Segmentation and morphmetry of vasucular bundle

Tsuyama et al: Quantitative morphological transformation of vascular bundles in the culm of moso bamboo (Phyllostachys pubescens) DOI: 10.1371/journal.pone.0290732

Directory structure

The original dataset, pairs of image and the correponding mask, for building u-net model should be placed in '_original_pdg' directory. Under the job number (ex. 003), following directory will be generated.The 'test' directory contains the target images to be segmented.

.
├── _original_png
├── _run
│   └── 003
│       ├── extracted_VB_images
│       ├── model
│       ├── morphology
│       ├── segmentation
│       └── train
│           ├── image
│           │   ├── 0.png
│           │   ├── 1.png
│           │   └── 2.png
│           └── mask
│               ├── 0.png
│               ├── 1.png
│               └── 2.png
└── _test

Jupyter notebook

001_model_builing.ipynb: Building u-net model

002_segment_analysis.ipynb: Segmentation using u-net model

003_morphometry.ipynb: Measurement of morphological parameters

Flow chart

Step 1. The original microscopic image and the corresponding mask image are divided into 512x512 pixel images. A set of around 200 images was used as a training dataset, followed by augmentation such as shift, scale, rotate, The U-net was successfully trained and a model with 98% accuracy was constantly obtained. Phase 2 The actual test microscopic images were divided into image patches, and for each image patch segmentation was performed by U-net. Finally, patches were stitched into a single image again to complete the process.

Extracted vascular bundles are exemplified in the following images. Those images are sequentially numbered with respect to the from the relative radial distance from the outer surface of the column.

Finally, morphological parameters were obtained using scikit-image package. Typical output was exemplified as follows.

Data used in the paper

26 microscope images and corresponding mask images (1575 × 6150) are available upon request.

Please mail to the corresponding author.

References

The gerenative network for segmentation was referenced from U-Net

U-Net: Convolutional Networks for Biomedical Image Segmentation

Olaf Ronneberger, Philipp Fischer, Thomas Brox: Medical Image Computing and Computer-Assisted Intervention (MICCAI), Springer, LNCS, Vol.9351: 234--241, 2015, available at [arXiv:1505.04597 cs.CV]

U-Net coding

The network structure was constructed by referring to the following website.https://github.com/zhixuhao/unet/blob/master/README.md