Evaluation of Parameters for Fetal Behavioural State Classification

Evaluation of Parameters for Fetal Behavioural State Classification | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article Evaluation of Parameters for Fetal Behavioural State Classification Lorenzo Semeia, Katrin Sippel, Julia Moser, Hubert Preissl This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-966925/v1 This work is licensed under a CC BY 4.0 License Status: Published Journal Publication published 01 Mar, 2022 Read the published version in Scientific Reports → Version 1 posted 9 You are reading this latest preprint version Abstract Fetal behavioural states (fBS) describe periods of fetal wakefulness and sleep and are commonly defined by features such as body and eye movements and heart rate. Automatic state detection through algorithms relies on different parameters and thresholds derived from both the cardiogram and the actogram, which are highly dependent on the specific datasets and are prone to artefacts. Furthermore, the development of the fetal states is dynamic over the gestational period and the evaluation usually only separated into early and late gestation (before and after 32 weeks). In the current work, fBS were classified in 120 fetal magnetocardiographic data segments, between 27 and 39 weeks of gestational age, by both a classification algorithm and visual inspection. To identify how automated fBS detection could be improved, we first identified commonly used parameters for fBS classification in both the cardiogram and the actogram, and investigated their distribution across the different fBS. Then, we calculated a Receiver Operating Characteristics (ROC) curve to determine the performance of each parameter in the fBS classification. Finally, we investigated the development of parameters over gestation through linear regression. As a result, the parameters derived from the cardiogram have a higher classification accuracy compared to those derived from the body movement as defined by the actogram. However, the overlapping distributions of several parameters across states limit a clear separation of states based on these parameters. The changes over gestation of the cardiogram parameters reflect the maturation of the fetal autonomous nervous system. Given the higher classification accuracy of the cardiogram in comparison to the actogram, we suggest to focus further research on the cardiogram. Furthermore, we propose to develop a probabilistic fBS classification approaches to improve classification in less prototypical datasets. Internal Medicine General Biochemistry Evaluation parameters fetal behavioural classification body and eye movements and heart rate actogram cardiogram Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 1. Introduction In 1974 Prechtl described for the first time ‘behavioural states’ in infants [ 1 ], which is a concept that was later applied to fetuses [ 2 ]. Starting from 32 weeks of gestational age (GA), four fBS were identified, namely quiet sleep (1F), active sleep (2F), quiet awake (3F) and active awake (4F). The definition criteria of these states rely on body/eye movement and heart rate patterns that are stable for at least 3 minutes (see Table 1 ). Fetal states can be detected even before 32 weeks of GA but, in this case, it is only possible to define periods of rest (quiet states, QS) and periods of activity (active states, AS; [ 3 ]. The distribution of these fBS has been found to be stable over daytime[ 4 ] but depends on maternal position [ 5 ]. Table 1 Fetal behavioural states original criteria defined by Nijhuis and colleagues in 1982 [ 2 ]. State 1F State 2F State 3F State 4F Body movements Incidental Periodic Absent Continuous Eye movements Absent Present Present present Heart rate patterns Stable heart rate, with small oscillation bandwidth. Accelerations strictly related to movement Wider oscillation bandwidth compared to 1F. Frequent accelerations related to body movements Stable heart rate. No acceleration and higher oscillation bandwidth compared to 1F Unstable heart rate. Long lasting accelerations, often with tachycardia Fetal states were investigated in relation to fetal health [ 6 ], fetal maturation [ 7 ] and the development of the autonomic nervous system (ANS) [ 8 ], [ 9 ], as well as in fetal cognitive processing [ 10 ]. With the advancement of technology, it was possible to correctly classify states even without the usage of all defined indicators. Table 2 Fetal behavioural states definition criteria by Brändle and colleagues from 2015 [ 9 ]. 1F 2F 4F Baseline <160 bpm 160 bpm possible Oscillation bandwidth ±15 bpm Accelerations no >15 bpm/>15s > 30 bpm/>30s Movement no yes yes For example, Maeda described [ 11 ] how it was possible to define fBS without considering eye movements, adopting only the fetal heart rate and the gross body movement as available in the actocardiogram. This also enabled the investigation of fBS using fetal magnetocardiography (fMCG). This device allows to non-invasively record the magnetic signal of the fetal heart directly through the maternal abdomen. Compared to classical CTG and ultrasound, fMCG provides a higher temporal resolution and it is less susceptible to maternal interferences [ 12 ]. These recordings provide more accurate fetal heart dynamic parameters, like heart rate variability (HRV) measures. The gold standard for fBS classification using fMCG is a visual inspection and classification of the fetal cardiogram and actogram by experts. To facilitate this classification, the original Nijhuis criteria ([ 2 ], see Table 1 ) were adapted by Brändle and colleagues in 2015 ([ 9 ], see Table 2 ) . Due to the low occurrence of state 3F in this adaption only states 1F, 2F, and 4F were included. The fetal actogram describes the movement of the fetal body by calculating the spatial location of the highest heart activity amplitude (heart vector) for every single heartbeat, and follows the changes of this location within the sensor array over time. Recently, some algorithms have been developed to automatically classify fBS, both in CTG [ 13 ] and in fMCG [ 14 ]. Therefore, the original criteria defined by Nijhuis and colleagues [ 2 ] had to be translated into values that could be used as classification thresholds by those algorithms. To achieve the quality and accuracy of a visually-inspected evaluation performed by experts, automatic state detection algorithms rely on the combination of several heart and movement parameters and their more or less strictly defined thresholds. These thresholds are highly dependent on the specific datasets, and are applicable only to those datasets in which no artefacts are occurring. Although this is a practical solution for the study of prototypical fetal physiology during the states, these thresholds prevent the applicability of existing algorithms on datasets, for example where a transitions between states occurs. Moreover, given the dynamic development of fBS over gestation, for classification purposes researchers usually separate the early and the late gestation (before and after 32 weeks of GA). Altogether, these factors make it difficult to obtain a satisfactory classification of fBS. In the current work we aim to perform an exploratory analysis of commonly used fBS classification parameters, like variability in the heart-rate or in the actogram, to determine the reliability of these parameters. Secondly, the development of these parameters over gestation is investigated. The first part (Step1, “Parameter distribution and thresholds”) of the current work will investigate the accuracy of the fBS classification parameters. Therefore, we analysed the distribution of the main parameters commonly used in fBS algorithms. Second (Step2, “Best classifier”), we want to verify which of the heart-rate (HR) or movement parameters are the best parameters for state classification performance both in early as in late gestation. Finally (Step3, “Changes over time”), we investigate changes and development of these parameters over the last trimester of gestation, in order to highlight possible age-dependent trends. 2. Methods 2.1. Fetal Magnetoencephalography (fMEG) fMEG is a non-invasive device for the recording of brain and heart activity of fetuses in the last trimester of gestation, and of newborns in the first weeks of life [ 15 ]. The datasets included in the current study were recorded using a SARA (SQUID Array for Reproductive Assessment, VSM MedTech Ltd., Port Coquitlam, BC, Canada) system available at the fMEG Center at the University of Tuebingen, Germany. All the recordings were sampled with a frequency of 610,3516 Hz. This device comprehends 156 magnetic sensors and 29 additional reference sensors. These sensors are placed in a concave array, designed to match the maternal abdomen. In order to limit the influence of external magnetic fields, the device is placed in a magnetically shielded room (Vakuumschmelze, Hanau, Germany). 2.2. Datasets The 169 original datasets (117 for fetuses younger than 32 weeks of GA) used in the current analysis come from previously published studies [ 16 ], [ 17 ]. All recordings were performed in the morning. In addition, previous data showed that the fBS distribution since is stable over daytime [ 4 ]. These datasets were previously screened for good quality of the raw data in terms of absence of large amplitude artefacts. These studies were approved by the Ethical Committee of the Medical Faculty at the University of Tuebingen (No. 476/2008MPG1 and 339/2010BO1). All the participants to these studies gave their informed consent in accordance to the Declaration of Helsinki and agreed for the usage of the data in further studies. The datasets include recordings of spontaneous activity as well as recordings during an auditory paradigm. The datasets length varies between 6 and 15 minutes. 2.3. Preprocessing The data preprocessing was performed using Matlab R2016b (The MathWorks, Natick, MA, USA). R-peak detection of maternal and fetal heart has been performed using FLORA [ 18 ]. Maternal heart interference was removed by FAUNA [ 19 ]. HRV parameters in the time domain (mean heart-rate (HR), the standard deviation of normal to normal R-R intervals (SDNN) and the root mean square of successive differences between normal heartbeats (RMSSD) were calculated for the fMCG signals [ 20 ]. Finally, fetal actogram and cardiogram were calculated as described by Govindan and colleagues [ 21 ]. 2.4. Fetal states analysis The fetal behavioural state underlying the datasets was automatically detected using an algorithm adapted from the work of Vairavan and colleagues [ 14 ] and implemented in Matlab R2019b. We tested the algorithm on three groups based of the fetal gestational age (GA), an early, a middle, and a late group. The early group includes datasets from fetuses between 27-32 (27 0/1-32 6/7) weeks of GA, the middle group fetuses between 33-36 (33 0/1-36 6/7) weeks of GA, and the late group includes fetuses between 37-39 (37 0/1-39 6/7) weeks GA. According to the work of Vairavan and colleagues [ 14 ], the percentage of accelerations in the HR, the standard deviation of the HR (σ(HR)) and the percentage of points above 160 bmp were computed. The thresholds for these values used in our algorithm are also from Vairavan and collaborators [ 14 ]. To differentiate between active and passive states in fetuses from 27-29 (27 0/1-29 6/7) weeks of GA we applied the threshold of 5.54% of accelerations for dividing between passive (≤ 5.54%) and active (> 5.54%) states. This threshold is based on Vairavan and colleagues [ 14 ], where it is used for dividing between active and passive states between 30-36 (30 0/1- 36 6/7) weeks of GA. Out of the included 169 datasets, a state was classified for approximately 80% of the total data length. In the remaining 20% of data, the variability of parameters used for classification (e.g. the percentage of acceleration in the HR) was exceeding the tolerance of the detection algorithm, thus preventing a fBS classification. This variability is likely due to transitions between states or to signal noise. To maintain a reasonable amount of data for each group, we considered the middle and late group as a whole late group for all the further analyses. From the algorithm outputs, 120 data segments (from 85 different recordings performed in 52 subjects) were selected based on a clear detection of the behavioural state: 31 for active and 27 for passive in the early group, 31 for 1F and 31 for 2F in the late group. The goal of the selection was to have an equivalent amount of datasets in each group however, no more than 27 passive state windows were available. Yet, a lower occurrence of passive states is expected [ 22 ]. 