Introduction

Parkinson’s disease (PD) and other alpha-synucleinopathies are characterized by a prodromal phase that precedes the onset of classic motor symptoms (e.g., akinesia, rest tremor, rigidity), often spanning years, if not decades, during which neurodegeneration progresses silently. Among the various premotor manifestations, rapid eye movement (REM) sleep behavior disorder (RBD) is one of the most prevalent ones1. RBD is a disturbing parasomnia marked by loss of normal muscle atonia during REM sleep, referred to as REM sleep without atonia (RSWA), and the occurrence of dream-enacting behaviors2. Isolated RBD (iRBD) has gained significant attention in recent years, due to its strong association with alpha-synucleinopathies such as PD, dementia with Lewy bodies (DLB), and the rare multiple system atrophy (MSA)3,4,5,6,7. The term ‘isolated’ refers to the occurrence of RBD without overt clinical signs of neurodegenerative disease at the time of diagnosis and without any other apparent explanation, including the lack of any obvious secondary cause for the RBD phenotype. Remarkably, over 90% of individuals diagnosed with iRBD will progress to PD, DLB, or MSA within 15 years of RBD onset8. This makes iRBD an ideal candidate for inclusion in population-based screening efforts aimed at identifying individuals at risk for developing neurodegenerative diseases such as PD. Timely recognition of PD is of great importance, particularly in anticipation of potential disease-modifying therapies9,10.

Video-polysomnography (vPSG) remains the gold standard for diagnosing RBD11. REM sleep on vPSG is characterized by several distinct features. First, there is a low-amplitude, mixed-frequency electroencephalographic (EEG) activity, with the absence of K-complexes and sleep spindles. Additionally, sharp, irregular, and conjugate ocular movements are observed, measured by electrooculography (EOG). Lastly, there is a reduction in muscle tone, as measured by chin-, arm-, and leg-electromyography (EMG)2. According to the third edition of the International Classification of Sleep Disorders (ICSD-3-TR)12, the diagnostic criteria for RBD include: (1) recurrent episodes of behavior or vocalizations, either observed during vPSG as originating from REM sleep or inferred from reports of dream enactment and (2) evidence of RWSA on a PSG. vPSG involves the overnight monitoring of patients in a controlled sleep laboratory. From a simple logistical point, access to an accredited sleep laboratory is limited, and vPSG is resource-intensive. Since typically conducted over one or two nights, vPSG is affected by night-to-night variability, limiting its reliability. Additionally, the visual analysis of vPSG data is time-consuming, requires specialized expertise, and presents challenges in accurately quantifying RSWA, even for experienced sleep experts. Consequently, taking all these aspects into account, vPSG is ill-suited for large-scale population screening initiatives13.

The growing demand for early detection of RBD underscores the urgent need for developing alternative screening tools. Questionnaires, such as the RBDSQ14, are commonly used as an alternative to PSG. However, they are significantly less reliable15. Home-based, cost-effective, and scalable modalities have the potential to overcome the limitations of vPSG by providing more representative assessments of natural sleep and enabling broader accessibility to the general population. This systematic review aims to explore emerging tools and modalities as alternatives to vPSG for detecting RBD (isolated RBD and in patients with alpha-synucleinopathies) in home settings, with a focus on their potential application in population-based screening for PD and related disorders.

Results

Selection of studies

A total of 1474 citations were found through the multiple databases search. After the removal of duplicate citations, 1065 citations remained. After abstract screening, 64 manuscripts were retrieved and analyzed in full text. Eventually, a total of 17 studies (i.e., 15 original articles, one conference paper, and one preprint) were included in this systematic review. Further details of the study selection process are shown in the PRISMA Flow Diagram (Fig. 1).

Fig. 1: Search Strategy According to PRISMA guidelines.
figure 1

PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analysis) flow diagram describing our search strategy of databases to retrieve and qualify publications of relevance for this systematic review.

Full size image

Systematic review characteristics

Multiple technologies have been developed to detect RBD using various approaches. All 17 studies were categorized into three main groups: actigraphy devices (n = 9)16,17,18,19,20,21,22,23,24, novel tools (n = 5)25,26,27,28,29 (Fig. 2), and emerging modalities (n = 3)30,31,32 (Table 1). The findings for each category are presented below (Tables 2 and 3), highlighting trends, similarities, and differences among the studies. The results of the quality assessment of the case-control studies indicate that the majority of included studies are of high quality (i.e., 11 out of 15), while four are of medium quality. Secondly, the two diagnostic accuracy studies generally exhibit a low risk of bias in key domains, supporting their classification as moderate to high-quality studies.

