Presentation during Kasprowisko 2019 (XXV Konferencja Szkoleniowa i XXI Konferencja Wspólna Sekcji Elektrokardiologii Nieinwazyjnej i Telemedycyny oraz ISHNE
3. 3
Causality analysis may be applied for sports medicine
Hypothesis
Training Prediction Parameterization
How could it happen in the
opposite case?
(counterfactuals)
Time-domain
Frequency-domain
Information-domain
Causal-domain
Determining the training
program (by interventions)
based on the knowledge of
causal structures
4. Background
4
Level Symbol
Typical
”activity”
Typical
questions
Examples
Association Seeing
What is?
How would seeing X
change my belief in Y?
What does
a symptom tell me
about the disease?
Intervention
Doing
Intervening
What if?
What if I do X?
What if I take aspirin,
will my headache be
cured?
Counterfactuals
Imagining
Retrospection
Why?
Was it X that caused Y?
What if I had acted
differently?
Was it the aspirin
that stopped my
headache?
J. Pearl ”The Seven Tools of Causal Inference with Reflections on Machine Learning”
5. Cardiorespiratory context
5
From general to detail, in data-driven manner
Parameters globally Signals globally Signals locally
Taking into account
temporal relationships
Assessment of the causal
connections stability in
narrow time intervals
Without considering the
impact of time
Causal structure needs to be primarily established
6. Cardiorespiratory context
6
Parameters globally Signals globally Signals locally
Taking into account
temporal relationships
Assessment of the causal
connections stability in
narrow time intervals
Without considering the
impact of time
Latest papers
Short-term bivariate temporal orders exploration for
causality analysis - a conceptual study
Marcel Mły´nczak
Warsaw University of Technology
Faculty of Mechatronics
Warsaw, Poland
marcel.mlynczak@pw.edu.pl
Abstract—Causality analysis of cardiorespiratory signals may
be carried out on different levels of detail, e.g., global time-
independent or global temporal. To analyze particular parts of
the data one also needs a local approach. Therefore, the concept
of short-term bivariate temporal orders (SBTOs) exploration was
introduced. Two different kernels, based on linear modeling and
transfer entropy, were proposed. The most important applications
of the method are the stability assessment of the temporal orders
across time and the detection of similar signals’ parts or changes
in signals’ dynamics. The algorithm was preliminarily tested
and the results appear to confirm theoretical predictions. SBTOs
based on linear modeling kernel is informative enough for regular
curves; more sophisticated transfer entropy approach is more
relevant when the complexity is higher.
Index Terms—Causality analysis, temporal succession, tidal
volume, RR intervals, stability
I. INTRODUCTION
The network physiology concept is widely accepted [1].
It lies in the assumption that several systems are combined
in a network of non-linear dependencies with various loops,
feedbacks, and delays in transmitting information. The car-
diovascular or respiratory systems may be also influenced by
many environmental, psychological or demographic factors
[2].
One of the possible and commonly analyzed combinations
is that of respiration and heart activity. Several effects have
been introduced and accepted. For instance, sinus respiratory
arrhythmia is a phenomenon evident in resting ECG as the
effect of successive inspirations and expirations [3]. The
baroreflex effect is based on adjusting neural responses and
thus affecting both heart and respiratory activity in a specific
way [4]. Finally, the so-called cardiorespiratory coupling is
that heartbeats coincide with the respiratory phases particularly
because of increased sympathetic nervous activity [5].
Many mathematical approaches were proposed to analyze
cardiorespiratory relationships. We also hypothesized that the
parameterization of the cause-and-effect relationships would
be promising. However, to evaluate different levels of con-
nections we stated the analysis should move from general
to detail; from the look at the global parameters without
considering the impact of time; then through the insight into
temporal relationships and causalities; up to the analysis in
narrow time intervals enabling the assessment of the stability
of causal connections within different time segments, e.g.
during successive states after an orthostatic maneuver.
The first two approaches have been already utilized [6], [7].
In the first, the discovery of time-independent causal paths sug-
gested different results depending on the body position. During
lying supine the values of tidal volume seemed to cause heart
activity variation, which affected average heart activity, which
finally influenced respiratory timing. Alternately for standing,
the relation went from normalized respiratory activity variation
to average heart activity.
In the second approach, temporal relationships examined by
Granger causality frameworks (with extensions that consider
zero-lag effects [8]) or Time Series using Restricted Structural
Equation Models (TiMINo) [9], suggested that the most promi-
nent combination appeared between tachogram (RR intervals
curve) and tidal volume signal.
However, when analyzing signals, instead of beat-by-beat
sequences, the results were weak and not stable, which is an
effect of considering relatively long segments of data. On the
other side, temporal causality analyses require long enough
data to work correctly, e.g., based on the definition and due
to the stationarity criterion.
Therefore, there is a need for a framework to causally
examine local short segments of data and to explore the
temporal orders of cardiorespiratory signals. The presentation
of the concept of such a technique is hence the main aim of
the study.
