Next Article in Journal
Systematic Review of Diagnostic Approaches for Human Giardiasis: Unveiling Optimal Strategies
Next Article in Special Issue
A Comprehensive Study on a Deep-Learning-Based Electrocardiography Analysis for Estimating the Apnea-Hypopnea Index
Previous Article in Journal
Self-Inflicted Laser-Induced Retinopathy
Previous Article in Special Issue
Potential of Machine Learning for Predicting Sleep Disorders: A Comprehensive Analysis of Regression and Classification Models
 
 
Search for Articles:
Title / Keyword
Author / Affiliation / Email
Journal
Article Type
 
 
Advanced Search
 
Section
Special Issue
Volume
Issue
Number
Page
 
Journals
Diagnostics
Volume 14
Issue 4
10.3390/diagnostics14040363
diagnostics-logo
Submit to this Journal Review for this Journal Propose a Special Issue
► Article Menu

Article Menu

share Share announcement Help format_quote Cite question_answer Discuss in SciProfiles thumb_up
...
Endorse textsms
...
Comment
first_page
settings
Order Article Reprints
Font Type:
Arial Georgia Verdana
Font Size:
Aa Aa Aa
Line Spacing:
Column Width:
Background:
Article

Periodic Leg Movements during Sleep Associated with REM Sleep Behavior Disorder: A Machine Learning Study

by
Maria Salsone
1,2,*,†,
Basilio Vescio
3,4,†,
Andrea Quattrone
5,
Sara Marelli
2,
Alessandra Castelnuovo
6,
Francesca Casoni
2,
Aldo Quattrone
7,* and
Luigi Ferini-Strambi
2,6
1
Institute of Molecular Bioimaging and Physiology, National Research Council, 20054 Segrate, Italy
2
Sleep Disorders Center, Division of Neuroscience, San Raffaele Scientific Institute, 20132 Milan, Italy
3
Neuroimaging Research Unit, Institute of Molecular Bioimaging and Physiology (IBFM), National Research Council (CNR), 88100 Catanzaro, Italy
4
Biotecnomed S.C.aR.L., c/o Magna Graecia University, G Building, lev.1, 88100 Catanzaro, Italy
5
Institute of Neurology, Magna Graecia University, 88100 Catanzaro, Italy
6
Sleep Disorders Center, Division of Neuroscience, Vita-Salute San Raffaele University, 20132 Milan, Italy
7
Neuroscience Research Center, Magna Graecia University, 88100 Catanzaro, Italy
*
Authors to whom correspondence should be addressed.
†
These authors contributed equally to this work.
Diagnostics 2024, 14(4), 363; https://doi.org/10.3390/diagnostics14040363
Submission received: 7 December 2023 / Revised: 20 January 2024 / Accepted: 1 February 2024 / Published: 7 February 2024
(This article belongs to the Special Issue Diagnosis of Sleep Disorders Using Machine Learning Approaches)
Browse Figures
Versions Notes

Abstract

:
Most patients with idiopathic REM sleep behavior disorder (iRBD) present peculiar repetitive leg jerks during sleep in their clinical spectrum, called periodic leg movements (PLMS). The clinical differentiation of iRBD patients with and without PLMS is challenging, without polysomnographic confirmation. The aim of this study is to develop a new Machine Learning (ML) approach to distinguish between iRBD phenotypes. Heart rate variability (HRV) data were acquired from forty-two consecutive iRBD patients (23 with PLMS and 19 without PLMS). All participants underwent video-polysomnography to confirm the clinical diagnosis. ML models based on Logistic Regression (LR), Support Vector Machine (SVM), Random Forest (RF), and eXtreme Gradient Boosting (XGBoost) were trained on HRV data, and classification performances were assessed using Leave-One-Out cross-validation. No significant clinical differences emerged between the two groups. The RF model showed the best performance in differentiating between iRBD phenotypes with excellent accuracy (86%), sensitivity (96%), and specificity (74%); SVM and XGBoost had good accuracy (81% and 78%, respectively), sensitivity (83% for both), and specificity (79% and 72%, respectively). In contrast, LR had low performances (accuracy 71%). Our results demonstrate that ML algorithms accurately differentiate iRBD patients from those without PLMS, encouraging the use of Artificial Intelligence to support the diagnosis of clinically indistinguishable iRBD phenotypes.

