CryoET technology is generating significant amounts of 3D tomogram data containing rich information about subcellular components, their structural appearance, and interactions. However, analysis of this data through manual annotation is very time-consuming, while the 3D nature of the data poses challenges for automated, machine learning-based feature segmentation and analysis. In this talk, I will present work on developing computer vision methods for automated segmentation of features in cryoET tomograms. I will discuss our computer vision approach for 3D segmentation and analysis of mitochondrial volumes and granules in iPSC-derived neurons differentiated from Huntington's Disease patient samples. I will additionally present ongoing work on developing methods for increasingly label-efficient segmentation of features, that can enable scaling computer vision analysis of 3D tomograms to significantly larger numbers of features.
Copyright. All Rights Reserved.