Interval Non-linear Model of Information Signal Characteristics Distribution for Detection of Recurrent Laryngeal Nerve during Thyroid Surgery Mykola Dyvaka, Volodymyr Manzhulaa, Yulia Trufanovaa a West Ukrainian National University, 11 Lvivska Str., Ternopil, 46009, Ukraine Abstract The work provides an analysis of known methods and technical means of identifying and visualizing the recurrent laryngeal nerve during thyroid surgery. There's proposed a method of building a non-linear model for detecting the location of the laryngeal nerve in the area of thyroid surgery. It's based on the characteristics of the information signal from the preliminary stimulation of the tissues of the surgical wound with an alternating current of a fixed frequency and the subsequent construction of the distribution function of the response to stimulation. The proposed method simplifies the procedures for identifying of the interval model, in particular, due to the analytical representation of the objective function of the optimization problem of their identification, and, accordingly, reduces the time spent on building the non-linear models based on interval data. Based on real experimental data obtained during thyroid surgery an interval non-linear model was built, which enables detection and visualization of the location of the laryngeal nerve in the thyroid surgery area and, accordingly, reducing the risk damage of its. Keywords 1 Thyroid surgery, recurrent laryngeal nerve, information signal amplitude, interval data, interval non-linear model, model identification, optimization problem 1. Introduction The main problem when conducting thyroid surgery is the identification of the recurrent laryngeal nerve, the damage of which leads to the patient losing his voice, as well as to other negative consequences related to the functioning of the human respiratory system [1, 2]. As a rule, the means used during such surgery make it impossible to visualize the process of identifying the laryngeal nerve, also it are based on the dangerous procedure of introducing the patient to the third stage of anesthesia, where there is a high risk of transition to a state of clinical death [3, 4]. The process of visualizing the laryngeal nerve is extremely complex and includes the procedure for its detection [5, 6, 7]. The analysis of known technical means of detecting the recurrent laryngeal nerve during thyroid surgery made it possible to establish the general principle of their surgery, which is based on stimulation with a constant electric current in the area of surgery and evaluating the results of this s on the vocal cords If the area of stimulation includes the recurrent laryngeal nerve, the vocal cords are shortened, but if the stimulation is done on the muscle tissue, the reaction to the stimulation will be insignificant. The basis of the method of identification of the laryngeal nerve from other tissues of the area thyroid surgery than proposed by the authors [8, 9, 10] is to ensure the accuracy of detection and visualization of the location of the laryngeal nerve in the surgical wound. The task is solved by the fact that the tissue stimulation in the area thyroid surgery is carried out by an alternating current of a fixed frequency which provides a low conductivity of the electrical signal through the IDDM-2022: 5th International Conference on Informatics & Data-Driven Medicine, November 18–20, 2022, Lyon, France EMAIL: mdy@wunu.edu.ua (A. 1); v.manzhula@wunu.edu.ua (A. 2); yu.trufanova@wunu.edu.ua (A. 3) ORCID: 0000-0002-9049-4993 (A. 1); 0000-0001-5222-8443 (A. 2); 2022 Copyright for this paper by its authors. Use permitted under Creative Commons License Attribution 4.0 International (CC BY 4.0). CEUR Workshop Proceedings (CEUR-WS.org) muscle tissues and a high conductivity of the electrical signal through the laryngeal nerve and muscles that control the tension of the vocal cords, followed by registration of the contraction of the vocal cords at a given frequency by a sound sensor installed in the breathing tube placed in the patient's larynx, followed by its conversion into an electrical signal, and the output information signal, which characterizes the proximity to the laryngeal nerve, is determined by the change in the amplitude of the electric current of a given frequency [8]. To visualize the location of the laryngeal nerve in the surgical wound, information signal processing tools are used [10]. A signal processing software module includes filtering the signal at the excitation frequency, determining the maximum amplitude of the filtered signal for each observation, and recording the received data in interval form, taking into account errors of various nature. Moreover, the measurement of the interval value of the information signal amplitude is carried out according to the coordinates on the thyroid surgery area, which are fixed on a sterile grid, respectively, placed on the wound. In papers [9, 10], the authors have proposed methods for constructing an interval models that describe the maximum amplitude of the information signal depending on the coordinates on the surgical wound, based on tolerance and guaranteed interval or ellipsoidal estimates of the parameters of the algebraic expression. However, the computational complexity of implementing these methods complicates online-visualization of the location of the recurrent laryngeal nerve during thyroid surgery. Therefore, the actual task is to develop a method for