Diagnosis of Compressed Nerve Root in Lumbar Disc Herniation Patients by Surface Electromyography

Abstract

Objective

To establish a logistic regression model using surface electromyography (SEMG) parameters for diagnosing the compressed nerve root at L₅ or S₁ level in patients with lumbar disc herniation (LDH).

Methods

This study recruited 24 patients with L₅ nerve root compression and 23 patients with S₁ nerve root compression caused by LDH from May 2014 to May 2016. SEMG signals from the bilateral tibialis anterior and lateral gastrocnemius were measured. The root mean square (RMS), the RMS peak time, the mean power frequency (MPF), and the median frequency (MF) were analyzed. The accuracy, sensitivity, and specificity values were calculated separately. The areas under the curve (AUC) of the receiver‐operating characteristic (ROC) curve and the kappa value were used to evaluate the accuracy of the SEMG diagnostic model.

Results

The accuracy of the SEMG model ranged from 85.71% to 100%, with an average of 93.57%. The sensitivity, specificity, AUC, and kappa value of the logistic regression model were 0.98 ± 0.05, 0.92 ± 0.09, 0.95 ± 0.04 (P = 0.006), and 0.87 ± 0.11, respectively (P = 0.001). The final diagnostic model was: $P=1-\frac{1}{1+e^y}$; y = 10.76 − (5.95 × TA_RMS Ratio) − (0.38 × TA_RMS Peak Time Ratio) – (5.44 × 44 × LG_RMS Peak Time Ratio). L₅ nerve root compression is diagnosed when P < 0.5 and S₁ nerve root compression when P ≥ 0.5.

Conclusions

The logistic regression model developed in this study showed high diagnostic accuracy in detecting the compressed nerve root (L₅ and S₁) in these patients with LDH.

Introduction

Lumbar disc herniation (LDH) is the most prevalent spine disorder requiring surgical intervention and is a common problem encountered by spine specialists1. The incidence of LDH is high, and the classic study conducted by Haley and Perry showed protrusions of intervertebral discs in 63% of 99 unselected cadavers2. LDH causes radicular pain, particularly at the L₄–L₅ and L₅–S₁ levels3. More than 90% of LDH cases occur in these two segments and compress the nerve root below the level of herniation, which causes symptoms of pain and dysfunction4, 5. Therefore, nerve root compression mainly involves L₅ and S₁, and in most cases of radiculopathy, only a single root (L₅ or S₁) is involved4, 6, 7. As a result, accurate diagnosis of compressed nerve root of L₅ or S₁ is highly important.

The initial diagnosis of LDH is based on medical history and neurophysiological examination8. For patients with history and neurophysiological examination consistent with radiculopathy, magnetic resonance image (MRI) is a regular noninvasive method to confirm the diagnosis of LDH and the compressed nerve root9, 10.

The accuracy of MRI in detecting LDH can reach 81.83%11. However, MRI does not provide information about physiological nerve functioning and has low accuracy in diagnosing a compressed nerve root12. Detection of abnormality by MRI occurs frequently in asymptomatic individuals and is frequently irrelevant to the patient’s symptoms13. Thus, an additional diagnostic method that has a higher accuracy in locating compressed nerve roots is important to clinical diagnosis and treatment14.

Electro diagnostic study (EDX) is a useful method for evaluating patients with lumbosacral radiculopathy, and the results of EDX correspond better with patients’ clinical manifestations than the results of MRI15. EDX includes needle electromyography (NEMG) and surface electromyography (SEMG). NEMG is not used routinely because it is invasive and static; meanwhile, the muscle contraction can be limited by pain and the accuracy of diagnosis may be affected by its static nature12, 16. Because of noninvasive, portable, and dynamic features, SEMG is widely used to evaluate muscle and nerve function17, 18. Van et al. used it to differentiate patients with non‐specific chronic low back pain from healthy subjects19. There are some machine learning algorithms that have been used to develop the diagnosis model, such as logistic regression analysis, support vector machines, and neural networks20. Among them, logistic regression analysis has been widely applied and proven to be highly efficient21, 22.

The aim of this study was to develop a logistic regression diagnostic model using SEMG signals for identifying the specific compressed nerve root (L₅ or S₁) in patients with sciatica caused by confirmed disc herniation.

