This pipeline takes in paths to folders containing the data from Withings Scanwatch watches, with no preprocessing of the data required. The PPG Heart Rate data is then processed, and the ECG signals are processed using a seperate file. All metrics are then added to a SQL Database to be easily accessed for analysis. Graphs for different metrics can be toggled on or off.
Summary of work completed can be found in the internship_write_up_pdf file.
Download and unzip the user data from the smartwatches, either via the app or the website.
Main Pipeline is within the raw_hr_analysis.py file
Update patient_data_path with the path to the folder and volunteer_data_path if relevent.
Update sys.path statement at beginning of code to the location of the patient_analysis3 python script for ECG analysis.
Update saving_path with the location you wish to save the plots generated to.
numpy
matplotlib.pyplot
pandas
pathlib
scipy
datetime
sqlite3
sys
traceback
collections
statsmodels.api
pyhrv.nonlinear
csv
warnings
This pipeline requires ECG analysis from patient_analysis3.py
The code runs from the main() function, but is modular and can be used in parts with minimal preprocessing - ECG analysis requires more preprocessing but is easily seen in the afformentioned function.
Flags is an instance of the named tuple flags which is used to control when plots should be created. It is changed by ammending the Bools in the instantiation at the beginning of main. Most functions have a Flags default but check the docstrings to avoid errors when using in a modular form.
The Database created from all the determined metrics is by default created in the same folder as the pipeline script. This can be changed simply by adding the path into the line defining the database: db_name = 'patient_metrics.db' if Flag else 'volunteer_metrics.db' on line 946.
The Database is a Relational Database normalised to 3NF, and the database diagram is seen bellow:
The DFA analysis uses R squared linear regression to fit 2 straight lines to the data,finding the point where the two lines produce the greatest weighted R squared. This method, whilst valid, is not necessarily the correct method, and it may in fact be better to define the crossover region either exactly or much more specifically than it currently i so this can be very simply changed in the function detecting_crossover.
There is a slight bug with the databasing due to I think some patients having only 1 nights data, so data is inputtted into dictionary as a float and then attempted to be looped over