Deep learning-driven automated high-content dSTORM imaging with a scalable open-source toolkit.

Abstract

Super-resolution microscopy offers the ability to visualize molecular structures in biological samples with unprecedented detail. However, the full potential of these techniques is often hindered by a lack of automated, user-independent workflows. Here, we present an open-source toolkit that automates dSTORM super-resolution microscopy using deep learning for segmentation and object detection. This standalone program enables reliable segmentation of diverse biomedical images, even in low-contrast samples, surpassing existing solutions. Integrated into the imaging pipeline, it rapidly processes high-content data in minutes, reducing manual labor. Demonstrated by biological examples, such as microtubules in cell culture and the βII-spectrin in nerve fibers, our approach makes super-resolution imaging faster, more robust, and easy to use, even by nonexperts. This broadens its potential applications in biomedicine, including high-throughput experimentation.