Materials and Methods

Inclusion and Exclusion Criteria

The inclusion criteria were: (i) patients have LDH with resultant sciatica; (ii) patient’s symptoms appearing for more than 3 weeks; and (iii) the compressed nerve root was limited to only one level, which was either L₅ or S₁.

Patients were excluded if they: (i) had a pacemaker or any implants made with ferromagnetic materials; (ii) had conduction block in the motor fibers of their peripheral nerve or other peripheral nerve diseases; (iii) had spastic paresis in the lower extremities; (iv) had any spinal tumors or prior surgical procedures on the spine; (v) had LDH combined with lumbar spinal stenosis (LSS); and (vi) had transitional vertebra.

Demographic Data of Patients

A prospective, randomized study was designed and participants were recruited from May 2014 to May 2016 in Tianjin Hospital. Forty‐seven patients with unilateral sciatica symptoms caused by LDH were finally recruited for the study: 24 subjects had intervertebral disc protrusion at the L₄–L₅ level with only L₅ nerve root compression; 23 subjects had intervertebral disc protrusion at the L₅–S₁ level with only S₁ nerve root compression. Another 24 healthy participants were included as a comparison group. Patients reported their first sciatica symptoms appearing between 1.5 and 12 months before their first examination (average, 4.8 months). This study was approved by the Ethics Committee of Capital Medical University and Tianjin Hospital. All subjects signed informed consents.

The diagnosis of the compressed nerve root was made by MRI and SEMG separately after the patient was hospitalized. The final diagnosis was confirmed by the operative findings23, 24.

Magnetic Resonance Image and Surface Electromyography Measurement

Magnetic resonance image scans were performed using a 1.5 Tesla machine with a standardized protocol (Philips Eletronics, NV, USA). Scans were read by a neurosurgeon and radiologist separately23. If there was disagreement regarding the level of a herniated disc or the compressed nerve root, the MRI study was reviewed by the senior neurosurgeon or a university neuroradiologist23.

Surface electromyography was measured using a DELSYS wireless dynamic EMG tester (EMGworks 4.3, Delsys, Massachusetts, USA). Electrodes were pasted on patients’ skin after scraping, sanding, and cleaning (Fig. 1). The SEMG signals were collected and transmitted via Bluetooth wireless connection. The EMG signals were then processed and analyzed by EMG works Analysis (EMGworks 4.3, Massachusetts, USA) and MATLAB (R2016b, MathWorks, Massachusetts, USA). The EMG signals were sampled at 1200 Hz with a band pass filter of 20–500 Hz. Four parameters were analyzed: root mean square (RMS) and RMS peak time were used to observe muscular activity25; mean power frequency (MPF) and median frequency (MF) were used to observe the muscle fatigue26. Data for the symptomatic side and asymptomatic sides were calculated separately. Previous studies have already shown that L₅ nerve root compression can result in tibialis anterior dysfunction, and S₁ nerve root compression can result in lateral gastrocnemius dysfunction27, 28. Therefore, the SEMG signals from the bilateral tibialis anterior and lateral gastrocnemius were collected in this study.

Figure 1.

The picture shows the positions where the sensors were attached (26‐year‐old male patient with L₄–L₅ herniation).

In this study, each parameter was averaged from 20 gait cycles while subjects walked on a 10‐m flat floor back and forth at natural speed and cadence29, 30. A small, portable high‐speed camera (GoPro Hero3, California, USA) was used to identify the gait cycles with the shooting rate of 120 frames per second. A GoPro camera was used to record gait cycles and to capture the flash when the DELSYS wireless dynamic EMG tester was triggered by the Trigno Trigger Adapter (EMGworks 4.3, Delsys Inc., Massachusetts, USA). From the trigger action time, gait and EMG data can be synchronized.