1. Introduction

The loss of muscle atonia and abnormal behaviors, such as sleep-related vocalization and/or complex motor behaviors, during the rapid eye movement (REM) sleep phase, characterize the REM sleep behavior disorder (RBD) [1,2]. The scientific interest in RBD is that its idiopathic (or isolated) form (iRBD) is now considered a neuropathological precursor of synucleinopathies like Parkinson’s disease (PD), Dementia with Lewy bodies (DLB), and Multiple System Atrophy (MSA) [3]. Data from a multicentric study revealed that the rate of phenoconversion to overt neurodegenerative disorder significantly increases in the presence of specific biomarkers including the olfactory deficit, mild cognitive impairment, erectile dysfunction, motor symptoms, abnormal DAT scan, color vision abnormalities constipation, REM atonia loss, and age [4].
Interestingly, about 70% of patients with RBD exhibit periodic leg movements during sleep (PLMS) at all sleep stages, especially during REM sleep [5,6,7]. PLMS are clinically defined as stereotyped and recurring movements of the lower limbs, such as extension of the big toe, and partial flexion of the ankle, knee, and hip [6,7,8]. However, these peculiar movement disorders are not exclusive to RBD. Indeed, they are symptoms belonging to the clinical picture of several sleep disorders, such as narcolepsy and obstructive sleep apnea, and are significantly associated with restless leg syndrome (RLS). On the other hand, distinct electrophysiological features including distribution across the night, periodicity, duration, amplitude, and time structure characterize PLMS in sleep disorders, thus suggesting a disease-specific phenomenon [9]. Finally, PLMS were also frequently reported in apparently normal older adults [8]. Therefore, the question that remains unsolved until now is whether the presence of PLMS in iRBD identifies a clinical subtype with a higher risk of phenoconversion. In this context, differentiating iRBD patients from those without PLMS could be of clinical relevance for future neurodegenerative risk.
Although the clinical history is accurate, PLMS are often misdiagnosed. The gold standard for a correct diagnosis is polysomnography (PSG). A PLMS index on PSG recording, calculated as the number of periodic leg movements per hour of sleep equal to or greater than 15 events/h, allows for the detection and quantification of PLMS [10]. The literature regarding the development of alternative methods for the automatic detection, scoring, and analysis of PLMS, as well as for differential diagnosis among iRBD phenotypes, is poor. In recent years, some Artificial Intelligence (AI) tools including Deep Learning (DL) and Machine Learning (ML) have been developed with good results. A new Deep Learning model (deepPLM) based on a single-lead electrocardiogram (ECG) signal from polysomnographic recordings was used for the automatic detection of PLMS in 52 subjects (26 controls and 26 patients) with good performances [11]. An AI-enabled method for the automatic classification of sleep disorders, including PLMS using ECG alone, has also been proposed [12]. ML algorithms such as the Random Forest (RF) model and Extreme Gradient Boosting (XGBoost) have been widely developed and successfully applied for the correct differential diagnosis of neurological diseases. Recently, it has been demonstrated that the RF model had excellent accuracy (94%), sensitivity (95%), and specificity (92%) in distinguishing iRBD subjects from healthy subjects [13]. In the current study, we aimed to evaluate the performance of ML algorithms (Logistic Regression (LR), Supporter Vector Machine (SVM) XGBoost, and RF) in differentiating iRBD patients with and without PLMS. Furthermore, we tested the hypothesis that the phenotype might be different between the two iRBD groups, with a different risk profile for future neurodegeneration.

2. Materials and Methods

2.1. Participants

Forty-two patients who fulfilled the diagnostic criteria for iRBD were enrolled in this study. All subjects were thoroughly examined by neurologists to exclude cases in which other neurological diseases were present. A video-PSG was needed for the confirmation or exclusion of the presence of clinical/subclinical RBD and to detect the presence of PLMS. Significant muscle activity during REM sleep associated with abnormal movements and other behaviors detected in PSG recordings was required for the confirmation of the clinical or subclinical diagnosis of RBD [14], as required by the International Classification of Sleep Disorders—Third Edition (ICSD-3) criteria. Automatic labeling was used for scoring the sleep stages in the PSG recordings. Scored sleep stages were then checked manually by an expert sleep examiner in order to correct possible wrong scores. Regarding PLMS diagnosis, PLMS were scored according to the previously published criteria of the American Academy of Sleep Medicine [15] and were carefully differentiated from the phasic EMG activity on the anterior tibialis based on their regular periodicity [16]. A PLMS index value ≥ 15 events/h (abnormal value) [8,17] was used to divide our cohort into two iRBD subgroups: (i) iRBD with PLMS (iRBD-PLMS, value ≥ 15); (ii) iRBD without PLMS (iRBD, value < 15). To better characterize these two iRBD phenotypes, the presence of phenoconversion biomarkers has also been investigated, including (i) olfactory deficits/hyposmia; (ii) deficits in cognitive domains; (iii) cerebral hypometabolism showed on FDG-PET imaging; (iv) motor symptoms. In detail, olfactory deficits/hyposmia were considered when referred by the patients. Neuropsychological evaluation was performed by a specialist (SM) using the following criteria: Mini-Mental State Examination (MMSE) [18], Digit Span forward and backward [19]; Corsi block-tapping Test [20]; Rey’s List: learning, recall, and recognition [21]; Raven’s Progressive Matrices [22]; Attentive Matrices [22]; Verbal Fluency with Phonemic and Semantic cues [23]; Token Test [24]; and Copy of Rey–Osterrieth complex figure [25]. The presence of hypometabolism on FDG-PET imaging was evaluated on the basis of a previous neuroimaging visual investigation made by the patient. Finally, the Unified Parkinson’s Disease Rating Scale motor score (UPDRS-ME) [26] was used to evaluate the presence and severity of extrapyramidal signs. The exclusion criteria were as follows: (i) the diagnosis of other sleep disorders including RLS, obstructive sleep apnea syndrome (OSA), and narcolepsy; (ii) ongoing treatment with medications modifying REM sleep architecture and muscle tone, such as serotonin reuptake inhibitors. Before being included in the study, all participants provided their written informed consent. The experimental procedure was previously approved by the local Ethical Committee of the “Vita-Salute” San Raffaele University, Milan, Italy (no. 38, 8 March 2022).

2.2. Heart Rate Variability (HRV) Analysis

All study participants underwent a circadian HRV analysis to detect potential cardiac autonomic dysfunctions. HRV components during 24 h and related cardiac autonomic indices were calculated according to our previously published protocol [13,27].

2.3. Machine Learning Models

The discriminating performances of single HRV features were evaluated using a Receiver Operator characteristic (ROC) analysis. Logistic Regression (LR), Support Vector Machine (SVM) with linear kernel, Random Forest (RF), and eXtreme Gradient Boosting (XGBoost) ML models were trained on HRV features [13]. Leave-One-Out Cross-Validation (LOO-CV) was used during the training process to assess the training classification performances in discriminating iRBD with PLMS patients from iRBD without PLMS patients. Feature importances were computed and ML models were built using forward feature selection, starting from the feature with the highest importance and iteratively adding new features in descending order of importance. Each model was trained using LOO-CV, and the optimal model, with the optimal number of features, was chosen as the model with the highest classification accuracy. During the training phase, a grid search strategy was used to tune the hyperparameters of the ML models and maximize the performance.