II. MATERIALS AND METHODS
A. Description of the algorithm
The block diagram of the short-term bivariate temporal
orders (SBTOs) exploration is presented in Fig. 1.
First, the bivariate dataset should be loaded. One needs to
decide, which signal will not be moved, and which will be
shifted in time (backward, forward, in the given range). There
are several inputs parameters:
• the length of the signal part being used in the particular
step (the same length for both signals);
• the maximum shift in time (the same for backward and
forward direction);
• time resolution of the signal parts’ centers (by default
compatible with the sampling frequency);
ORIGINAL RESEARCH
published: 30 October 2018
doi: 10.3389/fphys.2018.01455
Frontiers in Physiology | www.frontiersin.org 1 October 2018 | Volume 9 | Article 1455
Edited by:
Toby Mündel,
College of Health, Massey University,
New Zealand
Reviewed by:
Jui-Lin Fan,
University of Otago, New Zealand
Cristina Blasco-Lafarga,
Universitat de Valencia, Spain
Anabel Forte Deltell,
Universitat de Valencia, Spain
*Correspondence:
Marcel Mły ´nczak
marcel.mlynczak@pw.edu.pl
Specialty section:
This article was submitted to
Exercise Physiology,
a section of the journal
Frontiers in Physiology
Received: 09 July 2018
Accepted: 25 September 2018
Published: 30 October 2018
Citation:
Mły ´nczak M and Krysztofiak H (2018)
Discovery of Causal Paths in
Cardiorespiratory Parameters:
A Time-Independent Approach in Elite
Athletes. Front. Physiol. 9:1455.
doi: 10.3389/fphys.2018.01455
Discovery of Causal Paths in
Cardiorespiratory Parameters:
A Time-Independent Approach in
Elite Athletes
Marcel Mły ´nczak1
* and Hubert Krysztofiak2
1
Institute of Metrology and Biomedical Engineering, Faculty of Mechatronics, Warsaw University of Technology, Warsaw,
Poland, 2
Department of Applied Physiology, Mossakowski Medical Research Centre, Polish Academy of Sciences, Warsaw,
Poland
Training of elite athletes requires regular physiological and medical monitoring to plan
the schedule, intensity and volume of training, and subsequent recovery. In sports
medicine, ECG-based analyses are well-established. However, they rarely consider
the correspondence of respiratory and cardiac activity. Given such mutual influence,
we hypothesize that athlete monitoring might be developed with causal inference
and that detailed, time-related techniques should be preceded by a more general,
time-independent approach that considers the whole group of participants and
parameters describing whole signals. The aim of this study was to discover general
causal paths among cardiac and respiratory variables in elite athletes in two body
positions (supine and standing), at rest. ECG and impedance pneumography signals
were obtained from 100 elite athletes. The mean heart rate, the root-mean-square
difference of successive RR intervals (RMSSD), its natural logarithm (lnRMSSD), the
mean respiratory rate (RR), the breathing activity coefficients, and the resulting breathing
regularity (BR) were estimated. Several causal discovery frameworks were applied,
comprising Generalized Correlations (GC), Causal Additive Modeling (CAM), Fast Greedy
Equivalence Search (FGES), Greedy Fast Causal Inference (GFCI), and two score-based
Bayesian network learning algorithms: Hill-Climbing (HC) and Tabu Search. The discovery
of cardiorespiratory paths appears ambiguous. The main, still mild, rules best supported
by data are: for supine - tidal volume causes heart activity variation, which causes
average heart activity, which causes respiratory timing; and for standing - normalized
respiratory activity variation causes average heart activity. The presented approach allows
data-driven and time-independent analysis of elite athletes as a particular population,
without considering prior knowledge. However, the results seem to be consistent with
the medical background. Causality inference is an interesting mathematical approach
to the analysis of biological responses, which are complex. One can use it to profile
athletes and plan appropriate training. In the next step, we plan to expand the study
using time-related causality analyses.
Keywords: athlete training adaptation biomarker, cardiac function, tidal volume, cardiorespiratory causality, elite
athletes
ORIGINAL RESEARCH
published: 05 February 2019
doi: 10.3389/fphys.2019.00045
Frontiers in Physiology | www.frontiersin.org 1 February 2019 | Volume 10 | Article 45
Edited by:
Toby Mündel,
College of Health, Massey University,
New Zealand
Reviewed by:
Luca Faes,
Università degli Studi di Palermo, Italy
Luke Charles Wilson,
University of Otago, New Zealand
*Correspondence:
Marcel Mły ´nczak
marcel.mlynczak@pw.edu.pl
Specialty section:
This article was submitted to
Exercise Physiology,
a section of the journal
Frontiers in Physiology
Received: 21 October 2018
Accepted: 16 January 2019
Published: 05 February 2019
Citation:
Mły ´nczak M and Krysztofiak H (2019)
Cardiorespiratory Temporal Causal
Links and the Differences by Sport or
Lack Thereof. Front. Physiol. 10:45.
