An Interpretable Rule-Based Clinical Decision Support Framework for Electrotherapy-Related Conditions using Simulink

An Interpretable Rule-Based Clinical Decision Support Framework for Electrotherapy-Related Conditions using Simulink

Yogita S. Salunkhe (1),*, Bharat T. Jadhav (2) , Sushant R. Pol (1), Shashikant K Shinde (1)

(1),*Department of Electronics, YCIS, Satara, KBP University Satara, Maharashtra, India

(2),Department of Electronics, Rayat Institute of Research and Development, Satara, Maharashtra, India

Corresponding Author: Yogita S. Salunkhe

GRAPHICAL ABSTRACT

Procedure of the suggested MATLAB/Simulink rule-based illness detection model. Clinical inputs (VAS, symptom duration, DN4, nerve damage, muscle strength, ROM and age) are analyzed by ifelse threshold rules employing priority-based conflict resolution to categorize patients into seven electrotherapy-related diseases. Validation is demonstrated via a fake dataset and accuracy derived from a confusion matrix.

ABSTRACT

Electrotherapy techniques, including Transcutaneous Electrical Nerve Stimulation (TENS) and Neuromuscular Electrical Stimulation (NMES), are extensively utilized in physiotherapy for pain alleviation and musculoskeletal rehabilitation. Treatment selection frequently relies on expert clinical judgment, which may not be uniformly accessible across all healthcare environments. This research introduces a comprehensible, rule-based mathematical model for disease identification, executed in MATLAB/Simulink, to categorize clinical situations pertinent to electrotherapy. The model correlates validated clinical inputsVisual Analog Scale (VAS), symptom duration, DN4 score, muscle strength percentage, range of motion (ROM), swelling, and ageinto seven output categories: acute pain, chronic pain, neuropathic pain, muscle weakness, soft tissue injury, osteoarthritis, and absence of disease. The innovation of the approach is the explicit operationalization of clinically recognized criteria into transparent decision equations that can be simulated and audited within Simulink. A synthetic dataset was created to verify the rule logic. Through confusion-matrix-based evaluation, the model attained an overall accuracy of 47.5% on the synthetic dataset, suggesting that symptom overlap, and threshold closeness may result in misclassification inside deterministic rule systems. These findings validate the model as a proof- of-concept decision-support framework and encourage the modification of rule priorities and further validation using actual clinical datasets. This work presents a proof-of-concept framework validated on a synthetic dataset, intended for future clinical integration.

Keywords: Disease Detection, Electrotherapy, Mathematical Model, Simulink, Rule-Based System, Clinical Decision Support.

1. Introduction

   Pain-related and musculoskeletal disorders represent a significant global health burden and are among the leading causes of disability worldwide [1], [2]. Accurate classification of patient conditions is essential for effective therapeutic planning and outcome improvement in rehabilitation and electrotherapy domains [3]. Clinical decision-making commonly relies on structured assessment parameters such as pain intensity, symptom duration, neurological screening scores, muscle strength evaluation, joint mobility, and age-related risk indicators [4], [5], [6], [7]. However, integrating these heterogeneous clinical indicators into consistent and reproducible diagnostic decisions remains challenging in routine clinical environments [8].

   The Visual Analog Scale (VAS) is widely validated for quantifying pain intensity and is routinely used in both acute and chronic pain assessment frameworks [4], [9] . Symptom duration is a critical diagnostic criterion, with chronic pain formally defined as pain persisting for three months or longer according to international classification standards [10]. Neuropathic pain identification is frequently supported by validated screening instruments such as the DN4 questionnaire, which demonstrates strong sensitivity and specificity in clinical populations[11]. Similarly, reductions in muscle strength and restricted range of motion are recognized indicators of functional impairment in musculoskeletal and rehabilitation medicine[12].

   Decision-support systems have increasingly been adopted in healthcare to enhance diagnostic consistency and reduce subjectivity in clinical practice [13]. Rule-based expert systems, in particular, offer interpretability and transparency, aligning closely with traditional clinical reasoning processes[14]. Unlike purely data-driven black-box models, rule-based systems allow explicit incorporation of validated clinical thresholds derived from established medical literature[15]. Such transparency is especially important in healthcare applications where explainability and safety are primary concerns [16].