2.4. Statistical Analysis

All categorical variables (sex, RBD symptoms, and phenoconversion biomarkers) were compared using the Χ2 test. Continuous variables were compared using either Student’s t-test or Mann–Whitney U test after assessing distribution properties using the Shapiro–Wilk test.
ECG processing, extraction of R peaks, and preprocessing of RR intervals were performed using SciPy (version 1.8.0), Nolds (version 0.5.2), hrv-analysis (version 1.0.4), and BiospPy (version 0.8.0) libraries in the Python programming language (version 3.9.2). HRV features were computed using the PhysioNet HRV Toolkit [28] written in the C programming language. ML and statistical analysis were performed with the help of the caret R package (version 4.0.0) [29] and the R programming environment [30] (version 4.0.4, 2021, The R Foundation for Statistical Computing, Vienna, Austria).

3. Results

The demographic, clinical, and neuropsychological features of the iRBD patients with and without PLMS are summarized in Table 1. No significant differences were found between the two groups in terms of age and sex distribution (Table 1). The iRBD-PLMS patients were slightly older and had a higher age at the onset of RBD symptoms than those with iRBD. The PSG evaluation revealed a statistically insignificant higher percentage of patients with PSG features without clinical signs in the iRBD-PLMS group. The education level was similar, although it was lower than that of iRBD in iRBD-PLMS patients. No significant differences were observed between the two RBD groups in terms of the neuropsychological battery performance (Table 1). A slight reduction within the normal range was detected in both groups regarding the attentive function. Finally, a similar distribution of phenoconversion biomarkers to synucleinopathies was detected in both groups (Table 2). Hyposmia and cognitive deficits were more frequently biomarkers reported in iRBD patients with and without PLMS (Table 2). Interestingly, the iRBD-PLMS group showed a significantly higher percentage of patients with at least one biomarker (Table 2).

3.1. HRV Analysis

In 24 h recordings, we observed that autonomic indices increased in iRBD patients with PLMS compared to those without PLMS (Table 1). However, this increase was not statistically significant, and the previously published cut-off levels [13,27] did not show good classification performances, as individual values in iRBD patients with and without PLMS overlapped (Figure 1a,b). The best discriminating accuracy of the autonomic indices was 0.71% for the sympathetic index and 0.69% for the parasympathetic index (Table 3).

3.2. Machine Learning

3.2.1. Feature Importance and Feature Selection

Random Forest, XGBoost, and ROC-AUC algorithms were used to evaluate the feature importance, as shown in Figure 2a–c. The normalized Hurst coefficient during sleep (H_norm_s), defined as the ratio between the mean and the standard deviation of the Hurst coefficient evaluated on sleep segments, showed the best discriminating power using ROC AUC importance, and its performances in classifying iRBD subjects with and without PLMS using a Logistic Regression model with LOO-CV were as follows: accuracy, 0.71 (0.55–0.84); AUC, 0.71; sensitivity, 0.70; specificity, 0.74; PPV, 0.76; NPV, 0.67. The eight most important features used in the SVM model, selected according to ROC-AUC importance, were as follows: the normalized Hurst coefficient, the standard deviation of the sleep Hurst coefficient, the sleep-to-wake ratio of mean AVNN, the normalized sleep triangular index, the sleep-to-wake ratio of mean SD1, the normalized sleep pNN50, the mean sleep triangular index, and the mean sleep SD1. The twelve most important features according to the Random Forest algorithm were as follows: the sleep-to-wake ratio of the mean AVNN, the normalized wake Lyapunov exponent, the sleep-to-wake ratio of the Lyapunov exponent, the normalized sleep triangular index, the mean wake Lyapunov exponent, the sleep-to-wake ratio of SD1, the mean sleep pNN50, the normalized sleep Hurst coefficient, the mean wake NN-to-RR ratio, the normalized sleep pNN50, the mean sleep SD1, and the mean wake SD2-to-SD1 ratio. XGBoost identified the sleep-to-wake ratio of the mean AVNN, the normalized sleep Hurst coefficient, the normalized sleep-to-wake ratio of the mean pNN50, the sleep-to-wake ratio of the mean rMSSD, the standard deviation of the wake AVNN, the mean wake NN-to-RR ratio, and the normalized sleep-to-wake ratio of the LF/HF ratio as the seven features with the highest importance. The subsets of the most important features identified by the ROC-AUC, RF, and XGBoost did not fully overlap. However, all the ML algorithms confirmed that the normalized sleep Hurst coefficient was one of the most important features selected. Sex and age did not contribute to the predictions, as their calculated importance in discriminating RBD with PLMS from those without PLMS was very low.

3.2.2. Classification Performance of ML Models in Distinguishing iRBD with and without PLMS

Figure 3 shows the ROC and calibration curves of the optimal ML models trained with LOO-CV for their best performing sets of features. The hyperparameters of the ML models, obtained from the tuning process, are listed in Table S1 (Supplementary Material). Table 3 reports the classification accuracies (with their 95% confidence interval), AUCs, sensitivities, specificities, PPVs, and NPVs for each ML model. The RF model trained using its twelve most important features achieved the best classification accuracy of 0.86. The sensitivity, specificity, PPV, and NPV were 0.96, 0.74, 0.81, and 0.93, respectively. The classification accuracy was a bit higher than the AUC corresponding to this optimal model (AUC = 0.85). The SVM model, trained on its eight most important features, reached a classification accuracy of 0.81, with a sensitivity of 0.83, specificity of 0.79, ppv of 0.83, and npv of 0.79. The XGBoost model, trained on its seven most important features, achieved a 0.78 classification accuracy and AUC of 0.84, with a sensitivity of 0.83, specificity of 0.72, ppv of 0.79, and npv of 0.76. A comparison between the classification accuracies of ML models and cardiac autonomic indices is also reported in Table 3.

