Fully automatic segmentation and quantification of lightsheet microscopy images

Lightsheet microscopy (LSM) allows 3D imaging of whole embryo samples at an early developmental stage with a spatial resolution that can display single cells. Due to the large number of acquired images by LSM, the large size of these images, and the different artifacts that can be present, manual analysis of LSM images is ineffective. The aim of this research project is to develop machine learning techniques for image quality enhancement, cell segmentation and classification, and registration of the images for morphometric analysis. We propose and evaluate convolutional neural networks (CNNs) for segmentation of tissues, cells and proliferative cells, providing accurate results. Then, morphological analysis is necessary for the quantitative assessment of the shape variation of the head and face, among different developmental stages and genetic alterations. Here, we propose to develop and evaluate ML techniques for registration of highly dynamic LSM images between different specimens, which remains an open problem in the field.