Research questions

Following the methodological rigor described in the [22] protocol, all the relevant studies were selected to answer the study’s research questions, which investigate different aspects of the use of sEMG in the assessment of individuals with ALS, considering clinical parameters, methodologies used and integration with other technologies, as shown in the Table 1.

Download:

• PNG
  larger image
• TIFF
  original image

Table 1. Research questions.

https://doi.org/10.1371/journal.pone.0350029.t001

Eligibility criteria

The eligibility criteria were previously defined and recorded in the protocol for this systematic review [22]. The selection of studies was guided by the PICOTS strategy (Participants, Interventions, Comparisons, Outcomes, Period, and Type of Study) [23], contemplating:

• Participants: individuals with a confirmed diagnosis of ALS, according to the revised El Escorial criteria [24].
• Interventions: use of sEMG to characterize motor symptoms, alone or combined with clinical scales or assistive technologies.
• Comparisons: different signal analysis approaches (time, frequency, or both), as well as comparing sEMG device configurations.
• Outcomes: motor parameters obtained through sEMG. Studies focusing exclusively on respiratory muscles, cognitive, or bulbar function were excluded.
• Evaluation periods: baseline, post-intervention (short term) and follow-up (medium or long term).
• Eligible study types: observational (cross-sectional and longitudinal), quasi-experimental and experimental studies, including randomized and non-randomized clinical trials.

Electronic research

The articles were selected through a systematic search conducted in various electronic databases, including MEDLINE (via PubMed), Web of Science, Embase, IEEE Xplore, Scopus, Google Scholar, SciELO, PEDro, Cochrane CENTRAL, and LILACS. Ongoing or unpublished clinical trials were identified through the ClinicalTrials.gov and World Health Organization International Clinical Trials Registry Platform. To complement the search, the reference lists of the included studies were manually analyzed, as well as the proceedings of relevant conferences, to locate potentially unindexed grey literature.

Search strategy

The search strategy used in this review was previously recorded and is detailed in the research protocol [22]. To ensure that the present manuscript is self-contained, we summarize the full search approach below. The strategy was developed using a combination of controlled descriptors (MeSH and DeCS) and free terms related to surface electromyography, Amyotrophic Lateral Sclerosis, clinical motor signs and symptoms, physical rehabilitation, and body segments.

Keywords and synonyms were grouped into broad conceptual categories—body segment, physical rehabilitation, clinical data, electromyography/mechanomyography, and ALS terminology—as presented in Table 2. Our multidisciplinary team identified relevant descriptors and synonyms, which were then adapted to the syntax and indexing structure of each database. Boolean operators (“AND”, “OR”), truncation symbols, and field tags were applied as appropriate.

Download:

• PNG
  larger image
• TIFF
  original image

Table 2. Main concepts and related keywords.

https://doi.org/10.1371/journal.pone.0350029.t002

We adopted a sensitive and non-specific search strategy to maximize retrieval of potentially eligible studies. Therefore, search terms were intentionally broad, and irrelevant articles were excluded during the selection phase. No restrictions were applied regarding language or year of publication.

Study selection

Three independent reviewers (APMF, MCFS, ACSML) analyzed the titles and abstracts to determine the eligibility of the studies. Any disagreements were resolved with the participation of a fourth reviewer (BHSB). The selected studies were imported into the Rayyan software [25], making it easier to exclude duplicates and apply the inclusion and exclusion criteria. In cases of doubt, the full text was obtained for a more detailed evaluation. Studies that did not meet the criteria of the review protocol concerning the type of study, participants, intervention, comparison group and study design were excluded.

To ensure that only relevant articles were included in the review, the following Selection Criteria (SC) were established when reading the title and abstract:

• SC1: Articles published in peer-reviewed scientific journals or peer-reviewed scientific conference proceedings;
• SC2: Studies using sEMG as an analytical tool in the investigation of ALS;
• SC3: Studies that have applied sEMG signal processing and analysis methods to the functional assessment of ALS patients;

The selection criteria were applied as a first filter to gather only relevant studies. CS1 ensures the reliability of the articles selected, guaranteeing scientific rigor in the description of methods, results, and analysis. CS2 and CS3 reinforce the relationship between the studies and the scope of the review, focusing on neuromuscular analysis in ALS patients.

