This repository contains the code of the prediction model used in the paper Automation of the Kidney Function Prediction and Classification Through Ultrasound-based Kidney Imaging Using Deep Learning. The paper has been accepted by npj Digital Medicine. The propsed automated system ensembles 10 trained models to predict estimated glomerular filtration rate (eGFR), and 10 trained gradient-boosted tree models to classify CKD stage. Go to project blog post for more information. Open access of the paper on npj Digital Medicine
This work uses Python 3.5.2. Before running the code, you have to install the following.
The above dependencies can be installed using pip by running:
pip install -r requirement.txt
bash get_models.sh
python3 ensemble_predict.py
use GPU: True
numbers of GPU: 1
Input image path:
Use --help to see usage of ensemble_predict.py:
usage: ensemble_predict.py [-h] [-g]
optional arguments:
-h, --help show help message and exit
-g, --gpu_id assign GPU ID, default 0
For ease of future clinical use, added a GUI for the predictiom module.
bash get_models.sh
python3 predict.py
Executable application for Windows environment. The full model version utilizes the 10-model-ensemble model used in our paper, it requires a GPU with 4GB VRAM. There is also a lite version with 5-model ensemble which requires 2GB VRAM, suitable for entry-level GPUs, yet at a cost of performance.
Dowload the file below and extract. Execute 'RenalFnXNet.exe'.
Full model version Requires GPU with 4GB of RAM
Lite version with 5-model ensembling Requires GPU with 2GB of RAM