4. Discussion

In this study, we investigated the performance of ML models in differentiating iRBD patients with PLMS from those without PLMS. Our study demonstrates that ML algorithms using a combination of HRV parameters can accurately differentiate between these two iRBD phenotypes. In detail, while the RF model performed the best with an 86% accuracy, SVM achieved an accuracy of 81%, and the XGBoost algorithm showed a LOO-CV accuracy of 78%. In contrast, LR, cardiac sympathetic index, and cardiac parasympathetic index did not perform with acceptable classification performances in differentiating between these two iRBD groups, if compared to the other ML models.
In our study, we also compared the demographic and clinical features of iRBD patients with and without PLMS. No between-group differences were detected, and the two groups were clinically indistinguishable. Indeed, only slight differences in age, age at onset of iRBD symptoms, and education level were detected. These findings are in accordance with previous evidence investigating the clinical significance of PLMS in iRBD, reporting similar clinical characteristics for both phenotypes [31]. On the other hand, our iRBD-PLMS patients showed a greater number of synucleinopathy phenoconversion biomarkers, in particular hyposmia and deficits in attentive and visuo-spatial domains, than those with iRBD. Cognitive dysfunctions have been widely described in patients with iRBD, especially in those with neurodegenerative disorders including DLB [4]. Conversely, patients with DLB exhibit PLMS with a higher index value than that found in patients with Alzheimer's Disease [32]. However, it remains to be elucidated whether iRBD converters in DLB also showed PLMS and whether DLB with PLMS also presented RBD. This last issue could likely be that RBD is a symptom in the core diagnostic criteria for DLB [33]. Taken together, these findings indicate that distinguishing patients with iRBD associated with PLMS from those without remains challenging. This question is of interest since it still remains to be determined whether PLMS modifies the rate and/or risk of phenoconversion of IRBD to neurodegenerative disorders and, for this reason, it may be considered as a biomarker.
This is the first study to investigate the cardiac autonomic changes in iRBD patients with and without PLMS using HRV analysis. Interestingly, during 24 h recordings, iRBD-PLMS patients showed an increase in cardiac autonomic indices compared to those without PLMS. Dynamic changes in the autonomic nervous system activity have been widely observed during nocturnal sleep in other sleep disorders associated with PLMS such as RLS and PLM disorder (PLMD) [34]. In PLMD, the PLMS-related cardiac sympathetic activation during the sleep stage has been detected several seconds before the occurrence of PLMS. In contrast, an increase in cardiac parasympathetic activity (index calculated as the ratio of HF components during night/day) has been reported in the group with PLMS. We believe that this discrepancy is due to two factors. Firstly, different disorders are correlated with different PLMS pathogenesis. Secondly, there are different modalities of investigation, such as long versus short recording. Supporting this, it has been demonstrated that an initial activation of sympathetic activity occurs in iRBD patients, later followed by a more prolonged bradycardia [5]. Finally, the classification performances of the cardiac autonomic indices for differentiating iRBD with and without PLMS have been also calculated. Both the cardiac sympathetic and parasympathetic indices showed a low accuracy of 71% and 69%, respectively. These findings are not surprising, since similar results have been previously published regarding the differentiation between iRBD and controls [13]. Overall, our results suggest that HRV parameters should be used with caution because of the possibility of overlap between the iRBD groups.
In the last few years, (AI) models have gained growing capabilities in assisting clinicians with the diagnosis of neurological disorders and in supporting the differential diagnosis between clinically indistinguishable phenotypes [34,35]. Interestingly, in this study, we present novel ML models built on HRV parameters for distinguishing iRBD patients with and without PLMS. In detail, we used the RF model trained on the twelve most important features (sleep-to-wake ratio of the mean AVNN, the normalized wake Lyapunov exponent, the sleep-to-wake ratio of the Lyapunov exponent, the normalized sleep triangular index, the mean wake Lyapunov exponent, the sleep-to-wake ratio of SD1, the mean sleep pNN50, the normalized sleep Hurst coefficient, the mean wake NN-to-RR ratio, the normalized sleep pNN50, the mean sleep SD1, and the mean wake SD2-to-SD1 ratio) to differentiate iRBD patients with and without PLMS. According to our findings, the RF model showed both high classification accuracy (86%) and AUC (85%), whereas the SVM with the linear kernel model achieved a cross-validated accuracy of 81% (AUC = 75%). The classification performance of the XGBoost model trained on the seven most important features (sleep-to-wake ratio of mean AVNN, the normalized sleep Hurst coefficient, the normalized sleep-to-wake ratio of mean pNN50, the sleep-to-wake ratio of the mean rMSSD, the standard deviation of the wake AVNN, the mean wake NN-to-RR ratio, and the normalized sleep-to-wake ratio of the LF/HF ratio) was also tested. The XGBoost model showed a lower accuracy (equal to 78%) than the RF and SVM models, probably due to its inclusion in the model of the LF/HF ratio features. However, the AUC corresponding to this optimal model was equal to 84%. Moreover, the sensitivity of the ML models was different, with XGBoost lower (83%) than the RF model (96%). Dataset imbalance may be an important problem, as pointed out in [36,37]. However, in our study, the imbalance was not actually an issue. Classes were represented with a minority-to-majority ratio of 0.45 to 0.55. There was a slight imbalance that did actually does not affect performances.
To the best of our knowledge, only one antecedent study applied ML algorithms using HRV features in iRBD patients without PLMS, with excellent accuracy in differentiating patients from controls [13]. The current study represents a further step forward in identifying iRBD with PLMS. A new method for the automatic detection of PLMS has been previously proposed for other sleep disorders; however, it requires electromyographic recording [38]. Finally, our ML models have been trained on the HRV parameters acquired from electrocardiographic recordings. Considering that the difference between electrocardiographic and photoplethysmography signals is small [39], in the future, we will investigate ML models trained on RR intervals obtained from pulse photoplethysmography.
This study, however, has some limitations. First, the sample size of our cohort was small. PLMS, however, are often clinically misdiagnosed, requiring polysomnographic confirmation for a correct diagnosis. Therefore, the results presented in this study need to be validated using a larger sample of iRBD subjects with and without PLMS to confirm the usefulness of ML models in clinical practice. Second, as our ML models were evaluated using only a sample extracted from an Italian population, further testing should be performed in other ethnic and geographical groups. Third, studies evaluating patients from other sleep centers are needed in order to ensure the generalizability of our findings. However, this study has clear strengths. First, our proposed ML models accurately identified iRBD patients with PLMS. Second, AI algorithms were trained using HRV features only, which, beyond being very easy to obtain, are non-invasive measures of particular practical value for differentiating among clinically indistinguishable phenotypes. Finally, although our results are preliminary and based on a small sample size, we found that iRBD patients with PLMS showed a greater number of synucleinopathy biomarkers than those without PLMS. Longitudinal studies are needed to determine whether these iRBD patients convert faster to defined neurodegenerative disorders over time. In conclusion, our results demonstrate that ML models based on features derived from HRV analysis are useful for identifying iRBD patients with and without PLMS. The implementation of AI tools in a software application could support clinicians in properly diagnosing iRBD patients with or without PLMS.