When the abstract did not provide enough information about the focus of the research, the article was temporarily kept for full reading. At this initial stage, the methodological quality of the studies was not assessed. The selected articles were then applied to the Exclusion Criteria (EC):

• EC1: The study did not investigate the use of sEMG signals to collect neuromuscular data from people with ALS or to identify clinical signs and symptoms;
• EC2: The study exclusively used a biological signal other than the neuromuscular signal;
• EC3: The study did not collect data from people diagnosed with ALS.

If an article met any of the Exclusion Criteria, it was eliminated from the review. The remaining articles were then assessed for Inclusion Criteria (IC):

• IC1: The study used sEMG to collect data from people with ALS;
• IC2: The study featured neuromuscular processing and analysis to assess clinical signs in people with ALS.

If an article did not meet any of the Inclusion Criteria, it was excluded. After this screening, the selected articles were subjected to a detailed assessment of methodological quality and risk of bias to determine their final inclusion in the systematic review.

Data extraction

Data extraction was conducted by five independent evaluators (APMF, MCFS, ACSML, LHBB, BHSB), based on a standardized form previously defined in the study protocol [22]. Any disagreements were initially resolved by discussion between reviewers or, if necessary, with the intervention of a sixth reviewer (GAMV).

The information extracted included general data on the studies (year, authors, location, evaluator group), characteristics of the participants (number, age, time since onset of symptoms and diagnosis), details of the control groups, and information on the instruments and protocols used. Technical aspects of sEMG signal collection, such as electrodes, sampling frequency, muscles analyzed, type of motor tasks, and integration with other technologies, were also recorded.

The sEMG signal processing was described in detail, including the filters applied, analysis domains (time, frequency, and time-frequency), computational tools, and algorithms used. The main findings were organized into three axes: clinical data, sEMG signal processing, and the effects of the interventions.

In cases of incomplete data, the corresponding authors of the included studies were contacted to obtain the missing information, as provided for in the protocol.

Data analysis

Study selection

A total of 30,478 records were identified. After removing duplicates, 23,245 articles were screened by title and abstract, with 531 selected for full-text review. Following eligibility assessment and methodological quality evaluation, 34 studies were included in the quantitative synthesis.

The summary of this process is illustrated in the PRISMA flow diagram in Fig 1.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 1. Flowchart of the study selection process according to PRISMA guidelines.

https://doi.org/10.1371/journal.pone.0350029.g001

Date and place of publication

Fig 2 below shows the results of the temporal analysis of the publications included in the review, illustrating the distribution of studies over the years.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 2. Distribution of included studies by year of publication.

https://doi.org/10.1371/journal.pone.0350029.g002

The studies included in this review were published between 1995 and 2023. The earliest works, from the mid-1990s, originated in the United States, such as the studies by Felice et al., [29] and Bromberg et al., [30]. Throughout the 2000s, research output remained moderate, with contributions from Kent et al., [31] in the United States, Sanjak et al., [32] in China, and Kleine et al., [33] in the Netherlands. However, from 2010 onwards, there was a noticeable increase in scientific research, with significant contributions from the Netherlands [6,34,35], Australia [36], Switzerland [37,38], South Korea [39,40], China [9,13,41,42], Brazil [43], and the United States [44,45]. In the past five years, this growth became even more evident, especially with new publications from the United Kingdom [46–51], Portugal [52–54], France [55], Spain [56], Japan [14,57], and again the United States [58]. This trend highlights the growing global interest in sEMG applications in the context of ALS.

Clinical data identified

RQ1: Which motor parameters, such as fatigue, fasciculations, paresis, or plegia, have been assessed using sEMG in individuals diagnosed with ALS?

The studies included in this review addressed clinical parameters such as fatigue, fasciculations, muscle weakness, loss of motor units, muscle coactivation, recruitment patterns, and changes in neuronal excitability, presented in the S1 Table – Motor Clinical Parameters Evaluated in the Included Studies.