   Computational modeling platforms such as MATLAB and Simulink are widely utilized in biomedical engineering research for system design, simulation, and performance evaluation under controlled conditions [17]. Simulation-based validation using structured datasets is an accepted preliminary evaluation approach when real-world clinical data are limited or restricted due to ethical considerations [18]. Therefore, the development of a structured rule-based mathematical model supported by validated clinical thresholds and evaluated using simulation techniques represents a methodologically justified approach for early-stage decision-support system design [14], [17].

   Despite extensive clinical evidence supporting electrotherapy, the decision-making process for selecting appropriate treatment modalities remains largely manual and experience-based. This highlights the need for structured, interpretable, and automated decision support systems. In this work, a mathematical rule-based disease detector is developed and implemented in Simulink to classify electrotherapy-related conditions using clinically validated thresholds derived from the literature. The proposed system aims to bridge the gap between clinical knowledge and computational modeling.

   Notwithstanding substantial data endorsing electrotherapy, the process of clinical decision- making for the selection of suitable electrotherapy interventions predominantly relies on manual methods and experiential knowledge. In standard practice, clinicians must analyze various clinical signs (pain severity, duration, neuropathic assessment, muscle strength, range of motion, and ) to ascertain the most probable underlying problem prior to selecting a treatment approach. This approach may become unreliable in the absence of professional physiotherapists or when documentation is deficient. Consequently, a structured and interpretable decision-support framework is required to transform proven clinical scales into clear, reproducible categorization logic appropriate for application in a simulation setting.

   We opted for a rule-based approach instead of machine-learning technologies due to the transparency, auditability, and direct alignment with established clinical thresholds that rule-based systems offer. In initial phases where authentic labeled clinical datasets may be scarce, deterministic rule models facilitate swift prototyping, error identification, and clinician-friendly interpretation. Future research may investigate machine-learning methodologies once real clinical data gathering and adequate ground-truth labeling are achieved.

2. Methodology

   The proposed disease detection system follows a rule-based mathematical modeling approach implemented using MATLAB and Simulink. Input parameters were selected based on commonly used clinical asessment metrics in pain management and rehabilitation. These include pain intensity measured using the Visual Analogue Scale (VAS), duration of symptoms (D), DN4 score for neuropathic pain (NDN4), nerve injury indicator (Ni), muscle strength percentage (Ms), presence of swelling (Sw), range of motion (ROM), age (A), and pain severity level (Pl).

   1. Mathematical Formulation of Classification Rules

      Each patient case is represented as an input feature vector:

      X = [VAS, D, NDN4, Ni, Ms, Sw, ROM, A]

      The disease classification function is defined as:

      y = f(X)

      where y {C1, C2, C3, C4, C5, C6} represents classified electrotherapy-related conditions.

   2. Deterministic Rule Structure

      The rules are evaluated sequentially in a priority-based order to avoid ambiguity. The execution order is designed according to clinical severity and diagnostic specificity:

      1. Neuropathic Pain

      2. Nerve Damage

      3. Acute Condition

      4. Chronic Pain

      5. Muscle Weakness

      6. Osteoarthritis / Degenerative Condition

         The ordered evaluation ensures that highly specific neurological conditions are detected before broader musculoskeletal classifications.

   3. Mathematical Rules

      Rule 1: Neuropathic Pain (C)

      C1 = { 1, if NDN4 4 VAS 4

      0, oterwise

      Rule 2: Nerve Damage (C)

      C2 = { 1, if Ni 1 Ms < 60% 0, oterwise

      . (1)

      . (2)

      Rule 3: Acute Condition (C)

      C3 = {1, if D 90 days Sw 1 0, oterwise

      . (3)

      Rule 4: Chronic Pain (C)

      C4 = {1, if D 90 Days VAS 3 0, oterwise

      . (4)

      Rule 5: Muscle Weakness (C)

      C5 = {1, if Ms 70% ROM < 75%

      0, oterwise

      . (5)

      Rule 6: Osteoarthritis (C)

      C6 = {1, if A > 50 ROM < 60% D 90 (6)

      0, oterwise

   4. Conflict Resolution Strategy

      In cases where more than one condition is satisfied, the system applies a priority-based conflict resolution mechanism. The first satisfied rule in the predefined execution order assigns the final class label.

      Mathematically:

      C1, if C1 = 1

      C2, else if C2 = 1

      C3, else if C3 = 1

      y = C4, else if C4 = 1

      C5, else if C5 = 1

      C6, else defaut

      . (7)

      eq. (7) structured rule execution prevents overlapping classifications and ensures deterministic output generation.

