Quantitative image processing.
Light microscopy is a fundamental technology in life sciences, facilitating quantitative and dynamic assessments of fluorescent molecule quantities and their cellular localization with minimal perturbation. The development of linear detectors for digital imaging allowed for direct access to quantitative data and lead to an explosive growth in the field of bioimage informatics. With the recent breakthroughs in machine learning and computer vision and ubiquitously available computational power, the instant recognition of even complex phenotypes is theoretically accessible to anyone. However, the refinement of these techniques to suit individual experimental assays still requires dedicated efforts involving often preprocessing of image data and elaborate training and refinement of the model. Also, the bottleneck towards experimental results remains the extraction of quantitative data from the images after phenotypes or specific cellular structures have been identified in the deep-learning procedure. In practice, image data are often channeled through a suite of several software solutions dedicated to individual steps, leading to an inefficient and not always reproducible image analysis process.
Few laboratories have the expertise to assemble complete image to data pipelines for every assay they aim to use, even fewer have the resources to generate true end-to-end data processing solutions that automate key assays in a reproducible way that is also portable to other laboratories. Both these approaches we aim to offer with the Z02 project through a central facility by providing tailored assistance for image processing. We will inform about new solutions becoming available in the field and we will train and coach researchers in their application. Finally, a major focus will be the generation of a GUI-based end-to-end image analysis software as it is mandated by several projects in TRR186.
The Z02 project thus aims to connect advanced imaging techniques and assays as they are ubiquitously used in TRR186 with the latest in practical applications for the analysis of microscopy data. Furthermore, as the volume and complexity of microscopy data grow, the project provides necessary user-friendly tools and systems that require minimal user interaction. Therefore, leveraging the advancements in artificial intelligence and machine learning, a primary objective of the Z02 project is to devise machine-learning-driven solutions for high-throughput analysis of extensive multidimensional datasets, especially in scenarios where the need of dedicated solutions is paramount.
The Z02 project is hosted by the Ewers laboratory, which has longstanding expertise in multidimensional live cell imaging, single molecule tracking, and super-resolution microscopy. In this central platform of this expertise has been extended to the development of novel machine learning-based data analysis solutions and other high throughput data analysis approaches developed in the last funding period. Many more solutions we aim to implement in the coming funding period as detailed in the work program.
In summary, the project Z02 aims to:
i) assist in the development of analysis routines for image analysis and to develop such sophisticated routines for TRR186 projects,
ii) provide image analysis and programing training courses to TRR186 collaborators,
iii) establish the necessary technical infrastructure for image analysis, encompassing computational resources and software,
iv) provide the technological background for secure and seamless data sharing and storage,
v) distribute information and updates about pioneering image processing solutions and available resources, with a specific emphasis on solutions powered by machine learning.