3F and 4F were not included because we were not able to detect enough data segments. The mean duration of these 120 data segments was 266±44 seconds. The data segments were then visually evaluated by two experts to gain security about the result of the automatically classified states. There was agreement between the two raters in 96 windows (80% agreement). 87 out of these 96 windows were also in agreement with the automatic states classification algorithm. Following this procedure, our further analysis only contains those 87 datasets in which there was accordance between all 3 ratings (see Table 3 ). Table 3 Characteristics of the 87 data segments that were chosen for further evaluation [mean ± standard deviation]. Passive Active 1F 2F Windows length, in seconds 288.63 ± 30.83 252.81 ± 43.98 283.60 ± 34.07 250.00 ± 52.98 Number of windows 19 31 15 22 GA, in weeks 29.21 ± 1.47 29.52 ± 1.43 35.13 ± 1.41 34.95 ± 1.65 2.5. Parameters definition The following parameters, which are based on the states definition criteria reported in Table 1 and Table 2 , are included in the further analysis. For a summary of the parameters see Table 4 . HRV parameters (mean HR, SDNN and RMSSD) are calculated in order to verify whether their values are comparable to previous work which investigated fBS [ 9 ]. Therefore, their classification accuracy and development over time are not further investigated. In the cardiogram, we defined A) Baseline standard deviation ( Baseline std ). The baseline is defined as smoothing the HR in a two minutes moving window with 1 second shift. We considered the baseline standard deviation as an index of stability in the heart rate. B) Percentage of points outside the ± 5 bpm oscillation bandwidth ( % points outside ± 5 bpm ), based on Nijhuis and colleagues [ 2 ]. The oscillation bandwidth is defined as the area between the baseline and the baseline ± 5 bpm. C) Percentage of points outside the ± 7.5 bpm oscillation bandwidth ( % points outside ± 7.5 bpm ) based on Brändle and colleagues [ 9 ]. The oscillation bandwidth is defined as the area between the baseline and the baseline ±7.5 bpm. D) Root mean square of successive differences in the HR ( RMSSD in the HR ). This parameter has been chosen as an index of variance in the heart rate. In contrast to the HRV parameter RMSSD, which is calculated on the time differences between successive R peaks, this parameter is directly calculated on HR values. E) Heart rate std ( HR std ). This is an index of variability in the heart rate. In the actogram we defined F) Root mean square of successive differences in the actogram ( RMSSD in the actogram ). This parameter has been selected as an index of short term variability in the actogram. G) Actogram standard deviation ( actogram std ). This is an index of variability in the actogram. Finally, since in the original definition by Nijhuis and colleagues [ 2 ] fBS are also defined based on the relation between heart-rate and movement accelerations, we calculated the correlation between cardiogram and actogram to explore whether this parameter could be used for states classification. 2.6. Statistics Statistics were performed with Matlab R2019b for Windows. Before proceeding, outliers for each of the parameters, defined as a value exceeding the mean ± 3 standard deviations, were removed. Then, the Kolmogorov-Smirnov test was used to test the normality assumption required by the following statistical analysis. Table 4 Summary of the parameters used in the current work. Cardiogram Actogram Baseline std RMSSD in the actogram % points outside ± 5 bpm Actogram std % points outside ± 7.5 bpm RMSSD in the HR HR std Correlation between cardiogram-actogram Step 1. Parameter distribution and thresholds. We tested the alternative hypothesis that there is a difference between HRV, cardio- or actogram parameters between active and passive (H1: µActive - µPassive ≠ 0), or between 1F and 2F (H1: µ2F - µ1F ≠ 0). A t-test was performed for the analysis of the different parameters between active and passive, and between 1F and 2F. After Bonferroni correction for N=10 different parameters, the significance level was set to p < 0.005. Step 2. Best classifier. Receiver operating characteristics (ROC) curves were defined to verify the performance of each parameter in classifying a state. Step 3. Changes over time. Finally, to detect the development of parameters during gestation, linear models were built for each parameter. In the models, we combined passive (early passive +1F) and active (early active + 2F) states between 27 and 39 weeks of GA The model was defined as: 𝑦 = 𝛽0 + 𝛽1X + 𝜀 where, 𝑦 is one of the parameters included in the current analysis, 𝛽0 is the intercept, 𝛽1 the regression coefficient, X the gestational age, and 𝜀 the error in the estimate. Model values of p < 0.05 were considered statistically significant and effect size, defined by r 2 . 3. Results Step 1. Parameters distribution and thresholds HRV parameters The t-test between states revealed no difference in the mean HR both between passive and active in the early gestation (t(48) = 1.46, p = 0.15), and between 1F and 2F in the late gestation (t(35) = 0.38, p = 0.70) (Fig. 1 A). SDNN was higher in active compared to passive state (t(45) = 9.20, p < 0.001), and higher in 2F compared to 1F (t(35) = 7.36, p < 0.001) (Fig. 1 B ) . Finally, RMSSD was higher in active compared to passive state (t(48) = 10.43, p < 0.001), and higher in 2F compared to 1F (t(35) = 6.40, p < 0.001) (Fig. 1 C). Although two out of three HRV parameters showed a significant difference both in early and in late gestation, only the RMSSD values in the early gestation could be separated by a single threshold. The distributions of all other parameters overlap. Cardio- and actogram parameters In the early gestation, the t-tests revealed that all parameters were higher in active compared to passive states: ‘Baseline std’ (t(45) = 5.59, p < 0.001), ‘% points outside ± 5 bpm’ (t(46) = 13.47, p < 0.001), ‘% points outside ± 7.5 bpm’ (t(46) = 14.16, p < 0.001), ‘RMSSD in the HR’ (t(42) = 13.54, p < 0.001), ‘HR std’ (t(44) = 13.80, p < 0.001), ‘RMSSD in the actogram (t(43) = 3.46, p = 0.001), and ‘Actogram std’ (t(42) = 4.22, p < 0.001). Also in late gestation, the t-tests revealed that all parameters were higher in 2F compared to 1F: ‘Baseline std’ (t(34) = 5.28, p < 0.001), ‘% points outside ± 5 bpm’ (t(32) = 16.73, p < 0.001), ‘% points outside ± 7.5 bpm’ (t(35) = 15.91, p < 0.001), ‘RMSSD in the HR’ (t(32) = 6.97, p < 0.001), ‘HR std’ (t(32) = 11.65, p < 0.001), ‘RMSSD in the actogram (t(32) = 3.67, p < 0.001), and ‘Actogram std’ (t(34) = 4.07, p < 0.001). Although all of the seven cardio- and actogram parameters show significant differences in their distribution in early and late gestation, only three parameters in early and two in late gestation could be separated by a single threshold, while the distributions of all the other parameters overlap (see Fig. 2 ). For details about the mean parameters values, see supplementary table 1 and supplementary table 2. Correlation cardiogram-actogram Correlating the cardiogram and the actogram, the t-tests revealed no difference in the mean r-value between passive (M = -0.01, SD = 0.22) and active (M = 0.08, SD = 0.26) in the early gestation (t(48) = 1.32, p = 0.19). There was a significant difference between 1F (M = 0.09, SD = 0.21) and 2F (M = 0.28, SD = 0.27) in the late gestation (t(35) = 2.33, p = 0.03) (see Fig. 3 ). Step 2. Best classifier ROC analysis. In the early gestation, ‘% points outside ± 5 bpm’, ‘% points outside ± 7.5 bpm’, ‘HR std’ and ‘RMSSD in the HR’ are perfect classifier (AUC = 1), ‘Baseline std’ has a high performance (AUC = 0.93), ‘RMSSD in the actogram’ (AUC = 0.80) and ‘Actogram std’ (AUC = 0.83) perform less compared to the other classifier (see Fig. 4 A). In the late gestation the results are very similar. ‘% points outside ± 5 bpm’, ‘% points outside ± 7.5 bpm’, ‘HR std’ are perfect classifier (AUC = 1), ‘Baseline std’ (AUC = 0.92) and ‘RMSSD in the HR’ (AUC = 0.98) have a high performance, ‘RMSSD in the actogram’ (AUC = 0.87) and ‘Actogram std’ (AUC = 0.86) perform less compared to the other classifier (see Fig. 4 B). Step 3. Changes over time Development of parameters over gestational age, not divided by fBS First of all, we tested whether the parameters change with GA regardless the fBS (see supplementary table 3). The parameters did not show any general change during development. Development of parameters over gestational age, divided by fBS Furthermore, we investigated whether the parameters change over time depending on the fBS. For this analysis, we combined passive (early passive + 1F) and active (early active + 2F) states to detect their general evolution over time (see supplementary table 4). Only the RMSSD in the HR showed a difference between states. In particular, the decline can be seen in active states (p = 0.01, r 2 = 0.13) but not in passive states (p = 0.30; Fig. 5 ). Development of the correlation cardiogram-actogram over gestational age Thirdly, we looked at the evolution of the correlation between cardiogram and actogram over time (Fig. 6 A). We found an overall increase of the r-values over gestational age (p = 0.001, Effect size = 0.18). Introducing the behavioural state as further parameter, we found an increase in the correlation between actogram and cardiogram for active states (p < 0.001, r 2 = 0.22) but not for passive states (p = 0.2) (Fig. 6 B). 4. Discussion This work systematically investigated different parameters commonly used for fBS classification. Although nearly all investigated parameters differed significantly in their distribution between active and passive, or between 1F and 2F, only a few of these parameters had non-overlapping distributions, preventing a clear separation and classification of the states. Furthermore, we identified two parameters that continuously changed during fetal development. We therefore argue that they are not suitable for a general fBS classification. Step 1. Parameters distribution and thresholds The first aim of the current work was to investigate the distribution of properties of the acto- and cardiogram over fetal behavioural states. As expected, different parameter distributions are representative for the different states, as confirmed by the significant differences between parameters across states. Moreover, our HRV parameters (mean HR, SDNN and RMSSD) were in line with previous works [ 9 ]. Some emphasis should be put on the parameters ‘% points outside ± 5 bpm’ and ‘% points outside ± 7.5 bpm’. For example, in Brändle and colleagues [ 9 ], the authors operationalised the variability of the HR considering the oscillation bandwidth in which the HR could vary. Specifically, it is assumed that the HR does not exceed the [baseline ± 7.5] bpm in state 1F. Although this value could efficiently divide between active and passive states during early or late gestation, this threshold was exceeded in each data-window. In states classified as passive, up to 4% of the total data points exceeded this value. This raises the question, whether the threshold value should be increased (e.g. ± 10 bpm), or whether an increase would instead classify active states as passive and vice-versa. Our suggestion would be to define a certain tolerance around this threshold. For example, in our datasets this value could be exceeded in 5% of data points in a passive state and still provide the same results in terms of classification. Another point is the correlation between the cardiogram and the actogram. In Nijhuis and colleagues [ 2 ], the authors defined the fBS by means of the different ways heart-rate and body movement relate to each other. As a proxy for this coupling, in our work we explored how heart rate and body movement are related through analysis of the correlation between the two signals. Generally, the correlation was weak, with r-values both positive and negative (Fig. 3 ). Given the high variability and wide overlap of this parameter over fBS, this parameter does not appear specific of a particular state. However, as discussed later, this index has interesting developmental characteristics. Step 2. Best classifier The ROC analysis revealed that the parameters derived from the cardiogram have very high classification performance, as noted by the AUC of almost 1 in every case (Fig. 4 ). Even though movement plays an important role in the initial Nijhuis criteria, the actogram parameters, as defined in the current work, show the worst classification performance