   5. Simulink Implementation Details

      The model is implemented using a modular Simulink architecture consisting of:

      • Input blocks for parameter entry

      • Signal conditioning blocks for normalization (where required)

      • A MATLAB Function block containing the rule-based logic

      • Output display and data logging blocks

      • Each input parameter is fed as a separate signal into the MATLAB Function block. Within this block, the feature vector X is formed, and the sequential ifelse rule logic is executed exactly as defined in the mathematical formulation.

        The MATLAB Function block outputs a scalar disease label (Y), which is then routed to:

      • A Display block for real-time visualization

      • A Scope block for simulation monitoring

      • A Function block for exporting results into MATLAB for further analysis

      The simulation is executed using fixed-step discrete solver settings to ensure repeatability of classification outcomes. Disease classification is performed using deterministic ifelse logic, where each condition corresponds to a specific electrotherapy-related disease. The threshold values used in the model are derived from validated clinical studies and guidelines reported in the literature [10], [19], [20]. For performance evaluation, a synthetic dataset was generated within clinically acceptable ranges, and the Simulink model was executed iteratively using MATLAB scripts. Predicted disease labels were compared with ground truth labels to compute classification accuracy and confusion matrices.

      AS

      s

      N4

      ury

      System Inputs

      Duration Days

      V S

      Display

Model

Age

ROM

Swelling

Strength

Nerve Inj ry

D 4

Duration

1. System Architecture:

   Fig. 1.1 Overall system architecture

   Fig. 1.1 shows Overall architecture of the proposed rule-based electrotherapy disease detection system illustrating clinical input parameters, rule evaluation module, and final disease classification output.

2. Mathematical Model:

   Fig 1.2 Simulink implementation of the rule-based disease detection system showing input parameter blocks, MATLAB Function block, and output classification unit.

   The disease classification rules employed in the proposed model are derived from clinically validated thresholds reported in the literature. The Simulink implementation of rule-based decision model is shown in above fig. 1.2.

3. Model Justification:

   Acute pain is identified when pain intensity is severe (VAS 7) and symptom duration is short ( 14 days). Severe pain is commonly defined as VAS values between 7 and 10, and acute pain is typically associated with short-term duration[19], [21].

   Chronic pain is classified when VAS 6 and symptom duration exceeds three months. The three- month duration threshold is widely accepted as the defining criterion for chronic pain in epidemiological and clinical studies[20].

   Neuropathic pain is detected using a DN4 score 4 or the presence of nerve injury. The DN4 questionnaire has been validated extensively, with a score of 4 or higher indicating neuropathic pain with high sensitivity and specificity[22].

   Muscle weakness is defined when muscle strength falls below or equal to 50% of normal capacity. Rehabilitation literature commonly considers strength levels below 50% as clinically significant weakness warranting neuromuscular electrical stimulation[10].

   Soft tissue injury is identified based on the combined presence of swelling, restricted range of motion (ROM < 80%), and moderate pain (VAS 4). These clinical signs are widely reported as indicators of soft tissue damage and inflammation[11].

   Osteoarthritis is classified when patient age is 45 years or above, ROM is less than 70%, and pain level is moderate to severe (VAS 5). These criteria align with epidemiological and clinical findings that associate osteoarthritis with age-related degeneration, pain, and joint stiffness[23].

   If none of the above conditions are satisfied, the model assigns a no-disease classification. The rule-based model ensures transparency, interpretability, and ease of validation for electrotherapy decision support.

   Table 1.1. Disease with its Equations

   Disease	Equation
   Acute Pain	VAS 7, duration 14d
   Chronic Pain	VAS 6, duration 3 mo
   Neuropathic Pain	DN4 4 or Nerve Injury=1
   Muscle Weakness	Strength 50%
   Soft Tissue Injury	Swelling + ROM<80% + Pain4
   Osteoarthritis	Age45 + ROM<70% + Pain5

1. Dataset Ge ration

   A synthetic dataset was created due to the scarcity o publicly available labeled datasets specialized to electrotherapy. Each sample signifies a theoretical patient case with values limited to clinically

   relevant ranges based on the chosen clinical inputs. A total of N = 201 samples were produced across six categories: Acute pain (AC), Chronic pain (CP), Muscle weakening (MW), Neuropathic pain (NP), No illness (ND), and Soft tissue injury (STI).

   Class distribution: AC = 34, CP = 40, MW = 10, NP = 95, ND = 17, STI = 4 (total N = 200). The distribution is skewed, with NP constituting the predominant class and STI the least presented class.

