Skip to content

Latest commit

 

History

History
70 lines (53 loc) · 2.64 KB

README.md

File metadata and controls

70 lines (53 loc) · 2.64 KB

AMAP

Automatic Morphological Analysis of Podocytes (AMAP) - method for segmentation and morphology quantification of fluorescent mocroscopy images of podocytes https://www.biorxiv.org/content/10.1101/2021.06.14.448284v3

AMAP Results

  1. Installation

Inference uses BICO method to accelerate the clustering. bico_source contains the source code that you need to compile next copy BICO executable and CluE library to a folder bico next to the amap folder. The folder hierarchy should look as follows:

./amap
|---amap
|---bico
|  |---BICO_Experiments
|  |---BICO_Quickstart
|  |---cluster
|  |---libCluE.a
|---bico_source
|---sample_images
  1. Generation of train data

Running training requires binarized image and label files that can be efficiently used through the training iterations. We provide a set of sample images and their binar labels in the sample_images folder. Running

python generate_data.py

with default arguments will convert the images in the sample_images folder into the binary format and place them in a data folder.

  1. Training

Training can be performed on cpu or a given number of gpus, defined by the -g parameter. -g 0 will perform the training on the cpu. For example

python train_amap.py -g 2

will use 2 gpus in training. By default training set will be located in the data folder and checkpoints saved in the checkpoints folder according to such a folder structure:

./amap
|---amap
|---bico
|---bico_source
|---checkpoints
|---data
|   |---test
|---sample_images

with test dataset in the data/test folder. These parameters can be also passed as stript arguments -dd and -dc respectively.

  1. Inference

The same -g parameter will define the device for inference. Depending on your compute capacity modify variables in the amap_predict.py script and adapt NPROC_TILE and NPROC_CLUSTER variables. These variables define the number of processes that will postprocess the segmentation results.

By default images are read from sample_images folder and results saved in amap_res folder. You can change these with -i and -o parameters respectively.

  1. Morphometric parameters

Calling

python3 morphometry.py

will infer morphometric parameters on files in the folders set as default:

  • images -i ../sample_images
  • segmentation results -p ../amap_res
  • output folder -o ../amap_res/morphometry

This script estimate individuals foot process parameters, slit diaphragm parameters, and joint results that are saved in the output folder as all_params.xls file. Results of each step are saved in the same folder.