Fig. 2: Illustration of all novel devices.
figure 2

a Portable recording system (SleepGraph®, Proassist Co., Japan)27, reproduced with permission from the original author. b Soft EEG, EOG and EMG electrode array with Bluetooth data acquisition unit (DAU; i.e., the black box in the left figure)29, reproduced with permission from Journal of Sleep Research obtained via the Copyright Clearance Center29. c Dreem 2 EEG headband, with the layout of the device (left) and an illustration of an individual wearing it (right)26,48, reproduced from PLOS One, licensed under a Creative Commons Attribution 4.0 International License (CC BY 4.0) d Sleep Profiler28,49, reproduced with permission from Clinical Pharmacology & Therapeutics, obtained via the Copyright Clearance Center49. e 3D time-of-flight camera with the regions of interest investigated in this study (lower body (green), hands (white), head (blue), upper body (red)), reproduced with permission from the original author, and licensed under a Creative Commons Attribution 4.0 International License (CC BY 4.0)25.

Full size image
Table 1 Results table of all 17 studies
Full size table
Table 2 Results of quality assessment of diagnostic accuracy studies with QUADAS-2
Full size table
Table 3 Quality assessment with the Newcastle-Ottawa Scale
Full size table

Novel technologies for RBD detection

Our results first focus on tools capable of detecting key diagnostic criteria for RBD: RSWA and dream-enacting behavior. Additionally, other tools are discussed that employ alternative approaches, including the measurement of nocturnal activity, to identify RBD.

Portable biopotential systems (Fig. 2a), such as the device developed by Kataoka and colleagues27, have shown promise in detecting REM sleep abnormalities and RBD. This two-channel recording system measures EOG and EEG signals, enabling accurate detection of REM sleep. In addition, chin EMG activity was measured in order to automatically calculate the ratio RSWA to total REM sleep. Although EOG primarily records eye movements, in this system, it was also used as a proxy to detect muscle activity in the chin, complementing the EMG measurement. This dual function of EOG allows for the identification of RSWA during REM sleep. In this research, both PSG and the portable EEG/EOG recording system were performed simultaneously on the same night in a laboratory setting. Both the diagnostic PSG and portable device were manually and blindly scored by a certified sleep technologist. REM sleep staging was done by both the PSG and the portable system. The REM sleep staging of the portable device was then used to calculate the RSWA-to-total REM sleep ratio. Compared to PSG, the system showed moderate-to-high accordance for detection of RSWA, i.e., in six out of eight PD patients (p value: 0.686).

Secondly, the temporary-tattoo electrode arrays introduced by Shustak and colleagues33 are capable of recording EEG, EOG, and EMG signals for home-based sleep monitoring. This device was validated in a feasibility study33, with solely healthy controls (and therefore not included in our review), demonstrating acceptable signal quality and user comfort. The lightweight and flexible design of the tattoo electrodes ensures minimal disruption to natural sleep. Oz and co-researchers29 extended this work, i.e., the use of the temporary-tattoo, or soft electrode array (Fig. 2b), focusing on its ability to monitor sleep stages and detect RSWA at home. The cohort consisted of healthy controls (n = 21) and PD patients (n = 29). The study compared its performance to vPSG and validated the feasibility of identifying RSWA in non-clinical settings. They found a sensitivity of 85.7% and specificity of 58.3% for RSWA detection based on a single night of simultaneous recording with PSG and the soft electrode array in the sleep laboratory. An additional home-based recording night in a subgroup of 25 patients showed 100% sensitivity and 89.47% specificity, when comparing lab-based vPSG-confirmed RBD with RSWA detection using the wearable. Nevertheless, variability in signal quality across different environments highlights the need for robust algorithm refinement. Possti et al.32 further advanced the application of this tool by developing and evaluating a semi-automatic method for quantifying RSWA. The cohort consisted of patients with PD (with RBD (n = 18) and without RBD (n = 13)) and healthy controls (n = 24). This approach successfully measured RSWA across two separate nights, one in the sleep laboratory (with PSG and wearable) and one in natural sleep environments, e.g., at home (with the wearable only). A 60% correlation between RSWA indices from home and laboratory settings was achieved, based on automated scoring of the wearable data. At the same time, it significantly reduced the manual effort required for data analysis. By incorporating automation, the device demonstrated potential for scalable screening, though issues with data consistency across diverse scorers or data analysts were identified.