   Limitation: The synthetic nature of the dataset may fail to accurately reflect real-world comorbidities, measurement variability, and discrepancies in physician labeling. Consequently, the claimed performance should be regarded as an affirmation of rule execution logic and conflict resolution technique, rather than as conclusive clinical diagnostic accuracy. Subsequent efforts will involve validation utilizing actual patient records with clinician-verified labeling.

2. Accuracy Evaluation

   Performance was assessed utilizing a confusion matrix and derived metrics such as overall accuracy, per-class precision, recall (sensitivity), F1-score, and specificity. For each class k, we calculated:

   • Precisionk

     = TPk

     TPkandFPk

     . (8)

   • Recallk

     (Sensitivityk

     ) = TPk (9)

     TPkandFNk

   • F1k

     = 2PrecisionkRecallk

     PrecisionkandRecallk

     . (10)

   • Specificityk

   = TNk [24] (11)

   TNkandFPk

   The overall accuracy was 47.5% (94 out of 200). Table 1.1 presents per-class metrics derived from the confusion matrix.

   1. Per-class performance (Precision / Recall-Sensitivity / F1 / Specificity)

      Table 1.2. Per-class classification performance derived from the confusion matrix (N = 200). Precision, recall (sensitivity), F1-score, and specificity are reported for each class (AC: acute pain, CP: chronic pain, MW: muscle weakness, NP: neuropathic pain, ND: no disease, STI: soft tissue injury). Metrics are computed using a one-vs-rest formulation; class imbalance (e.g., STI

      = 4 samples) may influence the stability of minority-class estimates.

      Class	Precision	Recall (Sensitivity)	F1-score	Specificity
      AC	46.88%	42.86%	44.78%	89.76%
      CP	45.00%	45.00%	45.00%	86.34%
      MW	36.36%	40.00%	38.10%	96.34%
      NP	66.23%	53.68%	59.30%	75.47%
      ND	22.22%	35.29%	27.27%	88.59%
      STI	7.14%	25.00%	11.11%	93.40%

      The detailed per class performance metrics including precision, recall, F1 score, and specificity are shown in table 1.2.

3. Results and Discussion

   The model attained an overall accuracy of 47.26% on the synthetic dataset, with 95 correct predictions out of 201. Per-class analysis indicates that neuropathic pain (NP) exhibits the highest performance, with a precision of 66.23% and a recall of 53.68%, implying that the DN4/nerve- injury indicators offer relatively specific decision-making cues within the rule framework. Conversely, no-disease (ND) and soft tissue injury (STI) exhibit diminished precision and recall, signifying considerable ambiguity with clinical classifications. This phenomenon is anticipated in deterministic threshold systems when clinical traits intersect (e.g., pain intensity and diminished range of motion might fulfill several condition criteria) and when minority classes are under- represented (STI comprises only four samples), hence exacerbating instability in class-specific estimations.[24]

   The current validation used synthetic data; thus, these findings should be regarded as a proof-of-concept verification of rule execution, precedence order, and Simulink implementation, rather than as an assessment of clinical diagnostic performance. Subsequent efforts will concentrate on enhancing thresholds and conflict resolution through physician feedback, augmenting the representation of minority classes, and validating the model using real patient datasets with doctor-verified labels.

   Model performance was assessed using classification accuracy and a confusion matrix. The overall accuracy of the model was found to be 47.5 %, indicating reliable agreement between the predicted disease class and the reference labels.

   Fig. 1.3 Confusion matrix of disease classification results obtained using the proposed model.

   The confusion matrix demonstrates that the model correctly identifies most disease categories, with minimal misclassification observed between clinically overlapping conditions such as acute and chronic pain. Distinct conditions such as neuropathic pain and muscle weakness show high classification consistency due to well-defined diagnostic thresholds. This disease categories are shown with the confusion matrix showing performance classification across all classes in fig 1.3.

   These results confirm that the proposed mathematical model can effectively classify electrotherapy-relevant diseases using symptom-based rule logic and is suitable for decision- support applications in rehabilitation and electrotherapy planning. The current study focuses on validating rule logic and system implementation rather than clinical diagnostic performance.

4. Conclusion, Limitations, and Future Work

This article introduced an interpretable rule-based mathematical model executed in MATLAB/Simulink to identify electrotherapy-related clinical conditions utilizing validated clinical inputs, including VAS, symptom duration, DN4 score, muscle strength, ROM, , and age. The model offers clear decision equations that are apparent and verifiable, facilitating the initial construction of decision-support systems for electrotherapy applications.