Supplementary Materials

The following supporting information can be downloaded at: https://www.mdpi.com/article/10.3390/diagnostics14040363/s1, Table S1: Hyperparameters tuning of Random Forest and eXtreme Gradient Boosting models.

Author Contributions

Conceptualization, M.S.; Methodology, B.V. and A.Q. (Andrea Quattrone); Validation, S.M.; Investigation, A.C. and F.C.; Writing—review & editing, A.Q. (Aldo Quattrone) and L.F.-S. All authors have read and agreed to the published version of the manuscript.

Funding

This research received no external funding.

Institutional Review Board Statement

Ethical Committee of the “Vita-Salute” San Raffaele University, Milan, Italy (no. 38, 8 March 2022).

Informed Consent Statement

Before being included in the study, all participants provided their written informed consent.

Data Availability Statement

The data presented in this study are available on request from the corresponding author. The data are not publicly available due to privacy restrictions.

Acknowledgments

The authors are grateful to the patients who kindly participated in the study.

Conflicts of Interest

The authors declare no conflicts of interest.

References

  1. Dauvilliers, Y.; Schenck, C.H.; Postuma, R.B.; Iranzo, A.; Luppi, P.-H.; Plazzi, G.; Montplaisir, J.; Boeve, B. REM sleep behaviour disorder. Nat. Rev. Dis. Prim. 2018, 4, 19. [Google Scholar] [CrossRef] [PubMed]
  2. Zhang, F.; Niu, L.; Liu, X.; Li, S.; Yu, H.; Le, W. Rapid Eye Movement Sleep Behavior Disorder and Neurodegenerative Diseases: An Update. Aging Dis. 2020, 11, 315–326. [Google Scholar] [CrossRef] [PubMed]
  3. Högl, B.; Stefani, A.; Videnovic, A. Idiopathic REM sleep behaviour disorder and neurodegeneration-An update. Nat. Rev. Neurol. 2018, 14, 40–55. [Google Scholar] [CrossRef] [PubMed]
  4. Postuma, R.B.; Iranzo, A.; Hu, M.; Högl, B.; Boeve, B.F.; Manni, R.; Oertel, W.H.; Arnulf, I.; Ferini-Strambi, L.; Puligheddu, M.; et al. Risk and predictors of dementia and parkinsonism in idiopathic REM sleep behaviour disorder: A multicentre study. Brain 2019, 142, 744–759. [Google Scholar] [CrossRef] [PubMed]
  5. Fantini, M.L.; Michaud, M.; Gosselin, N.; Lavigne, G.; and Montplaisir, J. Periodic leg movements in REM sleep behavior disorder and related autonomic and EEG activation. Neurology 2002, 59, 1889–1894. [Google Scholar] [CrossRef] [PubMed]
  6. Pollmacher, T.; Schulz, H. Periodic leg movements (PLM): Their relationship to sleep stages. Sleep 1993, 16, 572–577. [Google Scholar] [PubMed]
  7. Montplaisir, J.; Boucher, S.; Poirier, G.; Lavigne, G.; Lapierre, O.; Lespérance, P. Clinical, polysomnographic, and genetic characteristics of restless legs syndrome: A study of 133 patients diagnosed with new standard criteria. Mov. Disord. 1997, 12, 61–65. [Google Scholar] [CrossRef] [PubMed]
  8. Bixler, E.O.; Kales, A.; Vela-Bueno, A.; Jacoby, J.A.; Scarone, S.; Soldatos, C.R. Nocturnal myoclonus and nocturnal myoclonic activity in the normal population. Res. Commun. Chem. Pathol. Pharmacol. 1982, 36, 129–140. [Google Scholar]
  9. Manconi, M.; Ferri, R.; Zucconi, M.; Fantini, M.L.; Plazzi, G.; Ferini-Strambi, L. Time structure analysis of leg movements during sleep in REM sleep behavior disorder. Sleep 2007, 30, 1779–1785. [Google Scholar] [CrossRef]
  10. AASM International Classification of Sleep Disorders: Diagnostic and Coding Manual, 2nd ed.; American Academy of Sleep Medicine: Westchester, IL, USA, 2005.
  11. Urtnasan, E.; Park, J.K.; Lee, J.H.; Koh, S.B.; Lee, K.J. Deep Learning for Automatic Detection of Periodic Limb Movement Disorder Based on Electrocardiogram Signals. Diagnostics 2022, 12, 2149. [Google Scholar] [CrossRef]
  12. Urtnasan, E.; Joo, E.Y.; Lee, K.H. AI-Enabled Algorithm for Automatic Classification of Sleep Disorders Based on Single-Lead Electrocardiogram. Diagnostics 2021, 11, 2054. [Google Scholar] [CrossRef] [PubMed]
  13. Salsone, M.; Quattrone, A.; Vescio, B.; Ferini-Strambi, L.; Quattrone, A. A Machine Learning Approach for Detecting Idiopathic REM Sleep Behavior Disorder. Diagnostics 2022, 12, 2689. [Google Scholar] [CrossRef] [PubMed]
  14. American Academy of Sleep Medicine. International Classification of Sleep Disorders, 3rd ed.; American Academy of Sleep Medicine: Darien, IL, USA, 2014. [Google Scholar]
  15. AASM: American Academy of Sleep Medicine. The AASM Manual for the Scoring of Sleep and Associated Events: Rules; Terminology and Technical Specifications: Westchester, IL, USA, 2007. [Google Scholar]
  16. Frauscher, B.; Iranzo, A.; Hogl, B.; Casanova-Molla, J.; Salamero, M.; Gschliesser, V.; Tolosa, E.; Poewe, W.; Santamaria, J.; SINBAR. Quantification of electromyographic activity during REM sleep in multiple muscles in REM sleep behavior disorder. Sleep 2008, 31, 724–731. [Google Scholar] [CrossRef] [PubMed]
  17. Hallegraeff, J.; de Greef, M.; Krijnen, W.; van der Schans, C. Criteria in diagnosing nocturnal leg cramps: A systematic review. BMC Fam. Pract. 2017, 18, 29. [Google Scholar] [CrossRef] [PubMed]
  18. Measso, G.; Cavarzeran, F.; Zappalà, G.; Lebowitz, B.D.; Crook, T.H.; Pirozzolo, F.J.; Amaducci, L.A.; Massari, D.; Grigoletto, F. The Mini-Mental state examination. Normative study of an Italian random sample. Dev. Neuropsychol. 1991, 9, 77–85. [Google Scholar] [CrossRef]
  19. Orsini, A.; Grossi, D.; Capitani, E.; Laiacona, M.; Papagno, C.; Vallar, G. Verbal and spatial immediate memory span: Normative data from 1355 adults and 1112 children. Ital. J. Neurol. Sci. 1987, 8, 539–548. [Google Scholar] [CrossRef] [PubMed]