Another wearable device is the Sleep Profiler28 (Fig. 2d), an in-home patient-applied sleep test that is capable of measuring sleep stages. Levendowski and colleagues (A)28 evaluated the Sleep Profiler for automated RSWA detection and its reliability in home settings for RBD (n = 66) and healthy controls (n = 21). In-laboratory validation showed excellent agreement between visual and automated RSWA scoring, with intra-class correlations of 89% (chin) and 95% (arm). In-home recordings achieved specificities of 88% (chin), 93% (arm), and 90% (chin or arm), with corresponding sensitivities of 81%, 81,% and 86%. Night-to-night consistency was high (ICC: 0.81-0.84), though some variability in abnormal RSWA detection was observed. Automated RSWA detection closely matched expert visual scoring.

In 2023, Cesari and coworkers25 introduced and validated a 3D time-of-flight camera for detecting iRBD (Fig. 2e). This contactless device uses machine learning (ML)-based automated movement analysis to track subtle body movements during REM sleep. In a controlled sleep-laboratory setting, the ML model achieved a high sensitivity of 83.6% for classifying individuals with iRBD, using the clinical diagnosis based on vPSG as the reference standard. While these findings underscore the potential of this technology as a screening tool, its application in home environments remains untested. Nonetheless, the authors emphasize this technology’s promise for enabling the detection of a larger number of individuals with iRBD in the future. The ML model reached an overall accuracy of up to 86.6% for distinguishing RBD (n = 53) from other sleep conditions (n = 128), including sleep-related breathing disorders, restless legs syndrome (RLS), insomnia and NREM parasomnias, using only two features of rate (frequency) and ratio (burden) of movements25.

Building on these advancements, Abdelfattah and co-authors30 refined video-based RBD detection by developing an ML classifier using a simpler 2D camera, commonly available in clinical sleep laboratories. They compared iRBD patients (n = 81) with a random sample of non-RBD controls from their sleep clinic (n = 91). By incorporating three features, (1) velocity of movements, (2) magnitude of movements and (3) immobility fragmentation, this approach improved upon earlier methods, achieving the highest classification accuracy of 91.9% when focusing on short movements (0.1–2 seconds). To the authors’ knowledge, the model outperformed previous video-based 3D systems and classifiers relying on traditional PSG signals (EEG, EOG, and EMG). Notably, seven out of 11 iRBD patients without noticeable movements during vPSG were correctly identified, highlighting its sensitivity to subtle motor abnormalities. The authors propose that integrating this 2D camera system with automated REM sleep detection could further facilitate in iRBD diagnosis and RBD monitoring30. However, similar to the study by Cesari et al., this technology has yet to be tested in home environments.

Alternative approaches, such as measuring nocturnal activity and rest-activity cycles, have been extensively studied for the detection of RBD. In particular, actigraphy has emerged as a non-invasive, cost-effective alternative to PSG for identifying RBD in home settings. Actigraphy involves the use of devices worn on the wrist equipped with an accelerometer to record movement over extended periods, typically spanning multiple days and nights. By analyzing rest activity cycles, patterns of both daytime activity and nocturnal rest, these devices estimate sleep patterns and nocturnal motor activity, with sleep being deduced from sustained periods of low movement. This method is often accompanied by a sleep diary, in which the patient indicates periods of wakefulness during the night. Combining actigraphy with subjective sleep reports can enhance the accuracy of sleep assessments34. These tools hold significant promise due to their accessibility and scalability, making them valuable for the early detection of sleep-related movement disorders.

The earliest study in this area was conducted in 2010 by Naismith et al. 22, who examined the association between actigraphy-defined sleep disturbances and RBD in PD patients. Twenty-two PD patients underwent neurological assessment, completed sleep questionnaires and wore an actigraphy device for two weeks. Based on RBD questionnaire results, they were classified as RBD negative (RBD-, n = 9) or positive (RBD + , n = 13). Actigraphy data showed that RBD+ patients had significantly more wake bouts (40.56 ± 12.3) than RBD- patients (27.22 ± 8.9). In line with the original terminology used in RBD actigraphy studies, wake bouts refer to periods of increased nocturnal activity detected by the device, which were initially interpreted as wakefulness but are now considered to also reflect dream-enacting behaviors. In conclusion, this study demonstrated a correlation between increased nocturnal motor activity and the presence of RBD, suggesting that actigraphy could serve as a useful tool for RBD detection. However, the sample size was small, and there is a lack of PSG validation22. In 2014, Louter and colleagues21 investigated this further to confirm the findings of Naismith and coworkers in a larger group of well-defined PD patients (i.e., 23 individuals RBD+ and 22 individuals RBD-). Actigraphy-derived nocturnal motor activity (also defined as wake bouts) was compared with PSG data. The reported sensitivity was 20.1%, with a specificity of 95.5%. The authors suggested to combine actigraphy with RBD questionnaires to improve sensitivity21.