Development of Classifier

A logistic regression classifier was developed to estimate the probability (P) for each patient to be L₅ or S₁, and a diagnosis model was established as31:

$$P=1-\frac{1}{1+e^y}$$

$$\mathrm{y}={\mathrm{\beta }}_0+{\mathrm{\beta }}_1{\mathrm{X}}_1+{\mathrm{\beta }}_2{\mathrm{X}}_2+{\mathrm{\beta }}_3{\mathrm{X}}_{3\dots }{\mathrm{\beta }}_{\mathrm{m}}{\mathrm{X}}_{\mathrm{m}},\mathrm{m}=1,2,\dots ,8.$$

P is the probability of the presence of a compressed nerve root. β are logistic regression coefficients. X_m are independent variables. To reduce the effect of individual differences, the ratio of the data on symptomatic and asymptomatic sides was used to establish the diagnostic model21, 22. The ratios of RMS, RMS peak time, MPF, and MF in the tibialis anterior of the symptomatic side over that of the asymptomatic side were represented as X₁, X₂, X₃, X₄; the ratios of RMS, RMS peak time, MPF, and MF in the lateral gastrocnemius of the symptomatic side over that of asymptomatic side were represented as X₅, X₆, X₇, X₈. The number of variables is represented by m (1–8). e = 2.71828 is the base of the system of natural logarithms.

Data Analysis

The SEMG data for each subject was processed and analyzed using MATLAB software after elimination of noise. For time domain data, the RMS of the SEMG signal for each gait cycle was calculated using the 30‐ms window and the 20‐ms step duration32. For frequency domain data: we used a fast Fourier transform for SEMG signals and calculated the MPF and the MF of each gait cycle33.

Data collected from two lower limbs of each subject were compared using a paired Student’s t‐test or a nonparametric Wilcoxon test depending on whether data was normally distributed. The MRI results were recorded as either positive or negative. The accuracy, sensitivity and specificity were calculated from a 2‐by‐2 crosstabs, and the area under the curve (AUC) of the receiver‐operating characteristic (ROC) was used to evaluate the accuracy of the diagnosis model. The kappa value was used to evaluate the concordance between the final diagnosis and the SEMG test. The following concordance criteria were applied: kappa > 0.8: excellent; kappa = 0.6–0.8: good; kappa = 0.4–0.6, moderate; kappa = 0.2–0.4: fair; and kappa < 0.2: poor34.

A repeat hold‐out test procedure was used to analyze the performance of the classifier. The model development and verification process was repeated 10 times as data were divided randomly to model the development group (70% of the sample size) and the model validation group (30% of the sample size) each time35, 36. The data of the model development group were used to build the classifier and to calculate the P‐value. The cut‐off point was obtained from the ROC curves with the maximum value for the Youden index (YI). Then, the data of the validation group were calculated through the classifier to obtain the P‐value. The S₁ nerve root was considered as compressed when the P‐value was greater than the cut‐off point, and the L₅ nerve root was confirmed as compressed when P was less than the cut‐off point.

The diagnostic accuracy was calculated and averaged. The final diagnostic model was built using the entire sample. ROC curves were generated, and the AUC, Sensitivity, Specificity were used to investigate the validity of SEMG diagnostic models. Data was analyzed using SPSS 19 software (IBM SPSS Statistics, Version 19.0 for Windows, IBM Corporation, New York, USA). The degree of the accuracy followed the criteria as: non‐informative (AUC = 0.5), less accurate (0.5 < AUC ≤ 0.7), moderately accurate (0.7 < AUC ≤ 0.9), highly accurate (0.9 < AUC < 1), and perfect tests (AUC = 1)37.

Results

Patients Information

Forty‐seven patients with unilateral sciatica symptoms caused by LDH and 24 healthy participants were recruited for the study.

Twenty‐four patients (5 females and 19 males) with only L₅ nerve root compression were selected as the L₅ group (age: 53.13 ± 9.37 years, ranging from 39 to 73 years; height: 167.04 ± 8.48 cm, ranging from 150 to 185 cm; weight: 72.71 ± 16.09 kg, ranging from 53 to 110 kg).

Twenty‐three patients (8 females and 15 males) with only S₁ nerve root compression were selected as the S₁ group (age: 52.34 ± 11.92 years, ranging from 26 to 74 years; height: 164.39 ± 5.81 cm, ranging from 150 to 180 cm; weight: 69.48 ± 8.17 kg, ranging from 55 to 88 kg).

Twenty‐four healthy participants (8 females and 16 males) were included as a comparison group (age: 54.49 ± 12.21 years, ranging from 34 to 82 years; height: 166.88 ± 7.09 cm, ranging from 150 to 176 cm; weight: 68.21 ± 8.09 kg, ranging from 45 to 83 kg).

