Detecting negative myoclonus during long-term home measurements using wearables

First study to perform long-term home monitoring of negative myoclonus with wearable sensors. (cid:1) Silent periods associated with negative myoclonus can be successfully detected and quantiﬁed from surface electromyography data. (cid:1) Silent period ﬁndings correlate with clinically assessed patients’ myoclonus severity and functional status.


Introduction
Myoclonus is a hyperkinetic movement disorder.It can occur in two forms: positive and negative myoclonus.Positive myoclonus is defined as a shock-like jerky involuntary movement caused by a sudden contraction of the muscles, whereas negative myoclonus (NM) is a sudden loss of muscle tone, often seen as a lapse of posture (Blume et al., 2001).Some neurological diseases, mainly some epilepsies (Caviness, 2007;Daida et al., 2021;Gong et al., 2019;Noachtar et al., 1997) but also Parkinson's disease (Sifoglu et al., 2017), have been associated with NM.Furthermore, several cases have been reported in which a drug induced NM as an adverse effect (Hellwig and Amtage, 2008;Kim et al., 2017;Van Samkar et al., 2017).Negative myoclonus can be classified as cortical or subcortical based on its origin (Rubboli and Tassinari, 2006).It can involve multiple muscles in the body simultaneously or locally in a single limb.If NM is strong enough and affects the muscles of the lower limbs, it can severely affect a patient's ability to walk or stand (Hellwig and Amtage, 2008).Based on earlier descriptions, NM often occurs in combination with positive myoclonus, which is referred as positive-negative myoclonus (Shibasaki and Hallett, 2005a).Positive myoclonus is described as a jerky shock-like movement.In positive-negative myoclonus positive precedes the negative myoclonus (Cassim and Houdayer, 2006;Shibasaki and Hallett, 2005b).
Surface electromyography (EMG) and electroencephalography are the gold standards for the electrophysiological assessment of NM (Belcastro et al., 2011;Cerminara et al., 2004;Hellwig and Amtage, 2008;Li et al., 2015).In general, the EMG equivalent of NM is a silent period (SP) with a duration shorter than 500 ms (Blume et al., 2001;Rubboli and Tassinari, 2006).However, even longer SPs have been reported in the EMG signal associated with NM (Fotedar et al., 2022).Specifically, cortical NM appears irregular and is associated with a longer SP of 100-400 ms, whereas subcortical NM can be seen as a rhythmic phenomenon with a shorter SP (Rubboli and Tassinari, 2006).The EMG equivalent of positive myoclonus is a group of brief potentials.An SP with a preceding discharge in EMG resembles a positive-negative myoclonus (Cassim and Houdayer, 2006;Shibasaki and Hallett, 2005a).
Earlier studies of NM have examined spontaneous SPs and SPs created by applying stimuli and receiving feedback from EMG (Ohnari et al., 2006;Rubboli et al., 2006).Artificially created SPs by applying stimuli to cortical areas of the brain are called cortical silent periods (CSP).CSP duration is shown to be prolonged in EPM1 patients compared to healthy subjects (Danner et al., 2013).The relationship between artificial CSP and spontaneous SP has not been studied to our knowledge.In a clinical setup for NM assessment, a participant was asked to maintain tonic activation of the upper limb muscles (holding hands straight in front).From this setting, lapses in posture can be evaluated visually or using EMG.By its definition, NM doesn't require preceding tonic muscle activity, but it is a necessity for the detection of the SP (Belcastro et al., 2011;Cerminara et al., 2004;Daida et al., 2021;Fotedar et al., 2022;Li et al., 2015;Matsunaga et al., 2000;Noachtar et al., 1997;Ohnari et al., 2006;Yoshimura et al., 2015).The duration of an SP is proposed to be a parameter that describes an NM event (Noachtar et al., 1997).A large variability exists in reported SPs, with durations ranging from 30 ms to 600 ms (Fotedar et al., 2022;Matsunaga et al., 2000;Noachtar et al., 1997;Ohnari et al., 2006).
Progressive myoclonus epilepsy type 1 (EPM1, Unverricht-Lundborg disease) is a rare neurodegenerative disease caused by mutations in the gene encoding cystatin B. The hallmarks of the disease are early age of onset (6-14 years of age), epileptic seizures, and debilitating myoclonic activity, including NM even though EPM1 is mostly known for positive myoclonus (Acharya and Acharya, 2023).The disease is progressive, and over time, patients' motor function becomes impaired, which may ultimately lead to the patient being bound to a wheelchair and other forms of functional dependence (Kälviäinen et al., 2008;Sipilä et al., 2020).The role of NM in EPM1 has not been clear, partly because of the lack of a useful means for detecting and quantifying NM.We previously reported that wearable surface EMG and accelerometer monitoring of EPM1 patients can reliably detect both positive and negative myoclonus (Rissanen et al., 2021).
Movement disorders are characterized by fluctuations in the severity of symptoms.During a short visit at the clinic, the symp-toms may be mild or not present at all, which can lead to a misinterpretation of motor functions.To minimize the error induced by symptom severity fluctuation, long-term electrophysiological measurements using wearable sensors and home monitoring systems have been applied successfully to movement disorders and have proven to be valuable tools in assessing a patient's motor state despite symptom fluctuation (Delrobaei et al., 2018;Fisher et al., 2016;Jones et al., 2017;Ossig et al., 2016;Rissanen et al., 2020;Roland et al., 2013).In a home monitoring setting, the subject performs everyday activities in an unsupervised environment.Evidence shows that in EPM1 patients, myoclonus severity also fluctuates during the day (Rissanen et al., 2021).However, to our knowledge, there are no studies on home monitoring or longterm measurements of NM.
The aims of the present study were as follows: 1) Develop and present a novel method for the quantification of NM by detecting SPs from long-term surface EMG measurements 2) Correlate the EMG results with the disease severity, functional status, and other clinical characteristics of EPM1 patients 3) Examine the association between the spontaneous SP and the CSP induced by transcranial magnetic stimulation (TMS).

