Important
We thank the patients who participated in this project and generously share their data. We genuinely respect their privacy, so we will not release our dataset or the weights of our DNN.
This project involves Professor Feng-Tsun Chien from the Institute of Electronics and the Institute of Artificial Intelligence Innovation, National Yang Ming Chiao Tung University (NYCU), his two affiliated research assistants Ethan Lai and Sam Wang, and the Department of Internal Medicine -- Nephrology at National Taiwan University Hospital.
Our goal is to build a powerful artificial intelligence-assisted diagnosis system. Specifically, we train deep neural networks (DNNs) to predict whether a patient will experience kidney failure and, if so, when it will occur.
In utils, we implement three imputation methods to mitigate the missing data problem. The missing data arises because patients require various medical tests. We may not be able to collect all the above-mentioned metrics from every patient.
The DNN we adopt is in model/net.py.
We use various biometrics as inputs for prediction:
B_CRE, B_K, B_NA, B_UN, Hemoglobin, MCHC, PLT, WBC, Albumin, B_P, B_UA, Calcium, Triglyceride, DL, UPCR.
These are common metrics a doctor will refer to when diagonizing a patient. While experiencing a severe data missing problem with these metrics, we applied 3 types of imputation strategies to our dataset and compare their results. We also implemented an average moving filter to achieve data smoothing and reduce noise.
A 10-layer MLP with an input dimension of 16 and an output dimension of X. X refers to the number of classes we'd like to divide our result into. For example, we categorize patients based on whether they will develop kidney disease within two years. Those who will are placed in one category, while those who won't are placed in another. In this case, we set X=2.
To get started with our kidney project, clone this repository and install the required dependencies:
$git clone git@github.com:LaiEthanLai/Project_Kidney.git
$cd ./Project_Kidney
$pip install -r requirements.txt
$python train.py