  20. Kessels, R.P.; Van den Berg, E.; Ruis, C.; Brands, A.M. The backward span of the Corsi Block-Tapping Task and its association with the WAIS-III Digit Span. Assessment 2008, 15, 426–434. [Google Scholar] [CrossRef]
  21. Carlesimo, G.A.; Caltagirone, C.; Gainotti, G. Mental Deterioration Battery: Normative data, diagnostic reliability and qualitative analyses of cognitive impairment. The Group for the Standardization of the Mental Deterioration Battery. Eur. Neurol. 1996, 36, 378–384. [Google Scholar] [CrossRef]
  22. Spinnler, H.; Tognoni, G. Standardizzazione e taratura italiana di test neuropsicologici. Ital. J. Neurol. Sci. 1987, 8, 21–120. [Google Scholar]
  23. Novelli, G.; Papagno, C.; Capitani, E.; Laiacona, M.; Vallar, G.; Cappa, S.F. Tre test clinici di ricerca e produzione lessicale. Taratura su soggetti normali. Arch. Psicol. Neurol. Psichiatr. 1986, 47, 477–506. [Google Scholar]
  24. De Renzi, E.; Faglioni, P. Normative data and screening power of a shortened version of the Token Test. Cortex 1978, 14, 364–381. [Google Scholar] [CrossRef] [PubMed]
  25. Caffarra, P.; Vezzadini, G.; Dieci, F.; Zonato, F.; Venneri, A. Rey-Osterrieth complex figure: Normative values in an Italian population sample. Neurol. Sci. 2002, 22, 443–447. [Google Scholar] [CrossRef] [PubMed]
  26. Fahn, S.; Elton, R.L. Unified Parkinson’s disease rating scale. In Recent Developments in Parkinson’s Disease II; MacMillan: New York, NY, USA, 1987; pp. 153–163. [Google Scholar]
  27. Salsone, M.; Vescio, B.; Fratto, A.; Sturniolo, M.; Arabia, G.; Gambardella, A.; Quattrone, A. Cardiac sympathetic index identifies patients with Parkinson’s disease and REM behavior disorder. Park. Relat. Disord. 2016, 26, 62–66. [Google Scholar] [CrossRef] [PubMed]
  28. PhysioNet: Heart Rate Variability Analysis with the HRV Toolkit. Available online: https://archive.physionet.org/tutorials/hrv-toolkit/ (accessed on 1 December 2023).
  29. Kuhn, M. Building Predictive Models in R Using the caret Package. J. Stat. Softw. 2008, 28, 1–26. [Google Scholar] [CrossRef]
  30. R Core Team. R. A Language and Environment for Statistical Computing; The R Foundation for Statistical Computing: Vienna, Austria, 2021; Available online: https://www.R-project.org/ (accessed on 2 November 2023).
  31. Sasai, T.; Inoue, Y.; Matsuura, M. Clinical significance of periodic leg movements during sleep in rapid eye movement sleep behavior disorder. J. Neurol. 2011, 258, 1971–1978. [Google Scholar] [CrossRef] [PubMed]
  32. Hibi, S.; Yamaguchi, Y.; Umeda-Kameyama, Y.; Yamamoto, H.; Iijima, K.; Momose, T.; Akishita, M.; Ouchi, Y. The high frequency of periodic limb movements in patients with Lewy body Dementia. J. Psychiatr. Res. 2012, 46, 1590–1594. [Google Scholar] [CrossRef] [PubMed]
  33. Ferman, T.J.; Boeve, B.F.; Smith, G.E.; Lin, S.-C.; Silber, M.; Pedraza, O.; Wszolek, Z.; Graff-Radford, N.; Uitti, R.; Van Gerpen, J.; et al. Inclusion of RBD improves the diagnostic classification of dementia with Lewy bodies. Neurology 2011, 77, 875–882. [Google Scholar] [CrossRef] [PubMed]
  34. Sasai, T.; Matsuura, M.; Inoue, Y. Change in heart rate variability precedes the occurrence of periodic leg movements during sleep: An observational study. BMC Neurol. 2013, 13, 139. [Google Scholar] [CrossRef]
  35. Quattrone, A.; Sarica, A.; Buonocore, J.; Morelli, M.; Bianco, M.G.; Calomino, C.; Aracri, F.; De Maria, M.; Vescio, B.; Vaccaro, M.G.; et al. Differentiating between common PSP phenotypes using structural MRI: A machine learning study. J. Neurol. 2023, 270, 5502–5515. [Google Scholar] [CrossRef]
  36. Li, Y.; Zhang, S.; Yin, Y.; Xiao, W.; Zhang, J. Parallel one-class extreme learning machine for imbalance learning based on Bayesian approach. J. Ambient. Intell. Human. Comput. 2018. [Google Scholar] [CrossRef]
  37. Xiao, W.; Zhang, J.; Li, Y.; Zhang, S.; Yang, W. Class-specific cost regulation extreme learning machine for imbalanced classification. Neurocomputing 2017, 261, 70–82. [Google Scholar] [CrossRef]
  38. Ferri, R.; Zucconi, M.; Manconi, M.; Bruni, O.; Miano, S.; Plazzi, G.; Ferini-Strambi, L. Computer-assisted detection of nocturnal leg motor activity in patients with restless legs syndrome and periodic leg movements during sleep. Sleep 2005, 28, 998–1004. [Google Scholar] [CrossRef] [PubMed]
  39. Vescio, B.; Salsone, M.; Gambardella, A.; Quattrone, A. Comparison between Electrocardiographic and Earlobe Pulse Photoplethysmographic Detection for Evaluating Heart Rate Variability in Healthy Subjects in Short- and Long-Term Recordings. Sensors 2018, 18, 844. [Google Scholar] [CrossRef] [PubMed]
Diagnostics 14 00363 g001
Figure 1. Box plots of (a) cardiac sympathetic (p = 0.38) and (b) cardiac parasympathetic (p = 0.19) index in iRBD patients with and without PLMS (distributions have been compared using Wilcoxon’s rank sum test). Please refer to [13] for the interpretation of the box plots and the definitions of the LF and HF power bands.
Figure 1. Box plots of (a) cardiac sympathetic (p = 0.38) and (b) cardiac parasympathetic (p = 0.19) index in iRBD patients with and without PLMS (distributions have been compared using Wilcoxon’s rank sum test). Please refer to [13] for the interpretation of the box plots and the definitions of the LF and HF power bands.
Diagnostics 14 00363 g001
Diagnostics 14 00363 g002
