & Word, K. (2024) The Carpentries Collaborative Lesson Development Training. Software Carpentry Foundation, doi: 10.5281/zenodo.8410625, url: https://carpentries.github.io/lesson-development-training/. Bankhead, P. (2024) Introduction to Bioimage Analysis. doi, url: https://bioimagebook.github.io/. Brown, S. M., Dennis, T., Perez-Suarez, Porter, N., Wheeler, J. & Word, K. (2023) The Carpentries Instructor Training. Software Carpentry Foundation, doi: https://doi.org/10.5281/zenodo.7612756, url: https://carpentries.github.io/instructor-training/. EMBL-EBI (2024) BioImage Archive. doi, url: https://www.ebi.ac.uk/bioimage-archive/. GloBIAS (2024) Global BioImage Anlaysts's Society. doi, url: https://www.globias.org. Haase, R. (2020) Lecture BioImage Analysis. Youtube, doi, url: https://www.youtube.com/playlist?list=PL5ESQNfM5lc7SAMstEu082ivW4BDMvd0U. Haase, R., Fazeli, E., Legland, D., Doube, M., Culley, S., Belevich, I., Jokitalo, E., Schorb, M., Klemm, A. & Tischer, C. (2022) A Hitchhiker's guide through the bio-image analysis software universe. FEBS Lett, 596, 2472-2485, doi: 10.1002/1873-3468.14451, url: https://www.ncbi.nlm.nih.gov/pubmed/35833863. Haase, R., Tischer, C., Bankhead, P., Miura, K. & Cimini, B. (2024) A Call for FAIR and Open-Access Training Materials to advance Bioimage Analysis. OSF Preprints, doi: 10.31219/osf.io/2zgmc, url. Image.sc (2019) Bioimage Analysis: Recommended Reading and Viewing - Image Analysis., doi, url: https://forum.image.sc/t/bioimage-analysis-recommended-reading-and-viewing/28051. Jekyll (2024). 4.3.4 ed., doi, url: https://jekyllrb.com/. Jones, S., Burns, I., Armstrong, B., Wang, A., Rich, M. & Thompson, M. (2022) Image Processing for Engineering and Science Specialization. Mathworks, Coursera, doi, url: https://www.coursera.org/specializations/image-processing. Meysenburg, M., Durham Brooks, T., Kutra, D., Pape, C. & Becker, E. (2023) Image Processing with Python. doi, url: https://datacarpentry.github.io/image-processing/. NEUBIAS (2022) NEUBIAS Academy. doi, url: https://eubias.org/NEUBIAS/training-schools/neubias-academy-home/. NEUBIAS (2024) Network of European BioImage Analysts. doi, url: https://eubias.org/NEUBIAS/. Seitz, A., Guiet, R., Burri, O. & Chiaruttini, N. (2019) EPFLx: Image Processing and Analysis for Life Scientists. EPFL, edX, doi, url: https://www.edx.org/learn/image-analysis/ecole-polytechnique-federale-de-lausanne-ima ge-processing-and-analysis-for-life-scientists. Sivagurunathan, S., Marcotti, S., Nelson, C. J., Jones, M. L., Barry, D. J., Slater, T. J. A., Eliceiri, K. W. & Cimini, B. A. (2024) Bridging imaging users to imaging analysis - A community survey. J Microsc, 296, 199-213, doi: 10.1111/jmi.13229, url: https://www.ncbi.nlm.nih.gov/pubmed/37727897. Sweller, J. (1988) Cognitive Load During Problem Solving: Effects on Learning. Cognitive Science, 12, 257-285, doi: https://doi.org/10.1207/s15516709cog1202_4, url. Tischer, C., Politi, A., Buchholz, T., Fazeli, E., Gritti, N., Halavatyi, A., Gonzalez Tirado, S., Hennies, J., Hodges, T., Khan, A., Kutra, D., Marcotti, S., Oezdemir, B., Schneider, F., Schorb, M., Stokkermans, A., Sun, Y. & Vakili, N. (2024) Modular training resources for 16 bioimage analysis. 1.0.1 ed. Zenodo, doi: 10.5281/zenodo.14264885, url: https://github.com/NEUBIAS/training-resources/. ZIDAS (2024) Switzerland's Image and Data Analysis School. doi, url: https://www.zidas.org/. Supplemental material The supplemental material is available at: Preprint Supplemental Table 1: List of courses 2021-2024 The table lists courses that have being held in the last 4 years using the bioimage analysis training resources. Supplemental Table 2: List of modules The table lists all the consolidated modules in the resource. 17 Supplemental Video 1: Bioimage analysis training resource: How to A video illustrating how to navigate through the website of the training resource. Supplemental Video 2: Digital image basics: Introduction Video showing the teaching of the module Digital image basics. Supplemental Video 3: Digital image basics: 2d python Video schowing the teaching of the first activity in the module Digital image basics. 18