The maximum value of the correlation between Sm(f) and Sr(f) (ρmax) determines the lower and upper limit [amax, bmax] of the redefined HF band (HFSC). The number of subjects analyzed for each parameter and each pair of elicitations is indicated in parentheses.
Motivation
Also, in the framework of this thesis, it is intended to expand the study of linear analysis with techniques based on the analysis of nonlinear dynamics, with the aim of studying the complexity of cardiac signals related to them. The study of the nonlinear dynamics of these signals can provide very significant information on the characterization of the ANS.
Work hypothesis
Therefore, it will be necessary to adapt the developed algorithms to be robust methods for analyzing short-term signals. The combination of information from different physiological signals, such as HRV and respiration (RSP), improves the characterization of the ANS for the discrimination of emotions as proposed in [47].
Physiological aspects
- The autonomic nervous system
- The autonomic nervous system control of the heart
- Heart rate variability
- Physiology of emotions
- Heart rate variability and emotions
SDANN (ms) SD of NN averages in all 5-min segments of the entire recording. SDNN Index (ms) Mean SD of all NNs for all 5-minute segments of the entire recording.
Objectives and outline of the thesis
The ability of the parameters derived from AMIF and CMIF to discriminate emotions will be evaluated on a database of video-induced emotion elicitation. In Chapter 5: it is proposed to use a linear classifier to identify the linear and non-linear characteristics that distinguish between pairs of emotions and between.
Registered emotions
Therefore, each of the 25 subjects was elicited with 2 videos of the same emotion, resulting in a total of 50 recordings per emotion. Regarding the analysis of the relaxation, only the first relaxation of each session was analyzed as a basal state during the entire session.
Emotion database validation
Most of them define the HF band centered at respiratory frequency and use a fixed bandwidth. Recently, spectral coherence between respiration and HRV has been used to define the HF band [27,56]. First, a simulation study is designed to evaluate the ability of the proposed HF band to quantify RSA.
The performance of the proposed HF band is compared with other commonly used HF band definitions.
Methods and materials
- Signal preprocessing
- Frequency band definition
- Performance measurement
- Statistical analysis
Spectral HRV indices were estimated from the power spectral density (PSD) of m(t) (Sm(f)), calculated by means of the Welch Periodogram. The HF band is redefined based on the correlation between Sm( f ) and Sr(f ) as given in Eq. 3.1), where a and b are the lower and upper limits of the analyzed frequency range. A total of 50 realizations of the LF component were generated for each considered subject, yielding mLFik(t) with k.
The breathing frequency of the recordings was also taken into account, meeting all the constraints specified in section 3.2.2 Definition of the FRSC-labeled frequency band.
Results
Evaluation of the methods for synthetic data
Evaluation of the methods for real data
Note that parameters with p ≤ 0.05, AUC index ≥ 0.70, sensitivity, specificity, accuracy values ≥ 70% are noted are in bold. The statistical differences between the pair of emotions are indicated by ∗ for p-value. It can be noted that the indices not marked with a † have an AUC ≥ 0.70. Together with the label of the pair of emotions studied in parentheses, it is the number of comparisons. all negative valences and F-S were significantly different.
No statistically significant differences were found in the comparison between neutral condition vs. negative valence and anger vs. 3.6 shows two examples where the SCHF method is particularly useful:. a) FR is below 0.15.
Discussion
With a redefinition of the HF band in these cases, a more refined description of the physiological information could be extracted from the signals. AMIF has been studied as an indicator of increased cardiac mortality in depressed patients [12] and in patients with multiple organ dysfunction syndrome [48], and CMIF has been applied to electroencephalographic signals for stress assessment [1]. The ability of the parameters derived from the AMIF and the CMIF to distinguish evoked states is evaluated on a database of video-induced emotion arousal, described in [100].
Here, we aim to study the discriminative ability of the nonlinear AMIF and CMIF emotion techniques that complement the information with linear features.
Methods and materials
- Signal preprocessing
- Auto-Mutual Information Function
- AMIF-based measures
- Cross-Mutual Information Function
- CMIF-based measures
- Selection of the number of bins
- Statistical analysis
The AMIF is a non-linear equivalent of the autocorrelation function, based on the Shannon entropy. The CMIF is a non-linear equivalent of the cross-correlation function, based on the Shannon entropy similar to the AMIF, but quantifies the coupling between two. The non-linear analysis of the coupled signals using the CMIF was as described for the AMIF, that is, the CMIF at τ = 0 represents the common maximum information of both time series and describes the decay of this function over a forecast time the loss of information about this τ [49].
The following parameters were calculated from the CMIF of the synchronized cardiac and respiratory signals: CMIF0 defined as the CMIF value at τ = 0, which represents the amount of shared information between the two time series without time lag;.
Results
Selection of the number of bins
A T-test, or Wilcoxon test as appropriate, depending on the results of the normality test, was then used to assess differences for the following paired conditions: relaxation and each emotion, and each emotion was compared with each other. The statistical significance level was p-value ≤ 0.05, as this threshold provides a reliable value for statistical discrimination [86]. In addition, the AUC index was examined to analyze the ability of the parameters to distinguish between the studied statements, and AUC ≥ 0.70 was used to determine statistically significant differences for each studied parameter.
