aueb-wim / ehr-datafactory-template

EHR Data Factory folder template for deployment in hospital server

datafactory-wrapper

This repo contains a wrapper script for running DataFactory EHR and MRI pipelines.

Prerequisites

• Python 2.x (see: https://www.python.org/)
• docker
• docker-compose
• Matlab (see: here) (MRI pipeline)
• pandas 1.24.2 for python version 2 (MRI pipeline)
• SPM12 deployed in /opt folder (see: here)(MRI pipeline)
• Matlab engine for Python must be installed (see: here) (MRI pipeline)

Deployment and Configuration

It is suggested to use the ansible script here for installing and creating the datafactory data folders. Alternatively, you can manually clone this repo in /opt/DataFactory.

sudo git clone --recurse-submodules <repo_url> /opt/DataFactory

DataFacotry user settings

Create user group datafactory:

sudo groupadd datafactory

Datafactory User must be in the user group docker, so the scripts will run without the “sudo” command. To do that, follow the below instructions:

Add the docker group if it doesn't already exist:

 sudo groupadd docker

Add the connected user "$USER" to the docker and datafactory group. Change the user name to match your preferred user if you do not want to use your current user:

sudo gpasswd -a $USER docker
sudo gpasswd -a $USER datafactory

log out/in to activate the changes to groups.

You can use $ docker run hello-world to check if you can run docker without sudo.

Install python packages

In /opt/DataFactory folder run:

pip3 install -r requirements.txt 

In case of locale.Error: unsupported locale setting give the below command and rerun the above pip command:

export LC_ALL=C

Data Factory reserved ports

• 55432: PostgreSQL container
• 8082: MySQL 5.7 container for LORIS-for-MIP (DF optional feature)
• 8088: apache-php container hosting LORIS (DF optional feature)

Those ports can be changed in the configuration. The PostgreSQL container port can be changed in the config.json. The LORIS-for-MIP ports can be changed in the docker-compose.yml file (see LORIS-for-MIP readme file)

Data Factory Folders

The config.json contains the following folder structure as default.

Path	Description
/data/DataFactory	DataFactory data folder
/data/DataFactory/EHR	EHR data root input folder
/data/DataFactory/MRI/dicom/raw	DICOM raw data root input folder
/data/DataFactory/MRI/nifti/organized	DICOM root input folder with organized files
/data/DataFactory/MRI/nifti/raw	NIFTI raw data root input folder
/data/DataFactory/MRI/nifti/organized	NIFTI root input folder with organized files
/data/DataFactory/imaging	imaging data root input folder
/data/DataFactory/output/	output root folder
/data/DataFactory/anonymized_output/	anonymized output root folder
/opt/DataFactory/dbproperties	db properties folder
/opt/DataFactory/mipmap_mappings/preprocess_step	preprocess step config root folder
/opt/DataFactory/mipmap_mappings/capture_step	capture step config root folder
/opt/DataFactory/mipmap_mappings/harmonize_step	harmonize step config root folder
/opt/DataFactory/mipmap_mappings/imaging_step	imaging mapping config folder
/opt/DataFactory/export_step	export sql scripts folder

Change the group ownership of DataFactory installation folder /opt/DataFactory to the datafactory group and give write and execute rights:

sudo chgrp -R datafactory /opt/DataFactory
sudo chmod -R g+xw /opt/DataFactory

Update the parameteres in config.json file (Hospital name, docker postgres container details etc). Then run:

$ sudo ./update_files.py

Then change the group ownership of the DataFactory data folder to the datafactory group and give writing permitions:

sudo chgrp -R datafactory <datafactory data folder>
sudo chmod -R g+w <datafactory data folder>

The default datafactory data folder is /data/DataFactory

LORIS-for-MIP setup (optional)

If we want to add MRI Quality Control functionality to the MIP's DataFactory (as an extra step prior to the MRI volumetric brain feature extraction pipeline), we have the option of installing LORIS-for-MIP. Please refer to the repo's README for further information and installation instructions.

After installing the LORIS-for-MIP, we must create a soft link between DataFactory's folder that contains the batches of organized NIFTI files, and the output folder of LORIS-for-MIP. Do do so, we give in the command line:

ln -s /data/DataFactory/MRI/nifti/organized /data/LORIS/nifti_out

Create the DataFactory databases

Caution! The following creation script drops any prexisting DataFactory database. Skip this step if there is no such need i.e. when importing a second batch of hospital data.

for dropping and creating all databases at once

$ sh build_dbs.sh all

Ignore any warning messages about variables mipmap_map, mipmap_source and mipmap_pgproperties.