Patients
Twenty-three genetically confirmed EPM1 patients were enrolled in this study.Patients with mild, moderate, and severe myoclonus were selected from the cohort based on earlier evaluations (Hyppönen et al., 2015).Four recently diagnosed patients were also enrolled.The basic clinical characteristics are presented in Table 1.Written informed consent was obtained from every participating patient.This study was approved by the ethics committee of the North Savo Hospital District (4/10/2019).

Clinical evaluation
The clinical histories of the patients, including anti-seizure medication use, were collected during the visit.The Singer (Scores of Independence for Neurologic and Geriatric Rehabilitation) scoring model was used by an experienced neurologist (KS) to assess the patients' functional status.The Singer score is an assessment tool based on Independent Classification of Functioning (ICF) and has 20 measures of ''independence in daily activities" (Gerdes et al., 2012).Item number 10 of the Singer assesses walking with a score of 0 (complete inability to walk) to 5 (independent for > 300 m on uneven ground without breaks or walking aids).F = female, M = male, ambulatory status 1 = walks with or without an aid, 2 = uses a wheelchair but can walk 100 meters, 3 = not capable of walking.
All individuals underwent assessment of the Unified Myoclonus Ranking Scale (UMRS) during the study visit.An experienced specialist in clinical neurophysiology (JH) scored action myoclonus and NM using previously published UMRS scoring guidelines (Frucht et al., 2002).Visible episodes of muscle tone loss in the UMRS video-recorded arm tasks or during walking were indicative of NM.If no sudden muscle tone losses were observed, NM severity was scored as 0. If the probability that NM is present is higher than 50%, NM severity was scored on a scale of 1-3 (1 = mild, 2 = moderate, 3 = severe).After the clinical assessment, the patients went home with the measurement devices attached.For action myoclonus, the total score of the action myoclonus task panels of the UMRS was used (Frucht et al., 2002).

