STAGETOOL: Deploying deep learning for a wider audience

Many Euro-BioImaging Nodes are working on innovative image analysis and image data management solutions and the next Euro-BioImaging User Forum is designed to highlight these. Join us for an afternoon of presentations from Euro-BioImaging users and image data/image analysis experts at our Nodes that will provide a compelling overview of the state-of-the-art in Image Data. At this event, Oliver Meikar, University of Tartu, and Junel Solis, image analyst at the Finnish Advanced Microscopy Node, describe their work on StageTOOL, a project aimed to bridge the gap between a research tool intended for a small number of users, and scaling up its deployment to a production environment that increases accessibility while catering to a wider audience.

What: Euro-BioImaging User Forum “Image Data”

When: March 26, 2024, from 14:00-17:00 CEST Where: Online

Full program

Register

Abstract

STAGETOOL: Deploying deep learning for a wider audience

Oliver Meikar, University of Tartu

Junel Solis,

Turku Bioimaging, Finnish Advanced Microscopy Node

STAGETOOL is a convolutional neural network-based approach that automates the classification of seminiferous tubules into developmental stages and epithelial cells into 9 cell types. It was created by reproductive biology researchers and further developed by Turku BioImaging of the Finnish Advanced Microscopy Node (FiAM). Today, a substantial portion of deep learning-based tools remain confined to a limited-use state, usable only as command-line tools with complex dependencies. This FiAM Node project aimed to bridge the gap between a research tool intended for a small number of users, and scaling up its deployment to a production environment that increases accessibility while catering to a wider audience. While a detailed histological analysis of spermatogenesis from DAPI-stained testis cross-sections traditionally requires special training, expertise and laborious manual scoring, STAGETOOL automates this process by classifying tubule stages with an accuracy of 99.1%, and classifying epithelial cell types with F1 scores ranging from 0.80 to 0.98. STAGETOOL can be used to analyze knockout mouse models with spermatogenic defects, for profiling protein expression patterns, and is the first fluorescent labeling-based automated method for mouse testis histological analysis that enables both stage and cell-type recognition. While STAGETOOL qualitatively parallels an experienced human histologist, it outperforms humans time-wise, and represents a major advancement in male reproductive biology research. The FiAM Node standardized the data flow to and from STAGETOOL to allow communication with a bespoke server-based application, developed a web-based user interface, built Docker images, and deployed the entire application stack onto cloud servers. The work performed by FiAM allowed STAGETOOL to progress from a tool with limited command-line usage into an easily accessible web service for researchers, usable from any computer with a browser, without the need for setting up a complex computing environment.