The Effect of Cut-Off Frequency on Signal Features When Filtering Equine sEMG Signal from Selected Extensor Muscles

Abstract

The use of surface electromyography (sEMG) in equine locomotion research has increased significantly due to the essential role of balanced, symmetrical, and efficient movement in riding. However, variations in sEMG signal processing for forelimb extensor muscles across studies have made cross-study comparisons challenging. This study aims to compare the sEMG signal characteristics from carpal extensor muscles under different filtering methods: raw signal, low-pass filtering (10 Hz cut-off), and bandpass filtering (40–450 Hz cut-off and 7–200 Hz cut-off). sEMG signals were collected from four muscles of three horses during walking and trotting. The raw signals were normalized and filtered separately using a 4th-order Butterworth filter: low-pass 10 Hz, bandpass 40–450 Hz, or bandpass 7–200 Hz. For each filtered signal variant, eight activity bursts were annotated, and amplitude, root mean square (RMS), median frequency (MF), and signal-to-noise ratio (SNR) were extracted. Signal loss and residual signal were calculated to assess noise reduction and data retention. For m. extensor digitorum lateralis and m. extensor carpi ulnaris, bandpass filtering at 40–450 Hz resulted in the lowest signal loss and the highest amplitude, RMS, MF, and SNR after filtering. However, variations were observed for the other two carpal extensors. These findings support the hypotheses that the characteristics of myoelectric activity in equine carpal extensors vary depending on the filtering method applied and differ among individual muscles, thereby guiding future research on sEMG signal processing and, consequently, equine biomechanics. Since both noise and its reduction alter raw sEMG signals, potentially affecting data analysis, this study provides valuable insights for improving the reliability and reproducibility of equine biomechanics research across different sEMG studies.

1. Introduction

Surface electromyography (sEMG) is a non-invasive technique used to assess muscle activity in both humans and animals. Among animal species, horses are of particular interest due to the essential role that balanced, symmetrical, and efficient locomotion plays in racing, sports, and pleasure riding [1].

However, the comparability and reproducibility of equine sEMG research are limited due to the lack of standardized guidelines, such as those established for human sEMG studies [2,3]. These human guidelines cover aspects such as muscle fiber direction, surface electrode placement, and sEMG sensor characteristics, as well as signal processing and analysis. In equine medicine, efforts have been made to standardize horse skin preparation and habituation to sEMG measurements [2], surface electrode placement along specific muscles [1,4], and signal processing from large muscles [5,6].

To date, equine studies focusing exclusively on sEMG signal processing have examined signal filtering and normalization from the m. biceps femoris during canter [5], as well as the effects of high-pass filter cut-off frequencies on signals recorded from the m. biceps femoris and m. triceps brachii during trot and canter [6]. However, these studies did not consider the smaller, distally located extensor muscles, which play a crucial role in the forelimb motion [7]. Among these muscles, four major extensors—the m. extensor carpi radialis, m. extensor digitorum communis, m. extensor digitorum lateralis, and m. extensor carpi ulnaris—are responsible for extending and stabilizing the carpal joints and/or extending the digit [7,8,9]. Of these extensor muscles, the m. extensor carpi radialis [4,10,11,12], m. extensor digitorum communis [4,10,13,14,15], and m. extensor digitorum lateralis [10] have been investigated using sEMG, while the myoelectric activity of the m. extensor carpi ulnaris remains unexplored.

To date, the myoelectric activity of the m. extensor carpi radialis was measured during walk and trot and filtered using low-pass filtering with a 10 Hz cut-off frequency to assess the effects of Kinesio Taping on forelimb motion [11]. During trotting, it was analyzed using bandpass filtering with a 40–450 Hz cut-off frequency to investigate the effects of clenbuterol on muscle myoelectric activity [12] and to develop recommendations for electrode placement to measure muscle myoelectric activity [4]. In research focused on standardizing electrode placement, the m. extensor digitorum communis was also investigated [4]. Among carpal extensors, three muscles—the m. extensor carpi radialis, m. extensor digitorum communis, and m. extensor digitorum lateralis [10], or the m. extensor digitorum communis alone [13,14,15]—were studied in the context of the nociceptive withdrawal reflex evoked by transcutaneous electrical stimulation. In these specific sEMG studies, signals were filtered using bandpass filtering with a 7–200 Hz cut-off frequency [10,13,14,15].

However, the comparability of previous equine sEMG studies is limited due to the different filtering methods employed. Even when assessing the myoelectric activity of a group of muscles with similar functions, such as carpal extensors, the signal filtering methods used in published sEMG studies vary considerably. These included low-pass filtering with a 10 Hz cut-off frequency [11], band-pass filtering with a 40–450 Hz cut-off frequency [4,12], and band-pass filtering with a 7–200 Hz cut-off frequency [10,13,14,15]. It is important to note that filtering with different cut-off frequencies reduces noise and artifact contamination in various ways while also influencing the preservation of the myoelectric signal [5,16]. High-pass filtering is used to remove low-frequency noise and movement artifact contamination, preserving signal components above the chosen cut-off frequency. Conversely, low-pass filtering is applied to eliminate high-frequency contamination, retaining signal components below the cut-off frequency [16]. As a result, the frequency ranges analyzed in previous studies on carpal extensors [4,10,11,12,13,14,15] differed, potentially altering the characteristics of the signals, and thus affecting the final outcomes of the equine biomechanics studies.

Considering the impact of filtering on human sEMG signals from small abaxial muscle [16] and on equine sEMG signals from large muscles [5,6], we hypothesize that the characteristics of the myoelectric activity of equine carpal extensors vary depending on the filtering method applied. Furthermore, due to the differences in noise and movement artifact contamination among various muscles, we also hypothesize that filtering-dependent characteristics of the myoelectric activity differed between individual equine carpal extensor muscles. If the results of this preliminary observational study confirm the proposed hypotheses, further systematic investigations into sEMG signal-filtering methods for different equine muscles will be justified. Conversely, if the hypotheses are rejected, it will be reasonable to compare previous studies on the myoelectric activity of carpal extensors without limitations, and future equine sEMG studies may be conducted without specific signal processing restrictions.

Since both noise and its reduction introduce modifications to the raw sEMG signal [16], potentially affecting data analysis [5,6], this observational study demonstrates how the same sEMG signals from the equine carpal extensors are altered by processing with previously published filtering methods [4,10,11,12,13,14,15]. To address the current gap in knowledge regarding equine sEMG signal processing, this preliminary proof-of-concept study aims to compare the sEMG signal features and signal-loss metrics across four equine carpal extensor muscles using different, previously applied in the literature, filtering approaches: raw (unfiltered) signal, low-pass filtering with a 10 Hz cut-off frequency, bandpass filtering with a 40–450 Hz cut-off frequency, and bandpass filtering with a 7–200 Hz cut-off frequency.

