Deep Learning for automatic signal enhancement in biological imaging

Object detection in biological imaging has recently made tremendous progress, mainly through the use of Deep Learning. However, there are still many difficulties related to the volume of images and the quality of the signal in these images. We are particularly interested in the challenge of detecting small objects such as RNA spots in single molecule FISH microscopy images. These are very small objects, only a few pixels in size, for which automatic detection methods are difficult to implement due to the lack of context.

More specifically, we focus our efforts on enhancing the biological signal present in smFISH microscopy images, i.e. the fluorescent probes, to facilitate a posteriori detection. To this end, we are developing neural networks that allow the integration of classical image analysis methods such as wavelets to automate the analysis process of biological images. For smFISH microscopy, we have developed DeepSpot, a neural network that enables seamless detection of mRNA spots, streamlining studies of localized mRNA expression in cells.