Maxime Délitroz: Automated Segmentation of High-content Fluorescent Microscopy Data of Developing MN in Culture
Posted on Tue 30 August 2022 in theses
Visual comparison between Little U-Net model predictions and ground truth annotation. (Image): contrast-enhanced image. (Ground truth): ground truth segmentation. (Model segmentation): segmentation generated by Little U-Net model. (Errors): error image where false positives and false negatives are represented in blue and red respectively.
In this thesis, we sought at developing precise neuron segmentation in order to support future study on the morphological phenotype of amyotrophic lateral sclerosis. We tackled this task by developing two different segmentation pipelines: the first one based on classical image and signal processing, and the second one based on a fully convolutional neural network architecture, capitalizing on our first pipeline segmentation's to bypass the need of labeled training data. We shown that both our method outperformed other segmentation tools used in the literature and shed light on important requirements for training neural network on segmentation tasks.
Available Materials