YeastTube: Revolutionizing Microscopy Data Management with AI and Web Technologies

Session

Poster Session

Authors

Andrey Aristov (1), Madison M Lenormand (1), Guy-Franck Richard (1), Charles Baroud (1)

Affiliations

1. Institut Pasteur

Keywords

webapp, timelapse, segmentation, filtering, yeast

Abstract text

The processing of extensive timelapse microscopy datasets from multi-well microfluidics devices presents a formidable challenge in the field, primarily due to the substantial size of image stacks that are not only time-consuming to read but also difficult to annotate. To address this bottleneck, YeastTube, a cutting-edge web application, has been developed to transform the workflow associated with these datasets.

YeastTube automates the preprocessing steps, such as single-well stack extraction, cell segmentation, using Cellpose and intensities measurements. These complex processes are conducted in the background, streamlining the initial stages of data handling without user intervention. The application, however, prioritizes user interaction for refining the outcomes of automated segmentation. It introduces intuitive tools for user-friendly filtering of segmentation results, including a lasso tool for selecting only the cells from a single lineage and removing unwanted segmentations using the plot of cell flurescence intensities. This approach enables precise refinement of automated analyses, ensuring the accuracy and relevance of data for research purposes.

By seamlessly integrating automated processing with interactive tools for data refinement, YeastTube significantly reduces the time required for data annotation from weeks to merely an afternoon per acquisition. This efficiency gain not only accelerates the pace of research in our lab but also fosters a collaborative environment. Each user's input, including detailed annotations, is automatically saved alongside the raw data, ensuring that every piece of information is readily available for future reference. 

YeastTube's development also highlights the instrumental role of ChatGPT in rapidly assembling a custom-tailored web application. By leveraging LLM capabilities, we were able to adopt a 'Lego-like' approach to software development, assembling the necessary components with precision and speed. This innovative use of AI not only facilitated the swift creation of a highly specialized tool for microscopy data management but also exemplifies the transformative potential of AI in scientific research.