Synthesis and Segmentation of 3D Fluorescence Microscopy Image Data
Motivation
Developmental biology focuses on studying the processes of how cells migrate, interact and divide. Specifically, cell location and morphology, analysed by detection and segmentation approaches, constitute one of the key tasks in understanding the underlying principles of development. Technological advances allow biologists to capture and store unprecedented amounts of highly detailed microscopy image data, which renders the application of automated approaches essential. However, the potential of recent deep learning-based approaches is severely limited by the scarcity of fully-annotated image data sets. Consequently, this project focuses on two different research topics:
- Robust and generalist deep learning-based instance segmentation approaches, allowing to obtain accurate results with limited available data.
- Simulation and synthesis approaches for an automated generation of fully-annotated data sets rendering human annotation efforts obsolete.
Scientific Questions
To increase robustness and generalizability of instance segmentation approaches, new principles of deep learning-based approaches are integrated into current state-of-the-art approaches and new concepts to increase robustness are developed.Simulation and synthesis approaches with improved sampling speed and optimized control over generated structures are studied, allowing for the generation of fully-annotated image data sets on demand. Generated data and their practical usage are assessed by determining if they can replace manually annotated image data sets in segmentation training pipelines.
Partners
- G. U. Nienhaus and Dr. A. Kobitski, Institute of Applied Physics, Karlsruhe Institute of Technology (KIT)
- Prof. U. Strähle and Dr. M. Takamiya, Institute of Biological and Chemical Systems – Biological Information Processing,
- Karlsruhe Institute of Technology (KIT)
- apl. Prof. R. Mikut, Institute for Automation and Applied Informatics, Karlsruhe Institute of Technology (KIT)
External Funding
- DFG Research Grant, “On-the-fly data synthesis for deep learning-based analysis of 3D+t microscopy experiments”, Project nr. 447699143