with an AUC around 0.80. Therefore, at least in the frame of fMCG recordings, we argue that the actogram does not improve an automatic fBS classification. However, it should be considered that the actogram only represents gross-body movements and does not represent the movements of extremities. Step 3. Changes over time We found a decrease over gestation in the parameter RMSSD in the HR in active states but not in passive states (Fig. 5 ). Even though there is no accordance on the development of short term variability in the heart patterns during the last trimester of gestation, with some works reporting an increase [ 23 ] and others no changes [ 9 ], in our case we found a decrease which depends on the activity state (Fig. 5 ). The progressive decline in short-term variability in the heart-rate, as expressed in our findings by the RMSSD in the HR, makes the distinction between active and passive states, based on this parameter, less clear as gestational age increases. This is compatible with evidence for no substantial differences in short term variability in the heart-rate across different sleep stages later in childhood and adolescence [ 24 ]. Furthermore, we found an increasing correlation following gestational age between cardiogram and actogram only in active states (Fig. 6 ). One reason for the developmental change of this relationship could be a change in fetal respiratory movements. For example, as Schneider and colleague[ 23 ] report, the increase in high frequency HRV close to term could be explained taking into account the fetal respiratory sinus arrhythmia, which is associated with thoracic movements and respiratory efforts [ 25 ]. Even if these respiratory movements are not exclusive of a particular fBS, they are indeed more common in periods of active sleep. These developmental trends, found in the correlation between cardiogram and actogram, and in the RMSSD in the HR, could reflect the maturational changes of the fetal autonomous nervous system and makes these parameters less suitable for fBS analysis over different gestational ages. Limitations and Outlook In the present analysis wake states, like 3F and 4F were excluded due to their low occurrence. Including a substantially higher number of datasets could shed light on the parameter distribution over these fBS. Moreover, in contrast to Vairavan and colleagues [ 14 ], we grouped together fetuses from the middle (32-36 weeks of GA) and late (37-39 weeks of GA) in a whole late group. Furthermore, the data segments included in the current analysis were selected by a states detection algorithm and validated by two external raters, which results in selecting very prototypical data segments. In the algorithm used in this study (adapted from Vairavan and colleagues [ 14 ]) data segments with a parameter distribution that does not fit the prototypical characteristic of a certain fBS, are discarded. Therefore, an algorithm with clearly defined thresholds has a problem with classifying less clear datasets, or datasets with a transition between states. Classification of transitions between wake and sleep states is not only relevant for understanding basic sleep physiology, but also for the study of cognitive processing [ 26 ]. To improve the detection of fBS, one option could be to consider the probability of the fetuses being in one of the fBS compared to the others. In fact, considering the distribution of parameters reported before, it appears that a certain value for a parameter is more likely to occur in a certain state compared to another state. This likelihood could be used for a more dynamic state detection, accounting for transitions. Another approach could then be a dynamic definition of the fBS using the fetal brain data extracted during the measurement. fMEG allows, theoretically, to also use the fetal brain activity as an additional parameter. Even in preterm infants, for example, a state classification based on the electroencephalographic data is at the moment the frontier for automatic state classification [ 27 ]. Conclusion Our results indicate that cardiogram parameters have a higher classification accuracy in comparison to those extracted from the actogram. Therefore, we propose for the automation of fBS detection is to concentrate on the cardiogram instead of body movement as defined by the actogram. Moreover, the changes over time in some of the parameters in our opinion disqualify them from being part of a fBS classification algorithm that simply divides fetuses into early and late groups. In fact, given the parameters continuous change over time, we instead suggest to consider fetal gestational age in weeks. State of the art algorithms, like the one from Vairavan and colleagues [14], are easily applicable and partly reflect the original fBS definition criteria. On the other side, these algorithms may be too simplistic, thus failing to grasp more complex physiological events such as transitions between states. Based on these conclusions, our suggestion is therefore the development of a probabilistic approach for the fBS detection that is more dynamic in detecting the alternating patterns between wake and sleep, which could possibly include fetal brain parameters. Declarations 5. Data availability The datasets adopted in the current study are available from the corresponding author on a reasonable request. 7. Acknowledgment This work was partially supported by the DAAD (Deutscher Akademischer Austauschdienst) and the International Max Planck Research School for the Mechanisms of Mental Function and Dysfunction (IMPRS-MMFD). We also thank Dr. Franziska Schleger and Dr. Isabelle Kiefer-Schmidt for their support during the fetal state evaluation. 8. Author contributions L.S. conceptualized and conducted the analysis, drafted and revised the manuscript, and prepared the figures. L.S. is also the corresponding author. K.S. and J.M. assisted during the analysis, and reviewed and edited the manuscript. H.P. supervised the work, reviewed and edited the manuscript, and managed the project. All authors read and approved the revised manuscript. 9. Conflict of interest statement The authors declare no conflicts of interest. 6. References 1. Prechtl, H. F. The behavioural states of the newborn infant (a review). Brain research 76(2), 185–212 (1974). 2. Nijhuis, J. G., Prechtl, H. F. R., Martin Jr, C. B. & Bots, R. S. G. M. Are there behavioural states in the human fetus? Early human development 6(2), 177–195 (1982). 3. Pillai, M. & James, D. The development of fetal heart rate patterns during normal pregnancy. Obstetrics and Gynecology 76, 812–816 (1990). 4. Sonanini, A. et al. Fetal behavioral states are stable over daytime - evidence by longitudinal and cross-sectional fetal biomagnetic recordings. Journal of perinatal medicine 42, 307–314 (2014). 5. Stone, P. R. et al. Effect of maternal position on fetal behavioural state and heart rate variability in healthy late gestation pregnancy. The Journal of physiology 595, 1213–1221 (2017). 6. Martin, C. B. Normal fetal physiology and behavior, and adaptive responses with hypoxemia. Seminars in perinatology 32, 239–242 (2008). 7. Wakai, R. T. Assessment of fetal neurodevelopment via fetal magnetocardiography. Experimental neurology 190 Suppl 1, S65-71 (2004). 8. Hoyer, D. et al. Fetal development of complex autonomic control evaluated from multiscale heart rate patterns. American journal of physiology. Regulatory, integrative and comparative physiology 304, R383-92 (2013). 9. Brändle, J. et al. Heart rate variability parameters and fetal movement complement fetal behavioral states detection via magnetography to monitor neurovegetative development. Frontiers in human neuroscience 9, 147 (2015). 10. Moser, J. et al. Magnetoencephalographic signatures of conscious processing before birth. Developmental cognitive neuroscience 49, 100964 (2021). 11. Maeda, K. Fetal monitoring and actocardiogram in the evaluation of fetal behavior. The Ultrasound Review of Obstetrics and Gynecology 4, 12–25 (2004). 12. Queyam, A. B., Pahuja, S. K. & Singh, D. N on - Invasive Feto - Maternal Well - Being Monitoring: A Review of Methods. JESTR 10, 177–190 (2017). 13. Jezewski, M., Czabanski, R., Leski, J. M. & Jezewski, J. Fuzzy classifier based on clustering with pairs of ε-hyperballs and its application to support fetal state assessment. Expert Systems with Applications 118, 109–126 (2019). 14. Vairavan, S. et al. A computer-aided approach to detect the fetal behavioral states using multi-sensor Magnetocardiographic recordings. Computers in biology and medicine 69, 44–51 (2016). 15. Preissl, H., Lowery, C. L. & Eswaran, H. Fetal magnetoencephalography: current progress and trends. Experimental neurology 190 Suppl 1, S28-36 (2004). 16. Morin, E. C. et al. Functional brain development in growth-restricted and constitutionally small fetuses: a fetal magnetoencephalography case-control study. BJOG : an international journal of obstetrics and gynaecology 122, 1184–1190 (2015). 17. Linder, K. et al. Maternal insulin sensitivity is associated with oral glucose-induced changes in fetal brain activity. Diabetologia 57, 1192–1198 (2014). 18. Sippel, K. et al. Fully Automated R-peak Detection Algorithm (FLORA) for fetal magnetoencephalographic data. Computer methods and programs in biomedicine 173, 35–41 (2019). 19. Sippel, K. et al. Fully Automated Subtraction of Heart Activity for Fetal Magnetoencephalography Data. Annual International Conference of the IEEE Engineering in Medicine and Biology Society. IEEE Engineering in Medicine and Biology Society. Annual International Conference 2019, 5685–5689 (2019). 20. Mat Husin, H. et al. Maternal Weight, Weight Gain, and Metabolism are Associated with Changes in Fetal Heart Rate and Variability. Obesity (Silver Spring, Md.) 28, 114–121 (2020). 21. Govindan, R. B. et al. A novel approach to track fetal movement using multi-sensor magnetocardiographic recordings. Annals of biomedical engineering 39, 964–972 (2011). 22. Kiefer-Schmidt, I. et al. Is there a relationship between fetal brain function and the fetal behavioral state? A fetal MEG-study. Journal of perinatal medicine 41, 605–612 (2013). 23. Schneider, U. et al. Developmental milestones of the autonomic nervous system revealed via longitudinal monitoring of fetal heart rate variability. PloS one 13, e0200799 (2018). 24. Kontos, A. et al. The Inconsistent Nature of Heart Rate Variability During Sleep in Normal Children and Adolescents. Frontiers in cardiovascular medicine 7, 19 (2020). 25. Gustafson, K. M., May, L. E., Yeh, H., Million, S. K. & Allen, J. J. B. Fetal cardiac autonomic control during breathing and non-breathing epochs: the effect of maternal exercise. Early human development 88, 539–546 (2012). 26. Lang, A., Del Giudice, R. & Schabus, M. Sleep, Little Baby: The Calming Effects of Prenatal Speech Exposure on Newborns' Sleep and Heartrate. Brain sciences 10 (2020). 