2. Materials and Methods

2.1. Signal Collection

In this preliminary study, a sample of sEMG signals was collected from three warmblood horses (the Polish half-bred; geldings; range age: 6–12 years; range body mass 480–560 kg; female). The horses were selected from the available pool of didactic horses at the Didactic Stable of the Horse Breeding Division at the Warsaw University of Life Sciences (WULS) based on their similarity in age, body weight, height at the withers, and body conformation, as well as being free from lameness and any history of neurological disorders. Horses were deemed sound by an experienced veterinarian (B.T.) and had no known history of neurological disorders. Horses were fed individually calculated rations of hay and concentrate, with feed distributed across three meals per day. A mineral salt block and fresh water were available at all times. Horses were regularly exercised and housed in boxes with daily turnout. Procedures were approved by the II Local Ethical Committee on Animal Testing in Warsaw on behalf of the National Ethical Committees on Animal Testing (protocol code WAW2/089/2020 approved on 29 July 2020).

On the day of signal collection, horses were clipped (0.8 mm hair length) and cleaned (ethanol 80%) on the proximal aspect of the right forelimb from the carpal joint area to the elbow joint area (Figure 1A). In this area of the forelimb, four extensor muscles were identified by palpation: the m. extensor carpi radialis, m. extensor digitorum communis, m. extensor digitorum lateralis, and m. extensor carpi ulnaris. For each muscle, specific anatomical landmarks were identified by palpation (performed by an experienced equine surgeon and certified specialist in equine diseases, B.T.) and confirmed ultrosonographically (performed by an equine surgery resident who had completed training in equine ultrasound, E.S.). Ultrasound examination was performed using a SonoSite Edge II ultrasound machine (Fujifilm, Valhalla, NY, USA) with a 6–13 MHz linear probe (Fujifilm, Valhalla, NY, USA). For the m. extensor carpi radialis, the proximal landmark was the lacertus fibrosus, the distal landmark was the medial styloid process of the radius [4], and two electrodes were placed at 30–40% of the proximal section of the distance between landmarks. For the m. extensor digitorum communis, the proximal landmark was the lateral tuberosity of the radius, the distal landmark was the lateral styloid process of the radius, and two electrodes were placed at 30–40% of the proximal section of the distance between landmarks [4]. For the m. extensor digitorum lateralis, the proximal landmark was the lateral epicondyle of the humerus, the distal landmark was the ulnar carpal bone [7], and two electrodes were placed at 20–30% of the proximal section of the distance between landmarks. For the m. extensor carpi ulnaris, the proximal landmark was the lateral epicondyle of the humerus, the distal landmark was the accessory carpal bone [7], and two electrodes were placed at 20–30% of the distal section of the distance between landmarks.

In each location, two surface electrodes (pre-gelled Ag/AgCl; 4 cm diameter electrode; 4 cm inter-electrode distance; 270Bx, Noraxon, Scottsdale, AZ, USA) were placed next to each other, parallel to the direction of the muscle fibers (Figure 1B). Tiny drops of conductive gel (Indiba, Las Rozas, Spain) were applied between the electrodes and the skin. Two electrodes were connected to one wireless sEMG sensor (DTS, Noraxon, Scottsdale, AZ, USA), composing bipolar derivations (Figure 1C). The surface electrodes and sEMG sensors were fixed on the skin by a tiny layer of Rudaspray medical glue (Nobamed, Bielefeld, Germany). The sEMG sensors were additionally stabilized using Pino Tape Sport kinesiology tape (Pino, Ahrensburg, Germany). One piece of kinesiology tape was applied horizontally, perpendicular to the long axis of the limb and the direction of the studied muscle fibers, encircling the entire limb at the level of the sEMG sensors and adhering to the skin between the sensors.

The sEMG signal was recorded using the Ultime System electromyograph (Noraxon, Scottsdale, AZ, USA). Before the measurement was started, electrode-skin impedance levels were checked using a built-in function of the MR software version 3.18.98 (Noraxon, Scottsdale, AZ, USA) to ensure proper electrode adhesion. The sEMG signal was recorded during the tested exercise during a four-beat walk and a two-beat trot. The transitions between gaits were assessed visually based on the characteristic movement patterns. The tested exercise was performed in a straight line and lasted 10 s. The signal was sampled at a frequency of 2000 Hz. The tested exercise was repeated six times. During each measurement, each horse took at least eight strides. Data were stored using MR software version 3.18.98 (Noraxon, Scottsdale, AZ, USA). The measurement free of artifacts and loss of electrode contact was selected for further analysis.

2.2. Signal Processing

Data stored in MR software version 3.18.98 (Noraxon, Scottsdale, AZ, USA) were exported as CSV files. CSV files were imported to MATLAB software version R2024b (MathWorks, Natick, MA, USA) as a raw sEMG signal and processed in the scripts, including normalization by DC-offset removal and notch filtering by removing 50 Hz frequency. The signals were filtered with a 4th order Butterworth filter with a low-pass 10 Hz cut-off frequency [11], a bandpass 40–450 Hz cut-off frequency [4,12], and a bandpass 7–200 Hz cut-off frequency [10,13,14,15], separately. Following Zellner et al. [11], the low-pass filtered signal was rectified. As a result, four variations of the same sEMG signals, marked as raw, low-pass 10 Hz, bandpass 40–450 Hz, and 7–200 Hz, were obtained and analyzed using the same protocol.

For each signal variation, eight segments of burst (activity burst) and eight segments of noise (no activity burst) were manually selected by an experienced evaluator (M.D.) over eight strides using the Signal Analyzer application in MATLAB [17]. Each segment was marked between two time-cursors and extracted as a separate region of interest (ROI). For each signal, eight burst ROIs and eight noise ROIs were saved and exported to the MATLAB workspace.

2.3. Signal Features Extraction

For each burst ROI, the signal features were calculated in MATLAB [17] considering amplitude [11], root mean square (RMS) [15], median frequency (MF) [12], and signal-to-noise ratio (SNR) [4].

Amplitude represents the maximum absolute value of the signal expressed in [mV]. RMS represents the mean square of signal values used to determine the magnitude of a signal, varying and expressed in [mV]. RMS is calculated using Formula (1):

$$RMS=\sqrt{{\displaystyle \frac{1}{N}}{\sum }_{n=1}^N{\left(x\left(n\right)\right)}^2}$$

(1)

MF represents a frequency value at which the sEMG power spectrum is divided into two regions with equal integrated power. MF is expressed in Hz and was calculated using a Formula (2):

$$MF={\sum }_{j=1}^{MF}{P}_j={\sum }_{j=MF}^M{P}_j={\displaystyle \frac{1}{2}}{\sum }_{j=1}^M{P}_j$$

(2)

where $P_j={\displaystyle \frac{1}{N^2}}{\left|X_k\right|}^2$ is the signal power spectrum at the frequency bin j and $M$ is the length of the frequency bin, defined as the next power of 2 from the length of the signal in the time domain.