The fatigue, both peripheral and central, has been widely investigated and assessed through changes in strength, activation patterns, and functional performance [31,32,37,39,43,56].

The muscle fasciculations were characterized in terms of frequency, morphology, amplitude, firing pattern, neuronal origin, and electromechanical latency, using different detection approaches and tools [9,33,34,44–49,51].

Muscle weakness was analyzed using measures such as maximum voluntary strength, the Medical Research Council (MRC) Muscle Strength Scale, and functional decline [6,30,35,38,43,58].

The loss of motor units was investigated using techniques such as MUNE and MUNIX, with a focus on analyzing the morphology of motor unit action potentials (MUAPs) and neuronal hyperexcitability [13,29,35–37,57].

Aspects such as muscle coactivation and activation rhythm were also explored [56], as well as motor unit firing patterns [14,42,50], the variability of the interspike interval, and the dispersion of the innervation zone [44], as well as reduction in recruitment and overlapping of action potentials [13,42].

In addition, synchrony-related measures derived from sEMG were identified, particularly those based on coherence analysis. Intermuscular coherence (IMC) was used to quantify the degree of common oscillatory drive and synchronization between muscle pairs, especially in the beta-band frequency range, providing indirect information about corticospinal and upper motor neuron integrity. Studies reported reduced coherence values in individuals with ALS compared to healthy controls, reflecting impaired neural coupling and altered motor control strategies [52].

Changes in the morphology of sEMG signals aiming at automated diagnosis were also analyzed [53], as well as the application of the cutaneous silent period (CutSP) as an indirect measure of upper motor neuron dysfunction [54]. Finally, characteristics of upper and lower motor neurons were explored based on muscle activation patterns [52,55].

Characteristics of the participants

RQ2: What participant characteristics were reported?

The data describing the characteristics of the participants is presented in the S2 Table – Participant Characteristics. The average age of ALS patients varied between the studies, ranging from approximately 58–65 years. Some reported averages with standard deviation, such as 59,3 ± 11,3 years [38], 58 ± 12 years [36], 60,1 ± 11,2 years [40], 62,3 ± 8,5 years [43], 64,9 ± 9,6 years [58] and 65,4 ± 10,3 years [14]; while others showed simple average values or ranges, such as 60,5 years (55,2–65,4) [49] and 59,5 years [54]. Additional studies have indicated averages of around 64 years [6,33,35].

The average duration since the onset of symptoms was also variable, with reports ranging from around 9–28 months. Some studies have indicated values such as 13,4 ± 5,8 months [38], 8,9 ± 7,8 months [40], 17,9 ± 12,3 months [58], 19 ± 11 months [43], 23 months [49], about 20,25 months [36], and median 28 months (IQR: 17–49) [51]. The average time to diagnosis was 11,4 ± 4,4 months [14] and 13,2 months (IQR: 7,6–19,8) [49]. Many studies, however, did not provide this information.

Various instruments were used for clinical assessment. The ALS Functional Scale-Revised (ALSFRS-R) has been widely used [6,14,35,37–40,47,49,51,54,57,58]. The MRC muscle strength scale was also widely used [6,35,39,43,46,51,54].

Other methods included the Forced Vital Capacity (FVC) [49] and the Ashworth scale for spasticity [54]. Some studies have proposed specific classifications of participants, such as the distinction between predominantly upper or lower motor neuron involvement [55], and categorization according to the type of onset of ALS [34].

Targeted muscles

RQ3: Which muscle groups were evaluated using sEMG? Did the evaluation include muscles of the upper limbs, lower limbs, trunk, cervical region, or head and neck?

Information on all muscles assessed is provided in S3 Table – Muscles or muscle groups evaluated in the included studies and Fig 3.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 3. Distribution of the muscles analyzed.