In 2018, Stefani et al.24 evaluated this combination, using the MicroMini-Motionlogger actigraphy as a screening tool for iRBD. The study compared patients with iRBD (n = 20) to patients with sleep apnea (n = 20), RLS (n = 20), the combination thereof (n = 10), and controls (n = 20). Actigraphy data were analyzed both automatically, based on different nocturnal activity indices, and visually by seven blinded sleep medicine experts, who rated the recordings as “no,” “possible,” or “probable” RBD. They reported that quantitative actigraphy analysis could be useful to distinguish RBD patients from controls, but not from patients with other sleep disorders. However, visual analysis of the actigraphy data (based on specific REM sleep activity patterns) by blinded sleep experts could differentiate between RBD and other sleep disorders with a sensitivity of 65-90% and a specificity of 64-86%, based on actigraphy data alone. The addition of clinical information, including RBD questionnaires, further improved the area under the curve for each rater and the diagnostic performance, resulting in a sensitivity of 85-95%, specificity of 79–91% and diagnostic accuracy of 81-91%. These findings suggest that the combination of visual analysis of the MicroMini-Motionlogger combined with clinical information, could serve as a RBD screening tool24. Sandala and co-researchers20 used a similar study design, using the combination of the Motionwatch actigraphy and the RBDSQ14. They found that the automatically calculated short immobile bouts (SIB) percentage, the number of non-registered activity episodes less than or equal to one minute divided by the total of non-registered activity episodes, differentiated best between RBD (n = 45) and other sleep-related motor disorders (n = 30) and controls (n = 20). Combining actigraphy data (SIB) with RBD-SQ scores allowed correct classification of RBD status in 87.6% of cases. It should be highlighted that Sandala et al.20 and Stefani et al.24 found that the percentage of SIB was the most predictive feature for RBD detection among all actigraphy-derived metrics. In 2020, Filardi et al.18 expanded the study of Stefani and colleagues24 and used the MicroMini-Motionlogger actigraphy to study 24-hour rest-activity patterns in iRBD during two weeks. The study identified distinct activity patterns in iRBD patients, such as lower sleep efficiency, increased estimated wakefulness after sleep onset (WASO), and increased occurrence of naps during the day. These sleep metrics were inferred from rest-activity patterns rather than direct sleep recordings. The researchers measured I < O, a 24 h measure which quantifies the ratio between nocturnal (I, i.e., inactive) and diurnal (O, i.e., active) motor activity intensity. This metric effectively distinguished iRBD patients (n = 19) from those with RLS (n = 20), sleep apnea (n = 19) and controls (n = 16). A threshold of 98.32 provided the optimal balance between sensitivity (63.2%) and specificity (89.1%), with lower values indicating increased nocturnal motor activity, a hallmark of iRBD18.

In the last few years, several studies have focused on ML-based analyses of actigraphy data. Building on the advancements described above, Brink-Kjaer et al. (A)16, explored a combined approach to enhance the detection of iRBD by integrating home actigraphy data with a brief questionnaire and data from a sleep diary. The study consisted of 42 iRBD patients and 42 controls, consisting of 21 sleep clinic patients and 21 community-dwelling healthy controls. The actigraphy parameters were a combination of the parameters of Stefani et al.24 and Sandala et al.20, combined with new parameters found by visual interpretation, i.e., twitching activity, long immobile bouts, activity counts. The questionnaire included the RBD Innsbruck Inventory, along with three prodromal symptoms of an alpha-synucleinopathy: subjective hyposmia, constipation, and orthostatic dizziness. An ML classifier, which used different combinations of datasets, was evaluated using a leave-one-out cross-validation method. This ensured an unbiased performance estimate by iteratively holding out each participant’s data for validation while fitting the model on the remaining data. The classifier showed the highest precision with the combination of a multi-feature actigraphy prediction score and all questionnaire features, achieving a sensitivity of 88.1% and specificity of 100%. The authors proposed a stepwise approach for population screening, e.g., starting with the questionnaire, followed by 10 days of actigraphy for individuals with positive questionnaire results and a confirmatory vPSG. The selection of a 10-day actigraphy window is supported by findings showing that classifier performance reaches a plateau after seven to 10 nights. Brink-Kjaer et al. (B)17, also expanded the study with 24-hour rhythm features to distinguish iRBD (n = 42) from control groups (n = 42), including 21 clinical controls with other sleep disorders and 21 healthy community controls. In addition, the 24-hour rhythm actigraphy model was further refined using data from a subgroup of 38 iRBD patients with daytime measurements, alongside 110 control participants from the UK Biobank matched for age, sex, and BMI. The detection model achieved high performance with a sensitivity of 78.9%, specificity of 96.4%, and an area under curve (AUC) of 0.954 when combining the nighttime and 24-hour rhythm classifiers. Unlike prior actigraphy-based studies, this method did not rely on sleep diaries, increasing its practicality for large-scale screening. The authors explain that at least four nights of recording are required for optimal performance of the combined classifier.