SNR, measured in decibels [dB], determines the ratio of the signal (segments of burst named signal ROIs) to the background signal (segments of noise named noise ROIs). SNR is approximated using Formula (3):

$${SNR}_i=10·\log 10{\displaystyle \frac{{RMS\left(signal {ROIs}_i\right)}^2}{{RMS\left(noise {ROIs}_i\right)}^2}},$$

(3)

where $i$ is the $i^{th}$ extracted segment of signal.

2.4. Signal Loss and Residual Signal Calculation

For each filtered signal, mean signal loss (%) and residual signal (%) were calculated using the RMS following De Luca et al. [16] and St. George et al. [5], respectively. Signal loss was determined using the Formula (4), while residual signal (%) was calculated using the Formula (5):

$$Signal Loss \left(\%\right)={\displaystyle \frac{\left({RMS}_{full}-{RMS}_{filter}\right)}{{RMS}_{full}}}$$

(4)

$$Residual Signal \left(\%\right)={\displaystyle \frac{{RMS}_{filter}}{{RMS}_{full}}}$$

(5)

where RMS_full is the mean RMS value of the raw signal and RMS_filter is the mean RMS value of the filtered signal (low-pass 10 Hz, bandpass 40–450 Hz, and bandpass 7–200 Hz).

2.5. Statistical Analysis

Statistical analysis was performed using Prism version 6 (GraphPad Software Inc., San Diego, CA, USA).

The dataset was structured as a series of data points representing each sEMG signal feature, with eight realizations representing segments of a burst for each of the three horses, resulting in 24 realizations per feature. The Shapiro–Wilk normality test was used to assess the distribution of each data series. Since not all signal feature data series followed a normal distribution, numerical data were presented as medians and ranges (lower and upper quartiles). In contrast, all data series for signal loss and residual signal were normally distributed. Therefore, their numerical data were presented as means ± SD.

Data series were compared between sEMG signal variation as paired data using a Repeated Measures ANOVA summary, when all compared data series were normally distributed, or a Friedman test, when at least one data series was not normally distributed. If significant differences were found, the post-hoc Tukey’s multiple comparisons test was performed after a Repeated Measures ANOVA summary, or a Holm–Sidak’s multiple comparisons test was performed after the Friedman test. Statistical significance was set at p < 0.05.

3. Results

3.1. Signal from M. Extensor Carpi Radialis

For the m. extensor carpi radialis, sEMG signal features varied across signal variations in both gaits, as shown in

Table 1. Median and ranges (lower quartile; upper quartile) signal feature data (amplitude, root mean square (RMS), median frequency (MF), signal-to-noise ratio (SNR)) compared between signal variants (low-pass filtered with a 10 Hz cut-off frequency, bandpass filtered with a 40–450 Hz cut-off frequency, bandpass filtered with a 7–200 Hz cut-off frequency) for signals obtained from the m. extensor carpi radialis during walking and trotting.

Gait	Signal Variation	Amplitude [mV]	RMS [mV]	MF [Hz]	SNR [dB]
Walk	Raw	95.3 (76.3; 109.1) a	21.6 (19.6; 22.7) a	42.9 (39.2; 47.4) a	11.6 (10.3; 12.9) a
	Low-pass 10 Hz	29.2 (26.0; 32.3) b	17.0 (15.6; 18.3) b	3.1 (3.0; 3.2) b	11.0 (10.2; 13.0) a
	Bandpass 40–450 Hz	78.9 (65.9; 88.4) c	14.9 (13.2; 16.5) c	72.5 (69.9; 78.5) c	13.2 (11.5; 14.1) ab
	Bandpass 7–200 Hz	88.4 (72.0; 99.9) ac	20.6 (18.9; 21.9) d	43.9 (40.6; 47.9) a	13.4 (11.8; 14.2) b
	p–value	<0.0001	<0.0001	<0.0001	<0.0001
Trot	Raw	426.8 (381.5; 487.4) a	105.6 (97.4; 110.8) a	51.0 (46.4; 57.8) a	13.6 (12.5; 14.7) a
	Low-pass 10 Hz	152.9 (136.1; 171.1) b	83.9 (78.3; 90.8) b	4.2 (3.9; 4.4) b	12.6 (11.3; 14.2) b
	Bandpass 40–450 Hz	366.5 (334.2; 454.7) c	78.8 (67.0; 82.8) c	82.3 (74.4; 88.2) c	16.6 (15.5; 17.3) c
	Bandpass 7–200 Hz	399.3 (346.6; 444.9) c	101.7 (93.1; 107.1) d	47.0 (43.3; 56.4) d	14.9 (13.9; 16.6) d
	p-value	<0.0001	<0.0001	<0.0001	<0.0001

Lowercase letters (a–d) indicate differences in signal features across signal variations. Statistical significance was set at p < 0.05.

. During walking, amplitude was lower than in the raw signal after bandpass filtering at 40–450 Hz and lowest after low-pass filtering at 10 Hz; while during trotting, amplitude was lower than in the raw signal after bandpass filtering at 40–450 Hz and 7–200 Hz, and lowest after low-pass filtering at 10 Hz, as clearly shown in Figure 2. During both gaits, RMS was lower than in the raw signal after bandpass filtering at 7–200 Hz, even lower after low-pass filtering at 10 Hz, and lowest after bandpass filtering at 40–450 Hz. During walking, MF was lower than in the raw signal after low-pass filtering at 10 Hz and higher after bandpass filtering at 40–450 Hz. During trotting, MF was lowest after low-pass filtering at 10 Hz, moderate after bandpass filtering at 7–200 Hz, higher in the raw signal, and highest after bandpass filtering at 40–450 Hz. During walk, SNR was higher than in the raw signal only after bandpass filtering at 40–450 Hz, while during trotting, SNR was higher than in the raw signal after bandpass filtering at 7–200 Hz and highest after bandpass filtering at 40–450 Hz, as well as lower after low-pass filtering at 10 Hz.

For the m. extensor carpi radialis, mean signal loss decreased, and mean residual signal increased across bandpass filtering at 40–450 Hz, low-pass filtering at 10 Hz, and bandpass filtering at 7–200 Hz in both gaits, as illustrated in Figure 2A,E. During both gaits, signal loss was highest after bandpass filtering at 40–450 Hz, lower after low-pass filtering at 10 Hz, and lowest after bandpass filtering at 7–200 Hz, as shown in

Table 2. Mean signal loss (%) and residual signal (%) data compared between signal variants (low-pass filtered with a 10 Hz cut-off frequency, bandpass filtered with a 40–450 Hz cut-off frequency, and bandpass filtered with a 7–200 Hz cut-off frequency) for signals obtained from the m. extensor carpi radialis during walking and trotting.