You can use the psql terminal tool or pgAdmin to recheck if the three DataFactory databases have been created (mipmap, i2b2_capture, i2b2_harmonized).

If you want to create a specific database give the following command:

for creating an empty database used by MIPMapReduce

$ sh build_dbs.sh mipmap

for creating am empty i2b2 capture database

$ sh build_dbs.sh capture

for creating an empty i2b2 harmonize database

$ sh build_dbs.sh harmonize

Mapping configuration files creation and upload

For creating the mapping configuration files, we use a mapping-desinging-template in a local machine (not in the hospital node) with a graphical desktop environment. Please refer to the DataFactory's User Guide for further instructions.

When, the mapping configuration files are ready, we upload them to the hospital node and placed them in the designated subfolders inside the folder /opt/DataFactory/mipmap_mappings.

For the preprocess step create a new folder in /preprocess_step and name it accordingly (ie config1 if is the first version of this configuration) and then place the configuration files in.

For the capture step create a new folder in /capture_step and name it accordingly (ie config1 if is the first version of this configuration) and then place the configuration files in.

For the harmonize step create a new folder in /harmonize_step and name it accordingly (ie config1 if is the first version of this configuration) and then place the configuration files in.

Please change the user group and the rights of the folders in order to be readable, writable and executable for the datafactory user group.

sudo chgrp -R datafactory  <configuration foler>
sudo chmod g+wrx -R <configuration foler>

Caution When using Filezilla for ftp uploading the configuration files, change the tranfer setting to binary! Otherwise the line endings are changed and scripts will be broken!

DataFactory data folders

The ehr files must be placed in a subfolder in the folder /data/DataFactory/EHR and named accordingly (ie batch1 if is the first batch of data)

The NIFTI files must be placed in a subfolder in the folder /data/DataFactory/MRI/nifti/raw and named accordingly (For example, in case we have a first batch of MRIs, we place them into the folder /data/DataFactory/MRI/dicom/raw/batch1.). The files must contain full-brain T1-weighted MRI data in nifti format and named <patient_id>_<visit_id>.nii.

When LORIS-for-MIP is installed, we could use DICOM files instead of NIFTI as raw MRI input. Please refer to LORIS-for-MIP github repo for further information.

Running DataFactory pipeline

MRI pipeline

Running the Ashburner's NeuroMetric scripts

In DataFactory folder run

./df.py mri [--loris] --input_folder <input folder>

--input folder we give the subfolder name in /data/DataFactory/MRI/nifti/raw/ and not the whole path. For example just batch1, batch2 etc. When this flag is skipped the user will be prompted to select one of the existing subfolders in /data/DataFactory/MRI/nifti/raw/.

--loris is a boolean flag and when is present, we give as input folder the subfolder name in /data/DataFactory/MRI/nifti/organized (just the name and not the whole path).

Note It is assumed that the NIFTI files are already organized by the LORIS-for-MIP module.

The output of this step is a volumes.csv file with the volumetric data of all the MRIs. This file is stored in a subfolder with the same name given as input_folder in the folder /data/DataFactory/imaging. For example /data/DataFactory/imaging/batch1.

Importing the volumetric brain features into the i2b2 capture database

In the folder where the output file volumes.csv is stored from the previous step, we place a csv file with the name mri_visits.csv. This file must have the headers PATIENT_ID, VISIT_ID, VISIT_DATE and be filled with this MRI information. The column VISIT_DATE must have the dd/mm/yyyy date format.

Then, in DataFactory folder run

./df.py ingest imaging  --input_folder <input folder>

--input folder we give just the subfolder name in /data/DataFactory/imaging and not the whole path. For example just batch1, v1 or batch2 etc. When this flag is skipped the user will be prompted to select one of the existing subfolders in /data/DataFactory/imaging.

EHR pipeline

Preprocess step

In DataFactory folder run

./df.py ingest ehr preprocess --input_folder <input folder> --config_folder <mapping config folder>

--input_folder we give just the corresponding subfolder name (i.e batch1, v1 or batch2), where the ehr data files are stored, and not the whole path. When this flag is skipped the user will be prompted to select one of the existing subfolders in /data/DataFactory/EHR/input/. --config_folder we give just the corresponding subfolder name (i.e. config1, config2 etc), where the preprocessing configuration files are stored, and not the whole path. When this flag is skipped the user will be prompted to select one of the existing subfolders in /opt/DataFactory/mipmap_mappings/preprocess_step/.

