PySiology: A python package for physiological feature extraction

Abstract

Introduction to Pysiology and advanced example (estimating images' valence from viewers physiological measurements)

Full text

PySiology
A python package for
physiological feature
extraction
Gabrieli G., Azhari A., Esposito G.
WIRN 2018

2
Feature estimation
Classification
Raw signal

GUI
Feature estimation is
done using a GUI. Easier
for non-experts.
Low customizability
Hidden parameters

CL / Scripting
Feature estimation is
done by (hard)-coding
your own set of functions
and classes.
(ex: MNE-python)
Errors
Time-consuming
3

Suitable for novice and expert users
Low learning curve
Highly customizable
Up-to-date with latest techniques

4

PySiology
Features estimation from ECG, EMG
and EDA raw signals
pysiology.rtfd.io
git.io/vh0PB
5

Python
Division by signals and not by stage of analysis
Scripting vs OOP
Clear documentation
Tutorials and sample data
Dummy pipelines for preprocessing and analysis
Open source package

6


7
Features in v. 0.0.9
ECG IBI, BPM,SDNN, SDSD, RMSSD, PNN50, PNN20, PNN50 / PNN20,
frequency analysis (high, low, very low)
EDA Rise time, latency, amplitude, half amplitude, EDA at apex, decay
time, SCR width
EMG
IEMG, MAV, SSI, VAR, TM, LOG, RMS, WL, AAC, DASDV, AFB, MYOP,
WAMP, SSC, MAVSLPk, HIST, MNF, MDF, peak frequency, MNP, TTP, SM,
FR, PSR, VCF

“
pysiology.signals.getFeatureNames(signal,
p1=stdalue1, …, pN=stdvalueN)
pysiology.signals.analyzeSignal(signal, sr)
8

“
import pysiology
ECG = pysiology.sampledata.loadsampleECG() #load the sample ECG Signal
sr = 1000 #samplerate in hZ
events = [["A",10], ["B",20]] #we can define the events the way we prefer
eventLenght = 8 #in seconds
for event in events:
startS, endS = [sr * event[1],startSample + (sr * eventLenght)
results = pysiology.electrocardiography.analyzeECG(ECG[startS:endS],sr)
9

Advanced
Example
Estimating images’ valence
through physiological
measurements


58 university students (age = 21.5+2,3)
50 IAPS images (25 low val., 25 high val.)
Images were presented for 8s (6s interval)
Physiological measurements:
ECG
EDA
EMG (Corrugator supercilii)
11


12

 !
13

"#
Preprocessing (using PySiology, standard parameters)
Feature estimation (using PySiology, standard
parameters)
Principal component analysis (using Scikit-learn, 6
components)
Classification using MLPNN (using Scikit-learn,
standard parameters)
Classification using Decision Tree (using Scikit-learn,
standard parameters)
Bootstrapping:
◉test = 100 repetition
◉train = 45 images (10% of the dataset)
14

$
 %
&'' 65%
$( 66%
15

Conclusions
PySiology is an open source python package for
physiological features estimation.
Suitable for novice users
Highly customizable
Tutorials and documentation
Reliable method for physiological features estimation
16

Thanks!
Any questions?
You can find me at:
giulio.gabrieli@studenti.unitn.it
17

$%)
 %
&'' 97%
18
N = 1361 epochs, bootstrapping with 100 repetition, 90% Train
Reference vs estimated valence
Reference valence
Estimated valence