Squatting exercises are performed primarily for training the muscles of the hip and thigh. Biomechanical analysis indicates that squatting increases intra-abdominal pressure and recruits the pelvic floor muscles (PFMs). The PFMs in women have a role in urinary and fecal continence, support of pelvic organs, and sexual function. Therefore, understanding the activity of these muscles during exercise can prevent dysfunctions. One strategy to analyze the PFMs is using computational models based on fuzzy logic as decision-making tools. The objective of this original study was to develop a fuzzy logic model to assess the activity of the PFMs during squatting with a barbell

This cross-sectional study evaluated adult women. The study population was recruited from universities and gyms and was also referred by volunteers who participated in data collection. Women with self-reported orthopedic lesions or heart disease were excluded. Adult volunteers signed an informed consent form and answered a questionnaire. In addition, the volunteers underwent individualized perineal evaluation, a physical squatting test, electromyography (EMG) of the surface of the PFMs, and electrogoniometry of the knee during squatting with a barbell. Fuzzy models were developed using data collected from the volunteers. The models were analyzed using data from volunteers with and without symptoms of urinary incontinence. 
Computational model based on fuzzy logic
For constructing the model, studies that used fuzzy logic in the muscular system,.
The model was built by a specialist in women's health and the authors of this study. The fuzzy model of indicators of PFM activity during squatting with barbell was generated by defining input and output variables. The input variables were selected from the electromyographic records of the PFMs of ten full squats, physical evaluation of the strength of the PFMs by a physical therapist, and barbell weight used to perform the squat, which was determined after the 1RM test.

The variable mv PFM was created in the fuzzy model using the electromyographic records. The minimum and maximum values of this variable were determined using the minimum and maximum mv PFM values used by volunteers from the training group of the sample.

The data regarding the variable “muscle endurance” from the PERFECT classification were used in the variable FM PFM in the fuzzy model. The minimum and maximum values of this variable were based on literature data and expert experience. The value varied from 0 to 15 seconds, with 0 indicating absence of contraction and 15 indicating maximum contraction time.

The value obtained by the volunteers using 70% of the 1RM load was used to feed the variable "WeightAg" in the fuzzy model. The minimum and maximum values of this variable were determined using the minimum and maximum load used by volunteers of the training group of the sample.

The output variable was the “fuzzy PFM activity index” created by the authors to indicate the activity of the PFMs in the proposed squatting exercise. The minimum and maximum values of this variable ranged from 1 to 5, with 1 indicating poor activity and 5 indicating excellent activity.

Fuzzy sets and linguistic variables were associated with each input and output variable. The parameters of each fuzzy set (X and Y values) and their associated linguistic variables were elaborated using expert experience, considering the minimum and maximum values of each input and output variable.

Diffuse rules were defined by asking a specialist to identify the type of relationship between the input linguistic variables of each fuzzy set, indicate an output linguistic variable, and indicate how these variables were correlated with clinical practice. The correlation between the input linguistic variables could be established using the logical connectives “and” and “or” and the negation or noninterference of that variable. In this study, the rules were elaborated using the conditional types “if” (antecedent) and “then” (consequent). The correlation between the input linguistic variables was established using the logical connective “and.”

After these steps, fuzzy logic models were elaborated using the Fuzzy Logic Toolbox (The MathWorks, Inc., Natick, MA, USA) provided by the MATLAB software package, version 7.0. In this fuzzy model, the type of inference used was of the type Mamdani and centroid defuzzification.

The models could be modified according to changes in the pertinence functions and fuzzy rules. Models with the trapezoidal, triangular, and Gaussian functions and different amounts of fuzzy rules were elaborated. A function was created in the MATLAB environment to optimize data analysis and generate the output variable of the fuzzy models more quickly. The function was designated "calcfuz". This function read the input variables, processed these variables in the created fuzzy model, and generated the fuzzy output variable. The MATLAB toolbox was used to create this function.

2.5 Testing, validation, and system settings

The sample was separated into a training group and a test group during data collection to test the fuzzy models and choose the model that best described the PFM activity. The fuzzy model was chosen by using 70.0% of the sample to create the training group and 30.0% of the sample to create the test group. A draw was made using the random number generator of Microsoft Excel. The choice of the fuzzy model depended on its correlation with the ICIQ-SF questionnaire, which evaluates urinary symptoms.

A descriptive statistical analysis was conducted using the Shapiro-Wilk test, Student’s t-test, and Pearson correlation at a level of significance of 5%.

A total of 76 women were invited to participate, but only 37 subjects were included in the analysis. Eight fuzzy models were developed, and a fuzzy Mamdani model was selected. In addition, three input variables, one output variable, trapezoidal and triangular functions, “and" logic connectors, and centroid defuzzification were used. The mean PFM activity using the fuzzy logic was 2.45 ± 0.72 in the symptomatic group and 3.26 ± 0.78 in the asymptomatic group (p = 0.03). The output language variables were “poor,” “good,” and “excellent.” The fuzzy model quantified the activity of PFM muscles during squatting. There was a significant moderate correlation of the fuzzy model (I-MAP fuzzy) with the International Consultation on Incontinence Questionnaire–Short Form (ICIQ-SF) and smallest knee angle during squatting in the groups with and without symptoms of urinary incontinence.
The adopted model was Mamdani with “and” logic connectors and centroid defuzzification (Figure 1).

This original models based on expert experience may serve as educational tools and diagnostic and/or therapeutic decision support systems for health professionals. In fuzzy logic models, the decimal values ​​indicate the degree of pertinence between the groups.

The fuzzy model quantified the activity of the PFMs in women during squatting exercises. The model correlated the quantitative output variable with behaviors “poor,” “good,” and “excellent,” closing the gap between decision-making and the interventions performed by health professionals. Therefore, the model adequately provided indicators of the activity of the PFMs during squatting.

In addition, there were significant moderate correlations between the fuzzy model (I-PFM fuzzy) and both the ICIQ-SF and smallest knee angle during squatting in the groups with and without symptoms of urinary incontinence. A lower fuzzy value indicated a lower ability to squat and higher rate of incontinence symptoms among the study participants. This model may be used as a reference for developing tools for clinical applications.

LONGPRÉ, H. S.; ACKER, S. M.; MALY, M. R. Muscle activation and knee biomechanics during squatting and lunging after lower extremity fatigue in healthy young women. Journal of Electromyography and Kinesiology, v. 25, n. 1, p. 40–46, 2015.
POTA, M.; ESPOSITO, M.; DE PIETRO, G. Designing rule-based fuzzy systems for classification in medicine. Knowledge-Based Systems, v. 124, p. 105–132, 2017.
SUBASI, A. Medical decision support system for diagnosis of neuromuscular disorders using DWT and fuzzy support vector machines. Computers in Biology and Medicine, v. 42, n. 8, p. 806–815, 2012.