Semi-supervised segmentation and tracking algorithms for cell segmentation
We present a novel weakly supervised 3D nuclei segmentation method that consists of deep learning based nuclei detection, watershed segmentation, and a boundary correction algorithm using supervoxels. Additionally, we present a simple and efficient graph-based tracking algorithm utilizing relative nuclei location information extracted from the adjacency graph.
For more details about our methodology, please refer to our paper.
The performance of our proposed method on CTC 2020 dataset is shown in the following table:
| Dataset | DET | SEG | TRA | OP_CSB | OP_CTB |
|---|---|---|---|---|---|
| Fluo-N3DH-CE | 0.927 | 0.705 | 0.895 | 0.816 | 0.800 |
| Fluo-C3DL-MDA231 | 0.839 | 0.545 | 0.795 | 0.692 | 0.670 |
The system was employed for our research presented in [1], where we propose a novel semi supervised nuclei segmentation method utilizing Simple linear Iterative Clustering (SLIC) boundary adherence and a graph-based tracking algorithm utilizing relative cell location information. If the use of the software or the idea of the paper positively influences your endeavours, please cite [1].
[1] S. Shailja, Jiaxiang Jiang, and B.S. Manjunath, "Semi supervised segmentation and graph-based tracking of 3D nuclei in time-lapse microscopy." Submitted to IEEE ISBI 2021.
The command
./Fluo-MDA231.sh indir outdirseg outdirtrack datatype
runs the segmentation and tracking pipeline on all tif stacks in indir and saves the label masks in outdirseg and outdirtrack respectively. The dataset can be passed through datatype argument.
./Fluo.sh ./01 ./01_RES_SEG ./01_RES_TRACK "N3DCHCE
./01/
├── t003.tif
├── t008.tif
├── t013.tif
├── t018.tif
├── t023.tif
├── t028.tif
./01_RES_SEG/
├── mask003.tif
├── mask008.tif
├── mask013.tif
├── mask018.tif
├── mask023.tif
├── mask028.tif
./01_RES_TRACK/
├── res_track.txt
├── mask008.tif
├── mask013.tif
├── mask018.tif
├── mask023.tif
├── mask028.tif