Ko et al.19 explored the use of an ASUS VivoWatch BP smartwatch, which utilizes activity and heart rate variability (HRV) to automatically classify sleep stages, including REM sleep, and detect sleep disturbances associated with RBD in patients with PD (RBD status unknown). An algorithm was developed for REM sleep classification based on HRV-derived features and accelerometer data. The percentage of abnormal motor activity during REM sleep, considered an indirect marker of RBD, was compared between patients with PD (n = 20) and healthy controls (n = 18). A significantly higher percentage of abnormal REM sleep was observed in patients with PD (p value = 0.007). Abnormal REM sleep was quantified as the percentage of REM sleep in which abnormally high G-values occurred, reflecting muscle activity or movements that are typically absent during REM sleep. No vPSG confirmation was performed19. Raschellà et al.23 combined the GENEActiv Original wrist actigraph (GENEActivTM; Activinsight Ltd., Kimbolton UK)35 with ML classification algorithms for home-based screening of RBD in PD patients. Initially, classifiers were trained and validated on data collected during in-laboratory recordings (PD patient with RBD (PD RBD+) n = 18, PD patients without RBD (PD RBD-) n = 8), using fourfold cross-validation repeated 100 times to avoid overfitting. Subsequently, for home-based evaluation, classifiers were trained on data from 6 PD patients (n = 3 RBD, n = 3 non-RBD) and tested on an independent cohort of the remaining 20 PD patients who were not included in model training. The models were based on the most relevant actigraphic features (e.g., individual movement episodes, length and magnitude of movements). Classification performance was then assessed in relation to clinical diagnosis. The study achieved remarkable results, with an accuracy of 92.9 ± 8.16% during in-clinic tests. More impressively, an accuracy of 100% was achieved for the identification of RBD over the two-week home recording period in the independent PD test cohort (n = 20). Additionally, the approach was tested in a control group of non-PD controls (n = 18), yielding an accuracy of 94.4% for identifying RBD-like behaviors.

In 2024, Levendowski et al. (B)31 explored the use of nine automatically derived sleep biomarkers other than RSWA, extracted using the Sleep Profiler, to support the profiling of neurodegenerative disorders, including iRBD. An ML algorithm was trained on sleep data from patients with various neurodegenerative conditions to differentiate between diagnostic groups. The study included iRBD patients (n = 33), healthy controls (n = 186), Alzheimer’s disease (n = 37), Lewy Body dementia (n = 18), PD (n = 29), and mild cognitive impairment (MCI) (n = 78) patients. These investigated biomarkers included REM sleep time, sleep efficiency, supine time, non-REM (NREM) relative theta, NREM theta/alpha, autonomic activation index, NREM hypertonia, atypical N3, and spindle duration. While the primary aim of the study was not to directly detect RBD or RSWA, the authors reported the diagnostic performance of these individual sleep biomarkers for NDD profiling. Sensitivity ranged from 63.0% for supine sleep to 41.0% for sleep efficiency, while specificity ranged from 59.0% for supine sleep to 89.0% for sleep efficiency. Moderate specificity was observed for the autonomic activation index (75.0%) and NREM hypertonia (79.0%), with sensitivity values of 54.0% and 70.0%, respectively. However, the study also showed that 27% of iRBD patients were misclassified as having normal sleep, highlighting limitations in sensitivity. The authors suggested that sleep profiling could help distinguish iRBD from other conditions, with possible additional automated RSWA detection.

