Deep Learning Image Classification

An important aspect of biomedical image analysis is the classification of sample material such as different cell types. Deep learning (DL) has emerged as a powerful tool for training highly accurate classification models. However, accessing and using DL requires distinct programming skills. AIDeveloper is a software that implements a graphical user interface (GUI) for training, evaluating, and applying neural nets for image classification. AIDeveloper has a focus on biomedical applications, provides a user-friendly interface, and automatically documents the workflow to allow for reproducibility. Similar to napari, AIDeveloper provides interactive visualization tools to generate a tight link between user settings and their effect on image data or on model training behavior. Since AIDeveloper is optimized for classifying images, translating it into a plugin for napari would be ideal, as napari offers performant image visualization of large datasets. The combination of interactive Deep Neural network (DNN) training and image visualization has great potential as users can immediately observe the effects after modifying hyper-parameters or image processing settings. Such a tool can help to uncover errors and could potentially lead to new discoveries.

The team will employ napari-AIDeveloper to train DNNs for discrimination of different mutants, based on their subtle morphological differences. The models will help identify properties of specific subpopulations and refine the follow-up image analysis pipeline. Based on these datasets, which contain millions of images, the project will introduce the plugin to the community and prepare corresponding tutorials and a publication for a scientific journal.