seg-to-nephrometry

Estimating nephrometry scores based on segmentations of tumors and affected kidneys.

Installation

cd /path/to/seg-to-nephrometry/
pip3 install -e .

Usage

Class definitions

All voxels labeled with class 1 are treated as kidney
All voxels labeled with class 2 are treated as tumor
All voxels labeled with class 3 are treated as cyst which is handled identically to kidney
All other values (0, 4, 5, ...) are treated as background

Command Line Interface

scripts/get_nephrometry.py [-h] --img-pth IMG_PTH --seg-pth SEG_PTH --out-json-pth OUT_JSON_PTH

options:
  -h, --help            show this help message and exit
  --img-pth IMG_PTH     Path to the image pixel data
  --seg-pth SEG_PTH     Path to the segmentation of the image
  --out-json-pth OUT_JSON_PTH
                        Path to the output file (will be overwritten if exists)

Example Usage

The following shell commands...

cd /path/to/seg-to-nephrometry/
python3 scripts/get_nephrometry.py \
    --img-pth /path/to/kits23/dataset/case_00000/imaging.nii.gz \
    --seg-pth /path/to/kits23/dataset/case_00000/segmentation.nii.gz \
    --out-json-pth ./case_00000_nephro.json

Should produce a ./case_00000_nephro.json file like the following:

{
    "cindex": {
        "composite": 2.4379700208996407,
        "radius": 13.905738186793728,
        "z_distance": 33.5,
        "plane_distance": 5.2038639677947485
    },
    "renal": {
        "composite": "5x",
        "R": 1,
        "E": 2,
        "N": 1,
        "A": "x",
        "L": 1,
        "R_cont": 27.811476373587457,
        "E_cont": 0.5973511875671977,
        "N_cont": 13.083594706327965,
        "A_cont": 5.0422282582651965,
        "L_cont": 0.0
    },
    "padua": {
        "composite": "8x",
        "a": 1,
        "b": 2,
        "c": 1,
        "d": 1,
        "e": 2,
        "f": 1,
        "a_cont": 0.0,
        "b_cont": 9.756078138135848,
        "c_cont": 13.112275994990298,
        "d_cont": 13.083594706327965,
        "e_cont": 0.5973511875671977,
        "f_cont": 27.811476373587457,
        "A_cont": 5.0422282582651965
    }
}

Output Format

Each nephrometry score component is returned both as a continuous value (e.g., a_cont) and as a discretized value (e.g., a). The continuous values are the raw measurements that were used by the algorithm to determine which discretized category to use for the composite score.

References

Original paper introducing the segmentation based RENAL score (JUrol) [html]
Paper introducing continuous version of RENAL score (Gold Journal) [html]
Paper describing the KiTS19 challenge (MedIA) [pdf] [html]
Paper describing the C-Index and PADUA scores (BJU International - in press)

neheller/seg-to-nephrometry