LightMyCells

LightMyCells predicts fluorescence images from label-free transmitted light inputs using self-supervised Vision Transformers, addressing phototoxicity and limited spectral channels in fluorescence microscopy.

Key Features:

• Self-supervised pretraining: Four specialized models, each targeting a different organelle, are pretrained in a self-supervised manner to enhance model generalization.
• Image-to-image prediction: Predicts optimally focused fluorescence images from label-free transmitted light inputs.
• BiaPy integration: The method is integrated within the open-source BiaPy library.

Scientific Applications:

• Cell Painting: Addresses the Cell Painting problem within the LightMyCells challenge at ISBI 2024.
• Cellular analysis: Contributes to image-to-image deep-learning techniques for robust and accurate fluorescence image prediction in cellular analysis.

Methodology:

Self-supervised Vision Transformers are used for image-to-image labeling, with four organelle-specific models pretrained in a self-supervised manner.