Gait	Signal Variation	Signal Loss (Residual Signal) per Horse			Common Signal Loss (Residual Signal)
		Horse 1	Horse 2	Horse 3
Walk	Low-pass 10 Hz	19.9% (80.1%)	20.4% (79.6%)	20.7% (79.3%)	20.3% (79.9%) a
Walk	Bandpass 40–450 Hz	30.2% (69.8%)	30.9% (69.1%)	31.4% (68,6%)	30.9% (69.1%) b
Walk	Bandpass 7–200 Hz	4.3% (95.7%)	4.6% (95.4%)	5.2% (94.8%)	4.7% (95.3%) c
	p-value				<0.0001
Trot	Low-pass 10 Hz	21.0% (79.0%)	19.2% (80.8.%)	19.2% (80.8.%)	19.8% (80.2%) a
Trot	Bandpass 40–450 Hz	24.9% (75.1%)	26.8% (73.2%)	28.7% (71.3%)	26.8% (73.2%) b
Trot	Bandpass 7–200 Hz	3.7% (96.3%)	4.2% (95.8%)	4.2% (95.8%)	4.1% (95.9%) c
	p-value				<0.0001

Lowercase letters (a–c) indicate differences in signal loss and residual signal across signal variations. Statistical significance was set at p < 0.05.

.

3.2. Signal from M. Extensor Digitorum Communis

For the m. extensor digitorum communis, sEMG signal features varied across signal variations in both gaits, as shown in

Table 3. Median and ranges (lower quartile; upper quartile) signal-feature data (amplitude, root mean square (RMS), median frequency (MF), signal-to-noise ratio (SNR)) compared between signal variants (low-pass filtered with a 10 Hz cut-off frequency, bandpass filtered with a 40–450 Hz cut-off frequency, bandpass filtered with a 7–200 Hz cut-off frequency) for signals obtained from the m. extensor digitorum communis during walking and trotting.

Gait	Signal Variation	Amplitude [mV]	RMS [mV]	MF [Hz]	SNR [dB]
Walk	Raw	142.5 (110.5; 160.0) a	28.8 (24.9; 33.8) a	72.4 (58.7; 84.8) a	13.3 (12.3; 14.0) a
	Low-pass 10 Hz	34.2 (29.3; 39.8) b	21.7 (19.2; 24.5) b	5.6 (5.0; 6.1) b	12.5 (11.0; 13.2) b
	Bandpass 40–450 Hz	117.6 (96.3; 143.0) c	23.4 (18.0; 28.4) b	109.1 (100.4; 123.3) c	16.6 (14.7; 17.9) c
	Bandpass 7–200 Hz	114.2 (91.9; 146.3) c	26.2 (21.3; 30.7) c	60.9 (50.2; 73.5) a	13.3 (12.1; 13.9) a
	p-value	<0.0001	<0.0001	<0.0001	<0.0001
Trot	Raw	1040.0 (841.0; 1227.0) a	191.7 (154.4; 236.8) a	62.6 (53.7; 73.1) a	13.9 (11.8; 15.1) a
	Low-pass 10 Hz	312.1 (230.5; 354.0) b	144.0 (118.2; 179.7) b	7.0 (5.9; 8.0) b	12.6 (10.8; 14.9) b
	Bandpass 40–450 Hz	844.0 (653.8; 1026.0) c	156.3 (116.0; 199.1) c	89.0 (79.3; 100.7) c	16.9 (14.6; 18.5) c
	Bandpass 7–200 Hz	924.3 (738.6; 1116.0) d	186.7 (149 0; 227.3) d	58.2 (51.5; 71.7) d	14.2 (12.2; 15.7) d
	p-value	<0.0001	<0.0001	<0.0001	<0.0001

Lowercase letters (a–d) indicate differences in signal features across signal variations. Statistical significance was set at p < 0.05.

. During both gaits, amplitude was lower than in the raw signal after bandpass filtering at 40–450 Hz and 7–200 Hz, and lowest after low-pass filtering at 10 Hz, as clearly shown in Figure 3. Additionally, during trotting, a difference in amplitude was noted between bandpass filtering at 40–450 Hz and 7–200 Hz. RMS was lower than in the raw signal after bandpass filtering at 7–200 Hz and lowest after low-pass filtering at 10 Hz and bandpass filtering at 40–450 Hz. During trotting, a difference in RMS was also noted between low-pass filtering at 10 Hz and bandpass filtering at 40–450 Hz. During walking, MF was lower than in the raw signal after low-pass filtering at 10 Hz and higher after bandpass filtering at 40–450 Hz, while during trotting, MF was lowest after low-pass filtering at 10 Hz, moderate after bandpass filtering at 7–200 Hz, higher in the raw signal, and highest after bandpass filtering at 40–450 Hz. SNR was higher than in the raw signal after bandpass filtering at 40–450 Hz and lower than in the raw signal after low-pass filtering at 10 Hz. Additionally, during trotting, SNR was also higher than in the raw signal also after bandpass filtering at 7–200 Hz.

For the m. extensor digitorum communis, mean signal loss decreased, and mean residual signal increased across low-pass filtering at 10 Hz, bandpass filtering at 40–450 Hz, and bandpass filtering at 7–200 Hz in both gaits, as illustrated in Figure 3A,E. During both gaits, signal loss was highest after low-pass filtering at 10 Hz, lower after bandpass filtering at 40–450 Hz, and lowest after bandpass filtering at 7–200 Hz, as shown in

Table 4. Mean signal loss (%) and residual signal (%) data compared between signal variants (low-pass filtered with a 10 Hz cut-off frequency, bandpass filtered with a 40–450 Hz cut-off frequency, and bandpass filtered with a 7–200 Hz cut-off frequency) for signals obtained from the m. extensor digitorum communis during walk and trot.

Gait	Signal Variation	Signal Loss (Residual Signal) per Horse			Common Signal Loss (Residual Signal)
		Horse 1	Horse 2	Horse 3
Walk	Low-pass 10 Hz	24.2% (75.8%)	24.1% (75.9%)	26.5% (73.5%)	24.9% (75.1%) a
Walk	Bandpass 40–450 Hz	21.7% (78.3%)	24.7% (75.3%)	16.4% (83.6%)	20.9% (79.1%) b
Walk	Bandpass 7–200 Hz	12.8% (87.2%)	13.6% (86.4%)	9.1% (90.9%)	11.8% (88.2%) c
	p-value				<0.0001
Trot	Low-pass 10 Hz	25.3% (74.7%)	24.4% (75.6%)	22.7% (77.3%)	24.1% (75.9%) a
Trot	Bandpass 40–450 Hz	19.5% (80.5%)	20.6% (79.1%)	19.8% (80.2%)	20.1% (79.9%) b
Trot	Bandpass 7–200 Hz	4.5% (95.5%)	3.2% (96.8%)	3.3% (96.7%)	3.7% (96.3%) c
	p-value				<0.0001

Lowercase letters (a–c) indicate differences in signal loss and residual signal across signal variations. Statistical significance was set at p < 0.05.