Surface EMG
Surface EMG was recorded using two small, lightweight (48 mm Â 29 mm Â 12 mm, weight 13 g), and wearable FarosEMG TM measurement units (Bittium Biosignals Ltd).The FarosEMG TM device records single-channel surface EMG.Surface EMG was measured from the biceps brachii of the patient's dominant arm and from the extensor digitorum communis of the nondominant arm.Muscles were chosen to represent both large movements (biceps brachii) and fine movements (extensor digitorum communis).Furthermore, both muscles are easily located and measured with surface EMG equipment.The same measurement setup was used in our previous study (Rissanen et al., 2021).Furthermore, EMG from the upper limb muscles has been used in several earlier NM studies (Belcastro et al., 2011;Fotedar et al., 2022;Matsunaga et al., 2000;Noachtar et al., 1997;Yoshimura et al., 2015).Disposable Ag-AgCl electrodes (Ambu BlueSensor MÀ00ÀS, TM) were used.Two measurement electrodes were placed parallel to the muscle fibers with an inter-electrode distance of 3 cm.The third (reference) electrode was placed laterally from the measuring electrodes to a location with minimal electrical activity (e.g., over a bone).The measurement unit was fixed with a skin tape on the lateral side of the upper limb when measuring biceps brachii, and distally with respect to the electrodes when measuring extensor digitorum communis.A sampling frequency of 1,000 Hz was used for surface EMG.Long-term EMG measurements were started before the UMRS-testing and data were recorded for approximately 48 h from both muscles during the patients' daily activities.

Detection of SPs
An algorithm was written with Matlab TM R2020b (Mathworks Ò ) for the detection of SPs from long-term surface EMG data.To establish the criteria for the SP detection, a small portion of the surface EMG data was visually interpreted beforehand.Based on the UMRS test results, our patients who had an NM severity of 2 were selected for manual data inspection.The case of a patient with an NM severity of 0 was also interpreted as a reference.
The algorithm was carried out with the following steps: 1. Abrupt change in EMG data: First-order difference filtering was used to detect sudden changes in the EMG data.The firstorder difference can be calculated as where x t and x tÀ1 are EMG data points, and t is discrete time index.
Dt equals 1 ms when the sampling frequency is 1 kHz.The values of the filtered signal correspond to the difference between consecutive data points of the original EMG signal.If the value of the filtered EMG signal is larger than the preset threshold, an abrupt change in the EMG signal is established.The threshold value was derived from the root mean square (RMS) value of the signal.The RMS value is calculated as where N is the number of data points, and x i refers to the data points of the signal.The threshold value was set as Threshold ¼ min RMS ; 50 f g After an abrupt change in the EMG data, the duration of the following SP was determined using the same first-order difference filter.When the value between two consecutive points was larger than 10 lV, the end point of the SP was reached.If the SP duration was between 50 and 500 ms, it was accepted.This interval was based on the definition of NM by Rubboli et al. (Rubboli and Tassinari, 2006) and on previously reported SP durations (Daida et al., 2021;Fotedar et al., 2022;Li et al., 2015;Matsunaga et al., 2000;Noachtar et al., 1997;Yoshimura et al., 2015).2. Preceding and following muscle activities: Based on its definition, NM does not require preceding or following muscle activity (Rubboli and Tassinari, 2006).However, it is extremely difficult to determine a loss of muscle tone from the EMG signal if there is no muscle tone to begin with (Cassim and Houdayer, 2006).This is the main reason why preceding and following muscle activities were established as requirements.Both activities were calculated as the average of the moving average filtered EMG signal with a window size of 40 ms.The values were compared to a preset threshold to ensure adequate preand post-SP muscle activity.Both pre-and post-SP muscle activities were calculated from a 300 ms window.3. Differentiation from tremor: It has been proposed that hyperkinetic symptoms as myoclonus and tremor, are difficult to distinguish from one another (Carr, 2012;Van Der Salm et al., 2013).To eliminate rhythmic SP, two SPs that were within a 1 s time window were disqualified.4. Parameters: Finally, as an output, two parameters were calculated for each patient.The first parameter, SP number, was calculated as the number of all detected SPs and scaled with the total duration of measurement:

