This repository is the official version of ecSeg, a tool used to quantify ecDNA in DAPI-stained images. It also has an extension to analyze FISH probes. Please cite (Bibtex at the bottom) the following reference if using ecSeg in your work:

Rajkumar, U. et al. ecSeg: Semantic Segmentation of Metaphase Images containing Extrachromosomal DNA. iScience. 21, 428-435. (2019)

...

This platform requires a modern processor with support for AVX instructions, python 3.5+, and conda installed.

git clone https://github.com/ucrajkumar/ecSeg
cd ecSeg
conda env create -f env.yml
conda activate ecseg

Always make sure the ecseg environment is activated before executing any tasks:

conda activate ecseg

1. Software will only read the .tif images in the input folder.
2. FISH analysis will only consider green and red FISH

Segment metaphase images stained with DAPI. Identified background, nuclei, chromosomes, and ecDNA.

Set parameters in config.yaml under metaseg:

inpath : path to folder containing images (make sure path is incapsulated inside single quotes. e.g. path\to\folder)

1. labels folder - Contains the RGB version of the post-processed segmentation and the raw values saved as a .npy file.
2. dapi folder - Contains the gray-scale dapi version of the images.
3. ec_quantifications.csv - Contains a list of all the images that were processed and the number of ecDNA in each image.

To extract individual classes:

seg_I = np.load('example_results/seg.npy')
background = (seg_I==0)
nuclei = (seg_I==1)
chromosome = (seg_I==2)
ecDNA = (seg_I==3)

Analyze fish interaction on metaphase images stained with DAPI. Supports green and red FISH. make metaseg must be run before make meta_overlay can be executed.

Set parameters in config.yaml under meta_overlay:

inpath : path to folder containing images
color_sensitivity : Sensitivity to FISH color. Value between 0 (most sensitive) and 255 (least sensitive)

2. green folder will contain the gray-scale version of the green fish signal.
3. red folder will contain the gray-scale version of red fish signal.
4. fish_quantification.csv will contain fish interaction stats for each image. Column headers are as follows:
   1. “image_name” - Name of image
   2. “# of ecDNA” - # of ecDNA based on DAPI only.
   3. “# of ecDNA (FISH_color)” - # of ecDNA based on that FISH color only.
   4. “# of ecDNA(DAPI and FISH_color)” - # of ecDNA based on DAPI colocated with that FISH_color
   5. “# of HSR (FISH)” - # of homogeneously stained regions based on FISH_color.

Identifies nuclei in the image and analyzes ratio of fish to dapi pixels. Provides rough approximation of oncogene amplification per cell in interphase images. Supports green and red FISH.

Recommended folder structure:

ecseg
|
|--models
|--|  metaseg.h5
|--|--nuset
|     |  foreground.ckpt.data-00000-of-00001
|     |  foreground.ckpt.index
|     |  ...
|--src
|  |  ...

Set parameters in config.yaml under stat_fish:

inpath : path to folder containing images

normal_threshold: Local Brightness Threshold (Recommended value: 15)
color_sensitivity: Sensitivity to FISH color. Value between 0 (most sensitive) and 255 (least sensitive)
scale: average square root of ratio of nuclei size relative to target_median_nuclei_size (Recommended value: auto)
  
cell_size_threshold_coeff: Minimum (nuclei size / median nuclei size) to consider for min-cut refinement (Recommended value: 1.25)
flow_limit: Min-cut algorithm threshold to seperate neighboring nuclei (Recommended value: 60)

min_score : 0.95
nms_threshold : Threshold to suppress oversegmentation. (Recommended value: 0.01)
scale_ratio : Ratio of cell size to image size. (Recommended value: 0.4)
nuclei_size_t : Size threshold for finding nuclei. 

Note: Nuclei smaller than nuclei_size_t will be considered erroneous signals. 
Rec: cultured images, use nuclei_size_t=5000. For tissue, use nuclei_size_t=500.

1. stat_fish_lsq.csv - Each row represents a single nucleus. Column headers are as follows:

   1. “image_name” - Name of image
   2. “nuclei_center” - Center of each nucleus
   3. “#_FISH_pixels (FISH_color)" - # of FISH pixels inside nucleus
   4. “#_FISH_blobs (FISH_color)" - # of FISH connected components.
   5. “Avg fish intensity (FISH_color)" - Avg intensity of the corresponding FISH pixels inside nucleus
   6. “Max fish intensity (FISH_color)" - Max intensity of the corresponding FISH pixels inside nucleus
   7. “#_DAPI_pixels” - # of DAPI pixels, i.e. size of nucleus.

2. annotated/img_name/img_name_lsq.tif

   1. Visualization of fish foci calls for each fish probe with cell segmentation.

3. annotated/img_name/img_name_original_with_segmentation.tif

   1. Visualization of nuclei segmentation outline overlayed with original image.

4. annotated/img_name/img_name_segmentation.tif

   1. NuSet Binary Segmentation visualization.

5. annotated/img_name/img_name_segmentation_corrected_min_cut.tif

   1. NuSet Binary Segmentation after Min Cut algorithm refinement, where each color represents a distinct nuclei.

Predicts the probability of each nucleus having no amplification, HSR amplification, and ecDNA amplification for the oncogene (i.e. FISH probe) of interest.

Recommended folder structure:

ecseg
|
|--models
|--|  metaseg.h5
|--|  interseg
|--|--nuset
|     |  foreground.ckpt.data-00000-of-00001
|     |  foreground.ckpt.index
|     |  ...
|--src
|  |  ...

Set parameters in config.yaml under interseg:

inpath : path to folder containing images
FISH_color : Fish probe of interest ('green' or 'red')

1. interphase_prediction.csv - Each row represents a single nucleus. Column headers are as follows:
   1. “image_name” - Name of image
   2. “nuclei_center” - Center of each nucleus
   3. “Predictions" - Prediction value in the form of [P(no-amp), P(ecDNA), P(HSR)]

@article{Rajkumar2019,
    author = {Rajkumar, Utkrisht and Turner, Kristen and Luebeck, Jens and Deshpande, Viraj and Chandraker, Manmohan and Mischel, Paul and Bafna, Vineet},
    journal = {iScience},
    title = {{EcSeg: Semantic Segmentation of Metaphase Images Containing Extrachromosomal DNA}},
    url = {https://github.com/ucrajkumar/ecseg},
    year = {2019}
}