AbyssEye
Microscopy workflow tool for TIFF datasets, ROI extraction, and AI-based analysis, developed with JAMSTEC and the authors of Nishimura et al. (Scientific Reports, 2025).
A browser app for scientists to manage microscopy images, run AI classification on detected cell candidates, manually review and correct results, and export labeled data to retrain better models, all stored locally.
AbyssEye is a browser-based application for scientists who work with microscopy images. It was developed with JAMSTEC (Japan's oceanography research agency) as part of a study on detecting microbial cells in deep seafloor sediment samples, and it is built around the workflow described in that 2025 Scientific Reports paper. The core problem it addresses is that managing microscopy images, running AI-based analysis on them, reviewing the results, and preparing data for further model training are steps that normally happen across separate tools. AbyssEye puts all of them in one operator-facing interface.
The workflow moves through several stages. You start by organizing your microscopy image files (TIFF format) into projects. The app extracts regions of interest from those images, meaning it identifies candidate areas worth examining, and stores them in a local database. You then run inference, which means asking a trained AI model to classify each candidate. After that, a review step called DeepScan lets you go through the AI's results, correct any mistakes manually, and label the data properly. The corrected data can then be exported to train an improved model, which feeds back into the next round of analysis.
The application has a Python backend (using FastAPI) and a React-based frontend. It stores data locally in SQLite databases rather than sending anything to external servers. Trained AI models, image files, and databases all stay on the machine running the app. Docker deployment files are included for teams who want a more controlled setup.
The repository does not include any pretrained models or sample images. You supply your own TIFF files and your own models through the app's model manager. The software license is listed as pending, so the terms are not yet finalized.
The app requires Python 3.11 and Node.js 20 or later. A quick-start section in the README walks through starting the backend and frontend locally.
Where it fits
- Organize deep-sea microscopy TIFF images into projects and run an AI model to classify candidate microbial cells
- Review AI predictions one-by-one using the DeepScan interface and manually correct misclassified cells
- Export corrected, labeled microscopy data to retrain and improve your cell-detection model in the next round