.

3.3. Signal from M. Extensor Digitorum Lateralis

For the m. extensor digitorum lateralis, sEMG signal features varied across signal variations in both gaits, as shown in

Table 5. Median and ranges (lower quartile; upper quartile) signal feature data (amplitude, root mean square (RMS), median frequency (MF), signal-to-noise ratio (SNR)) compared between signal variants (low-pass filtered with a 10 Hz cut-off frequency, bandpass filtered with a 40–450 Hz cut-off frequency, and bandpass filtered with a 7–200 Hz cut-off frequency) for signals obtained from the m. extensor digitorum lateralis during walking and trotting.

Gait	Signal Variation	Amplitude [mV]	RMS [mV]	MF [Hz]	SNR [dB]
Walk	Raw	651.8 (598.2; 795.4) a	122.5 (110.6; 128.2) a	189.0 (164.8; 205.4) a	19.1 (18.3; 19.6) a
	Low-pass 10 Hz	142.3 (130.5; 164.3) b	85.9 (77.1; 92.5) b	5.4 (5.1; 5.8) b	17.3 (16.1; 18.0) b
	Bandpass 40–450 Hz	648.4 (586.5; 714.4) c	112.8 (100.1; 119.1) c	189.3 (168.1; 203.6) c	24.8 (22.8; 26.7) c
	Bandpass 7–200 Hz	427.8 (354.1; 485.0) d	81.5 (73.7; 94.1) b	107.3 (102.0; 114.4) d	18.3 (16.8; 19.0) b
	p-value	<0.0001	<0.0001	<0.0001	<0.0001
Trot	Raw	1390.0 (1145.0; 1615.0) a	234.3 (210.4; 289.7) a	178.5 (165.1; 192.8) a	16.6 (15.5; 17.6) a
	Low-pass 10 Hz	282.3 (244.6; 324.8) b	170.3 (150.3; 196.7) b	5.3 (5.0; 5.7) b	14.9 (14.0; 16.0) b
	Bandpass 40–450 Hz	1258.0 (950.0; 1563.0) c	213.9 (184.9; 265.7) c	190.0 (168.3; 198.9) c	21.8 (20 7; 23.7) c
	Bandpass 7–200 Hz	913.6 (622.2; 1101.0) d	163.8 (145.4; 198.0) b	91.1 (87.1; 102.5) d	13.8 (12.5; 15.5) b
	p-value	<0.0001	<0.0001	<0.0001	<0.0001

Lowercase letters (a–d) indicate differences in signal features across signal variations. Statistical significance was set at p < 0.05.

.

During both gaits, amplitude was lower than in the raw signal after bandpass filtering at 40–450 Hz, even lower after bandpass filtering at 7–200 Hz, and lowest after low-pass filtering at 10 Hz, as clearly shown in Figure 4. RMS was lower than in the raw signal after bandpass filtering at 40–450 Hz and lowest after low-pass filtering at 10 Hz and bandpass filtering at 7–200 Hz. MF was lower than in the raw signal after bandpass filtering at 40–450 Hz, even lower after bandpass filtering at 7–200 Hz, and lowest after low-pass filtering at 10 Hz. SNR was higher than in the raw signal after bandpass filtering at 40–450 Hz and lower than in the raw signal after low-pass filtering at 10 Hz and bandpass filtering at 7–200 Hz.

For the m. extensor digitorum lateralis, mean signal loss, and mean residual signal varied across cut-off frequencies, as illustrated in Figure 4A,E. During both gaits, signal loss was highest after bandpass filtering at 7–200 Hz and low-pass filtering at 10 Hz, and lower after bandpass filtering at 40–450 Hz, as shown in

Table 6. Mean signal loss (%) and residual signal (%) data compared between signal variants (low-pass filtered with a 10 Hz cut-off frequency, bandpass filtered with a 40–450 Hz cut-off frequency, bandpass filtered with a 7–200 Hz cut-off frequency) for signals obtained from the m. extensor digitorum lateralis during walking and trotting.

Gait	Signal Variation	Signal Loss (Residual Signal) per Horse			Common Signal Loss (Residual Signal)
		Horse 1	Horse 2	Horse 3
Walk	Low-pass 10 Hz	30.5% (69.5%)	29.5% (70.5%)	28.3% (71.7%)	29.4% (70.6%) a
Walk	Bandpass 40–450 Hz	7.3% (92.7%)	8.6% (91.4%)	7.7% (92.3%)	7.9% (92.1%) b
Walk	Bandpass 7–200 Hz	32.0% (68.0%)	29.8% (70.2%)	34.4% (65.6%)	32.1% (67.9%) a
	p-value				<0.0001
Trot	Low-pass 10 Hz	30.4% (69.6%)	28.0% (72.0%)	28.3% (71.7%)	28.9% (71.1%) a
Trot	Bandpass 40–450 Hz	8.9% (91.1%)	9.7% (90.3%)	10.6% (89.4%)	9.7% (90.3%) b
Trot	Bandpass 7–200 Hz	32.2% (67.8%)	29.4% (70.6%)	28.9% (71.1%)	30.1% (69.9%) a
	p-value				<0.0001

Lowercase letters (a,b) indicate differences in signal loss and residual signal across signal variation. Statistical significance was set at p < 0.05.

.

3.4. Signal from M. Extensor Carpi Ulnaris

For the m. extensor carpi ulnaris, sEMG signal features varied across signal variations in both gaits, as shown in

Table 7. Median and ranges (lower quartile; upper quartile) signal feature data (amplitude, root mean square (RMS), median frequency (MF), signal-to-noise ratio (SNR)) compared between signal variants (low-pass filtered with a 10 Hz cut-off frequency, bandpass filtered with a 40–450 Hz cut-off frequency, bandpass filtered with a 7–200 Hz cut-off frequency) for signals obtained from the m. extensor carpi ulnaris during walking and trotting.