SP number ¼ number of detected SPs total duration of measurement
The second parameter, the average SP duration, was calculated as an average of all detected SPs average SP duration ¼ 1 n where n is the number of detected SPs, and SP(i) is the i'th SP duration.
Finally, to combine the two parameters, the SP index SP index ¼ average SP duration Â SP number was examined.
The criteria described above are based on the literature (Carr, 2012;Cassim and Houdayer, 2006;Daida et al., 2021;Fotedar et al., 2022;Li et al., 2015;Matsunaga et al., 2000;Noachtar et al., 1997;Rubboli and Tassinari, 2006;Van Der Salm et al., 2013;Yoshimura et al., 2015) and the visual interpretation of the EMG data.All thresholds used in the algorithm were experimentally obtained and patient specific.Thresholds were derived from the RMS values of the EMG signal.Resting and sleeping periods were disqualified using preceding and following muscle activity as a requirement for SP detection.Furthermore, long lasting atonia periods were disqualified by the definition of NM which is a < 500 ms in SP duration.Silent periods were detected using data recorded during the entire home monitoring period.These SPs were divided based on duration into short (50-69 ms), intermediate (70-100 ms), and long (101-500 ms).

TMS
Monophasic TMS pulses were targeted to the first dorsal interosseus muscle hotspot at the hand representation area using a neuronavigation system (version 3.1, Nexstim TMS).Two Magstim 200 stimulators connected via a Bistim unit was used since paired pulse tests were performed during the same measurement visit (Silvennoinen et al., 2022).For the CSP measurement one of the stimulators was set to zero and the other one was the active stimulator unit.The resting motor threshold was determined, and the CSP was measured using 120% of the resting motor threshold during stable muscle activation by squeezing soft balls with approximately half of the maximum voluntary contraction.Cortical silent period duration was measured by two researchers from the motor evoked potential onset to the end of silence (Säisänen et al., 2008).

Statistical analysis
Three parameters from the algorithm, namely, average SP duration, SP number, and SP index, were tested for correlation with the NM severity score and action myoclonus score obtained from the UMRS test.Correlations were calculated for disease duration, age, age at EPM1 onset, and genetic variant.Additionally, correlation between CSP and different duration of SPs was tested by calculating SP number using only short, intermediate, and long SPs.Because of the non-normal distribution of the data tested using the Kolmogorov-Smirnov normality test, the non-parametric tests of Spearman correlation, independent samples Kruskal-Wallis test, and pairwise comparisons with the Bonferroni correction median test were used.

Feasibility
Long-term EMG monitoring proved feasible.Electromyography data were successfully obtained from 22/23 EPM1 patients who participated in the study.Average duration of EMG measurement was 50.5 ± 7.4 h.Only one patient did not tolerate the measurement units attached, and their total duration of the measurement was too short.All data of this patient were discarded from the analysis.

Detection of SPs
Based on the algorithm, SPs were detected in all 22 analyzed patients.The individual results, SP number, the average SP duration, and the SP index are presented in Table 2. Out of the 22 patients, 13 were scored with an NM severity of 0, four were scored with an NM severity of 1, and five were scored with an NM severity of 2. Examples of SPs detected from EMG are presented in Fig. 1.The general daily variation in the number of detected SPs for all patients is shown in Fig. 2. Clearly, more SPs appeared during the day than during the night, which was expected because SP detection needs muscle activity to establish silence in an EMG signal.The number of SPs detected by duration intervals for each NM severity category is presented in Fig. 2. Short duration SPs were found in all patients.However, the number of SPs decreases with an increase in SP duration in patients with no or mild NM.In patients with moderate NM, most SPs were 100 ms or longer.

Associations between long-term monitoring assessments and clinical parameters
The UMRS test results and SP parameters for each patient are presented in Table 2.The SP number during monitoring was clearly higher in EPM1 patients, who had a clinically evaluated NM severity of 2 compared to patients with NM severity of 0 (Fig. 2).The average SP duration, the SP number, and the SP index for each patient during the 48 h measurement grouped in the three UMRS NM severity categories, as well as according to ambulatory status, are shown in Fig. 3.Those with mild NM had average SP duration similar to those with no NM.Those patients who were unable to walk differed from the other ambulatory groups in each SP parameter.The patients characterized with no or mild NM in the UMRS test showed shorter durations and fewer SPs overall.The correlations between the UMRS and SP parameters are shown in Table 3.All SP parameters (duration, number, and index) correlated strongly with UMRS NM severity and action myoclonus total score (rho 0.8, except for average SP duration and NM severity rho = 0.663).None of the SP parameters correlated with disease duration or age at diagnosis.The average SP duration and the SP number correlated strongly and significantly with the overall function of the patients (rho > -0.8) and with the item evaluating mobility (Singer score 10) (rho > -0.7).