General Information

For the SEMG model development and verification process, the average diagnostic accuracy is 93.6%, ranging from 85.7% to 100% (Table 1). The sensitivity, specificity, AUC and kappa value of the model were 0.98, 0.92, 0.95, and 0.87, respectively (Table 2). The diagnostic accuracy by MRI is 87.23%.

Table 1.

The diagnosis accuracy of 10 random sampling tests on the SEMG diagnostic model

Times	SEMG diagnosis	Final diagnosis			Accuracy (%)
		L5	S1	Total
1	L5	6	0	6	92.86
	S1	1	7	8
	Total	7	7	14
2	L5	7	1	8	92.86
	S1	0	6	6
	Total	7	7	14
3	L5	7	0	7	100
	S1	0	7	7
	Total	7	7	14
4	L5	7	1	8	92.86
	S1	0	6	6
	Total	7	7	14
5	L5	7	1	8	92.86
	S1	0	6	6
	Total	7	7	14
6	L5	7	2	9	85.71
	S1	0	5	5
	Total	7	7	14
7	L5	6	0	6	92.86
	S1	1	7	8
	Total	7	7	14
8	L5	7	0	7	100
	S1	0	7	7
	Total	7	7	14
9	L5	7	2	9	85.71
	S1	0	5	5
	Total	7	7	14
10	L5	7	0	7	100
	S1	0	7	7
	Total	7	7	14
Average					93.57

SEMG indicates surface electromyography; “SEMG diagnosis” means the diagnosis made through SEMG and the logistic regression model; “final diagnosis” means the combined diagnosis made through physical examination, magnetic resonance image and confirmed by the operative findings.

Table 2.

The sensitivity, specificity, AUC and kappa value of 10 random sampling tests on the SEMG diagnostic model

Times	Sensitivity	Specificity	AUC	AUC P‐value	Kappa value	Kappa P‐value
1	0.88	1.00	0.94	0.007	0.86	0.001
2	1.00	0.88	0.94	0.007	0.86	0.001
3	1.00	1.00	1.00	0.002	1.00	<0.001
4	1.00	0.88	0.94	0.007	0.86	0.001
5	1.00	0.88	0.94	0.007	0.86	0.001
6	1.00	0.78	0.89	0.020	0.71	0.005
7	0.88	1.00	0.94	0.007	0.86	0.001
8	1.00	1.00	1.00	0.002	1.00	<0.001
9	1.00	0.78	0.89	0.002	0.71	0.005
10	1.00	1.00	1.00	0.002	1.00	<0.001
Mean ± SD	0.98 ± 0.05	0.92 ± 0.09	0.95 ± 0.04	0.006	0.87 ± 0.11	0.002

AUC, areas under curves; SD, standard deviation; SEMG, surface electromyography.

Final Diagnostic Model

The final diagnostic model was derived with the entire sample:

$$\mathrm{P}=1-\frac{1}{1+e^y}$$

$$\mathrm{y}=10.76-\left(5.95\times {\mathrm{X}}_1\right)-\left(0.38\times {\mathrm{X}}_2\right)-\left(5.44\times {\mathrm{X}}_6\right).$$

The ROC curve is shown in Fig. 2. The AUC, sensitivity, and specificity were 0.97, 1.00, and 0.96, respectively when the cut‐off point P value was calculated as 0.5. It is diagnosed as L₅ nerve root compression when P < 0.5; and diagnosed as S₁ nerve root compression when P ≥ 0.5.

Figure 2.

The receiver‐operating characteristic curves of the final predicted diagnostic model. The area under the curve of the predicated probability is satisfied and larger than the other three parameters separately. LG, lateral gastrocnemius; RMS, root mean square; TA, tibialis anterior.

Surface Electromyography Information

There are no significant differences between SEMG data from two lower limbs in healthy subjects (Table 3).

Table 3.