Auxiliary files are created in the same folder where the ehr csv files are located (for example in /data/DataFactory/EHR/input/batch1 if we have used as input folder the name batch1)

Capture step

In DataFactory folder run

./df.py ingest ehr capture --input_folder <input folder> --config_folder <mapping config folder>

--input_folder we give just the corresponding subfolder name, where the ehr data files are stored, and not the whole path. When this flag is skipped the user will be prompted to select one of the existing subfolders in /data/DataFactory/EHR/input/. --config_folder we give just the corresponding subfolder name (i.e. config1, config2 etc), where the mapping configuration files are stored, and not the whole path. When this flag is skipped the user will be prompted to select one of the existing subfolders in /opt/DataFactory/mipmap_mappings/capture_step/.

Harmonization step

In DataFactory folder run

./df.py ingest ehr harmonize --config_folder <mapping config folder>

--config_folder we give just the corresponding subfolder name (i.e. config1, config2 etc), where the mapping configuration files are stored, and not the whole path. When this flag is skipped the user will be prompted to select one of the existing subfolders in /opt/DataFactory/mipmap_mappings/harmonize_step/.

Export flat csv

In DataFactory folder run

./df.py export  -s <flattening method> -d <string value> [--csv_name <flat csv name>] [--local] <output folder>

-s, --strategy we declare the csv flattening method. The choices (defined in config.json) are the following(if left blank, the user will be prompted to select the following options):

       1. 'mindate': For each patient, export one row with all information related to her first visit
       
       2. 'maxdate': For each patient, export one row with all information related to her last visit
       
       3. '6months': For each patient, export one row with the information according to the 6-month window selection strategy defined by CHUV for clinical data related to MRI's. The criteria in detail:
           - For a patient to have a row in the output CSV she has to have an MRI and a VALID Diagnosis (etiology1 !=“diagnostic en attente” and etiology1 != “NA”) in a 6-month window to the MRI. If there are more than one MRIs choose the first one. If there are more than one Diagnosis, choose the closest VALID to the MRI.
           - The age and the visit date selected are the ones of the Diagnosis.
           - Having information about MMSE and MoCA is optional. Has to be within a 6-month window to the Diagnosis date.
           
       4. 'longitude': For each patient, export all available information.

-d, --dataset we declare the value that is going to be filled in the final csv under the column 'Dataset'. --csv_name (optional) if it is given, it overrides the defalut strategy's flat csv name (this is declared in config.json) with the given value. --local is a boolean option and when is present, the flat csv is exported from the i2b2 capture database with the hospital's local variables only without any CDEs. As output folder we give just the folder name and not the whole path where the flat csv file is created.

Anonymization

For Anonymization we have 2 options:

1. (db) Copy the i2b2 harmonized db and anonymize it, and then export the flat csv with a given strategy.

2. (csv) Anonymize a previously exported flattened csv.

Also, there are currently 2 anonymization hash methods: md5 and sha224. We can declare which of those 2 methods is going to be used in the DataFactory anonymization step by updating the hash_method field in config.json file.

For the case 1 (db) , in DataFactory folder run:

./df.py anonymize db -s <flattening method>  --hash_function <hash_method> -d <string value> [--csv_name <csv file name>] <output_folder>  

-s, --strategy we declare the csv flattening method (look export step). If left blank, the user will be prompted to select a strategy.

--hash_function we declare the hash funtion that is goint to be used for anonymization [md5 | sha224]. If the user, the user will be prompted to select hash_function.

-d, --dataset we declare the value that is going to be filled in the final csv under the column 'Dataset'.

--csv_name (optional) if it is given, it overrides the defalut strategy's flat csv name (this is declared in config.json) with the given value.

output_folder we give just the folder name (not the whole path) where the output anonymized csv will be placed. This folder is located in /data/DataFactory/anonymized_output/

Important Note Make sure that there is no active connection to any DataFactory Databases

For the case 2 (csv), in DataFactory folder run:

./df.py anonymize csv  --hash_function <hash method> -d <string value> --csv_name <anonymized csv file name> <source csv path> <output_folder>

--hash_function we declare the hash funtion that is going to be used for anonymization [md5 | sha224]. If the user, the user will be prompted to select hash_function.

-d, --dataset we declare the value that is going to be filled in the final csv under the column 'Dataset'.

--csv_name we give the anonymized csv file name

source csv path the filepath of the input csv that is going to be anonymized

output_folder we give just the folder name (not the whole path) where the output anonymized csv will be placed. This folder is located in /data/DataFactory/anonymized_output/

Interactive mode

The user can execute the ehr pipeline steps (preprocess, capture, harmonize, export) interectively through user prompts in terminal. Currently, the interactive mode does not support the mri step.

./df.py interactive