2020-2021, Daniel C. Elton

Code is provided to ensure our method can be reproduced. The files containing the trained weights for the two deep learning models, normally stored under log/, are not included.

This method is designed specifically for non-contrast CT.

If you use or build off this code, please cite the following work:

Daniel C. Elton, Evrim B. Turkbey, Perry J. Pickhardt, Ronald M. Summers. "A deep learning system for automated kidney stone detection and volumetric segmentation on non-contrast CT scans". Medical Physics. 2022.

@article{Elton2022KidneyStones,
title = {A deep learning system for automated kidney stone detection and volumetric segmentation on non-contrast CT scans},
journal = {Medical Physics},
volume = {},
issue = {},
pages = {},
year = {2022},
doi = {},
author = {Daniel C. Elton and Evrim B. Turkbey and Perry J. Pickhardt and Ronald M. Summers},
}

Before using, please read the licence and terms of use in "Software Terms of Use-Kidney Stones.pdf"

to run on multiple files:

python detector.py --test_list filename_list.txt

to run on a single file:

python detector.py --single_file file.nii.gz

--test_list TEST_LIST run on list of test images
--single_file SINGLE_FILE path to single .nii.gz file to run on
--result_root *some_directory* , if specified saves results to directory specified (default is results/)
--offset X apply offset to Houndsfeld units (ie --offset -1024) (default is 0 offset)
--compute_test_metrics for testing (requires ground truth kidney stone data in .json files)
--save_seg saves the calculi segmentation to calculi_seg_output/ directory (using 0.5 threshold)
--save_kidney_seg saves the kidney segmentation to kidney_seg_output/ directory (using 0.5 threshold)
--threshold X set the initial threshold (default is 130 HU)
--offset OFFSET HU offset to apply (default = 0 )
--filename_tag FILENAME_TAG filename tag for testing results
--save_all_calculi_seg save all calculi prior to passing to CNN
--max_components MAX_COMPONENTS denoising is done interatively until the number of detections (connected components) is less than this number. (Default 200)
--run_CNN Run CNN for false positive removal?
--operating_point OPERATING_POINT Operating point for the CNN (default is 0.5)
--min_calculi_volume MIN_CALCULI_VOLUME minimum volume for calculi in mm^3 (default is 0.25)

The kidney segmentations are automatically saved in kidney_seg_output.

There are two files outputted for each case:

contains a single line with the format:

filename, max_score_in_case, binary_variable

where max_score is the max score among kidney stone detections (if no detection it is 0)

binary_variable is either 0 or 1 depending on whether max_score was > 0.5

contains one line per kidney stone detected with the format: filename, score, centroid_x, centroid_y, centroid_z, volume (in mm^3), median_HU, average_HU, std_HU, max_HU, pole, l_or_r

pole = 1 if the stone is in the upper half of the kidney, and = 0 if in the lower half l_or_r is either "l" or "r" depending on whether the stone is in the patient's left or right

run

python measure_kidney_sizes.py --test_list filename_list.txt

or

python measure_kidney_sizes.py --single_file file.nii.gz

It will run inference_kidney_only.py and then measure volume (in cc) and output to a file. Use --save_csv_header to add the csv header to the output file.