Gait	Signal Variation	Amplitude [mV]	RMS [mV]	MF [Hz]	SNR [dB]
Walk	Raw	726.8 (548.8; 819.6) a	121.9 (107.4; 133.6) a	169.0 (155.8; 182.0) a	19.2 (18.0; 20.2) a
	Low-pass 10 Hz	168.2 (128.1; 188.6) b	90.6 (78.7; 97.0) b	5.5 (5.1; 6.0) b	18.0 (16.6; 19.1) b
	Bandpass 40–450 Hz	630.1 (523.4; 796.2) a	113.4 (98.3; 126.8) c	177.0 (161.0; 186.9) c	24.4 (23.1; 26.1) c
	Bandpass 7–200 Hz	473.5 (359.9; 553.7) c	87.4 (75.6; 94.6) b	101.3 (94.5; 114.5) d	18.0 (16.7; 19.1) b
	p-value	<0.0001	<0.0001	<0.0001	<0.0001
Trot	Raw	1317.0 (1089.0; 1672.0) a	248.1 (214.3; 304.6) a	160.6 (150.4; 180.2) a	14.5 (14.0; 15.5) a
	Low-pass 10 Hz	368.8 (280.0; 422.4) b	190.4 (165.1; 226.3) b	5.4 (5.0; 6.1) b	13.6 (12.8; 14.5) b
	Bandpass 40–450 Hz	1171.0 (923.3; 1491.0) a	218.4 (190.4; 273.7) c	176.9 (161.6; 191.8) c	20.3 (19.2; 21.2) c
	Bandpass 7–200 Hz	875.4 (702.2; 1055.0) c	189.0 (154.6; 241.0) b	95.7 (86.6; 106.9) d	13.1 (12.2; 13.9) b
	p-value	<0.0001	<0.0001	<0.0001	<0.0001

Lowercase letters (a–d) indicate differences in signal features across signal variations. Statistical significance was set at p < 0.05.

. During both gaits, amplitude was lower than in the raw signal after bandpass filtering at 40–450 Hz, even lower after bandpass filtering at 7–200 Hz, and lowest after low-pass filtering at 10 Hz, as clearly shown in Figure 5. RMS was lower than in the raw signal after bandpass filtering at 40–450 Hz and lowest after low-pass filtering at 10 Hz and bandpass filtering at 7–200 Hz. MF was lower than in the raw signal after bandpass filtering at 40–450 Hz, even lower after bandpass filtering at 7–200 Hz, and lowest after low-pass filtering at 10 Hz. SNR was higher than in the raw signal after bandpass filtering at 40–450 Hz and lower than in the raw signal after low-pass filtering at 10 Hz and bandpass filtering at 7–200 Hz.

For the m. extensor carpi ulnaris, mean signal loss, and mean residual signal varied across cut-off frequencies, as illustrated in Figure 5A,E. During both gaits, signal loss was highest after bandpass filtering at 7–200 Hz and low-pass filtering at 10 Hz, and lower after bandpass filtering at 40–450 Hz, as shown in

Table 8. Mean signal loss (%) and residual signal (%) data compared between signal variants (low-pass filtered with a 10 Hz cut-off frequency, bandpass filtered with a 40–450 Hz cut-off frequency, bandpass filtered with a 7–200 Hz cut-off frequency) for signals obtained from the m. extensor carpi ulnaris during walking and trotting.

Gait	Signal Variation	Signal Loss (Residual Signal) per Horse			Common Signal Loss (Residual Signal)
		Horse 1	Horse 2	Horse 3
Walk	Low-pass 10 Hz	27.0% (73.0%)	27.7% (72.3%)	26.6% (73.4%)	27.1% (72.9%) a
Walk	Bandpass 40–450 Hz	7.7% (92.3%)	7.6% (92.4%)	6.1% (93.9%)	7.2% (92.8%) b
Walk	Bandpass 7–200 Hz	27.8% (72.2%)	27.8% (72.2%)	28.3% (71.7%)	27.5% (72.5%) a
	p-value				< 0.0001
Trot	Low-pass 10 Hz	26.4% (73.6%)	22.9% (77.1%)	23.6% (76.4%)	24.3% (75.7%) a
Trot	Bandpass 40–450 Hz	11.1% (88.9%)	9.6% (90.4%)	11.2% (88.8%)	10.7% (89.3%) b
Trot	Bandpass 7–200 Hz	25.9% (74.1%)	28.8% (71.2%)	23.1% (76.9%)	25.9% (74.1%) a
	p-value				< 0.0001

Lowercase letters (a,b) indicate differences in signal loss and residual signal across signal variations. Statistical significance was set at p < 0.05.

.

4. Discussion

4.1. Key Findings of the Study

Focusing on the key findings, one may observe that the characteristics of the myoelectric activity of equine carpal extensors, represented by sEMG signal features and signal loss metrics, vary depending on the filtering method applied. These results support acceptance of the first hypothesis and warrant verification of the second. Specifically, for two of the four muscles examined, the extensor digitorum lateralis and the extensor carpi ulnaris, bandpass filtering at 40–450 Hz achieved the lowest signal loss, and consequently, the highest residual signal, along with the highest amplitude, RMS, MF, and SNR values. In contrast, for the other two carpal extensors, the m. extensor carpi radialis and the m. extensor digitorum communis, the lowest signal loss and highest residual signal were observed after bandpass filtering at 7–200 Hz, with some variations in sEMG signal features. These results indicate that characteristics of the myoelectric activity of equine carpal extensors differ among individual muscles, thereby supporting acceptance of the second hypothesis. Consequently, these results justify further systematic investigations into sEMG signal-filtering methods for different equine muscles.

4.2. Signal Features After Filtering with the Studied Cut-Off Frequencies

Previous studies on the myoelectric activity of equine carpal extensor muscles employed varying signal filtering methods, ranging from low-pass filtering with a 10 Hz cut-off frequency [11], to bandpass filtering with a 40–450 Hz cut-off frequency [4,12], and bandpass filtering with a 7–200 Hz cut-off frequency [10,13,14,15]. This study has shown that applying different filtering methods to the same sEMG signal results in varying signal features and levels of signal loss. Therefore, direct comparisons between previous studies on equine carpal extensor muscles [4,10,11,12,13,14,15] are limited due to methodological differences in sEMG signal processing among research groups [1]. However, comparisons between studies using the same filtering methods, such as bandpass filtering with a 40–450 Hz cut-off frequency [4,12] or with a 7–200 Hz cut-off frequency [10,13,14,15], remain valid and justified.

In a study on the effects of Kinesio Taping on forelimb motion, the sEMG signals from the m. extensor carpi radialis and m. brachiocephalicus were measured during walking and trotting, which were filtered using low-pass filtering with a 10 Hz cut-off frequency, rectified, and represented by amplitude, referred to as maximum muscle activity [11]. In the control group (without taping), the authors reported a mean amplitude of 29.71 mV during walking and 61.80 mV during trotting [11]. Current studies found a median amplitude of 29.2 mV during walking and 152.9 mV during trotting for the rectified, low-pass filtered (10 Hz) sEMG signal from the m. extensor carpi radialis. Although Zellner et al. [11] reported mean values, this study presents median values due to data distribution. Nevertheless, the similarity of the results from these two research groups indicates a satisfactory level of reproducibility in sEMG studies, especially since both utilized the same Noraxon electromyograph. However, Zellner et al. [11] did not examine the remaining three carpal extensor muscles or other signal features such as RMS, MF, and SNR, and the current study did not analyze amplitude from the m. brachiocephalicus.