Figure 2. Feature importance evaluated by ML algorithms: (a) Random Forest model, (b) XGBoost model, and (c) ROC-AUC.
Figure 2. Feature importance evaluated by ML algorithms: (a) Random Forest model, (b) XGBoost model, and (c) ROC-AUC.
Diagnostics 14 00363 g002
Diagnostics 14 00363 g003
Figure 3. ROC curves (a) and calibration plots (b) from LOO-CV of optimal models with the best hyperparameters tuning: Logistic Regression (AUC = 0.71), Support Vector Machine (AUC = 0.75), Random Forest (AUC = 0.76), and XGBoost (AUC = 0.84). ROC: Receiver Operating Characteristic; LOO-CV: Leave-one-Out Cross Validation; AUC: Area Under the Curve.
Figure 3. ROC curves (a) and calibration plots (b) from LOO-CV of optimal models with the best hyperparameters tuning: Logistic Regression (AUC = 0.71), Support Vector Machine (AUC = 0.75), Random Forest (AUC = 0.76), and XGBoost (AUC = 0.84). ROC: Receiver Operating Characteristic; LOO-CV: Leave-one-Out Cross Validation; AUC: Area Under the Curve.
Diagnostics 14 00363 g003
Table 1. Comparisons of demographic, clinical, and electrophysiological features of patients affected by iRBD with and without PLMS.
Table 1. Comparisons of demographic, clinical, and electrophysiological features of patients affected by iRBD with and without PLMS.
VariablesTotal iRBD Group
(N = 42)		iRBD-PLMS
(N = 23)		iRBD
(N = 19)		p-Value
Demographics
Sex: No. men/women #31/1116/715/40.74
Age, years (mean ± SD) $69.52 ± 7.9071.26 ± 5.8567.42 ± 9.590.29
Education level, years (mean ± SD) $11.64 ± 4.0011.04 ± 4.2412.37 ± 3.670.21
Disease features
Disease duration, years (mean ± SD) $4.05 ± 3.304.57 ± 3.783.42 ± 2.570.33
Age at onset of iRBD, years (mean ± SD) $65.69 ± 8.3066.61 ± 7.4264.58 ± 9.340.65
RBD symptoms on Video-PSG recording, n (%)
-Motor Agitation and Vocalization #16 (38.1)9 (39.1)7 (36.9)1
-Motor Agitation #9 (21.4)4 (17.4)5 (26.3)0.75
-Vocalization #4 (9.5)2 (8.7)2 (10.5)1
-PSG features without clinical signs #13 (31)8 (34.8)5 (26.3)0.80
PLMS index (mean ± SD) $27.47 ± 27.4946.45 ± 23.664.49 ± 4.510
Motor Evaluation
-Part III of UPDRS scale (mean ± SD) $1.44 ± 1.891.80 ± 2.301.00 ± 1.200.58
Neuropsychological Battery
-MMSE (mean ± SD) $28.35 ± 1.4228.51 ± 1.3628.16 ± 1.510.35
-Token test (mean ± SD) $32.17 ± 2.1332.55 ± 1.6731.71 ± 2.560.14
-RAVLT D.R. (mean ± SD) &5.26 ± 0.765.27 ± 0.825.25 ± 0.700.96
-RAVLT I.R. (mean ± SD) &43.86 ± 7.7144.90 ± 6.7042.61 ± 8.800.36
-Raven’s Progressive Matrices (mean ± SD)30.21 ± 3.1930.35 ± 2.6930.05 ± 3.770.78
-Corsi block-tapping Test (mean ± SD) &4.46 ± 0.904.53 ± 0.834.39 ± 0.990.63
-Digit Span Forward (mean ± SD) $5.77 ± 0.865.66 ± 0.855.90 ± 0.880.65
-Digit Span Backward (mean ± SD) &4.26 ± 0.894.53 ± 0.824.39 ± 0.980.39
-Verbal Fluency with Phonemic cues (mean ± SD) &32.71 ± 9.8130.96 ± 7.9734.84 ± 11.530.22
-Verbal Fluency with Semantic cues (mean ± SD) &45.98 ± 7.5746.48 ± 7.2945.37 ± 8.050.65
-Attentive Matrices (mean ± SD) $47.95 ± 6.3848.43 ± 6.2547.37 ± 6.650.98
-Copy Rey–Osterrieth complex figure (mean ± SD) &32.30 ± 4.2632.60 ± 3.6231.93 ± 5.020.82
Cardiac Autonomic Evaluation
-Cardiac Sympathetic Index (mean ± SD) $3.47 ± 4.153.43 ± 3.133.51 ± 5.220.38
-Cardiac Parasympathetic Index (mean ± SD) $3.09 ± 3.153.95 ± 3.982.04 ± 1.060.19
IRBD: idiopathic REM sleep behavior disorder; PLMS: periodic leg movements during sleep; PSG: polysomnographic; MMSE: Mini-Mental State Examination; RAVLT I.R: Auditory–Verbal Learning Test Immediate; RAVLT-DR: Auditory–Verbal Learning Test Delayed; Italian Population Normal Values: MMSE: >23.80; Token test: >26.5; RAVLT I.R.: >28.53; RAVLT D.R.: >4.69; Raven’s Progressive Matrices: >18; Corsi block-tapping Test: >3.46; Digit Span Forward: >4.26; Digit Span Backward: >2.65; Verbal Fluency with Phonemic cues: >25; Verbal Fluency with Semantic cues: >17; Attentive Matrices: >31; Copy Rey–Osterrieth complex figure: >28.88. # Χ2 test; $ Mann–Whitney U test; & Student’s t-test.
Table 2. Comparisons of synucleinopathies phenoconversion biomarkers in iRBD patients with and without PLMS.
Table 2. Comparisons of synucleinopathies phenoconversion biomarkers in iRBD patients with and without PLMS.
VariablesiRBD-PLMS
(N = 23)		iRBD
(N = 19)		p-Value $
Phenoconversion Biomarkers, n (%)
–
Hyposmia/Olfactory deficits
13 (56.5)3 (15.7)0.02
–
Cognitive Impairment
11 (47.8)4 (21.0)0.14
–
Abnormal 18F-FDG PET
10 (43.5)2 (10.5)0.04
–
Motor Symptoms
4 (17.4)2 (10.5)0.85
Patients with phenoconversion biomarkers, n (%)
–
at least 1 biomarker
22 (95.6)9 (47.3)0.001
–
1 biomarker
10 (45.5)7 (77.8)0.90
–
2 biomarkers
9 (41)2 (22.2)0.08
–
3 biomarkers
2 (9)-0.56
–
4 biomarkers
1 (4.5)-1
iRBD: idiopathic REM sleep behavior disorder; PLMS: periodic leg movements during sleep; 8F-FDG PET imaging: Fluorodeoxyglucose PET. $ Χ2 test.
Table 3. Classification performances of Machine Learning (ML) models and cardiac autonomic indices.
Table 3. Classification performances of Machine Learning (ML) models and cardiac autonomic indices.
Accuracy
(95% conf. int.)		AUCSensitivitySpecificityppvNpv
ML Models
LR0.71 (0.55–0.84)0.710.700.740.760.67
SVM0.81(0.66–0.91)0.750.830.790.830.79
RF0.86 (0.71–0.95)0.850.960.740.810.93
XGBoost0.78 (0.62–0.89)0.840.830.720.790.76
Autonomic Indices
Sympathetic0.71 (0.60–0.83)0.700.700.740.780.67
Parasympathetic0.69 (0.55–0.81)0.630.700.680.730.65
LR: Logistic Regression; SVM: Support Vector Machine; RF: Random Forest; XGBoost: eXtreme Gradient Boosting; AUC: Area Under the (ROC) Curve.
Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content.