27. Dereymaeker, A. et al. An Automated Quiet Sleep Detection Approach in Preterm Infants as aGateway to Assess Brain Maturation. International journal of neural systems 27, 1750023 (2017). Additional Declarations No competing interests reported. Supplementary Files Supplementarymaterial.docx Supplementary Material Cite Share Download PDF Status: Published Journal Publication published 01 Mar, 2022 Read the published version in Scientific Reports → Version 1 posted Editorial decision: Major revision 05 Dec, 2021 Reviews received at journal 29 Nov, 2021 Reviewers agreed at journal 24 Nov, 2021 Reviewers agreed at journal 12 Nov, 2021 Reviewers invited by journal 08 Nov, 2021 Editor assigned by journal 08 Nov, 2021 Editor invited by journal 20 Oct, 2021 Submission checks completed at journal 20 Oct, 2021 First submitted to journal 13 Oct, 2021 You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-966925","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":57834881,"identity":"3bacb50b-9ccb-4c96-b93f-43641e85aefd","order_by":0,"name":"Lorenzo Semeia","email":"data:image/png;base64,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","orcid":"","institution":"IDM/fMEG Center of the Helmholtz Center Munich at the University of Tübingen, University of Tübingen, German Center for Diabetes Research (DZD), Tübingen","correspondingAuthor":true,"submittingAuthor":false,"prefix":"","firstName":"Lorenzo","middleName":"","lastName":"Semeia","suffix":""},{"id":57834882,"identity":"2ab27693-8812-4ef3-9891-1ca946a8021c","order_by":1,"name":"Katrin Sippel","email":"","orcid":"","institution":"IDM/fMEG Center of the Helmholtz Center Munich at the University of Tübingen, University of Tübingen, German Center for Diabetes Research (DZD), Tübingen","correspondingAuthor":false,"submittingAuthor":false,"prefix":"","firstName":"Katrin","middleName":"","lastName":"Sippel","suffix":""},{"id":57834883,"identity":"04bb44ef-07f7-4a22-8934-8e67e1506743","order_by":2,"name":"Julia Moser","email":"","orcid":"","institution":"IDM/fMEG Center of the Helmholtz Center Munich at the University of Tübingen, University of Tübingen, German Center for Diabetes Research (DZD), Tübingen","correspondingAuthor":false,"submittingAuthor":false,"prefix":"","firstName":"Julia","middleName":"","lastName":"Moser","suffix":""},{"id":57834884,"identity":"6aacdc02-fe87-45b4-8ad2-d71ebd46c166","order_by":3,"name":"Hubert Preissl","email":"","orcid":"","institution":"IDM/fMEG Center of the Helmholtz Center Munich at the University of Tübingen, University of Tübingen, German Center for Diabetes Research (DZD), Tübingen","correspondingAuthor":false,"submittingAuthor":false,"prefix":"","firstName":"Hubert","middleName":"","lastName":"Preissl","suffix":""}],"badges":[],"createdAt":"2021-10-13 12:59:13","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-966925/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-966925/v1","draftVersion":[],"editorialEvents":[{"content":"https://doi.org/10.1038/s41598-022-07476-x","type":"published","date":"2022-03-01T11:57:25+00:00"}],"editorialNote":"","failedWorkflow":false,"files":[{"id":14805649,"identity":"836ad8b9-3d26-4e13-bcf8-8e190994910c","added_by":"auto","created_at":"2021-10-22 15:18:23","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":148033,"visible":true,"origin":"","legend":"Box and whiskers plots of the HRV parameters in the four defined states. A) Heart rate. B) Standard deviation of normal to normal R-R intervals. C) Root mean square of successive differences between normal heartbeats. ","description":"","filename":"floatimage1.png","url":"https://assets-eu.researchsquare.com/files/rs-966925/v1/1d792e2bf783410b0c9e0338.png"},{"id":14805652,"identity":"a22c14e6-bf6f-48ee-be82-41ac69dc125c","added_by":"auto","created_at":"2021-10-22 15:18:23","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":736982,"visible":true,"origin":"","legend":"Parameters distributions divided by gestational age. A) Baseline std. B) % points outside ± 5 bpm. C) % points outside ± 7.5 bpm (%). D) RMSSD in the HR. E) HR std. F) RMSSD in the actogram. G) Actogram std.","description":"","filename":"floatimage2.png","url":"https://assets-eu.researchsquare.com/files/rs-966925/v1/c9eacfa3a16e013485705c6f.png"},{"id":14805653,"identity":"51dc1d5c-c2b8-4494-b9a6-9078e8d09349","added_by":"auto","created_at":"2021-10-22 15:18:23","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":158496,"visible":true,"origin":"","legend":"Correlations [mean r-value ± standard deviation] between cardiogram-actogram. ","description":"","filename":"floatimage3.png","url":"https://assets-eu.researchsquare.com/files/rs-966925/v1/2983f4daa3eeeae28c0a97c9.png"},{"id":14805650,"identity":"89efdcfb-c67f-4816-b20e-d583d5891b3b","added_by":"auto","created_at":"2021-10-22 15:18:23","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":318297,"visible":true,"origin":"","legend":"Receiver operating characteristics (ROC) curve for A) early gestation and for B) late gestation.","description":"","filename":"floatimage4.png","url":"https://assets-eu.researchsquare.com/files/rs-966925/v1/0580157f7e35aa34ebc6a00b.png"},{"id":14805655,"identity":"07a028f0-f8b7-4518-90d3-29adaa6d5603","added_by":"auto","created_at":"2021-10-22 15:18:23","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":345850,"visible":true,"origin":"","legend":"Decrease over gestational age of RMSSD in the HR in active but not passive fBS.","description":"","filename":"floatimage5.png","url":"https://assets-eu.researchsquare.com/files/rs-966925/v1/133e57499679242e6a37aeaa.png"},{"id":14806133,"identity":"551fba9f-dd2c-47d2-9f96-be0bc3817b5e","added_by":"auto","created_at":"2021-10-22 15:21:23","extension":"png","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":336442,"visible":true,"origin":"","legend":"A) Increase of the correlation cardiogram-actogram regardless the fBS. B) This increase is driven by the significant increase of correlation during active states.","description":"","filename":"floatimage6.png","url":"https://assets-eu.researchsquare.com/files/rs-966925/v1/63bf504cb576d34c75fcd49c.png"},{"id":18729214,"identity":"c4660489-5e8e-4736-9ecd-cd7fa80f97a4","added_by":"auto","created_at":"2022-03-01 11:57:29","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":1940368,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-966925/v1/8559a001-c0b6-4158-9266-85682f35b9dc.pdf"},{"id":14805651,"identity":"2cbc092c-c25a-4492-82ea-00d7f6df75ce","added_by":"auto","created_at":"2021-10-22 15:18:23","extension":"docx","order_by":1,"title":"","display":"","copyAsset":false,"role":"supplement","size":19847,"visible":true,"origin":"","legend":"Supplementary Material","description":"","filename":"Supplementarymaterial.docx","url":"https://assets-eu.researchsquare.com/files/rs-966925/v1/2b212a8e4adb4ab2fd0fc214.docx"}],"financialInterests":"No competing interests reported.","formattedTitle":"\u003cp\u003eEvaluation of Parameters for Fetal Behavioural State Classification\u003c/p\u003e","fulltext":[{"header":"1. Introduction","content":"\u003cp\u003eIn 1974 Prechtl described for the first time \u0026lsquo;behavioural states\u0026rsquo; in infants [\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e], which is a concept that was later applied to fetuses [\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e]. Starting from 32 weeks of gestational age (GA), four fBS were identified, namely quiet sleep (1F), active sleep (2F), quiet awake (3F) and active awake (4F). The definition criteria of these states rely on body/eye movement and heart rate patterns that are stable for at least 3 minutes (see Table \u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e). Fetal states can be detected even before 32 weeks of GA but, in this case, it is only possible to define periods of rest (quiet states, QS) and periods of activity (active states, AS; [\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e]. The distribution of these fBS has been found to be stable over daytime[\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e] but depends on maternal position [\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e].\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab1\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eFetal behavioural states original criteria defined by Nijhuis and colleagues in 1982 [\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e].\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"5\"\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eState 1F\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eState 2F\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eState 3F\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eState 4F\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eBody movements\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eIncidental\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003ePeriodic\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eAbsent\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eContinuous\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eEye movements\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eAbsent\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003ePresent\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003ePresent\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003epresent\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eHeart rate patterns\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eStable heart rate, with small oscillation bandwidth. Accelerations strictly related to movement\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eWider oscillation bandwidth compared to 1F. Frequent accelerations related to body movements\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eStable heart rate. No acceleration and higher oscillation bandwidth compared to 1F\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eUnstable heart rate. Long lasting accelerations, often with tachycardia\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003eFetal states were investigated in relation to fetal health [\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e], fetal maturation [\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e] and the development of the autonomic nervous system (ANS) [\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e], [\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e], as well as in fetal cognitive processing [\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e]. With the advancement of technology, it was possible to correctly classify states even without the usage of all defined indicators.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab2\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 2\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eFetal behavioural states definition criteria by Br\u0026auml;ndle and colleagues from 2015 [\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e].\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"4\"\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003e1F\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003e2F\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003e4F\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eBaseline\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u0026lt;160 bpm\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e\u0026lt;160 bpm\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e\u0026gt;160 bpm possible\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eOscillation bandwidth\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u0026lt;= \u0026plusmn; 7.5 bpm\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e\u0026plusmn; 7.5 \u0026ndash; \u0026plusmn; 15bpm\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e\u0026gt;\u0026plusmn;15 bpm\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAccelerations\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eno\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e\u0026gt;15 bpm/\u0026gt;15s\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e\u0026gt; 30 bpm/\u0026gt;30s\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMovement\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eno\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eyes\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eyes\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003eFor example, Maeda described [\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e] how it was possible to define fBS without considering eye movements, adopting only the fetal heart rate and the gross body movement as available in the actocardiogram.\u003c/p\u003e \u003cp\u003eThis also enabled the investigation of fBS using fetal magnetocardiography (fMCG). This device allows to non-invasively record the magnetic signal of the fetal heart directly through the maternal abdomen. Compared to classical CTG and ultrasound, fMCG provides a higher temporal resolution and it is less susceptible to maternal interferences [\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e]. These recordings provide more accurate fetal heart dynamic parameters, like heart rate variability (HRV) measures.\u003c/p\u003e \u003cp\u003eThe gold standard for fBS classification using fMCG is a visual inspection and classification of the fetal cardiogram and actogram by experts. To facilitate this classification, the original Nijhuis criteria ([\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e], see Table \u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e) were adapted by Br\u0026auml;ndle and colleagues in 2015 ([\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e], see Table \u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e\u003cb\u003e)\u003c/b\u003e. Due to the low occurrence of state 3F in this adaption only states 1F, 2F, and 4F were included. The fetal actogram describes the movement of the fetal body by calculating the spatial location of the highest heart activity amplitude (heart vector) for every single heartbeat, and follows the changes of this location within the sensor array over time.\u003c/p\u003e \u003cp\u003eRecently, some algorithms have been developed to automatically classify fBS, both in CTG [\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e] and in fMCG [\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e]. Therefore, the original criteria defined by Nijhuis and colleagues [\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e] had to be translated into values that could be used as classification thresholds by those algorithms.