SP correlation with TMS-induced CSP
The mean duration of CSP was 91 ± 33 ms, ranging from 52 ms to 131 ms.Individual CSP durations are shown in Table 2. Correlations between SP parameters, TMS-induced CSP, and ambulatory measures are shown in Table 3. Silent period duration correlated strongly with CSP duration (rho = 0.723, p = 0.003), as shown in Figure S1.The SP number calculated only using long SPs had the strongest correlation with CSP from all the SP duration intervals (rho = 0.682, p < 0.05).

Discussion
In this study, we present a new method for quantifying NM from long-term surface EMG monitoring by detecting and measuring SPs from EMG signals.Surface EMG data were recorded with two small measurement devices attached to both upper limbs.The method produced three parameters for each patient: SP number, average SP duration, and a combination of them, the SP index.These parameters were shown to correlate with the clinical metrics of myoclonus assessed with UMRS test scores and TMS-induced CSP duration.Silent period parameters also correlated with the patients' ability to function, especially walking.
Interestingly, in the clinical evaluation, NM was not observed in 13 EPM1 patients.However, based on the proposed method, every patient exhibited SPs during home monitoring.It is possible that although a patient exhibited NM during long-term follow-up, it was not present at the time of UMRS testing.Moreover, patients with no or mild NM had significantly shorter SPs.The clinical significance of short-duration SPs remains to be established.This study suggests that SP duration is associated with NM severity; most of the cases with long SPs were clinically scored as moderate NM.Our method provides an objective means of detecting and quantifying short-duration SPs.Clinically, lapses in posture during a longer SP (100-500 ms) are more visible.In other words, short disturbances in muscle activity can be inconspicuous to the physician.However, these episodes may still affect patients' functional abilities, resulting in posture and gait maintenance insecurity.It is also important to note that there is no absolute certainty that all detected SPs are NM.Visually validating every detected SP would be burdensome, and we can assume that an error margin  A. Sinokki, L. Säisänen, J. Hyppönen et al. Clinical Neurophysiology 156 (2023) 166-174 exists.Moreover, based on previous clinical experience, subjective reporting of NM using a symptom diary is difficult.For these reasons, there may be some limitations in the evaluation of the performance of the proposed method.The method was applied to data from both arms, thus increasing the diagnostic yield.Further studies are needed to establish whether changes in SP duration during follow-up could be used as clinical biomarkers to follow up disease progression or anti-seizure medication effects.
In addition to SP duration, we bring another parameter, SP number, to the discussion.In general, SP number had stronger correlations with UMRS results than average SP duration did.SP number describes the occurrence rate of the negative jerks.Since UMRS is evaluated in a 15-minute period, more frequent silent periods will more likely occur during the time of UMRS test.This explains SP number's overall better correlation with UMRS results.Finally, a simple product of the two discussed parameters, the SP index, was also examined.Based on correlations with the UMRS test and other clinical parameters, there was no improvement observed with the SP index compared to average SP duration or the SP number.Moreover, the SP index didn't improve the sensitivity of the proposed method, although it had the strongest correlation with walking item in Singer score from all SP parameters.EPM1 patients clinically evaluated as having moderate NM, showed both a higher SP number and average SP duration compared to patients evaluated as having no NM.Wearable EMG measurement can be considered more reliable for detecting NMs compared to subjective measures during clinical testing (Rissanen et al., 2021).Moreover, SP parameters were strongly associated with patients' functional status as assessed with the Singer questionnaire and the ability to walk.This further proves that long-term monitoring of NM reflects patients' disease severity.By contrast, we did not observe any correlation between EPM1 disease duration or disease onset age.This might be due to small patient population which is a notable limitation of this study.
In our study, fewer SPs were detected in the second 24 h period than in the first 24 h period.It is important to point out that for all subjects, the first 24 h period involved visiting the clinic, which included several different examinations and traveling.This implies that the second 24 h period is the monitoring period reflecting everyday life.In EPM1, myoclonus symptoms are strongly associated with activities and are aggravated by stress and different stimuli.Therefore, the length of home monitoring must be carefully selected and standardized for follow-up studies because the observed differences might be substantially impacted by patients' different environments and situations.
The classification of NM is demanding, as it is based on neurophysiological properties that are difficult to establish in different diseases with unique pathologies.In our study the duration of detected SPs might point to both cortical and subcortical myoclonus.Overall, myoclonus in EPM1 appears to be of both cortical and subcortical origins, with epileptic cortical myoclonus constituting only part of the myoclonus observed in EPM1 patients.Future studies combining EEG-EMG recordings in EPM1 patients are needed to delineate the cortical-subcortical origin of the NM.In some neurological disorders the NM affecting axial and lower limb muscles and resulting in falls is described to be of subcortical origin (Chandarana et al., 2021).EPM1 patients often lose their ability to walk without aids because of falls or fear to fall.
We found a strong correlation between the durations of the CSP and the SP, which suggests that they are related, either sharing the same mediator or originating in the same brain structures.SP number for long SP's showed an intermediate correlation with CSP duration, which further strengthens the possibility of cortical phenomena.Based on previous studies, the TMS-induced CSP is mediated via GABA B -and GABA A -mediated pathways and it is thought to reflect the state of intracortical inhibition (Silvennoinen et al., 2020).The CSP is prolonged in Finnish EPM1 patients (Danner et al., 2013).We observed a clear variation in detected SPs (Fig. 1), which reflects existence of both positive-negative myoclonus and pure negative myoclonus in EPM1.It remains speculative whether positive-negative myoclonus and pure negative myoclonus would likely to arise from the same pathophysiological mechanism.Especially, positive-negative myoclonus could reflect postexcitatory exhaustion and compensatory increase in intracortical inhibition.Pure negative myoclonus might be a result of increased inhibitory activity as such.It was previously suggested that hyperexcitable state might result into enhancement of inhibitory tonus in primary motor cortex (Rubboli and Tassinari, 2006).In such state the stimulus sensitivity, which is clearly known in EPM1, might lead to the suppression of muscle activity through inhibition of cortico-spinal connections.Overall, there is a clear background to suggest that the CSP and NM might share the same cortical mediators, such as GABA-A/B.However, further studies are needed to understand the pathophysiology of NM in EPM1.
In conclusion, both average SP duration and the SP number evaluated from long-term EMG can be used as indicators of NM severity and disease progression in EPM1 patients.This method can be used to confirm clinical findings or to test the effects of anti-myoclonic drugs in the future.Further development and optimization are needed for the algorithm and the accepted SP duration.