The SEMG results between double lower limbs in healthy group (24 cases)

	Left side		Right side
SEMG	Mean ± SD	Median (IQR)	Mean ± SD	Median (IQR)	P‐value (two tailed)
TA
RMS‐peak (μV)	41.99 ± 16.29	37.56 (32.62, 46.01)	46.88 ± 17.45	43.95 (31.13, 59.35)	0.186
RMS‐peak time (%)*	16.59 ± 22.21	6.50 (1.00, 25.30)	19.59 ± 25.20	5.00 (4.00, 46.50)	0.200
MPF (HZ)	75.27 ± 22.26	80.44 (54.63, 99.16)	81.89 ± 30.34	83.16 (57.67, 103.76)	0.340
MF (HZ)	96.54 ± 21.24	100.72 (80.56, 117.53)	102.69 ± 31.47	108.80 (78.51, 123.69)	0.339
LG
RMS‐peak (μV)	38.04 ± 16.25	34.78 (25.32, 46.48)	33.14 ± 10.98	29.42 (27.02, 37.18)	0.187
RMS‐peak time (%)*	48.09 ± 15.58	42.00 (40.75, 46.50)	45.91 ± 18.20	42.00 (39.00, 47.50)	0.839
MPF (HZ)	70.25 ± 21.38	71.93 (60.73, 83.47)	74.51 ± 20.27	78.46 (63.65, 88.65)	0.323
MF (HZ)	88.34 ± 22.21	95.42 (79.19, 100.64)	93.82 ± 20.25	96.20 (81.77, 110.88)	0.212

IQR, inter‐quartile range; LG, lateral gastrocnemius; MF, median frequency; MPF, mean power frequency; RMS, root mean square; SD, standard deviation; SEMG, surface electromyography; TA, tibialis anterior

^* The data was not normally distributed.

For subjects in the group with L₅ compression, the tibialis anterior’s maximum RMS value appeared at 34.50% of a whole gait cycle (100%) in the symptomatic side and 11.71% in the asymptomatic side. A 22.72% delay can be seen in the symptomatic side compared with the asymptomatic side (P = 0.002). The tibialis anterior’s MPF and MF are 55.85 Hz, 58.36 Hz of the symptomatic side and 71.61 Hz, 76.73 Hz of the asymptomatic side. The MPF on the symptomatic side decreased by 15.76 Hz compared to the asymptomatic side (P < 0.001) and the MF on the symptomatic side decreased by 18.37 Hz compared to the asymptomatic side (P < 0.001). The lateral gastrocnemius’s maximum RMS value was 34.11 μV on the symptomatic side and 45.49 μV of the asymptomatic side; the symptomatic side decreased by 11.38 μV compared to the asymptomatic side (P = 0.042, Table 4). There was no statistical significance for other parameters.

Table 4.

The SEMG results of double lower limbs in L₅ compressed patients (L₅ nerve root compressed, 24 cases)

	Symptomatic side		Asymptomatic side
SEMG	Mean ± SD	Median (IQR)	Mean ± SD	Median (IQR)	P‐value (two tailed)
TA
RMS‐peak (μV)	51.64 ± 34.09	49.08 (24.84, 70.25)	50.49 ± 37.45	48.98 (13.77, 72.40)	0.877
RMS‐peak time (%)*	34.50 ± 36.00	9.50 (5.00, 57.00)	11.71 ± 24.34	5.00 (1.00, 6.00)	0.002
MPF (HZ)	55.85 ± 26.83	55.47 (30.38, 75.75)	71.61 ± 33.13	63.50 (45.39, 97.71)	<0.001
MF (HZ)	58.36 ± 23.50	55.04 (38.97, 74.01)	76.73 ± 26.47	76.46 (63.47, 89.53)	<0.001
LG
RMS‐peak (μV)	34.11 ± 18.87	26.61 (18.26, 51.13)	45.49 ± 27.92	36.19 (21.14, 67.21)	0.042
RMS‐peak time (%)*	40.67 ± 17.09	43.00 (38.50, 44.75)	39.67 ± 15.46	43.00 (42.00, 46.00)	0.592
MPF (HZ)	63.71 ± 33.60	73.34 (23.61, 84.03)	70.15 ± 28.93	77.49 (42.40, 86.85)	0.412
MF (HZ)	62.89 ± 30.38	55.71 (39.62, 88.91)	74.53 ± 28.18	82.88 (49.35, 100.46)	0.141

IQR, inter‐quartile range; LG, lateral gastrocnemius; MF, median frequency; MPF, mean power frequency; RMS, root mean square; SD, standard deviation; SEMG, surface electromyography; TA, tibialis anterior

^* The data was not normally distributed.