\u003c/p\u003e \u003cp\u003eTo achieve the quality and accuracy of a visually-inspected evaluation performed by experts, automatic state detection algorithms rely on the combination of several heart and movement parameters and their more or less strictly defined thresholds. These thresholds are highly dependent on the specific datasets, and are applicable only to those datasets in which no artefacts are occurring. Although this is a practical solution for the study of prototypical fetal physiology during the states, these thresholds prevent the applicability of existing algorithms on datasets, for example where a transitions between states occurs. Moreover, given the dynamic development of fBS over gestation, for classification purposes researchers usually separate the early and the late gestation (before and after 32 weeks of GA). Altogether, these factors make it difficult to obtain a satisfactory classification of fBS. In the current work we aim to perform an exploratory analysis of commonly used fBS classification parameters, like variability in the heart-rate or in the actogram, to determine the reliability of these parameters. Secondly, the development of these parameters over gestation is investigated.\u003c/p\u003e \u003cp\u003eThe first part (Step1, \u0026ldquo;Parameter distribution and thresholds\u0026rdquo;) of the current work will investigate the accuracy of the fBS classification parameters. Therefore, we analysed the distribution of the main parameters commonly used in fBS algorithms.\u003c/p\u003e \u003cp\u003eSecond (Step2, \u0026ldquo;Best classifier\u0026rdquo;), we want to verify which of the heart-rate (HR) or movement parameters are the best parameters for state classification performance both in early as in late gestation.\u003c/p\u003e \u003cp\u003eFinally (Step3, \u0026ldquo;Changes over time\u0026rdquo;), we investigate changes and development of these parameters over the last trimester of gestation, in order to highlight possible age-dependent trends.\u003c/p\u003e"},{"header":"2. Methods","content":"\u003cdiv class=\"Section2\" id=\"Sec3\"\u003e\n \u003ch2\u003e2.1. Fetal Magnetoencephalography (fMEG)\u003c/h2\u003e\n \u003cp\u003efMEG is a non-invasive device for the recording of brain and heart activity of fetuses in the last trimester of gestation, and of newborns in the first weeks of life [\u003cspan class=\"CitationRef\"\u003e15\u003c/span\u003e]. The datasets included in the current study were recorded using a SARA (SQUID Array for Reproductive Assessment, VSM MedTech Ltd., Port Coquitlam, BC, Canada) system available at the fMEG Center at the University of Tuebingen, Germany. All the recordings were sampled with a frequency of 610,3516 Hz. This device comprehends 156 magnetic sensors and 29 additional reference sensors. These sensors are placed in a concave array, designed to match the maternal abdomen. In order to limit the influence of external magnetic fields, the device is placed in a magnetically shielded room (Vakuumschmelze, Hanau, Germany).\u003c/p\u003e\n\u003c/div\u003e\n\u003cdiv class=\"Section2\" id=\"Sec4\"\u003e\n \u003ch2\u003e2.2. Datasets\u003c/h2\u003e\n \u003cp\u003eThe 169 original datasets (117 for fetuses younger than 32 weeks of GA) used in the current analysis come from previously published studies [\u003cspan class=\"CitationRef\"\u003e16\u003c/span\u003e], [\u003cspan class=\"CitationRef\"\u003e17\u003c/span\u003e]. All recordings were performed in the morning. In addition, previous data showed that the fBS distribution since is stable over daytime [\u003cspan class=\"CitationRef\"\u003e4\u003c/span\u003e]. These datasets were previously screened for good quality of the raw data in terms of absence of large amplitude artefacts. These studies were approved by the Ethical Committee of the Medical Faculty at the University of Tuebingen (No. 476/2008MPG1 and 339/2010BO1). All the participants to these studies gave their informed consent in accordance to the Declaration of Helsinki and agreed for the usage of the data in further studies. The datasets include recordings of spontaneous activity as well as recordings during an auditory paradigm. The datasets length varies between 6 and 15 minutes.\u003c/p\u003e\n\u003c/div\u003e\n\u003cdiv class=\"Section2\" id=\"Sec5\"\u003e\n \u003ch2\u003e2.3. Preprocessing\u003c/h2\u003e\n \u003cp\u003eThe data preprocessing was performed using Matlab R2016b (The MathWorks, Natick, MA, USA). R-peak detection of maternal and fetal heart has been performed using FLORA [\u003cspan class=\"CitationRef\"\u003e18\u003c/span\u003e]. Maternal heart interference was removed by FAUNA [\u003cspan class=\"CitationRef\"\u003e19\u003c/span\u003e]. HRV parameters in the time domain (mean heart-rate (HR), the standard deviation of normal to normal R-R intervals (SDNN) and the root mean square of successive differences between normal heartbeats (RMSSD) were calculated for the fMCG signals [\u003cspan class=\"CitationRef\"\u003e20\u003c/span\u003e]. Finally, fetal actogram and cardiogram were calculated as described by Govindan and colleagues [\u003cspan class=\"CitationRef\"\u003e21\u003c/span\u003e].\u003c/p\u003e\n\u003c/div\u003e\n\u003cdiv class=\"Section2\" id=\"Sec6\"\u003e\n \u003ch2\u003e2.4. Fetal states analysis\u003c/h2\u003e\n \u003cp\u003eThe fetal behavioural state underlying the datasets was automatically detected using an algorithm adapted from the work of Vairavan and colleagues [\u003cspan class=\"CitationRef\"\u003e14\u003c/span\u003e] and implemented in Matlab R2019b.\u003c/p\u003e\n \u003cp\u003eWe tested the algorithm on three groups based of the fetal gestational age (GA), an early, a middle, and a late group. The early group includes datasets from fetuses between 27-32 (27 0/1-32 6/7) weeks of GA, the middle group fetuses between 33-36 (33 0/1-36 6/7) weeks of GA, and the late group includes fetuses between 37-39 (37 0/1-39 6/7) weeks GA. According to the work of Vairavan and colleagues [\u003cspan class=\"CitationRef\"\u003e14\u003c/span\u003e], the percentage of accelerations in the HR, the standard deviation of the HR (\u0026sigma;(HR)) and the percentage of points above 160 bmp were computed. The thresholds for these values used in our algorithm are also from Vairavan and collaborators [\u003cspan class=\"CitationRef\"\u003e14\u003c/span\u003e]. To differentiate between active and passive states in fetuses from 27-29 (27 0/1-29 6/7) weeks of GA we applied the threshold of 5.54% of accelerations for dividing between passive (\u0026le; 5.54%) and active (\u0026gt; 5.54%) states. This threshold is based on Vairavan and colleagues [\u003cspan class=\"CitationRef\"\u003e14\u003c/span\u003e], where it is used for dividing between active and passive states between 30-36 (30 0/1- 36 6/7) weeks of GA.\u003c/p\u003e\n \u003cp\u003eOut of the included 169 datasets, a state was classified for approximately 80% of the total data length. In the remaining 20% of data, the variability of parameters used for classification (e.g. the percentage of acceleration in the HR) was exceeding the tolerance of the detection algorithm, thus preventing a fBS classification. This variability is likely due to transitions between states or to signal noise.\u003c/p\u003e\n \u003cp\u003eTo maintain a reasonable amount of data for each group, we considered the middle and late group as a whole late group for all the further analyses. From the algorithm outputs, 120 data segments (from 85 different recordings performed in 52 subjects) were selected based on a clear detection of the behavioural state: 31 for active and 27 for passive in the early group, 31 for 1F and 31 for 2F in the late group. The goal of the selection was to have an equivalent amount of datasets in each group however, no more than 27 passive state windows were available. Yet, a lower occurrence of passive states is expected [\u003cspan class=\"CitationRef\"\u003e22\u003c/span\u003e]. 3F and 4F were not included because we were not able to detect enough data segments. The mean duration of these 120 data segments was 266\u0026plusmn;44 seconds.\u003c/p\u003e\n \u003cp\u003eThe data segments were then visually evaluated by two experts to gain security about the result of the automatically classified states. There was agreement between the two raters in 96 windows (80% agreement). 87 out of these 96 windows were also in agreement with the automatic states classification algorithm. Following this procedure, our further analysis only contains those 87 datasets in which there was accordance between all 3 ratings (see Table \u003cspan class=\"InternalRef\"\u003e3\u003c/span\u003e).\u003c/p\u003e\n \u003cdiv class=\"gridtable\"\u003e\u0026nbsp;\u003ctable border=\"1\" id=\"Tab3\"\u003e\n \u003ccaption language=\"En\"\u003e\n \u003cdiv class=\"CaptionNumber\"\u003eTable 3\u003c/div\u003e\n \u003cdiv class=\"CaptionContent\"\u003e\n \u003cp\u003e\u003cem\u003eCharacteristics of the 87 data segments that were chosen for further evaluation [mean\u003c/em\u003e \u0026plusmn; \u003cem\u003estandard deviation].\u003c/em\u003e\u003c/p\u003e\n \u003c/div\u003e\n \u003c/caption\u003e\n \u003cthead\u003e\n \u003ctr\u003e\n \u003cth align=\"left\"\u003e\u0026nbsp;\u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003ePassive\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eActive\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003e1F\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003e2F\u003c/p\u003e\n \u003c/th\u003e\n \u003c/tr\u003e\n \u003c/thead\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eWindows length, in seconds\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e288.63 \u0026plusmn; 30.83\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e252.81 \u0026plusmn; 43.98\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e283.60 \u0026plusmn; 34.07\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e250.00 \u0026plusmn; 52.98\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eNumber of windows\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e19\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e31\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e15\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e22\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eGA, in weeks\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e29.21 \u0026plusmn; 1.47\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e29.52 \u0026plusmn; 1.43\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e35.13 \u0026plusmn; 1.41\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e34.95 \u0026plusmn; 1.65\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n \u003c/table\u003e\n \u003c/div\u003e\n \u003cp\u003e\u003cbr\u003e\u003c/p\u003e\n\u003c/div\u003e\n\u003cdiv class=\"Section2\" id=\"Sec7\"\u003e\n \u003ch2\u003e2.5. Parameters definition\u003c/h2\u003e\n \u003cp\u003eThe following parameters, which are based on the states definition criteria reported in Table \u003cspan class=\"InternalRef\"\u003e1\u003c/span\u003e and Table \u003cspan class=\"InternalRef\"\u003e2\u003c/span\u003e, are included in the further analysis. For a summary of the parameters see Table \u003cspan class=\"InternalRef\"\u003e4\u003c/span\u003e.\u003c/p\u003e\n \u003cp\u003eHRV parameters (mean HR, SDNN and RMSSD) are calculated in order to verify whether their values are comparable to previous work which investigated fBS [\u003cspan class=\"CitationRef\"\u003e9\u003c/span\u003e]. Therefore, their classification accuracy and development over time are not further investigated. In the cardiogram, we defined A) Baseline standard deviation (\u003cstrong\u003eBaseline std\u003c/strong\u003e). The baseline is defined as smoothing the HR in a two minutes moving window with 1 second shift. We considered the baseline standard deviation as an index of stability in the heart rate. B) Percentage of points outside the \u0026plusmn; 5 bpm oscillation bandwidth (\u003cstrong\u003e% points outside \u0026plusmn; 5 bpm\u003c/strong\u003e), based on Nijhuis and colleagues [\u003cspan class=\"CitationRef\"\u003e2\u003c/span\u003e]. The oscillation bandwidth is defined as the area between the baseline and the baseline \u0026plusmn; 5 bpm. C) Percentage of points outside the \u0026plusmn; 7.5 bpm oscillation bandwidth (\u003cstrong\u003e% points outside \u0026plusmn; 7.5 bpm\u003c/strong\u003e) based on Br\u0026auml;ndle and colleagues [\u003cspan class=\"CitationRef\"\u003e9\u003c/span\u003e]. The oscillation bandwidth is defined as the area between the baseline and the baseline \u0026plusmn;7.5 bpm. D) Root mean square of successive differences in the HR (\u003cstrong\u003eRMSSD in the HR\u003c/strong\u003e). This parameter has been chosen as an index of variance in the heart rate. In contrast to the HRV parameter RMSSD, which is calculated on the time differences between successive R peaks, this parameter is directly calculated on HR values. E) Heart rate std (\u003cstrong\u003eHR std\u003c/strong\u003e). This is an index of variability in the heart rate.