cs-598-dlh-paper-32

Dependencies

• Python >= 3.8.11
• Please use conda to install all the dependencies using the command
  • conda env create -f environment.yml

How to run the code

Once you install the dependency, please run the following code to open Jupyter Notebook on your terminal (Mac OS/Linux system)

• jupyter notebook Then you can run the cell block one by one.

Data download instruction

The filtered data of MIMIC-III is dataset/t1 folder. We only use a subset of MIMIC-III and it is filtered based on the requirement in the paper. This subset of MIMIC-III is already decoded that does not contains any original information from the original MIMIC-III.

The file 'Extract MIMIC III Dataset.ipynb` contains code to extract patient information from the raw MIMIC III Dataset.

Table of results

Macro-Averging Measures of Disease Prediction

Method	Accuracy	Recall	Precision	F1 Score
Euclidean	0.3703	0.38209	0.4264	0.2907
Cosine	0.3925	0.4061	0.3968	0.3842
CNN-triple	0.4888	0.4832	0.4832	0.4824

Patient Clustering Performance Based on Learned Distance

Method	Rand index	Purity	NMI
Euclidean	0.4284	0.4	0.0190
Cosine	0.5473	0.4074	0.0162
CNN-triplet	0.5653	0.4888	0.0679

Citation:

1. Suo, Q., Ma, F., Yuan, Y., Huai, M., Zhong, W., Gao, J., Zhang, A. (2018). Deep Patient Similarity Learning for Personalized Healthcare. IEEE transac- tions on nanobioscience, 17(3), 219–227. https://doi.org/10.1109/TNB.2018.2837622