Author contributions
AS: development of the presented method, analysis, drafting of the manuscript, LS: interpretation of the results, statistical analy- sis, SR: study design, data collection, analysis, JH: interpretation of the results, KS, RK: data collection, EM, supervision, PK: supervision, fund-raising.All authors: critical review and approval of the manuscript.

Fig. 1 .
Fig. 1.A, B, C, D: Variety of examples of silent periods detected by the algorithm.Surface electromyography data are presented as blue, and the detected silent period starts at t = 0 ms.Silent period durations are presented in the upright corner in each subfigure.In Figures A and B, continuous muscle activity is seen before the silent period.In Figures C and D, large potentials are seen before the silent period, indicating a combination of positive and negative myoclonus.

Fig
Fig. 2. A. Visualization of duration distribution in three silent period (SP) duration intervals-short (50-69 ms), intermediate (70-100 ms), and long (101-500 ms)-in different negative myoclonus (NM) severity group.B: Daily variation in the number of detected silent periods (SP) for each negative myoclonus (NM) severity group.

Fig. 3 .
Fig. 3. Average silent period (SP) duration, SP number, and the SP index in separate categories of negative myoclonus severity (A-C) and ambulation (D-F).Horizontal bars show statistically significant differences between groups.

Table 2
Results from the detection of SP.From left to right: patient ID, SP number, average SP duration ± standard deviation, SP index, and CSP duration.The last two columns include the UMRS scores for NM severity and AM.SD = standard deviation, CSP = cortical silent period, UMRS = Unified Myoclonus Rating Scale, AM = action myoclonus, NM = negative myoclonus, NA = not applicable.