The number of bins I was chosen as the value that gave the highest number of parameters with statistically significant differences (p-value ≤ 0.001) between relax and each emotion and between pairs of emotions.
AMIF-based measures
Only comparable elicitations with statistically significant differences are presented: relaxation and joy (R-J), relaxation and fear (R-F), joy and fear (J-F), joy and sadness (J-S), joy and anger (J-A), fear and sadness (F-S ). ) and fear and anger (F-A).
CMIF-based measures
In this figure, boxes are presented in terms of median and interquartile range as first and third quartiles: CMIF0γ (Fig. 4.5a); CMIFmaxγ (Fig. 4.5b) and τmaxγ (Fig. 4.5c) for the coupling between each signal γ = {RR, SC} and r(t). In Table 4.4, p-values, AUC and precision values are marked in bold for those CMIF-based parameters that revealed statistically significant differences between the emotional states studied.
Discussion
As found by analyzing the AMIF technique, the pair of stimulus conditions that did not show statistically significant differences in CMIF by any of the parameters considered were: relaxation and sadness (R-S), relaxation and anger (R-A), joy and sadness (J-S) and sadness and anger (S-A). These results are consistent with those obtained in this study by AMIF and CMIF techniques, as statistically significant differences between the neutral state of relaxation. Consistent with the Dominant Lyapunov Exponent and Approximate Entropy, during fear elicitation, the nonlinear HRV parameters obtained in the present study revealed a reduced level of complexity.
These results are consistent with those obtained using the CMIF technique applied in the present study.
Methods
- Estimation of discriminant function
- Parameter selection
- Performance measures of a classifier
- Parameters considered in the analysis
In fact, the use of too large a number of parameters in relation to the number of cases leads to a biased estimate of the discriminant function, reducing their ability to classify new cases [4]. The F-statistic represents the increase in discrimination produced after inclusion of the parameter p + 1 relative to the total already reached with the p-parameters previously included. First step: If the input criterion is met (F-statistic is statistically significant. F>3.84)), the parameter with the lowest value of Wilks0's Lambda is included.
When none of the included parameters meet the input and output criteria, the parameter selection process ends.
Results
Evaluation of the analysis 2
The features that best classify joy and anxiety are: FR derived from the linear analysis, PDmSC derived from the AMIF nonlinear technique and taking into account the HFSC band, and CMIFmaxSC derived from the CMIF nonlinear technique and also taking into account with the HFSC band.
Evaluation of the analysis 3
Discussion
The dissertation proposed linear and non-linear methods based on HRV analysis for human emotion recognition. A joint analysis of HRV and respiration was proposed for the linear analysis methodology to improve the characterization of human emotions. For this purpose, the HF band was defined based on the maximum spectral correlation between HRV and respiration.
This method yielded the new index, ρmax, which provides additional information for emotion recognition, based on the relationship between HRV and respiration.
Conclusions for non-linear analysis methodology
Conclusions for the classification analysis
The parameter CMIFmaxSC, which considers the non-linear coupling between the HRV and the respiratory signal information in a HF band redefined by the SCHF method, is the recurring characteristic that occurs in all classifications between positive and all negative valences together, and between joy and every single negative valence. Another interesting parameter to mention is ρmax, as it was able to classify between joy vs. It can be noted that ρmax also takes into account the common information between HRV and respiration.
Future extensions
En esta tesis se propusieron métodos basados en análisis lineales y no lineales para analizar la variabilidad de la frecuencia cardíaca con el fin de identificar las emociones humanas. Para el análisis lineal se propuso analizar en conjunto la variabilidad de la frecuencia cardíaca y la respiración con el fin de mejorar la caracterización de las emociones humanas. Para ello, la banda de alta frecuencia se definió por la máxima correlación espectral entre la variabilidad de la frecuencia cardíaca y la respiración.
En el Capítulo 3, el reconocimiento de las emociones humanas se evaluó mediante el análisis de la variabilidad de la frecuencia cardíaca.
Conclusiones del an´alisis no lineal
Para aumentar la fiabilidad de las mediciones de la variabilidad de la frecuencia cardíaca, se propuso una nueva metodología basada en la correlación espectral máxima entre la señal de variabilidad de la frecuencia cardíaca y la respiración. Este nuevo método propuesto, la correlación espectral en la banda de alta frecuencia, mejoró la evaluación del equilibrio simpatovagal capaz de distinguir entre relajación y alegría, alegría y cualquiera de las valencias negativas y miedo y tristeza. Además, este método proporcionó un nuevo parámetro, ρmax, que brinda información adicional para el reconocimiento de emociones, en función de la relación entre la variabilidad del ritmo cardíaco y la respiración.
Las técnicas de función de información mutua mutua no lineal y función de información mutua cruzada proporcionaron información adicional para otros métodos lineales y no lineales.
Conclusiones del an´alisis de clasificaci´on
Extensiones futuras
Acronyms
List of abbreviations
The methodologies and results presented in this dissertation and elaborated during my PhD studies have been published in the following papers.
List of parameters
Analysis of heart rate variability using time-varying frequency bands based on respiratory frequency. Time-frequency analysis of heart rate variability during stress testing using a priori respiratory rate information. Analysis of heart rate variability in the presence of ectopic beats using the heart time signal.
Human emotion recognition using heart rate variability analysis with spectral bands based on respiration.