A pilot study26 evaluated the performance of a consumer EEG device, the Dreem 2 EEG headband (Fig. 2c), in assessing sleep stages compared to PSG in individuals with early-stage PD (n = 10). The Mayo Clinic Questionnaire (MSQ) was used to assess symptoms of RBD, and six out of 10 PD patients endorsed RBD. The headband achieved an overall sleep staging accuracy of 80.8%, comparable to previous studies in healthy adults. Sensitivity for sleep stage classification was generally low (15–57%), whereas specificity remained high (78–96%). PD duration and RBD symptoms were associated with more WASO26. Differences from prior studies may stem from variations in study populations, as PD is known to alter sleep architecture.

Practical implications of novel technologies

Some technologies are already commercially available and can be implemented in research, or clinical practice, with pricing and software influencing their feasibility for different settings. Therefore, Table 4 provides an overview and summarizes which technologies are currently available, their cost and operating software.

Table 4 Overview of commercial availability, pricing and used software of the novel technologies
Full size table

Discussion

vPSG remains the gold standard for diagnosing RBD, providing direct visualization of dream-enacting behaviors and objective quantification of RSWA, with diagnostic accuracies ranging from 80 to 95%36,37. However, its high cost, limited availability, and impracticality for large-scale screening underscore the urgent need for home-based alternatives. This need has driven the development of novel detection tools, including actigraphy, portable biopotential systems, wearable sensors, and video-based monitoring38. Our systematic review highlights significant advancements in non-PSG technologies, with some methods now demonstrating diagnostic accuracies comparable to traditional PSG-based methods. This underlines the importance of balancing sensitivity and specificity, and in the context of large-scale screening, achieving high sensitivity to minimize false negatives is often prioritized. Notably, 14 of the 17 reviewed16,18,20,21,23,24,25,26,27,28,29,31,32 studies used video-PSG as the reference standard, reinforcing the reliability of their findings.

Actigraphy-based studies demonstrated a clear evolution in their diagnostic performance for RBD. Early studies21,22 suggested that nocturnal motor activity could differentiate RBD patients from controls, though with limited sensitivity and a high specificity. Later research18,24 highlighted the importance of multi-night recordings to reduce variability and improve accuracy. Recent advancements in ML-enhanced actigraphy16,23 have significantly improved sensitivity and specificity, with some models reaching up to 100% accuracy in home-based RBD detection in PD patients23. Notably, studies16,24 consistently demonstrated that combining actigraphy with complementary diagnostic tools, such as the RBD Innsbruck Inventory, significantly enhances diagnostic precision. The combination might serve for large-scale population screening. However, as noted by Stefani et al., quantitative actigraphy may not reliably differentiate RBD from other sleep-related motor disturbances, such as periodic leg movements (PLM) and sleep apnea (SA) related movements24. In their study, actigraphy parameters distinguished RBD patients from patients with other motor-related sleep disorders, only with visual analysis by experts. Furthermore, actigraphy does not assess REM sleep or RSWA according to current diagnostic criteria based on EEG, EMG, and EOG. While actigraphy represents a promising, cost-effective alternative for home-based RBD detection, these limitations suggest that further research is needed to confirm these possibilities and should focus on standardizing protocols and validating findings across larger and more diverse populations. Special attention should be given to individuals with fragmented sleep, such as patients with insomnia, as this may impact the accuracy of rest-activity-based sleep estimation.

Portable biopotential systems, such as EEG/EOG/EMG-based devices27,29,32 and the Sleep Profiler28 showed strong potential for (semi)automatic RSWA detection with a moderate-to high correlation with RSWA on a vPSG. The portability and affordability of the EEG/EOG/EMG system make it a suitable tool for home-based diagnostics, but its limited electrode configuration may reduce sensitivity for complex cases. Additionally, temporary tattoo electrodes exhibit signal variability across environments, which is reported as an important limitation29,32. Furthermore, night-to-night inconsistencies remain a challenge28. Additionally, environmental factors (e.g., lighting, positioning, and external noise) may affect the reliability of video-based and biopotential recordings.

Studies using contactless video monitoring25,30 have redefined how dream-enacting behaviors may be objectively and automatically quantified. Cesari and colleagues25 introduced a 3D time-of-flight camera. Abdelfattah and co-authors30 extended this approach using a standard 2D camera system. Notably, their model outperformed prior 3D camera systems and EEG/EOG-based classifiers, suggesting video-based monitoring may offer a more sensitive and scalable method for detecting RBD-related motor abnormalities. However, both studies were conducted in controlled laboratory settings, and home-based validation remains a critical next step before these technologies can be widely implemented. Additionally, video-based systems do not quantify RSWA, limiting their ability to confirm an RBD diagnosis without complementary EMG analysis. Furthermore, the models in these studies required prior REM sleep staging by human experts through visual analysis, which undermines the potential for fully automated, at-home use, where expert input is not available.