Share and Cite

MDPI and ACS Style

Salsone, M.; Vescio, B.; Quattrone, A.; Marelli, S.; Castelnuovo, A.; Casoni, F.; Quattrone, A.; Ferini-Strambi, L. Periodic Leg Movements during Sleep Associated with REM Sleep Behavior Disorder: A Machine Learning Study. Diagnostics 2024, 14, 363. https://doi.org/10.3390/diagnostics14040363

AMA Style

Salsone M, Vescio B, Quattrone A, Marelli S, Castelnuovo A, Casoni F, Quattrone A, Ferini-Strambi L. Periodic Leg Movements during Sleep Associated with REM Sleep Behavior Disorder: A Machine Learning Study. Diagnostics. 2024; 14(4):363. https://doi.org/10.3390/diagnostics14040363

Chicago/Turabian Style

Salsone, Maria, Basilio Vescio, Andrea Quattrone, Sara Marelli, Alessandra Castelnuovo, Francesca Casoni, Aldo Quattrone, and Luigi Ferini-Strambi. 2024. "Periodic Leg Movements during Sleep Associated with REM Sleep Behavior Disorder: A Machine Learning Study" Diagnostics 14, no. 4: 363. https://doi.org/10.3390/diagnostics14040363

APA Style

Salsone, M., Vescio, B., Quattrone, A., Marelli, S., Castelnuovo, A., Casoni, F., Quattrone, A., & Ferini-Strambi, L. (2024). Periodic Leg Movements during Sleep Associated with REM Sleep Behavior Disorder: A Machine Learning Study. Diagnostics, 14(4), 363. https://doi.org/10.3390/diagnostics14040363

Note that from the first issue of 2016, this journal uses article numbers instead of page numbers. See further details here.

Article Metrics

Back to TopTop