Cell segmentation in 3D confocal images using supervoxel merge-forests with CNN-based hypothesis selection

Abstract

Automated segmentation approaches are crucial to quantitatively analyze large-scale 3D microscopy images. Particularly in deep tissue regions, automatic methods still fail to provide error-free segmentations. To improve the segmentation quality throughout imaged samples, we present a new supervoxel-based 3D segmentation approach that outperforms current methods and reduces the manual correction effort. The algorithm consists of gentle preprocessing and a conservative super-voxel generation method followed by supervoxel agglomeration based on local signal properties and a postprocessing step to fix under-segmentation errors using a Convolutional Neural Network. We validate the functionality of the algorithm on manually labeled 3D confocal images of the plant Arabidopis thaliana and compare the results to a state-of-the-art meristem segmentation algorithm.

Accurate 3D Cell Segmentation Using Deep Features and CRF Refinement

“…The L 1 layer of the training set is used to train the 3D U-Net, and L 1 , L 2 and L 3 layers of the testing set are used to evaluate the segmentation performance of the proposed algorithm. Table 1 to 3 show the comparison of the final result using our proposed method and other methods including ACME [12], MARS [11] and a supervoxel-based algorithm [2] on L1 to L3 respectively. With CRF refinement, our model improves approximately 0.028 in the F-score measure on average.…”

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

“…Finally, the transformed distance map and the corrected centroid map are utilized to obtain the cell instances by applying a seeded watershed algorithm [14], again excluding segments intersecting mostly with the identified background. In addition to the seeded watershed approach (SWS), we also adapted our previously presented supervoxel merging approach (SV) to be able to process 3D probability maps [5]. We directly provide the probability maps of the membrane signal as input to the algorithm.…”

“…Existing methods to accomplish this multi-instance segmentation problem use preprocessing strategies like alternate sequential filtering [1], Hessian-based filtering methods [2], anisotropic diffusion filtering [3] or combinations of these approaches [4], to emphasize the membrane structures for further processing. The preprocessed images are then segmented using the watershed transform with seed points located in hminima [1,2], using deformable models [3] or using supervoxel merging strategies [5,6]. To combine advantages of getting information from both nuclei and membranes, superposition of both experimental setups has improved generation of automated results [4,7], requiring twice the amount of data to be processed.…”

“…Moreover, membranes that are parallel to the detection plane are weakly visible. In our previous work [5], we tried to solve this problem by starting with a heavy oversegmentation followed by fusion of identified supervoxels to complete cells. This included CNN-based postprocessing steps to compensate undersegmentation errors.…”

SEGMENT3D: A web-based application for collaborative segmentation of 3D images used in the shoot apical meristem

“…C) An xz-slice of an enhanced (denoised and sharpened) meristem reveals the weak signal in lower parts. Panel D) shows missing cell boundaries, indicated by arrows, in the automatic 3D segmentation, in red, created by the method in [2]. the regulation mechanisms that govern meristem growth and consequently the crops that depend on it.…”