In the context of detecting RBD as a prodromal feature of PD and related disorders for population studies, a key distinction must be made between screening tools and diagnostic methods. Screening tools are used to identify at-risk individuals. Confirmatory diagnostic methods are used to establish a clinical diagnosis. High sensitivity is a key priority in screening tools39. The included devices that show the highest sensitivity might be the most suitable for this purpose. Furthermore, a stepwise approach, as suggested by Brink-Kjaer et al. (A)16, that uses one of the described devices (actigraphy, video-based movement tracking, and wearable biosensors) and a questionnaire as a screening tool with vPSG confirmation in positive screens, might give the best result. The studies in our review focused on RBD detection, isolated and in patients with alpha-synucleinopathies. In neuroprotective trials, wearables may also be important for outcome monitoring of RBD severity, as vPSG studies in PD suggest that both dream-enacting behavior event severity and RSWA severity increase over time10,40,41. From vPSG research, it is known that RSWA severity is the most consistent biomarker for phenoconversion10,42. Wearables capable of measuring RSWA could be most suitable for this. Future studies should evaluate this approach in larger multicenter study populations (Fig. 3). Combining wearables for RBD detection, with other prodromal PD biomarkers, such as olfactory testing, Dopamine Transporter (DAT) Single Photon Emission Computed Tomography (DAT-SPECT), biofluid biomarkers or wearables that detect other subtle prodromal PD signs (such as gait analysis, speech analysis)42, could also increase specificity for identifying patients with prodromal PD. Additionally, to address the lack of objective endpoints in symptomatic trials for RBD, particularly outside of a single night in the lab, video-based monitoring of REM movements emerges as a promising modality. This approach could provide a more continuous, real-world measure of symptom progression and could complement traditional lab-based assessments, enhancing the ability to track RBD-related motor abnormalities over time.

Fig. 3: Proposal for a stepwise approach of RBD screening, RBD confirmation, and RBD monitoring using novel technologies.
figure 3

RSWA: REM sleep without atonia, PSG: polysomnography, 3D: 3-dimension, EEG: Electroencephalography, EOG: Electrooculography.

Full size image

Several limitations must be acknowledged. First, it remains unclear whether some of these tools are detecting isolated RBD or PD-related motor symptoms. For example, research by Kataoka et al. and Gonzalez et al. exclusively included patients with PD, raising the question of whether these tools primarily detect RBD or whether they reflect PD-related motor symptoms. Some evidence suggests that nocturnal motor activity captured by actigraphy, or video-based tools, may reflect early PD-related motor symptoms rather than RBD-specific behaviors. Further research is needed to assess whether these technologies can differentiate between iRBD and manifest neurodegeneration. Secondly, although some studies report group-level diagnostic performance, recent studies employing actigraphy and video-based monitoring, particularly those using ML models, increasingly focus on individual-level diagnostic accuracy, emphasizing the models’ ability to correctly classify individual patients rather than just distinguishing between groups. However, these methods still face important challenges. Actigraphy, video-based monitoring, and wearable sensors rely on probabilistic classifications, meaning some individuals may be misclassified due to inter-individual variability in sleep movement patterns. Multi-night recordings and individualized diagnostic thresholds may enhance diagnostic accuracy. Moreover, ML-based methods are promising, but also create challenges. The performance of these algorithms depends on the quality and representativeness of the training data. Biases in datasets, overfitting, and lack of validation (external) can limit generalizability. Thirdly, this systematic review only included studies that evaluated subjects with RBD, e.g., isolated RBD and RBD in patients with alpha-synucleinopathies. Several other tools are under development but have not been tested in subjects with RBD yet. An example is the smart sleep mask43, which can track eye movements, head positions, and breathing patterns, which are crucial for assessing sleep stages and REM-related disorders. This might also be a candidate for future RBD detection studies. Fourthly, the included studies vary in how they define and categorize RBD and PD subgroups. Some investigate iRBD separately, while others include mixed populations of PD patients with and without RBD, or use other sleep-related disorders as controls. This heterogeneity complicates direct comparisons. Fifthly, adherence of patient acceptance of the different technologies was rarely reported. These factors are essential for the successful implementation of these technologies in clinical practice. Wearables such as wrist actigraphy are generally well-accepted due to their unobtrusiveness, while headbands and home-PSG systems might offer increased comfort and convenience compared to traditional in-lab PSG. However, video-based systems may raise privacy concerns. Future studies should therefore systematically evaluate usability, adherence, and patient acceptance of different technologies.