\u003c/p\u003e\n \u003cp\u003eIn the actogram we defined F) Root mean square of successive differences in the actogram (\u003cstrong\u003eRMSSD in the actogram\u003c/strong\u003e). This parameter has been selected as an index of short term variability in the actogram. G) Actogram standard deviation (\u003cstrong\u003eactogram std\u003c/strong\u003e). This is an index of variability in the actogram.\u003c/p\u003e\n \u003cp\u003eFinally, since in the original definition by Nijhuis and colleagues [\u003cspan class=\"CitationRef\"\u003e2\u003c/span\u003e] fBS are also defined based on the relation between heart-rate and movement accelerations, we calculated the correlation between cardiogram and actogram to explore whether this parameter could be used for states classification.\u003c/p\u003e\n\u003c/div\u003e\n\u003cdiv class=\"Section2\" id=\"Sec8\"\u003e\n \u003ch2\u003e2.6. Statistics\u003c/h2\u003e\n \u003cp\u003eStatistics were performed with Matlab R2019b for Windows. Before proceeding, outliers for each of the parameters, defined as a value exceeding the mean \u0026plusmn; 3 standard deviations, were removed. Then, the Kolmogorov-Smirnov test was used to test the normality assumption required by the following statistical analysis.\u0026nbsp;\u003c/p\u003e\n \u003ctable border=\"1\" id=\"Tab4\"\u003e\n \u003ccaption language=\"En\"\u003e\n \u003cdiv class=\"CaptionNumber\"\u003eTable 4\u003c/div\u003e\n \u003cdiv class=\"CaptionContent\"\u003e\n \u003cp\u003eSummary of the parameters used in the current work.\u003c/p\u003e\n \u003c/div\u003e\n \u003c/caption\u003e\n \u003cthead\u003e\n \u003ctr\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eCardiogram\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eActogram\u003c/p\u003e\n \u003c/th\u003e\n \u003c/tr\u003e\n \u003c/thead\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eBaseline std\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eRMSSD in the actogram\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e% points outside \u0026plusmn; 5 bpm\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eActogram std\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e% points outside \u0026plusmn; 7.5 bpm\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eRMSSD in the HR\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eHR std\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\" colspan=\"2\"\u003e\n \u003cp\u003eCorrelation between cardiogram-actogram\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n \u003c/table\u003e\n \u003cp\u003e\u003c/p\u003e\n \u003cp\u003e\u003cspan class=\"Underline\" name=\"Emphasis\" type=\"Underline\"\u003eStep 1. Parameter distribution and thresholds.\u003c/span\u003e\u003c/p\u003e\n \u003cp\u003eWe tested the alternative hypothesis that there is a difference between HRV, cardio- or actogram parameters between active and passive (H1: \u0026micro;Active - \u0026micro;Passive \u0026ne; 0), or between 1F and 2F (H1: \u0026micro;2F - \u0026micro;1F \u0026ne; 0). \u003cspan class=\"Underline\" name=\"Emphasis\" type=\"Underline\"\u003eA t-test was performed for the analysis of the different parameters between active and passive, and between 1F and 2F. After Bonferroni correction for N=10 different parameters, the significance level was set to p \u0026lt; 0.005.\u003c/span\u003e\u003c/p\u003e\n \u003cp\u003e\u003cspan class=\"Underline\" name=\"Emphasis\" type=\"Underline\"\u003eStep 2. Best classifier.\u003c/span\u003e\u003c/p\u003e\n \u003cp\u003eReceiver operating characteristics (ROC) curves were defined to verify the performance of each parameter in classifying a state.\u003c/p\u003e\n \u003cp\u003e\u003cspan class=\"Underline\" name=\"Emphasis\" type=\"Underline\"\u003eStep 3. Changes over time.\u003c/span\u003e\u003c/p\u003e\n \u003cp\u003eFinally, to detect the development of parameters during gestation, linear models were built for each parameter. In the models, we combined passive (early passive +1F) and active (early active + 2F) states between 27 and 39 weeks of GA The model was defined as:\u003c/p\u003e\n \u003cp\u003e𝑦 = 𝛽0 + 𝛽1X + 𝜀\u003c/p\u003e\n \u003cp\u003ewhere, 𝑦 is one of the parameters included in the current analysis, 𝛽0 is the intercept, 𝛽1 the regression coefficient, X the gestational age, and 𝜀 the error in the estimate. Model values of p \u0026lt; 0.05 were considered statistically significant and effect size, defined by r\u003csup\u003e2\u003c/sup\u003e.\u003c/p\u003e\n\u003c/div\u003e"},{"header":"3. Results","content":"\u003cp\u003e \u003cb\u003eStep 1. Parameters distribution and thresholds\u003c/b\u003e \u003c/p\u003e \u003cp\u003e \u003cspan type=\"Underline\" class=\"Underline\" name=\"Emphasis\"\u003eHRV parameters\u003c/span\u003e \u003c/p\u003e \u003cp\u003eThe t-test between states revealed no difference in the mean HR both between passive and active in the early gestation (t(48) = 1.46, p = 0.15), and between 1F and 2F in the late gestation (t(35) = 0.38, p = 0.70) (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003eA). SDNN was higher in active compared to passive state (t(45) = 9.20, p \u0026lt; 0.001), and higher in 2F compared to 1F (t(35) = 7.36, p \u0026lt; 0.001) (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003eB\u003cb\u003e)\u003c/b\u003e. Finally, RMSSD was higher in active compared to passive state (t(48) = 10.43, p \u0026lt; 0.001), and higher in 2F compared to 1F (t(35) = 6.40, p \u0026lt; 0.001) (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003eC).\u003c/p\u003e \u003cp\u003eAlthough two out of three HRV parameters showed a significant difference both in early and in late gestation, only the RMSSD values in the early gestation could be separated by a single threshold. The distributions of all other parameters overlap.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e \u003cspan type=\"ItalicUnderline\" class=\"ItalicUnderline\" name=\"Emphasis\"\u003eCardio- and actogram parameters\u003c/span\u003e \u003c/p\u003e \u003cp\u003eIn the early gestation, the t-tests revealed that all parameters were higher in active compared to passive states: \u0026lsquo;Baseline std\u0026rsquo; (t(45) = 5.59, p \u0026lt; 0.001), \u0026lsquo;% points outside \u0026plusmn; 5 bpm\u0026rsquo; (t(46) = 13.47, p \u0026lt; 0.001), \u0026lsquo;% points outside \u0026plusmn; 7.5 bpm\u0026rsquo; (t(46) = 14.16, p \u0026lt; 0.001), \u0026lsquo;RMSSD in the HR\u0026rsquo; (t(42) = 13.54, p \u0026lt; 0.001), \u0026lsquo;HR std\u0026rsquo; (t(44) = 13.80, p \u0026lt; 0.001), \u0026lsquo;RMSSD in the actogram (t(43) = 3.46, p = 0.001), and \u0026lsquo;Actogram std\u0026rsquo; (t(42) = 4.22, p \u0026lt; 0.001).\u003c/p\u003e \u003cp\u003eAlso in late gestation, the t-tests revealed that all parameters were higher in 2F compared to 1F: \u0026lsquo;Baseline std\u0026rsquo; (t(34) = 5.28, p \u0026lt; 0.001), \u0026lsquo;% points outside \u0026plusmn; 5 bpm\u0026rsquo; (t(32) = 16.73, p \u0026lt; 0.001), \u0026lsquo;% points outside \u0026plusmn; 7.5 bpm\u0026rsquo; (t(35) = 15.91, p \u0026lt; 0.001), \u0026lsquo;RMSSD in the HR\u0026rsquo; (t(32) = 6.97, p \u0026lt; 0.001), \u0026lsquo;HR std\u0026rsquo; (t(32) = 11.65, p \u0026lt; 0.001), \u0026lsquo;RMSSD in the actogram (t(32) = 3.67, p \u0026lt; 0.001), and \u0026lsquo;Actogram std\u0026rsquo; (t(34) = 4.07, p \u0026lt; 0.001).\u003c/p\u003e \u003cp\u003eAlthough all of the seven cardio- and actogram parameters show significant differences in their distribution in early and late gestation, only three parameters in early and two in late gestation could be separated by a single threshold, while the distributions of all the other parameters overlap (see Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eFor details about the mean parameters values, see supplementary table 1 and supplementary table 2.\u003c/p\u003e \u003cp\u003e \u003cspan type=\"ItalicUnderline\" class=\"ItalicUnderline\" name=\"Emphasis\"\u003eCorrelation cardiogram-actogram\u003c/span\u003e \u003c/p\u003e \u003cp\u003eCorrelating the cardiogram and the actogram, the t-tests revealed no difference in the mean r-value between passive (M = -0.01, SD = 0.22) and active (M = 0.08, SD = 0.26) in the early gestation (t(48) = 1.32, p = 0.19). There was a significant difference between 1F (M = 0.09, SD = 0.21) and 2F (M = 0.28, SD = 0.27) in the late gestation (t(35) = 2.33, p = 0.03) (see Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003e).\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e \u003cb\u003eStep 2. Best classifier\u003c/b\u003e \u003c/p\u003e \u003cp\u003e \u003cspan type=\"ItalicUnderline\" class=\"ItalicUnderline\" name=\"Emphasis\"\u003eROC analysis.\u003c/span\u003e \u003c/p\u003e \u003cp\u003eIn the early gestation, \u0026lsquo;% points outside \u0026plusmn; 5 bpm\u0026rsquo;, \u0026lsquo;% points outside \u0026plusmn; 7.5 bpm\u0026rsquo;, \u0026lsquo;HR std\u0026rsquo; and \u0026lsquo;RMSSD in the HR\u0026rsquo; are perfect classifier (AUC = 1), \u0026lsquo;Baseline std\u0026rsquo; has a high performance (AUC = 0.93), \u0026lsquo;RMSSD in the actogram\u0026rsquo; (AUC = 0.80) and \u0026lsquo;Actogram std\u0026rsquo; (AUC = 0.83) perform less compared to the other classifier (see Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003eA).\u003c/p\u003e \u003cp\u003eIn the late gestation the results are very similar. \u0026lsquo;% points outside \u0026plusmn; 5 bpm\u0026rsquo;, \u0026lsquo;% points outside \u0026plusmn; 7.5 bpm\u0026rsquo;, \u0026lsquo;HR std\u0026rsquo; are perfect classifier (AUC = 1), \u0026lsquo;Baseline std\u0026rsquo; (AUC = 0.92) and \u0026lsquo;RMSSD in the HR\u0026rsquo; (AUC = 0.98) have a high performance, \u0026lsquo;RMSSD in the actogram\u0026rsquo; (AUC = 0.87) and \u0026lsquo;Actogram std\u0026rsquo; (AUC = 0.86) perform less compared to the other classifier (see Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003eB).\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e \u003cb\u003eStep 3. Changes over time\u003c/b\u003e \u003c/p\u003e \u003cp\u003e \u003cspan type=\"ItalicUnderline\" class=\"ItalicUnderline\" name=\"Emphasis\"\u003eDevelopment of parameters over gestational age, not divided by fBS\u003c/span\u003e \u003c/p\u003e \u003cp\u003eFirst of all, we tested whether the parameters change with GA regardless the fBS (see supplementary table 3). The parameters did not show any general change during development.\u003c/p\u003e \u003cp\u003e \u003cspan type=\"ItalicUnderline\" class=\"ItalicUnderline\" name=\"Emphasis\"\u003eDevelopment of parameters over gestational age, divided by fBS\u003c/span\u003e \u003c/p\u003e \u003cp\u003eFurthermore, we investigated whether the parameters change over time depending on the fBS. For this analysis, we combined passive (early passive + 1F) and active (early active + 2F) states to detect their general evolution over time (see supplementary table 4). Only the RMSSD in the HR showed a difference between states. In particular, the decline can be seen in active states (p = 0.01, r\u003csup\u003e2\u003c/sup\u003e = 0.13) but not in passive states (p = 0.30; Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003e).\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e \u003cspan type=\"ItalicUnderline\" class=\"ItalicUnderline\" name=\"Emphasis\"\u003eDevelopment of the correlation cardiogram-actogram over gestational age\u003c/span\u003e \u003c/p\u003e \u003cp\u003eThirdly, we looked at the evolution of the correlation between cardiogram and actogram over time (Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003eA). We found an overall increase of the r-values over gestational age (p = 0.001, Effect size = 0.18).\u003c/p\u003e \u003cp\u003eIntroducing the behavioural state as further parameter, we found an increase in the correlation between actogram and cardiogram for active states (p \u0026lt; 0.001, r\u003csup\u003e2\u003c/sup\u003e = 0.22) but not for passive states (p = 0.2) (Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003eB).