Abbreviations: ADM: Abductor Digiti Minimi; AH: Abductor Hallucis; APB: Abductor Pollicis Brevis; BB: Biceps Brachii; EDB: Extensor Digitorum Brevis; EDC: Extensor Digitorum Communis; FDI: First Dorsal Interosseous; FPB: Flexor Pollicis Brevis; GM: Medial Gastrocnemius; OP: Opponens Pollicis; RF: Rectus Femoris; TA: Tibialis Anterior; TR: Triceps Brachii; VL: Vastus Lateralis.

https://doi.org/10.1371/journal.pone.0350029.g003

In the upper limbs, the most commonly evaluated muscle was the abductor pollicis brevis (APB), frequently analyzed across studies using different assessment techniques and clinical parameters [6,9,29,36,37,39–41,43,47,54]. The abductor digiti minimi was also widely studied, mainly in investigations focused on motor units and signal variability [36,37,39,40,43]. The first dorsal interosseous was included in analyses of signal morphology, pattern classification, and machine learning approaches [9,13,40,42,52,53]. Studies evaluating the extensor digitorum focused on muscle recruitment and activation [36,37,47,52,53]. The biceps brachii was investigated in terms of neuronal excitability, functional loss, and fasciculations [9,33,37,44,47,50,54,56], while the triceps brachii appeared less frequently, mainly in studies of coactivation and fatigue [56]. Muscles from the thenar and hypothenar groups were commonly used for MUNE/MUNIX and excitability characterization [9,29,30,35,41], and the lumbricals were assessed in longitudinal investigations [6].

In the lower limbs, the tibialis anterior (TA) was the most frequently examined muscle, especially in studies assessing strength, fatigue, and recruitment [31,32,37,43,54–56]. The medial gastrocnemius was analyzed primarily in studies focusing on fasciculations and action potential morphology [34,46,47,51]. The vastus lateralis appeared in analyses of firing patterns and motor units [14,33,57]. The abductor hallucis (AH) was included in studies estimating motor unit loss [37], while the extensor digitorum brevis (EDB) was examined in investigations of measurement reproducibility [36,37,47]. The rectus femoris (RF) was used in studies exploring coactivation and fatigue [56].

Although limb muscles predominated, a small number of studies evaluated trunk and cervical muscles [41,52]. These regions present additional challenges for sEMG analysis due to multilayered and deeper musculature, greater influence of respiratory and postural adjustments on the signal, and, in some cases, susceptibility to ECG artifacts, particularly when electrodes are placed near the thoracic cavity. These factors can reduce signal-to-noise ratio and increase variability compared to limb recordings. No studies were identified that directly assessed muscles of the head or neck.

Evaluation

RQ4: How was the evaluation performed using sEMG? Which activities were employed, what was the duration of data collection, and what rest periods were adopted?

During the collection of sEMG signals in the studies analyzed, we observed a wide variety of procedures involving different muscle activation and rest strategies, depending on the methodological objectives. This information is present in S4 Table – Data collection activities and rest/collection time in the included studies.

Several studies have collected data with participants at absolute rest, without any kind of voluntary contraction. This approach aimed to record fasciculations and spontaneous activity, and was used by Kleine et al., 2008 [33], Bashford et al., 2019 [46], Zhang et al., 2014 [45], Zhou et al., (2012, 2011) [9,41] Wannop et al., 2021 [49], Bashford et al., 2020a [47], Bashford et al., 2020b [48], Planinc et al., 2023 and [51], with recording sessions varying between 3 and 30 minutes per muscle. In these protocols, rest intervals were generally not specified because recordings were continuous.

Other studies have chosen to collect data during voluntary isometric contractions at different intensities. In general, the participants were instructed to perform contractions at minimum, intermediate and maximum levels of force, with visual or biofeedback control, according to Boekestein’s studies et al., 2012 [35], Ahn et al., 2010 [39], Kim et al., 2016 [40], Escorcio et al., 2016 [43], Neuwirth et al., 2010 [37], Neuwirth et al., 2017 [38], Alarcon et al., 2022 [56], Weddell et al., 2021 [50], Quintão et al., 2021 [52], Sanjak et al., 2004 [32], Chen et al., 2018 [42], Zhang et al., 2013 [13], Nishikawa et al., 2022 [14] and Noto et al., 2023 [57]. The protocols included progressive activation ramps, sustained contractions for up to 60 seconds, and rest periods typically ranging from 10 to 60 seconds when reported; however, several studies did not specify the exact rest duration, describing them only as “adequate” to prevent fatigue.