For subjects in the group with S₁ compression, the lateral gastrocnemius’s maximum RMS value appeared at 18.57% of a whole gait cycle (100%) on the symptomatic side and 44.35% on the asymptomatic side. That is, the position of maximum RMS value changes 25.78% on the symptomatic side compared with the asymptomatic side (P < 0.001). Meanwhile, the lateral gastrocnemius’s MPF and MF are 41.38 Hz, 44.73 Hz of the symptomatic side and 83.97 Hz, 90.56 Hz of the asymptomatic side. The MPF on the symptomatic side decreased by 42.59 Hz compared to the asymptomatic side (P < 0.001) and the MF on the symptomatic side decreased by 45.83 Hz compared to the asymptomatic side (P < 0.001). The tibialis anterior’s maximum RMS value is 46.27 μV on the symptomatic side and 64.98 μV on the asymptomatic side; the symptomatic side decreased by 18.71 μV compared to the asymptomatic side (P = 0.004, Table 5). There is no statistical significance for the other parameters.

Table 5.

The SEMG results of double lower limbs in S₁ compressed patients (S₁ nerve root compressed, 23 cases)

	Symptomatic side		Asymptomatic side
SEMG	Mean ± SD	Median (IQR)	Mean ± SD	Median (IQR)	P‐value (two tailed)
TA
RMS‐peak (μV)	46.27 ± 33.06	42.99 (28.87, 57.63)	64.98 ± 42.23	60.76 (29.91, 104.72)	0.004
RMS‐peak time (%)*	10.00 ± 19.76	5.00 (2.00, 6.00)	7.48 ± 15.51	5.00 (4.00, 6.00)	0.086
MPF (HZ)	57.90 ± 24.80	46.96 (38.17, 71.36)	58.38 ± 19.80	52.93 (43.67, 69.72)	0.923
MF (HZ)	67.06 ± 22.69	67.36 (52.47, 87.81)	66.98 ± 23.98	64.60 (45.19, 83.67)	0.987
LG
RMS‐peak (μV)	63.45 ± 43.59	53.50 (35.72, 81.77)	42.92 ± 31.02	35.63 (24.22, 44.48)	0.078
RMS‐peak time (%)*	18.57 ± 14.27	13.00 (8.00, 37.00)	44.35 ± 4.52	44.00 (41.00, 46.00)	<0.001
MPF (HZ)	41.38 ± 23.30	42.26 (20.95, 57.58)	83.97 ± 19.21	80.89 (70.54, 91.96)	<0.001
MF (HZ)	44.73 ± 22.06	33.74 (28.30, 66.46)	90.56 ± 23.37	94.77 (74.59, 102.66)	<0.001

IQR, inter‐quartile range; LG, lateral gastrocnemius; MF, median frequency; MPF, mean power frequency; RMS, root mean square; SD, standard deviation; SEMG, surface electromyography; TA, tibialis anterior

^* The data was not normally distributed.

Discussion

The logistic regression model developed in this study demonstrated high diagnostic accuracy (93.57%) in detecting the compressed nerve root (L₅ and S₁), and showed an excellent concordance with the final diagnosis (kappa = 0.87). The accuracy of detecting the compressed root by the SEMG model is also higher than by MRI, NEMG, and neurologic examination reported in other studies: MRI (57.6%–81.8%), NEMG (51%–86%), or neurologic examination (35%–64%)11, 38, 39. The model developing process is considered reasonable because the accuracy, sensitivity, specificity, and AUC are satisfied.

The higher diagnostic accuracy of the SEMG model may be attributed to the close relationship between lumbar nerve roots and lower limb muscles. There are different levels of nerve root control for different lower limb’s muscles, resulting in differences in SEMG features in patients with L₅ and S₁ (Tables 4 and 5). L₅ nerve root compression caused significant delay of the tibialis anterior’s RMS peak time of the symptomatic side when compared to the asymptomatic side (from 11.71% to 34.5%). Meanwhile, the MPF and MF in the tibialis anterior, and the maximum RMS value in the lateral gastrocnemius of the symptomatic side also became smaller (Fig. 3, Table 4). In contrast, S₁ nerve root compression caused the lateral gastrocnemius’s RMS peak time of the symptomatic side to appear earlier (from 44.35% to 18.57%) compared to the asymptomatic side. Meanwhile, the MPF and MF in the lateral gastrocnemius of the symptomatic side and the maximum RMS value in the tibialis anterior of the symptomatic side also became smaller (Fig. 4, Table 5).