\u003c/p\u003e \u003cp\u003e \u003c/p\u003e"},{"header":"4. Discussion","content":"\u003cp\u003eThis work systematically investigated different parameters commonly used for fBS classification. Although nearly all investigated parameters differed significantly in their distribution between active and passive, or between 1F and 2F, only a few of these parameters had non-overlapping distributions, preventing a clear separation and classification of the states.\u003c/p\u003e\n\u003cp\u003eFurthermore, we identified two parameters that continuously changed during fetal development. We therefore argue that they are not suitable for a general fBS classification.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eStep 1. Parameters distribution and thresholds\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe first aim of the current work was to investigate the distribution of properties of the acto- and cardiogram over fetal behavioural states. As expected, different parameter distributions are representative for the different states, as confirmed by the significant differences between parameters across states. Moreover, our HRV parameters (mean HR, SDNN and RMSSD) were in line with previous works [\u003cspan\u003e9\u003c/span\u003e].\u003c/p\u003e\n\u003cp\u003eSome emphasis should be put on the parameters \u0026lsquo;% points outside \u0026plusmn; 5 bpm\u0026rsquo; and \u0026lsquo;% points outside \u0026plusmn; 7.5 bpm\u0026rsquo;. For example, in Br\u0026auml;ndle and colleagues [\u003cspan\u003e9\u003c/span\u003e], the authors operationalised the variability of the HR considering the oscillation bandwidth in which the HR could vary. Specifically, it is assumed that the HR does not exceed the [baseline \u0026plusmn; 7.5] bpm in state 1F. Although this value could efficiently divide between active and passive states during early or late gestation, this threshold was exceeded in each data-window. In states classified as passive, up to 4% of the total data points exceeded this value. This raises the question, whether the threshold value should be increased (e.g. \u0026plusmn; 10 bpm), or whether an increase would instead classify active states as passive and vice-versa. Our suggestion would be to define a certain tolerance around this threshold. For example, in our datasets this value could be exceeded in 5% of data points in a passive state and still provide the same results in terms of classification.\u003c/p\u003e\n\u003cp\u003eAnother point is the correlation between the cardiogram and the actogram. In Nijhuis and colleagues [\u003cspan\u003e2\u003c/span\u003e], the authors defined the fBS by means of the different ways heart-rate and body movement relate to each other. As a proxy for this coupling, in our work we explored how heart rate and body movement are related through analysis of the correlation between the two signals. Generally, the correlation was weak, with r-values both positive and negative (Fig. \u003cspan\u003e3\u003c/span\u003e). Given the high variability and wide overlap of this parameter over fBS, this parameter does not appear specific of a particular state. However, as discussed later, this index has interesting developmental characteristics.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eStep 2. Best classifier\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe ROC analysis revealed that the parameters derived from the cardiogram have very high classification performance, as noted by the AUC of almost 1 in every case (Fig. \u003cspan\u003e4\u003c/span\u003e). Even though movement plays an important role in the initial Nijhuis criteria, the actogram parameters, as defined in the current work, show the worst classification performance with an AUC around 0.80.\u003c/p\u003e\n\u003cp\u003eTherefore, at least in the frame of fMCG recordings, we argue that the actogram does not improve an automatic fBS classification. However, it should be considered that the actogram only represents gross-body movements and does not represent the movements of extremities.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eStep 3. Changes over time\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eWe found a decrease over gestation in the parameter RMSSD in the HR in active states but not in passive states (Fig. \u003cspan\u003e5\u003c/span\u003e). Even though there is no accordance on the development of short term variability in the heart patterns during the last trimester of gestation, with some works reporting an increase [\u003cspan\u003e23\u003c/span\u003e] and others no changes [\u003cspan\u003e9\u003c/span\u003e], in our case we found a decrease which depends on the activity state (Fig. \u003cspan\u003e5\u003c/span\u003e).\u003c/p\u003e\n\u003cp\u003eThe progressive decline in short-term variability in the heart-rate, as expressed in our findings by the RMSSD in the HR, makes the distinction between active and passive states, based on this parameter, less clear as gestational age increases. This is compatible with evidence for no substantial differences in short term variability in the heart-rate across different sleep stages later in childhood and adolescence [\u003cspan\u003e24\u003c/span\u003e].\u003c/p\u003e\n\u003cp\u003eFurthermore, we found an increasing correlation following gestational age between cardiogram and actogram only in active states (Fig. \u003cspan\u003e6\u003c/span\u003e). One reason for the developmental change of this relationship could be a change in fetal respiratory movements. For example, as Schneider and colleague[\u003cspan\u003e23\u003c/span\u003e] report, the increase in high frequency HRV close to term could be explained taking into account the fetal respiratory sinus arrhythmia, which is associated with thoracic movements and respiratory efforts [\u003cspan\u003e25\u003c/span\u003e]. Even if these respiratory movements are not exclusive of a particular fBS, they are indeed more common in periods of active sleep.\u003c/p\u003e\n\u003cp\u003eThese developmental trends, found in the correlation between cardiogram and actogram, and in the RMSSD in the HR, could reflect the maturational changes of the fetal autonomous nervous system and makes these parameters less suitable for fBS analysis over different gestational ages.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eLimitations and Outlook\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eIn the present analysis wake states, like 3F and 4F were excluded due to their low occurrence. Including a substantially higher number of datasets could shed light on the parameter distribution over these fBS. Moreover, in contrast to Vairavan and colleagues [\u003cspan\u003e14\u003c/span\u003e], we grouped together fetuses from the middle (32-36 weeks of GA) and late (37-39 weeks of GA) in a whole late group.\u003c/p\u003e\n\u003cp\u003eFurthermore, the data segments included in the current analysis were selected by a states detection algorithm and validated by two external raters, which results in selecting very prototypical data segments. In the algorithm used in this study (adapted from Vairavan and colleagues [\u003cspan\u003e14\u003c/span\u003e]) data segments with a parameter distribution that does not fit the prototypical characteristic of a certain fBS, are discarded. Therefore, an algorithm with clearly defined thresholds has a problem with classifying less clear datasets, or datasets with a transition between states. Classification of transitions between wake and sleep states is not only relevant for understanding basic sleep physiology, but also for the study of cognitive processing [\u003cspan\u003e26\u003c/span\u003e].\u003c/p\u003e\n\u003cp\u003eTo improve the detection of fBS, one option could be to consider the probability of the fetuses being in one of the fBS compared to the others. In fact, considering the distribution of parameters reported before, it appears that a certain value for a parameter is more likely to occur in a certain state compared to another state. This likelihood could be used for a more dynamic state detection, accounting for transitions. Another approach could then be a dynamic definition of the fBS using the fetal brain data extracted during the measurement. fMEG allows, theoretically, to also use the fetal brain activity as an additional parameter. Even in preterm infants, for example, a state classification based on the electroencephalographic data is at the moment the frontier for automatic state classification [\u003cspan\u003e27\u003c/span\u003e].\u003c/p\u003e"},{"header":"Conclusion","content":"\u003cp\u003eOur results indicate that cardiogram parameters have a higher classification accuracy in comparison to those extracted from the actogram. Therefore, we propose for the automation of fBS detection is to concentrate on the cardiogram instead of body movement as defined by the actogram. Moreover, the changes over time in some of the parameters in our opinion disqualify them from being part of a fBS classification algorithm that simply divides fetuses into early and late groups. In fact, given the parameters continuous change over time, we instead suggest to consider fetal gestational age in weeks. State of the art algorithms, like the one from Vairavan and colleagues [14], are easily applicable and partly reflect the original fBS definition criteria. On the other side, these algorithms may be too simplistic, thus failing to grasp more complex physiological events such as transitions between states. \u0026nbsp;Based on these conclusions, our suggestion is therefore the development of a probabilistic approach for the fBS detection that is more dynamic in detecting the alternating patterns between wake and sleep, which could possibly include fetal brain parameters.\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003e5. Data availability\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe datasets adopted in the current study are available from the corresponding author on a reasonable request.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e7. Acknowledgment\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis work was partially supported by the DAAD (Deutscher Akademischer Austauschdienst) and\u0026nbsp;the International Max Planck Research School for the Mechanisms of Mental Function and Dysfunction (IMPRS-MMFD).\u003c/p\u003e\n\u003cp\u003eWe also thank Dr. Franziska Schleger and Dr. Isabelle Kiefer-Schmidt for their support during the fetal state evaluation.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e8. Author contributions\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eL.S. conceptualized and conducted the analysis, drafted and revised the manuscript, and prepared the figures. L.S. is also the corresponding author. K.S. and J.M. assisted during the analysis, and reviewed and edited the manuscript. H.P. supervised the work, reviewed and edited the manuscript, and managed the project. All authors read and approved the revised manuscript.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e9. Conflict of interest statement\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe authors declare no conflicts of interest.\u003c/p\u003e"},{"header":"6. References","content":"\u003cp\u003e1. Prechtl, H. F. The behavioural states of the newborn infant (a review). \u003cem\u003eBrain research\u0026nbsp;\u003c/em\u003e\u003cstrong\u003e76(2),\u0026nbsp;\u003c/strong\u003e185\u0026ndash;212 (1974).\u003c/p\u003e\n\u003cp\u003e2. Nijhuis, J. G., Prechtl, H. F. R., Martin Jr, C. B. \u0026amp; Bots, R. S. G. M. Are there behavioural states in the human fetus? \u003cem\u003eEarly human development\u0026nbsp;\u003c/em\u003e\u003cstrong\u003e6(2),\u0026nbsp;\u003c/strong\u003e177\u0026ndash;195 (1982).\u003c/p\u003e\n\u003cp\u003e3. Pillai, M. \u0026amp; James, D. The development of fetal heart rate patterns during normal pregnancy. \u003cem\u003eObstetrics and Gynecology\u0026nbsp;\u003c/em\u003e\u003cstrong\u003e76,\u0026nbsp;\u003c/strong\u003e812\u0026ndash;816 (1990).\u003c/p\u003e\n\u003cp\u003e4. Sonanini, A.\u003cem\u003e\u0026nbsp;et al.\u0026nbsp;\u003c/em\u003eFetal behavioral states are stable over daytime - evidence by longitudinal and cross-sectional fetal biomagnetic recordings. \u003cem\u003eJournal of perinatal medicine\u0026nbsp;\u003c/em\u003e\u003cstrong\u003e42,\u0026nbsp;\u003c/strong\u003e307\u0026ndash;314 (2014).\u003c/p\u003e\n\u003cp\u003e5. Stone, P. R.\u003cem\u003e\u0026nbsp;et al.\u0026nbsp;\u003c/em\u003eEffect of maternal position on fetal behavioural state and heart rate variability in healthy late gestation pregnancy. \u003cem\u003eThe Journal of physiology\u0026nbsp;\u003c/em\u003e\u003cstrong\u003e595,\u0026nbsp;\u003c/strong\u003e1213\u0026ndash;1221 (2017).