CellMimics: A microscopy image generator for AI image segmentation model training

Modern microscopy technologies are generating massive datasets of cellular dynamics at unprecedented resolution. However, we face a significant challenge in transforming these views into biological discoveries, and new and effective drug targets for curing disease. Current image processing methods struggle to handle large datasets efficiently, while AI models require extensive manually curated data for training.

This project introduces CellMimics, an innovative software program designed to generate synthetic microscopy images at scale, revolutionising the approach to AI model training in cell biology. CellMimics leverages the Unity 3D video game engine to create realistic synthetic cell microscopy videos with annotations, customisable for various microscopy modalities, cell types, and experimental settings.

By using these synthetic datasets and annotations for AI model training, CellMimics eliminates the need for time-consuming manual annotation, reducing the data preparation process from months to days.

^{How CellMimics will accelerate microscopy image analysis. After tuning CellMimics to generate synthetic data that mimic raw microscopy images, CellMimics generated images train state-of-the-art AI models to accurately segment and analyse cellular dynamics.}

The research team aims to develop CellMimics into a user-friendly, all-in-one application for generating synthetic data tailored to specific research needs. The software will support 2D and 3D movie generation for a wide array of microscopy techniques, including 4D lattice light sheet microscopy. This approach will significantly accelerate the segmentation and analysis of complex cellular dynamics from large biological microscopy data.

MDAP's expertise will be crucial in developing CellMimics’s underlying Unity + Python framework into a web deployable software program, enabling researchers without technical expertise to customise CellMimics for their specific applications. We will also explore integration of an AI model into CellMimics. This would enable users to train the model on custom synthetic data and automatically segment their microscopy images – significantly reducing data processing time.

The project's outcomes will benefit the broader cell biology community by providing a powerful tool for processing microscopy data and enabling new biological discoveries. CellMimics has the potential to revolutionise AI model training in biology, positioning the University of Melbourne at the forefront of biological AI research and opening new avenues for competitive research funding.