Some studies applied specific functional tasks during collection. For example, Antunes et al., 2023 [53] asked for coordinated movements of isometric lifting of the index finger against the other fingers, alternating with relaxation periods, although the exact duration of these rest periods was not reported. Quintão et al., 2021 [52] used cycles of contraction and rest of the extensor digitorum muscle, with approximately 10 seconds of rest between cycles. Saidane et al., 2021 [55] recorded electromyographic activity while the participants walked barefoot for five meters, a protocol that did not involve structured rest intervals.

Median or ulnar nerve stimulation techniques were applied to obtain compound potentials (CMAP) and estimates of the number of motor units (MUNE or MUNIX), as in Felice et al., 1995 [29], Baumann et al., 2012 [36], Nandedkar et al., 2022 [58], and Boekestein et al., 2012 [35]. The protocols included variations in stimulus intensity, number of repetitions, and electrode location, ensuring reproducible parameters that were sensitive to the progression of the disease. Rest periods between stimulations were generally brief (a few seconds), but most studies did not quantify them explicitly.

Finally, studies focusing on fatigue or prolonged recordings have adopted more extensive protocols. Kent et al., 2000 [31] performed intermittent isometric dorsiflexion exercises for 25 minutes, with contraction and relaxation cycles, although the exact duration of the relaxation phases was not reported. Alarcon et al., 2022 [56] evaluated muscle response during bilateral maximal contractions until fatigue, with individually determined rest intervals that were not numerically defined. Jahanmiri et al., 2015 [44] collected data at rest and during light contraction, with sessions between 10 and 30 minutes. Zhang et al., 2013 [13] and Chen et al., 2018 [42] performed isometric contractions with a minimum duration of 2–5 seconds and rest periods of 10–15 seconds between attempts.

sEMG properties

RQ5: What equipment and parameters were used for sEMG acquisition (e.g., electrode type, sampling frequency, portability, and data transmission method)?

The data describing the technical properties of the sEMG signals are presented in the S5 Table - sEMG device characteristics.

About the type of electrode, most studies have used surface electrodes, including self-adhesive, pre-frozen, and high-density models. Self-adhesive cut Electrocardiogram (ECG) electrodes, pre-frozen 20 mm diameter electrodes, disposable electrodes with diameters between 10 and 15 mm, 10×30 mm models, and 22×24 mm Ambu Blue Sensor Tabs were used. [29,36–38,40,43]. Disk electrodes have also been widely applied [39,58].

Several studies have used high-density electrodes (HD-sEMG), such as 8×15 grids with 4 mm spacing and 10×13 matrices with a 5 mm distance between electrodes [6,33–35]. Grids of 64 circular electrodes (8×8), each with a diameter of 4.5 mm, have also been widely used [46–48,50,51].

Few systems have been described as portable. Highlights include the Biosignalsplux, with 8 channels and a built-in filter, and the BTS FreeEMG, with Bluetooth transmission [53,56]. Most used laboratory equipment, such as Biosemi, TMS International BV, OT Bioelettronica, Natus Medical, Synergy, and Keypoint Classic.

As for the method of data transmission, several systems stored the signals locally for offline analysis [33,34,39,58]. In studies with HD-sEMG, it was common to use a wired connection with Refa64 or Refa128 amplifiers [9,41,42,46,51]. Other studies have reported wireless systems via Bluetooth or with on-board acquisition [53,56].

The sampling frequency varied widely. In general, HD-sEMG systems operated at 2048 Hz [6,34,35,46–48,51,57], while frequencies of 2000 Hz were also common [9,13,33,41,42,44]. Intermediate frequencies, such as 1000 Hz, have been reported [52,53,56], and lower values, such as 500 Hz, have also been used [32]. One study highlighted the application of 3000 Hz [54]. Some studies did not specify this information [29,30,36,37].

sEMG signal analysis

RQ6: What processing techniques were applied to the sEMG signal? Which software and analysis algorithms were used?