Figure 3.

The RMS values in one gait cycle of a patient with L₅ nerve root compression. A, The tibialis anterior’s RMS peak time of the symptomatic side appears later. B, The maximum RMS value in lateral gastrocnemius becomes smaller when compared to the asymptomatic side. LG, lateral gastrocnemius; RMS, root mean square; TA, tibialis anterior.

Figure 4.

The RMS values in one gait cycle of a patient with S₁ nerve root compression. A, The lateral gastrocnemius’s RMS peak time of the symptomatic side appearing earlier. B, The maximum RMS value in tibialis anterior become smaller when compared to the asymptomatic side. LG, lateral gastrocnemius; RMS, root mean square; TA, tibialis anterior.

For subjects with S₁ nerve root compression, the RMS peak time of lateral gastrocnemius appears earlier (Fig. 4). This may be because the mechanical compression of a healthy nerve causes motor deficits, which results in changes in nerve conduction velocity40, 41. Therefore, the lateral gastrocnemius and the tibialis anterior contracted at the same time on the symptomatic side, while they were alternate contraction in the asymptomatic side (Fig. 5). The nerve function of the tibialis anterior is normal, so the muscle strength of the tibialis anterior reduces to avoid ankle joint stiffness and appeared as the RMS maximum values of the tibialis anterior becoming smaller. The extended period of time for muscle contraction overlap in the symptomatic limb results in the consumption of more energy compared to the asymptomatic side. Therefore, the MPF and MF of the lateral gastrocnemius on the symptomatic side are smaller compared to the asymptomatic side. The same mechanism can also be used to explain the SEMG expression of patients with L5 nerve root compression.

Figure 5.

The RMS value in one gait cycle of a patient with S₁ nerve root compression. A, The lateral gastrocnemius and the tibialis anterior were alternate contraction, and the contraction periods rarely overlap. B, The earlier contraction of lateral gastrocnemius almost happens at the same time as for the tibialis anterior in the symptomatic side; and the co‐contraction period is extended. LG, lateral gastrocnemius; RMS, root mean square; TA, tibialis anterior.

Although the diagnostic accuracy of the model is satisfied, the model still failed to identify compressed nerve roots for 6.4% of patients in this study. There are three possible reasons for these misdiagnosed cases. First is the inability to detect pure sensory radiculopathies, especially for patients who present with either purely sensory complaints or primarily sensory complaints with minimal complaints of weakness29. This is because root pathologies cause changes in EMG only when there is axonal damage of motor fibers or when a proximal conduction block occurs42. Second, if denervation is balanced with reinnervation, or the denervation is old, no fibrillations will be seen, and the denervation will be missed43. Third, segmental motor overlaps and anatomical variations are other factors that may explain misdiagnosis. In the lower extremity, most muscles have two or three levels of roots innervating them, with one nerve root being dominant28, 44. The tibialis anterior is innervated by the L₅ root in 90% of cases; and in the remaining 10% of cases are innervated by the S₁ root44. In addition, there are nerve root variations in a considerable percentage of patients45.

In the present study, MRI failed to accurately diagnose the compression nerve root in 6 subjects. These misdiagnosis cases include 2 patients with extreme lateral lumbar disc herniation at the L₅–S₁ segment and 4 patients with multiple lumbar disc herniation at both L₄–L₅ and L₅–S₁ segments. The result showed that the SEMG model developed in this study has more advantages in these special types such as multiple lumbar disc herniation or extreme lateral lumbar disc herniation.

There is a limitation in this study. The diagnostic model was limited to affected L₅ or S₁ nerve roots, and patients with both L₅ and S₁ nerve root compression were excluded. Although Bartynski’s study demonstrated that the incidence of both L₅ and S₁ nerve root compression is only 4% in LDH patients, it would be ideal to include patients with both L₅ and S₁ nerve root compression in future studies7.

In conclusion, the logistic regression model developed in this study can identify the compressed nerve roots (L₅ or S₁) for patients with LDH with a high success rate (93.6%), and can be used as a supplementary diagnostic method with MRI. Different diagnostic results from SEMG and MRI may suggest that further evaluation should be undertaken. The SEMG method can also be used for patients who cannot take MRI or NEMG tests, such as patients with coagulopathies, patients with pacemakers, and pregnant women.