In a study on the effects of clenbuterol on muscle myoelectric activity, the sEMG signals from the m. extensor carpi radialis, m. longissimus dorsi, and m. semitendinosus were measured during trotting, filtered using bandpass filtering with a 40–450 Hz cut-off frequency, and analyzed using MF and the average rectified value [12]. In the control group (without clenbuterol treatment), the authors reported a mean MF of 11 Hz, whereas in this study, a median MF from the m. extensor carpi radialis during trotting after bandpass filtering at 40–450 Hz was 82.3 Hz. Despite both research groups using the same Noraxon electromyograph, Rankins et al. [12] calculated MF over a 100 ms period during muscle activation, whereas in this study, MF was calculated for an entire burst segment annotated according to Smit et al. [4]. Since the median burst duration of the m. extensor carpi radialis during trotting is 0.21 s [18], Rankins et al. [12] analyzed only a portion of the muscle activity, which could have significantly influenced MF values. While this limits the direct reproducibility of the studies, it does not diminish the reliability of Rankins et al.’s findings on the effects of clenbuterol, as both the control and treated groups in their study were processed using the same protocol. Similar to Zellner et al. [11], Rankins et al. [12] did not examine the remaining three carpal extensor muscles or other signal features, and this study did not consider m. longissimus dorsi and m. semitendinosus.

In a study focused on recommendations for electrode placement to measure muscle myoelectric activity, the sEMG signals were recorded from the extensor carpi radialis, extensor digitorum communis muscles, and 18 other superficial muscles, including the extensor digitorum lateralis in the hindlimb. These signals were measured during trotting, filtered using bandpass filtering with a 40–450 Hz cut-off frequency, and analyzed using SNR and the coefficient of variation. In supplementary data 3, the authors provided an SNR of 14.623 dB for the m. extensor carpi radialis and 16.579 dB for the m. extensor digitorum communis at electrode placements similar to those used in this study [4]. The current study found a median SNR of 16.6 dB for the m. extensor carpi radialis and 16.9 dB for the m. extensor digitorum communis during trotting after filtering with a 40–450 Hz cut-off frequency. Moreover, Smit et al. reported an SNR of 18.070 dB for the m. extensor digitorum lateralis in the hindlimb [4], while this study showed an SNR of 21.8 dB for the m. extensor digitorum lateralis in the forelimb. Despite the previous study using monopolar derivation and SAGA electromyographs [4], the similarity of results obtained by two different research groups in comparable anatomical locations supports the good reproducibility of sEMG studies when similar signal processing methods are applied. It is noteworthy that Smit et al. [4] did not investigate the extensor carpi ulnaris and extensor digitorum lateralis in the forelimb, as well as did not analyze signal features such as amplitude, RMS, and MF. Likewise, the current study did not examine the myoelectric activity of several muscles that were included in Smit et al.’s research [4], including the m. biceps femoris, m. brachiocephalicus, m. deltoideus, m. extensor digitorum longus, m. flexor carpi ulnaris, m. flexor digitorum profundus, m. gluteus medius, m. latissimus dorsi, m. longissimus, m. pectoralis decendens, m. rectus abdominis, m. semitendinosus, m. splenius, m. triceps brachii, m. ulnaris lateralis, and m. vastus lateralis.

In studies on the nociceptive withdrawal reflex (NWR), the sEMG signals of the m. extensor carpi radialis, m. extensor digitorum communis, and m. extensor digitorum lateralis [10], or the m. extensor digitorum communis and m. deltoideus [10,13,14,15], were measured following transcutaneous electrical stimulation. These signals were filtered using bandpass filtering with a 7–200 Hz cut-off frequency and analyzed using NWR [10,13,14,15], with RMS analysis applied in only one study [15]. In those studies, the highest RMS recorded from the m. extensor digitorum communis after electrical stimulation was approximately 80 μV [15], whereas in the current study, the RMS of the burst segment during walk after filtering with a 7–200 Hz cut-off frequency was 26.2 mV. The much higher RMS values obtained in this study likely result from the completely different research objectives and study protocols used. As indicated by the above comparison of previous sEMG studies on equine carpal extensor muscles, comparing research findings is challenging, not only due to differences in signal filtering but also because different research groups used different study protocols and considered different sEMG signal features. Therefore, in this study, the presented sEMG signal characteristics encompass all the signal features reported so far for equine carpal extensor muscles—amplitude [11], RMS [15], MF [12], and SNR [4]—along with signal-loss metrics, which have previously been studied in horses only for m. triceps brachii and m. biceps femoris [5].

4.3. Effect of the Studied Cut-Off Frequencies on Signal Features

The presented results underscore the importance of selecting appropriate signal filtering methods to enhance the reliability and interpretability of sEMG data in equine biomechanics research, particularly for applications in training, rehabilitation, and diagnosis. The first study to highlight the significance of equine sEMG-filtering methods investigated the effect of cut-off frequency on signal quality by comparing raw signals with four high-pass filtered signals (using cut-off frequencies of 20 Hz, 40 Hz, 60 Hz, and 80 Hz) to assess data loss [5]. The authors demonstrated that a 40 Hz high-pass filter completely attenuated the low-frequency peak while optimally preserving the non-artifact portion of the sEMG signal [5]. High-pass filtering is used to remove low-frequency components that are especially susceptible to baseline drift and movement artifacts [1,16]. St. George et al. [5] subsequently recommended bandpass filtering with a 40–450 Hz cut-off frequency for processing sEMG signals from large equine muscles during trotting and cantering [5,6]. A similar study was currently conducted on four small equine muscles during walking and trotting, evaluating additional signal features, amplitude, MF, and SNR, beyond the RMS analyzed by St. George et al. [5].

Both the study by St. George et al. [5] and the present study demonstrated that the specific choices of filtering method matter, and that future research is needed for unified signal-processing protocols and filtering constraints to ensure comparability across future equine sEMG studies. Efforts to standardize signal filtration methods in equine sport and medicine have already led to promising developments in subsequent biomechanics research. The filtering method recommended by St. George et al. [5]—bandpass filtering with a 40–450 Hz cut-off frequency—has since been adopted in various studies on equine training, including examination of movement and muscle-activation strategies in horses with different jumping techniques [19] and muscle activity of leading limbs during cantering [20]. It has also been applied in diagnostic studies examining muscle activity in horses with induced lameness [21,22], as well as research exploring the effects of β-hydroxy β-methylbutyrate [23] and clenbuterol [12] on equine muscle function. However, despite the foundational work of St. George et al. [5,6], more recent studies have continued to employ different filtering approaches, such as bandpass filtering with a 30–500 Hz cut-off frequency filtering when analyzing the infraspinatus, deltoids, splenius, and brachiocephalic muscles [24] and bandpass filtering with a 5–420 Hz cut-off frequency when measuring activity of the superficial gluteus muscle [25]. These inconsistencies underscore the ongoing need to raise awareness about the influence of filtering choices on sEMG signal characteristics and to further advance standardization in this area.