\u003c/p\u003e\n\u003cp\u003e6. Martin, C. B. Normal fetal physiology and behavior, and adaptive responses with hypoxemia. \u003cem\u003eSeminars in perinatology\u0026nbsp;\u003c/em\u003e\u003cstrong\u003e32,\u0026nbsp;\u003c/strong\u003e239\u0026ndash;242 (2008).\u003c/p\u003e\n\u003cp\u003e7. Wakai, R. T. Assessment of fetal neurodevelopment via fetal magnetocardiography. \u003cem\u003eExperimental neurology\u0026nbsp;\u003c/em\u003e\u003cstrong\u003e190 Suppl 1,\u0026nbsp;\u003c/strong\u003eS65-71 (2004).\u003c/p\u003e\n\u003cp\u003e8. Hoyer, D.\u003cem\u003e\u0026nbsp;et al.\u0026nbsp;\u003c/em\u003eFetal development of complex autonomic control evaluated from multiscale heart rate patterns. \u003cem\u003eAmerican journal of physiology. Regulatory, integrative and comparative physiology\u0026nbsp;\u003c/em\u003e\u003cstrong\u003e304,\u0026nbsp;\u003c/strong\u003eR383-92 (2013).\u003c/p\u003e\n\u003cp\u003e9. Br\u0026auml;ndle, J.\u003cem\u003e\u0026nbsp;et al.\u0026nbsp;\u003c/em\u003eHeart rate variability parameters and fetal movement complement fetal behavioral states detection via magnetography to monitor neurovegetative development. \u003cem\u003eFrontiers in human neuroscience\u0026nbsp;\u003c/em\u003e\u003cstrong\u003e9,\u0026nbsp;\u003c/strong\u003e147 (2015).\u003c/p\u003e\n\u003cp\u003e10. Moser, J.\u003cem\u003e\u0026nbsp;et al.\u0026nbsp;\u003c/em\u003eMagnetoencephalographic signatures of conscious processing before birth. \u003cem\u003eDevelopmental cognitive neuroscience\u0026nbsp;\u003c/em\u003e\u003cstrong\u003e49,\u0026nbsp;\u003c/strong\u003e100964 (2021).\u003c/p\u003e\n\u003cp\u003e11. Maeda, K. Fetal monitoring and actocardiogram in the evaluation of fetal behavior. \u003cem\u003eThe Ultrasound Review of Obstetrics and Gynecology\u0026nbsp;\u003c/em\u003e\u003cstrong\u003e4,\u0026nbsp;\u003c/strong\u003e12\u0026ndash;25 (2004).\u003c/p\u003e\n\u003cp\u003e12. Queyam, A. B., Pahuja, S. K. \u0026amp; Singh, D. N on - Invasive Feto - Maternal Well - Being Monitoring: A Review of Methods. \u003cem\u003eJESTR\u0026nbsp;\u003c/em\u003e\u003cstrong\u003e10,\u0026nbsp;\u003c/strong\u003e177\u0026ndash;190 (2017).\u003c/p\u003e\n\u003cp\u003e13. Jezewski, M., Czabanski, R., Leski, J. M. \u0026amp; Jezewski, J. Fuzzy classifier based on clustering with pairs of \u0026epsilon;-hyperballs and its application to support fetal state assessment. \u003cem\u003eExpert Systems with Applications\u0026nbsp;\u003c/em\u003e\u003cstrong\u003e118,\u0026nbsp;\u003c/strong\u003e109\u0026ndash;126 (2019).\u003c/p\u003e\n\u003cp\u003e14. Vairavan, S.\u003cem\u003e\u0026nbsp;et al.\u0026nbsp;\u003c/em\u003eA computer-aided approach to detect the fetal behavioral states using multi-sensor Magnetocardiographic recordings. \u003cem\u003eComputers in biology and medicine\u0026nbsp;\u003c/em\u003e\u003cstrong\u003e69,\u0026nbsp;\u003c/strong\u003e44\u0026ndash;51 (2016).\u003c/p\u003e\n\u003cp\u003e15. Preissl, H., Lowery, C. L. \u0026amp; Eswaran, H. Fetal magnetoencephalography: current progress and trends. \u003cem\u003eExperimental neurology\u0026nbsp;\u003c/em\u003e\u003cstrong\u003e190 Suppl 1,\u0026nbsp;\u003c/strong\u003eS28-36 (2004).\u003c/p\u003e\n\u003cp\u003e16. Morin, E. C.\u003cem\u003e\u0026nbsp;et al.\u0026nbsp;\u003c/em\u003eFunctional brain development in growth-restricted and constitutionally small fetuses: a fetal magnetoencephalography case-control study. \u003cem\u003eBJOG : an international journal of obstetrics and gynaecology\u0026nbsp;\u003c/em\u003e\u003cstrong\u003e122,\u0026nbsp;\u003c/strong\u003e1184\u0026ndash;1190 (2015).\u003c/p\u003e\n\u003cp\u003e17. Linder, K.\u003cem\u003e\u0026nbsp;et al.\u0026nbsp;\u003c/em\u003eMaternal insulin sensitivity is associated with oral glucose-induced changes in fetal brain activity. \u003cem\u003eDiabetologia\u0026nbsp;\u003c/em\u003e\u003cstrong\u003e57,\u0026nbsp;\u003c/strong\u003e1192\u0026ndash;1198 (2014).\u003c/p\u003e\n\u003cp\u003e18. Sippel, K.\u003cem\u003e\u0026nbsp;et al.\u0026nbsp;\u003c/em\u003eFully Automated R-peak Detection Algorithm (FLORA) for fetal magnetoencephalographic data. \u003cem\u003eComputer methods and programs in biomedicine\u0026nbsp;\u003c/em\u003e\u003cstrong\u003e173,\u0026nbsp;\u003c/strong\u003e35\u0026ndash;41 (2019).\u003c/p\u003e\n\u003cp\u003e19. Sippel, K.\u003cem\u003e\u0026nbsp;et al.\u0026nbsp;\u003c/em\u003eFully Automated Subtraction of Heart Activity for Fetal Magnetoencephalography Data. \u003cem\u003eAnnual International Conference of the IEEE Engineering in Medicine and Biology Society. IEEE Engineering in Medicine and Biology Society. Annual International Conference\u0026nbsp;\u003c/em\u003e\u003cstrong\u003e2019,\u0026nbsp;\u003c/strong\u003e5685\u0026ndash;5689 (2019).\u003c/p\u003e\n\u003cp\u003e20. Mat Husin, H.\u003cem\u003e\u0026nbsp;et al.\u0026nbsp;\u003c/em\u003eMaternal Weight, Weight Gain, and Metabolism are Associated with Changes in Fetal Heart Rate and Variability. \u003cem\u003eObesity (Silver Spring, Md.)\u0026nbsp;\u003c/em\u003e\u003cstrong\u003e28,\u0026nbsp;\u003c/strong\u003e114\u0026ndash;121 (2020).\u003c/p\u003e\n\u003cp\u003e21. Govindan, R. B.\u003cem\u003e\u0026nbsp;et al.\u0026nbsp;\u003c/em\u003eA novel approach to track fetal movement using multi-sensor magnetocardiographic recordings. \u003cem\u003eAnnals of biomedical engineering\u0026nbsp;\u003c/em\u003e\u003cstrong\u003e39,\u0026nbsp;\u003c/strong\u003e964\u0026ndash;972 (2011).\u003c/p\u003e\n\u003cp\u003e22. Kiefer-Schmidt, I.\u003cem\u003e\u0026nbsp;et al.\u0026nbsp;\u003c/em\u003eIs there a relationship between fetal brain function and the fetal behavioral state? A fetal MEG-study. \u003cem\u003eJournal of perinatal medicine\u0026nbsp;\u003c/em\u003e\u003cstrong\u003e41,\u0026nbsp;\u003c/strong\u003e605\u0026ndash;612 (2013).\u003c/p\u003e\n\u003cp\u003e23. Schneider, U.\u003cem\u003e\u0026nbsp;et al.\u0026nbsp;\u003c/em\u003eDevelopmental milestones of the autonomic nervous system revealed via longitudinal monitoring of fetal heart rate variability. \u003cem\u003ePloS one\u0026nbsp;\u003c/em\u003e\u003cstrong\u003e13,\u0026nbsp;\u003c/strong\u003ee0200799 (2018).\u003c/p\u003e\n\u003cp\u003e24. Kontos, A.\u003cem\u003e\u0026nbsp;et al.\u0026nbsp;\u003c/em\u003eThe Inconsistent Nature of Heart Rate Variability During Sleep in Normal Children and Adolescents. \u003cem\u003eFrontiers in cardiovascular medicine\u0026nbsp;\u003c/em\u003e\u003cstrong\u003e7,\u0026nbsp;\u003c/strong\u003e19 (2020).\u003c/p\u003e\n\u003cp\u003e25. Gustafson, K. M., May, L. E., Yeh, H., Million, S. K. \u0026amp; Allen, J. J. B. Fetal cardiac autonomic control during breathing and non-breathing epochs: the effect of maternal exercise. \u003cem\u003eEarly human development\u0026nbsp;\u003c/em\u003e\u003cstrong\u003e88,\u0026nbsp;\u003c/strong\u003e539\u0026ndash;546 (2012).\u003c/p\u003e\n\u003cp\u003e26. Lang, A., Del Giudice, R. \u0026amp; Schabus, M. Sleep, Little Baby: The Calming Effects of Prenatal Speech Exposure on Newborns' Sleep and Heartrate. \u003cem\u003eBrain sciences\u0026nbsp;\u003c/em\u003e\u003cstrong\u003e10\u0026nbsp;\u003c/strong\u003e(2020).\u003c/p\u003e\n\u003cp\u003e27. Dereymaeker, A.\u003cem\u003e\u0026nbsp;et al.\u0026nbsp;\u003c/em\u003eAn Automated Quiet Sleep Detection Approach in Preterm Infants as aGateway to Assess Brain Maturation. \u003cem\u003eInternational journal of neural systems\u0026nbsp;\u003c/em\u003e\u003cstrong\u003e27,\u0026nbsp;\u003c/strong\u003e1750023 (2017).\u003c/p\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":true,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"scientific-reports","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"scirep","sideBox":"Learn more about [Scientific Reports](http://www.nature.com/srep/)","snPcode":"","submissionUrl":"","title":"Scientific Reports","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Scientific Reports","inReviewEnabled":true,"inReviewRevisionsEnabled":true},"keywords":"Evaluation, parameters, fetal behavioural, classification, body and eye movements and heart rate, actogram, cardiogram","lastPublishedDoi":"10.21203/rs.3.rs-966925/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-966925/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eFetal behavioural states (fBS) describe periods of fetal wakefulness and sleep and are commonly defined by features such as body and eye movements and heart rate. Automatic state detection through algorithms relies on different parameters and thresholds derived from both the cardiogram and the actogram, which are highly dependent on the specific datasets and are prone to artefacts. Furthermore, the development of the fetal states is dynamic over the gestational period and the evaluation usually only separated into early and late gestation (before and after 32 weeks). In the current work, fBS were classified in 120 fetal magnetocardiographic data segments, between 27 and 39 weeks of gestational age, by both a classification algorithm and visual inspection. To identify how automated fBS detection could be improved, we first identified commonly used parameters for fBS classification in both the cardiogram and the actogram, and investigated their distribution across the different fBS. Then, we calculated a Receiver Operating Characteristics (ROC) curve to determine the performance of each parameter in the fBS classification. Finally, we investigated the development of parameters over gestation through linear regression. As a result, the parameters derived from the cardiogram have a higher classification accuracy compared to those derived from the body movement as defined by the actogram. However, the overlapping distributions of several parameters across states limit a clear separation of states based on these parameters. The changes over gestation of the cardiogram parameters reflect the maturation of the fetal autonomous nervous system. Given the higher classification accuracy of the cardiogram in comparison to the actogram, we suggest to focus further research on the cardiogram. Furthermore, we propose to develop a probabilistic fBS classification approaches to improve classification in less prototypical datasets.\u003c/p\u003e","manuscriptTitle":"Evaluation of Parameters for Fetal Behavioural State Classification","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2021-10-22 15:18:21","doi":"10.21203/rs.3.rs-966925/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"decision","content":"Major revision","date":"2021-12-06T04:40:06+00:00","index":"","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2021-11-29T16:20:24+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"94841ee6-c4d5-4aef-87bf-35d59e76a76c","date":"2021-11-24T18:01:26+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"16cb4807-36e4-4557-9af5-78947c33c0ac","date":"2021-11-12T14:36:12+00:00","index":"hide","fulltext":""},{"type":"reviewersInvited","content":"","date":"2021-11-08T15:09:11+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2021-11-08T14:46:25+00:00","index":"","fulltext":""},{"type":"editorInvited","content":"","date":"2021-10-20T08:13:21+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2021-10-20T08:01:20+00:00","index":"","fulltext":""},{"type":"submitted","content":"Scientific Reports","date":"2021-10-13T12:57:02+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"scientific-reports","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"scirep","sideBox":"Learn more about [Scientific Reports](http://www.nature.com/srep/)","snPcode":"","submissionUrl":"","title":"Scientific Reports","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Scientific Reports","inReviewEnabled":true,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"c2c62d6a-94ce-4d49-ba55-f4491e4eba67","owner":[],"postedDate":"October 22nd, 2021","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"published-in-journal","subjectAreas":[{"id":8022768,"name":"Internal Medicine"},{"id":8022769,"name":"General Biochemistry"}],"tags":[],"updatedAt":"2022-03-01T11:57:25+00:00","versionOfRecord":{"articleIdentity":"rs-966925","link":"https://doi.org/10.1038/s41598-022-07476-x","journal":{"identity":"scientific-reports","isVorOnly":false,"title":"Scientific Reports"},"publishedOn":"2022-03-01 11:57:25","publishedOnDateReadable":"March 1st, 2022"},"versionCreatedAt":"2021-10-22 15:18:21","video":"","vorDoi":"10.1038/s41598-022-07476-x","vorDoiUrl":"https://doi.org/10.1038/s41598-022-07476-x","workflowStages":[]},"version":"v1","identity":"rs-966925","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-966925","identity":"rs-966925","version":["v1"]},"buildId":"7rjqhiLT3MXkJMwkYKINL","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}