This microtubule does not exist: Super-resolution microscopy image generation by a diffusion model
Generative models, such as diffusion models, have made significant advancements in recent years, enabling the synthesis of high-quality realistic data across various domains. Here, we explore the adaptation and training of a diffusion model on super-resolution microscopy images from publicly available databases. We show that the generated images resemble experimental images, and that the generation process does not memorize existing images from the training set. Additionally, we compare the performance of a deep learning-based deconvolution method trained using our generated high-resolution data versus training using high-resolution data acquired by mathematical modeling of the sample. We obtain superior reconstruction quality in terms of spatial resolution using a small real training dataset, showing the potential of accurate virtual image generation to overcome the limitations of collecting and annotating image data for training. Finally, we make our pipeline publicly available, runnable online, and user-friendly to enable researchers to generate their own synthetic microscopy data. This work demonstrates the potential contribution of generative diffusion models for microscopy tasks and paves the way for their future application in this field.