It is worth noting that the bandpass filtering with a 40–450 Hz cut-off frequency was found to be favorable for signals recorded from the m. extensor digitorum lateralis and m. extensor carpi ulnaris during walk and trot. However, for signals recorded from the m. extensor carpi radialis and m. extensor digitorum communis, this filtering method resulted in significantly higher signal loss, a lower residual signal, and notable variations in signal features. Interestingly, the electrodes for these latter two muscles were placed closer to the proximal section of the distance between anatomical landmarks, compared to the placements for the former two muscles. This shift in electrode placement was necessitated by the considerable difficulty in obtaining high-quality signals for all four muscles at the same anatomical level. The sEMG signals from the m. extensor carpi radialis in a more distal position, as well as signals from the m. extensor digitorum lateralis and m. extensor carpi ulnaris in more proximal positions, were significantly more contaminated by noise and movement artifacts compared to signals recorded at the positions reported in this study. It is worth noting that the only study to date on optimizing electrode placement [5] did not identify a consistent safe zone for recording signals from the m. extensor carpi radialis; each of the three horses tested in that study showed a different optimal zone. Our electrode placement corresponded with the optimal zones reported for horses 1 and 3 but differed from that of horse 2 in Smith et al. [5]. For the m. extensor digitorum communis, our electrode placement aligned with the common safe zone proposed by Smith et al. [5], located between 30–40% of the proximal segment of the distance between anatomical landmarks. However, no similar recommendations were provided for the m. extensor digitorum lateralis of the forelimb or the m. extensor carpi ulnaris, as these muscles were not included in the referenced study [5]. Furthermore, no alternative recommendations for optimal electrode placement for these two muscles have been published to date.

The observed variation in signal features following distal shifts in electrode placement may suggest the presence of alternative sources of noise and artifacts that contaminate the raw signal and may affect the extraction of signal features. Potential contributors to these differences include skin displacement noise [1,26,27,28], movement artifacts [5,6] caused by high vertical forces during equine gaits [29,30], crosstalk from larger, proximally located forelimb muscles [4,31,32], and muscle fiber-type composition [33].

In human medicine, sEMG signals are typically filtered with a high-pass cut-off at 20 Hz and a low-pass cut-off in the range of 400–450 Hz to remove noise and movement artifacts, which generally appear in the spectral component between 0 and 20 Hz [16]. In equine sEMG studies, the higher body weight, faster movement, and unique gait characteristics of horses are suspected to generate greater perturbations at the electrode–skin interface compared to humans [5,6]. Consequently, movement artifacts in horses have been observed not only in the spectral component between 0 and 20 Hz but also between 20 and 40 Hz, leading to a shift to a high-pass cut-off frequency toward 40 Hz [5]. As a result, a recommended bandpass filtering range for equine sEMG signals has been proposed between 40 Hz [5] and 450 Hz [6]. Therefore, identifying the sources and types of noise and artifacts targeted for reduction is essential, as both noise and its reduction introduce modifications to sEMG signal characteristics, potentially affecting data analysis.

4.4. Future Perspectives and Limitations

Since human sEMG signal processing standards emphasize the importance of optimizing cut-off frequencies to strike a balance between noise reduction and signal preservation [16,34], further research should focus on determining the most appropriate filtering parameters for each equine muscle, gait, and specific noise sources, as well as on setting an acceptable threshold for signal loss. Moreover, in humans, the amplitude of sEMG signals shows a relatively linear decrease with increasing cut-off frequency [16]. However, no similar evidence has been published for horses. Therefore, further equine studies should also be conducted to confirm or refute a linear relationship between amplitude [19,20,22] and amplitude-dependent signal features—such as RMS [5], integrated electromyography (iEMG) [6], amplitude envelope, which is referred to as an average rectified value (ARV) [6,21], and SNR [4]—and cut-off frequencies, following consistent and gradual application of high-pass and low-pass cut-off filtering.

However, it should be emphasized that the generalizability of the discussed results is limited due to the small sample size used in this preliminary study. One may observe that, due to the complexity and challenges associated with performing sEMG recording in horses, study groups are often small and heterogeneous. For example, the study on the optimal placement of electrodes to measure equine myoelectric activity [4] was conducted on only three horses. Similarly, the investigation into the effects of trotting speed on muscle activity and kinematics included four horses ridden under saddle from two different breeds [35], and five horses of two breeds on a treadmill [36]. Other studies have included six [37], seven [15,38], or more horses [5,12,19,20,21,22], with some supporting their sample size through reliable power analysis. It is important to emphasize that this study is a preliminary study, aimed at generating empirical data necessary for accurate sample size calculation prior to more extensive methodological investigations on sEMG signal filtering. Therefore, while the number of horses examined in the present study is small, it aligns with previous preliminary work such as that by Smit et al. [4] and can be considered sufficient for obtaining initial findings. Interestingly, no previous equine sEMG studies have attempted to investigate breed-dependent differences, even when eight or more horses from two or more breeds were included [5,19,20,21,22]. In this study, only three horses representing one breed were examined, limiting breed-related conclusions. Since breed differences are an interesting aspect of equine sEMG signal characteristics, this research direction can be considered as a promising future perspective. As in all previous equine sEMG studies, a limitation of this work stems also from the manual annotation of burst and noise segments, which introduces potential bias due to the subjective nature of ROI selection. In this research, the annotation protocol used by Smit et al. [4] was followed, as it accounted for both burst and noise segments, which was essential for calculating the SNR. While this risk could be reduced by using an automatic burst activity detector, no such tool has yet been developed specifically for equine sEMG signals. Developing such a tool represents an important future research direction that could significantly advance the field of equine sEMG studies.

5. Conclusions

The myoelectric activity characteristics of equine carpal extensors vary depending on both the applied filtering method and the specific carpal extensor muscle examined. Probably due to a different noise source, the recommended bandpass filtering with a 40–450 Hz cut-off frequency results in less signal loss for the m. extensor carpi radialis and the m. extensor digitorum communis but greater signal loss for the m. extensor digitorum lateralis and the m. extensor carpi ulnaris. Therefore, obtained results justify further systematic investigations into sEMG signal-filtering methods for different equine muscles. Further research is required to match filtering cut-off frequency with the muscle being studied, considering the source and type of noise and artifacts intended to be reduced. Since both noise and its reduction introduce modifications to the raw (unfiltered) sEMG signal, potentially affecting data analysis, these findings may facilitate indirect comparisons across sEMG studies that have used different filtering methods. Thus, the significance of the work lies in enhancing the reliability and reproducibility of equine biomechanics research and guiding further studies using sEMG in the training, diagnosis, and medication of horses.