Limitations: The model was verified using synthetic data instead of actual patient records, and deterministic threshold criteria may result in misclassification when clinical symptoms intersect

among illnesses. The current evaluation presents overall accuracy and confusion matrix results; however, further per-class metric reporting is necessary for a comprehensive diagnostic assessment.

Future research will concentrate on authentic clinical validation utilizing clinician-verified labels, enhancement of rule precedence and threshold margins, extension to further electrotherapy- related conditions, and investigation of hybrid methodologies (rule-based and data-driven) upon the availability of adequate clinical data.

Funding:

The authors received no financial support for the research, authorship, and publication of this article.

Clinical trial number: Not applicable.

Clinical Trial Registration: This study does not involve human participants or real patient data. The dataset used in this study is synthetic and generated for model validation purposes. Therefore, Clinical Trial Registration was not Applicable.

Conflicts of Interest:

The authors declare that there is no conflict of interest regarding the publication of this paper.

Declaration:

Ethics and Consent to Participate declarations: not applicable

References

1. C. Abbafati et al., Global burden of 369diseases and injuries in 204 countries and territories, 19902019: a systematic analysis for the Global Burden of Disease Study 2019, The Lancet, vol. 396, no. 10258, pp. 12041222, Oct. 2020, doi: 10.1016/S0140-

   6736(20)30925-9.

2. C. Chen et al., Global years lived with disability for musculoskeletal disorders in adults 70 Years and older from 1990 to 2019, and projections to 2040, Heliyon, vol. 10, no. 15, Aug. 2024, doi: 10.1016/j.heliyon.2024.e35026.

3. J. D. Loeser and R. D. Treede, The Kyoto protocol of IASP Basic Pain Terminology, Pain, vol. 137, no. 3, pp. 473477, Jul. 2008, doi: 10.1016/j.pain.2008.04.025.

4. G. A. Hawker, S. Mian, T. Kendzerska, and M. French, Measures of adult pain: Visual Analog Scale for Pain (VAS Pain), Numeric Rating Scale for Pain (NRS Pain), McGill Pain Questionnaire (MPQ), Short-Form McGill Pain Questionnaire (SF-MPQ), Chronic Pain Grade Scale (CPGS), Short Form-36 Bodily Pain Scale (S, Arthritis Care Res. (Hoboken)., vol. 63 Suppl 11, no. SUPPL. 11, Nov. 2011, doi: 10.1002/acr.20543.

5. E. L. Nussbaum, P. Houghton, J. Anthony, S. Rennie, B. L. Shay, and A. M. Hoens, Neuromuscular electrical stimulation for treatment of muscle impairment: Critical review and recommendations for clinical practice, Physiotherapy Canada, vol. 69, no. 5 Special Issue, pp. 176, 2017, doi: 10.3138/ptc.2015-88.

6. D. J. Hunter and S. Bierma-Zeinstra, Osteoarthritis, The Lancet, vol. 393, no. 10182, pp. 17451759, Apr. 2019, doi: 10.1016/S0140-6736(19)30417-9.

7. Overview | Osteoarthritis in over 16s: diagnosis and management | Guidance | NICE.

8. I. Kononenko, Machine learning for medical diagnosis: history, state of the art and perspective, Artif. Intell. Med., vol. 23, no. 1, pp. 89109, Aug. 2001, doi: 10.1016/

   S0933-3657(01)00077-X.

9. H. Breivik et al., Assessment of pain, Br. J. Anaesth., vol. 101, no. 1, pp. 1724, 2008, doi: 10.1093/bja/aen103.

10. R. D. Treede et al., Chronic pain as a symptom or a disease: the IASP Classification of Chronic Pain for the International Classification of Diseases (ICD-11), Pain, vol. 160, no. 1, pp. 1927, Jan. 2019, doi: 10.1097/J.PAIN.0000000000001384.

11. D. Bouhassira et al., Comparison of pain syndromes associated with nervous or somatic lesions and development of a new neuropathic pain diagnostic questionnaire (DN4), Pain, vol. 114, no. 12, pp. 2936, 2005, doi: 10.1016/J.PAIN.2004.12.010.

12. A. C. Thomas and J. E. Stevens-Lapsley, Importance of Attenuating Quadriceps Activation Deficits after Total Knee Arthroplasty, Exerc. Sport Sci. Rev., vol. 40, no. 2, p. 95, Apr. 2012, doi: 10.1097/JES.0b013e31824a732b.