In S6 Table – Signal processing features in surface sEMG studies, the methods for processing and analyzing the sEMG signals are presented. Analog and digital filters between 10 and 450 Hz, combined with wavelet transforms, fractal analysis, and temporal and spectral feature extraction, were applied to classify signals from ALS patients and healthy controls, using an AdaBoost classifier [53]. A similar strategy was adopted for the extraction of metrics such as multiscale entropy (MSE), Detrended Fluctuation Analysis (DFA), and Lempel-Ziv complexity, feeding classifiers such as Random Forest, AdaBoost, Linear Discriminant Analysis (LDA), and K-Nearest Neighbors (KNN). [52].

The complexity of firing patterns was analyzed using multiscale entropy (MSE) and power spectral density (PSD), focusing on the linearity, Gaussianity, and sparsity of the signals [45]. In addition, complexity quantification was explored using Approximate Entropy (ApEn) and Fast Fourier Transform (FFT) to discriminate spontaneous firing patterns [41].

Time-frequency approaches have also been applied, such as the discrete wavelet transform, the Short Time Fourier Transform (STFT), and cepstrum analysis, combining these techniques with machine learning to classify healthy people and those with ALS, differentiating between upper and lower motor neuron involvement [55]. Extracted characteristics included mean amplitude, RMS, Willison Amplitude (WAMP), sample entropy, and median frequency.

Other studies have focused on the decomposition of high-density signals. The Automatic Progressive FastICA Peel-Off (APFP) algorithm was developed for blind source separation, with filters between 10–500 Hz and metrics such as Matching Rate and False Discovery Rate [42]. A peak detection routine was also used, exploring high-density signals and studying muscle fatigue after isometric contractions [56].

The detection and analysis of fasciculations was addressed with the SPiQE system, which uses 20–500 Hz band-pass filters, automatic exclusion of noisy channels, and a linear model for correlating fasciculation amplitude and depth [34,46,51].

Integration with other technologies

RQ7: Was sEMG used in conjunction with other data collection technologies, such as EEG, inertial sensors, or neuroimaging?

Our finding reveal that the integration of sEMG with other data collection technologies is still infrequent in research into ALS. This information can be found in the Table 3.

Download:

• PNG
  larger image
• TIFF
  original image

Table 3. Studies integrating sEMG with other technologies.

https://doi.org/10.1371/journal.pone.0350029.t003

sEMG was used in conjunction with magnetic resonance spectroscopy (MRS) to assess central fatigue during isometric exercise, demonstrating an integrated approach between peripheral muscle activity and cerebral metabolic parameters [31]. They have also been associated with muscle ultrasound (MUS) in the investigation of the electromechanical latency of fasciculations [51]. Another example includes the combination with transcranial magnetic stimulation (TMS) to assess upper motor neuron dysfunction [54].

In addition, mechanical support technologies have also been integrated. A force transducer was used to explore the dissociation between mechanical and electromyographic manifestations of fatigue [32], while customized dynamometers made it possible to measure muscle force accurately during signal acquisition [50,57].

Use of biofeedbacks

RQ8: Was sEMG used as a biofeedback tool at any stage of the study? If so, at what point in the evaluation or follow-up was it applied?

The use of biofeedback strategies has been reported as a resource to guarantee signal quality and control contraction force during data capture. Visual and auditory monitoring were used for this purpose, enabling real-time adjustments to muscle contraction [37,39,43]. Providing direct visual feedback of the sEMG signal during collection has also been reported [34].

Other forms of biofeedback include the use of sound signals to guide motor activity [53], combinations of visual and verbal instructions to standardize contractions [50], and real-time visual feedback related to contraction force [57]. Providing visual feedback of the force exerted during isometric tasks showed an association between sEMG and functional performance [31], while the use of auditory feedback was also described during the tests [54]. It is important to emphasize that none of these biofeedback procedures constitute therapeutic or experimental intervention. They were employed exclusively to standardize motor performance and ensure the reliable quality of the sEMG. This data is described in the Fig 4.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 4. Description of the feedback methods used in the studies.

https://doi.org/10.1371/journal.pone.0350029.g004

Characteristics of the intervention

RQ9: Which intervention or rehabilitation strategies were associated with sEMG analysis? Have any studies used sEMG as a support tool for motor rehabilitation?