doi: 10.3389/fphys.2019.00045
Cardiorespiratory Temporal Causal
Links and the Differences by Sport or
Lack Thereof
Marcel Mły ´nczak1
* and Hubert Krysztofiak2
1
Warsaw University of Technology, Faculty of Mechatronics, Institute of Metrology and Biomedical Engineering, Warsaw,
Poland, 2
Department of Applied Physiology, Mossakowski Medical Research Centre, Polish Academy of Sciences, Warsaw,
Poland
Fitness level, fatigue and adaptation are important factors for determining the optimal
training schedule and predicting future performance. We think that adding analysis of
the mutual relationships between cardiac and respiratory activity enables better athlete
profiling and feedback for improving training. Therefore, the main objectives were (1)
to apply several methods for temporal causality analysis to cardiorespiratory data; (2)
to establish causal links between the signals; and (3) to determine how parameterized
connections differed across various subgroups. One hundred elite athletes (31 female)
and a control group of 20 healthy students (6 female) took part in the study. All were
asked to follow a protocol comprising two 5-min sessions of free breathing - once
supine, once standing. The data were collected using Pneumonitor 2. Respiratory-related
curves were obtained through impedance pneumography, along with a single-lead ECG.
Several signals (e.g., tidal volume, instantaneous respiratory rate, and instantaneous
heart rate) were derived and stored as: (1) raw data down-sampled to 25 Hz; (2)
further down-sampled to 2.5 Hz; and (3) beat-by-beat sequences. Granger causality
frameworks (pairwise-conditional, spectral or extended), along with Time Series Models
with Independent Noise (TiMINo), were studied. The connections enabling the best
distinctions were found using recursive feature elimination with a random forest kernel.
Temporal causal links are the most evident between tidal volume and instantaneous
heart rate signals. Predictions of the “effect” variable were improved by adding preceding
“cause” samples, by medians of 20.3% for supine and 14.2% for standing body positions.
Parameterized causal link structures and directions distinguish athletes from non-athletes
with 83.3% accuracy on average. They may also be used to supplement standard
analysis and enable classification into groups exhibiting different static and dynamic
components during performance. Physiological markers of training may be extended
to include cardiorespiratory data, and causality analysis may improve the resolution of
training profiling and the precision of outcome prediction.
Keywords: granger causality framework, athlete training adaptation biomarker, cardiac function, tidal volume,
elite athletes
In
review
7. Cardiorespiratory context
7
Parameters globally
Without considering the
impact of time
Discovery of Causal Paths in
Cardiorespiratory Parameters:
A Time-Independent Approach
in Elite Athletes
Marcel Młyńczak, Hubert Krysztofiak
Published on 30th of October, 2018
8. Methodology
8
Device
• ECG signal to estimate heart rate
and tachogram
• Impedance signal relating to
main breathing activity
• Portable
• Recording on SD card
• Rechargeable battery
• Motion signal from 3-axis
accelerometer to indicate
subject’s activity and body position
• 14.2cm x 6.9cm x 2.3cm; 160g
• Sinusoidal application current amplitude
adjustable up to 1mA, with a single, adjustable
frequency (100kHz by default)
• Impedance range: 0-250 Ohms
• 250Hz sampling frequency, 100Hz pass
frequency, 10-bit resolution
• ECG amplifier has a gain of 100V/V,
10nV/sqrt(Hz) noise
• InvenSense’s MPU-6050 (accelerometer and
gyroscope unit, available commercially)
10. 5-minute
measurement
Free breathing
activity
Supine, then standing
10
Study performed on Polish elite athletes,
before Olympic Games in Rio de Janeiro 2016: 32 females and 68 males
Ethics Committee approval
AKBE/74/17 (WUM)
Methodology
Study protocol
14. 14
• The inspiration phase duration appears to cause the
expiration phase duration.
• The tidal volume during inspiration phase appears to cause
the tidal volume during expiration phase.
Results
Supine body position
Tidal Volume
Heart Activity Variation
Average Heart Activity
Respiratory Timing
Respiratory Sinus Arrythmia
Cardiorespiratory Coupling
The lack of an arrow coming out to the
breath parameter is in accordance
with the literature results.
Respiratory frequency is the most commonly
measured respiratory parameter,
and is at the end of the chain.
15. 15
• Definitely fewer connections found.
• The respiratory frequency does not appear to affect any
of the considered parameters.
Results
Standing body position
Normalized Respiratory
Activity Variation
Average Heart Activity
Respiratory Sinus Arrythmia?
16. Next paper
16
Signals globally
Taking into account
temporal relationships
Cardiorespiratory
Temporal Causal Links and
the Differences by Sport
or Lack Thereof
Marcel Młyńczak, Hubert Krysztofiak
Published on 5th of February, 2019
17. Next steps
17
• Benchmark of causal search techniques
• Causal analysis of IP + ECG + ABP signals
• Causal analysis of IP + ECG + REO signals
➡ Data from control group are already stored.
➡ All in ”pair-wise” and ”multivariate” approaches.
➡ All in time-independent, temporal and stability contexts.