CNN-based Preprocessing to Optimize Watershed-based Cell Segmentation in 3D Confocal Microscopy Images

• DOI: 10.48550/arxiv.1810.06933
• Publication Date: 2018-01-01
• Authors: Dennis Eschweiler; Thiago V. Spina; Rohan Choudhury; Elliot M. Meyerowitz; Alexandre Lobo‐da...; Johannes Stegmaier
• Keywords: Leverage (statistics); Discriminative model

The quantitative analysis of cellular membranes helps understanding developmental processes at the level. Particularly 3D microscopic image data offers valuable insights into cell dynamics, but error-free automatic segmentation remains challenging due to huge amount generated and strong variations in intensities. In this paper, we propose a new approach which combines discriminative power convolutional neural networks (CNNs) for preprocessing investigates performance three watershed-based postprocessing strategies (WS), are well suited segment object shapes, even when supplied with vague seed boundary constraints. To leverage full potential watershed algorithm, multi-instance problem is initially interpreted as three-class semantic problem, turn well-suited application CNNs. Using manually annotated confocal microscopy images Arabidopsis thaliana, show superior proposed method compared state art.