Finally, none of the included studies compared the different wearables described or the methods used to analyze them with each other. This makes it difficult to conclude which of the novel technologies has the highest diagnostic accuracy and should be recommended for large-scale RBD detection. Future studies should focus on the comparison of the different technologies in the same subjects and on whether a combination of different wearables that measure different variables, for example, actigraphy and an RSWA detector, could increase diagnostic accuracy. While sensitivity and specificity are key performance metrics, three of the 17 included studies did not report these values. In these cases, alternative results were used, and where possibl,e sensitivity and/or specificity were calculated based on the available data. We acknowledge that the absence of these key values limits direct comparability between studies.

Our systematic review discussed the ongoing efforts to make RBD detection more accessible and user-friendly, and demonstrates a comprehensive overview of significant advancements in relevant devices and technologies. The review included nine actigraphy studies that show improvement in diagnostic accuracy over the years, especially when combined with a RBD questionnaire. Furthermore, novel tools and modalities to detect RSWA and/or dream-enacting behavior, such as temporary tattoo EEG, 3D time-of-flight camera, and portable biopotential systems, look promising. Given that RBD is one of the strongest prodromal markers of PD, these advancements can play a significant role in earlier identification and risk stratification of PD. Most devices might function as screening tools for RBD detection as a prodromal Parkinson feature in population studies. Overall specificity may further be enhanced by adding an additional simple test, such as for olfactory function in the cascade of diagnostic steps. Future studies should focus on validating these technologies and results in well-defined iRBD cohorts, assessing their cost-effectiveness, and conducting multicenter studies to enhance their clinical applicability and ensure reliability across diverse clinical settings.

Methods

Search strategy

This systematic review was conducted in accordance with PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analysis) guidelines44. A comprehensive electronic search was performed across three major databases: PubMed, Web of Science, and the Cochrane Library. Only original articles were included, with the exception of conference papers and preprints, provided they met the criteria for sufficient quality assessment and presented clearly defined results. Furthermore, the reference lists of included articles were systematically reviewed to identify additional relevant studies. The search was conducted on January 11, 2025, with the detailed strategy outlined in Fig. 1.

The title and abstracts were independently screened for relevance by two reviewers, KC and A-MS. Subsequently, full-text articles were retrieved and independently reviewed by the same reviewers. Any disagreements or ambiguities were resolved through mutual discussion. Our systematic review was not registered, and a protocol was not predefined.

PICO model

The systematic review aims to address the following question:

Which non-polysomnographic tools are currently available and capable of detecting RBD and/or RSWA in home-settings?

  • Population (P): individuals with RBD, isolated or diagnosed in association with a neurodegenerative disease, either diagnosed by the ICSD-3 or RBD questionnaires (i.e., RBDSQ)14.

  • Intervention (I): Non-PSG tools or modalities potentially useful for home detection of RBD.

  • Comparison (C): PSG, the standard method for detecting RBD and RSWA, healthy control subjects (giving information about false negatives), or alternative methods such as questionnaires indicating suspected RBD.

  • Outcome (O):

    • Primary outcome: Sensitivity and specificity, or accuracy of the non-PSG tools in detecting RBD, or

    • Secondary outcome: Agreement, Cohen’s Kappa or accordance of a PSG and the non-PSG tools in detecting RBD or REM sleep staging.

All studies meeting these criteria were included. Articles published in English or Dutch were considered, as these are the languages spoken by the authors. However, all the articles found were in English.

It is important to note that this review focused on studies using novel tools for the home detection of RBD signs. Studies primarily targeting other prodromal Parkinson signs in RBD patients, such as gait analysis, balance assessment, or speech analysis, were excluded.

Data extraction and synthesis

Study data were extracted into Zotero45 by the main author, KC, and checked for accuracy and relevance by an independent reviewer, FD. Extracted data included demographic information, such as number, sex, age of participants, distribution of RBD, PD with or without RBD and healthy controls, type of device, type of study (monocentric, multicentric), aim of the study and results, such as accuracy, sensitivity and specificity (Table 1). A meta-analysis of the data was not possible due to the lack of homogeneity in the outcomes measured. Instead, we present a narrative synthesis of the reported data.

Quality assessment

Qualitative research was assessed using the Newcastle-Ottawa Scale46 (Table 3) for case-control studies, which applied to 15 out of 17 studies. For the two diagnostic accuracy studies, the QUADAS-247 (Quality Assessment of Diagnostic Accuracy Studies, see Table 2) tool was performed.