The data related to the interventions can be found in the Table 4. Electrical stimulation was used in different protocols involving the collection of sEMG for evaluation purposes. Stimuli on the median nerve and peripheral nerves were applied as part of protocols for estimating motor units, using specific equipment such as the Dantec Model 13L36 and the Nicolet Viking IV [29,36]. Supramaximal stimulation of the peroneal nerve was also performed before and after isometric exercises to investigate central fatigue, with no therapeutic purpose [31]. In addition, cutaneous stimuli were used to induce the cutaneous silent period (CutSP), with a view to functional evaluation [54].

Download:

• PNG
  larger image
• TIFF
  original image

Table 4. Summary of studies with interventions.

https://doi.org/10.1371/journal.pone.0350029.t004

None of the studies analyzed reported the application of sEMG as an active tool in rehabilitation protocols or as a control interface for assistive technologies. The interventions described focused predominantly on neurophysiological analysis.

Clinical results and processing

RQ10: What outcomes were observed in the analyzed motor parameters? Which sEMG processing methods demonstrated the best performance and applicability?

The main clinical motor parameters assessed using sEMG in individuals with ALS were muscle fatigue, the frequency and pattern of fasciculations, the estimated number of motor units (MUNE/MUNIX), and the firing rate of motor units, data presented in the S7 Table – Summary of clinical data and sEMG signal processing outcomes in ALS studies. These parameters have proved effective in differentiating clinical subtypes, characterizing disease progression, and predicting survival, offering objective support for functional assessment and longitudinal monitoring.

Muscle fatigue is predominantly central in ALS, with progressive failure of central activation observed during sustained isometric contractions, correlated with a smaller reduction in phosphocreatine levels [31]. The dissociation between the reduction in muscle strength and the discrete changes in myoelectric signals, observed through the median frequency, also reinforces the central origin of fatigue [32]. Recent studies, such as that by Alarcón et al., 2022 [56], have identified abnormal muscle activation patterns during maximum isometric contraction, highlighting the usefulness of dynamic sEMG in functional assessment.

The frequency and pattern of fasciculations are important biomarkers in ALS. Klein et al., 2008 [33] identified two distinct electromyographic patterns of fasciculations, while the SPiQE system, developed by Bashford et al., 2019 [46], allows automated detection with high accuracy. Subsequent studies by Bashford et al., 2020 [48] have shown that the frequency of fasciculations varies over time, with an initial increase and decline as the disease progresses. Planinc et al., 2023 [51] investigated the depth and latency of fasciculations, and Wannop et al., 2021 [49] introduced the Rate of Change of Fasciculation Frequency (RoCoFF) parameter, which proved promising as a prognostic indicator.

Estimating the number of motor units was explored using the MUNE and MUNIX techniques. Felice et al., 1995 [29] demonstrated the reduction of MUNE in ALS, with greater amplitude of unitary potentials due to compensatory reinnervation. Baumann et al., 2012 [36] applied a Bayesian model to describe the loss of motor units over time, varying according to the clinical subtype. Neuwirth et al., 2017 [38] observed decline rates of up to 5.6% per month, while Boekestein et al., 2012 [35] and Ahn et al., 2010 [39] confirmed the high reproducibility and reliability of MUNIX.

Analysis of the complexity of sEMG signals has proved useful in detecting early alterations. Zhou et al., 2011 [41] characterized spontaneous patterns based on approximate entropy (ApEn), showing greater regularity in the discharge of motor units in ALS patients. Zhang et al., 2014 [45] observed that multiscale entropy (MSE) is sensitive to changes specific to ALS, identifying less complexity at smaller scales. Saidane et al., 2021 [55] proposed an effective classification system between different types of ALS and controls, achieving up to 99.8% accuracy.

Advanced processing techniques, such as the APFP (Automatic Progressive FastICA Peel-Off) algorithm proposed by Chen et al., 2018 [42], and the use of morphological attributes extracted by Antunes et al., 2023 [53], have shown great accuracy in decomposing sEMG signals and classifying ALS patients. These approaches, combined with machine learning algorithms such as Support Vector Machine (SVM), KNN, and AdaBoost, have shown high performance in patient stratification and early assessment of neuromuscular degeneration.