13. H. P. Lundsgaarde, Evaluating medical expert systems, Soc. Sci. Med., vol. 24, no. 10,

    pp. 805819, Jan. 1987, doi: 10.1016/0277-9536(87)90182-1.

14. F. Magrabi et al., Artificial Intelligence in Clinical Decision Support: Challenges for Evaluating AI and Practical Implications A Position Paper from the IMIA Technology Assessment & Quality Development in Health Informatics Working Group and the EFMI Working Group for As, Yearb. Med. Inform., vol. 28, no. 1, pp. 128134, Aug. 2019, doi: 10.1055/s-0039-1677903.

15. L. A. Zadeh, Fuzzy logic and approximate reasoning – In memory of Grigore Moisil, Synthese, vol. 30, no. 34, pp. 407428, Sep. 1975, doi: 10.1007/BF00485052.

16. Y. Lecun, Y. Bengio, and G. Hinton, Deep learning, Nature, vol. 521, no. 7553, pp. 436 444, May 2015, doi: 10.1038/nature14539.

17. S. S. Haykin, Neural networks and learning systems for biomedical applications, IEEE Signal Processing Magazine, vol. 27, no. 4, pp. 1423, 2010. – Google Search. Accessed: Feb. 14, 2026. [Online]. Available:

    https://www.google.com/search?sca_esv=40662ef7c8bae166&sxsrf=ANbL- n6mNpr_ahyy98ThB0Dupn6y_- ndhw:1771051891364&q=S.+S.+Haykin,+Neural+networks+and+learning+systems+for+ biomedical+applications,+IEEE+Signal+Processing+Magazine,+vol.+27,+no.+4,+pp.+14

    %E2%80%9323,+2010.&sa=X&ved=2ahUKEwiP9fO2stiSAxV1cfUHHbc9HSEQgwN6 BAgSEAE&biw=1536&bih=730&dpr=1.25

18. A. Subasi, Application of adaptive neuro-fuzzy inference system for epileptic seizure detection using wavelet feature extraction, Comput. Biol. Med., vol. 37, no. 2, pp. 227 244, Feb. 2007, doi: 10.1016/j.compbiomed.2005.12.003.

19. M. M. ElMeligie et al., Neuropathic Pain Relief Through Transcutaneous Electrical Neuromuscular Stimulation: Insights From a Systematic Review and MetaAnalysis of Clinical Evidence, Biomed Res. Int., vol. 2025, no. 1, p. 5328365, 2025, doi: 10.1155/BMRI/5328365.

20. G. A. Hawker, S. Mian, T. Kendzerska, and M. French, Measures of adult pain: Visual Analog Scale for Pain (VAS Pain), Numeric Rating Scale for Pain (NRS Pain), McGill Pain Questionnaire (MPQ), Short-Form McGill Pain Questionnaire (SF-MPQ), Chronic Pain Grade Scale (CPGS), Short Form-36 Bodily Pain Scale (S, Arthritis Care Res. (Hoboken)., vol. 63 Suppl 11, no. SUPPL. 11, Nov. 2011, doi: 10.1002/ACR.20543.

21. E. G. VanDenKerkhof, E. G. Mann, N. Torrance, B. H. Smith, A. Johnson, and I. Gilron, An Epidemiological Study of Neuropathic Pain Symptoms in Canadian Adults, Pain Res. Manag., vol. 2016, p. 9815750, 2016, doi: 10.1155/2016/9815750.

22. A. Chiarotto, L. J. Maxwell, R. W. Ostelo, M. Boers, P. Tugwell, and C. B. Terwee, Measurement Properties of Visual Analogue Scale, Numeric Rating Scale, and Pain Severity Subscale of the Brief Pain Inventory in Patients With Low Back Pain: A Systematic Review, J. Pain, vol. 20, no. 3, pp. 245263, Mar. 2019, doi: 10.1016/J.JPAIN.2018.07.009.

23. D. A. Lake, Neuromuscular electrical stimulation. An overview and its application in the treatment of sports injuries, Sports Med., vol. 13, no. 5, pp. 320336, 1992, doi: 10.2165/00007256-199213050-00003.

24. D. Chicco and G. Jurman, The advantages of the Matthews correlation coefficient (MCC) over F1 score and accuracy in binary classification evaluation, BMC Genomics, vol. 21, no. 1, Jan. 2020, doi: 10.1186/s12864-019-6413-7.