identifying the model of the information signal characteristics distribution with minimal computational costs for visualizing the location of the laryngeal nerve in the thyroid surgery area. 2. Method for building an interval model of the distribution of information signal characteristics for the recurrent laryngeal nerve’s detection in the thyroid surgery area The process of building mathematical models includes solving two problems: structural and parametric identification [11]. At the same time, the task of identification the model structure is more difficult and primary, since it’s necessary to first define the basic functions, generate the structure of the model, and then calculate parameter estimates for selecting the optimal or "best". The most effective methods of structural identification of interval models are based on self-adaptation and self- organization procedures by analogy with the behavioral models of a bee colony. Complex optimization problems are solved for this [12, 13, 14, 15]. The distribution of the information signal characteristics in the thyroid surgery area we'll be described by interval mathematical models in the form of a nonlinear algebraic expression. Then we'll search for the resulting information signal characteristic (maximum amplitude) in the non-linear algebraic expression form of following kind: (1) where – is a set of unknown basis functions (of a known class), and the basis functions relate to unknown model parameters and the result of experimental measurements obtained in interval form: (2) where – is number of measurements. Let's set the conditions for the consistency of the model with experimental interval data as it's customary in interval analysis: (3) where – means the true value of the information signal output characteristic for a fixed model structure and for fixed input variables' values . Parameter values of the model only remain unknown in this case. Taking into account conditions (3) and replacing with expression (1) for fixed values of the input variables , we'll obtain the following system of interval non-linear algebraic equations (ISNAE): (4) Thus, the general form of the parametric identification problem of the interval model of the information signal characteristics when it's distributed to the thyroid surgery area in the interval system of non-linear algebraic equations form has been obtained. As is known [11], the solutions of this system they’re obtained as a result of the implementation of an iterative procedure at each iteration of its they calculate the function "quality" of estimation of mathematical model parameters. Accordingly, there’s the task of structural identification of the interval model of the distribution of information signal characteristics as a task of repeatedly solving the problems of parametric identification of this model. Let's assume that the solution of ISNAE (4) is obtained in the form of value intervals of model parameter estimates . Let's substitute the obtained interval estimates into expression (4) with the fixed values of the input variables (at the points of the experiment). Because of these substitutions, we'll get estimates of the information signal output characteristic in the interval form: . (5) Thus, the compatibility of ISNAE (4) means that the intervals of the values of the predictive characteristics of the information signal at the points of experimental measurements including to the intervals obtained experimentally, that is, if the following conditions are satisfied: (6) Stating the fact that the structural identification problem of interval models of an information signal characteristics is a problem of repeatedly solving parametric identification problems of this model, and therefore from a computational point of view, it is NP complete. The complexity of the problem related to the complexity of the objective function, which is given algorithmically, is discrete and does not have an analytical representation, therefore it complicates the calculation [11]. At the same time, in the vast majority of problems of both structural and parametric identification of mathematical models, the criterion of minimizing the mean square deviation is used. On the other hand-side, it's mostly sufficient to find at least one model even with the interval formulation of the problem in the sense of solving ISNAE (4). In this case, the interval model (5) will be able written in the following form: . (7) The task of identifying the interval model of the information signal characteristic distribution in the area of surgery we'll formulate in the optimization problem form: (8) , where (9) where – is set of potential model’s structural elements; – is the parameters vector components of the sth model; – is set minimum and maximum value for each model's parameter; – is the parameters number of the interval model. There's what the smaller value that the "better" structure of the interval model. If the equality is fulfilled (10) then the structure is guaranteed to allow building an adequate interval model of the information signal characteristic distribution, since the existence of the ISNAE (4) solution means that the condition (6) is satisfied, which in this case will have the following form: , (11) and it’s equivalent condition (12) since expression (12) is always a linear combination of limits of experimental values at measurement points . The advantage of using the objective function in the form (9) is that it's in an analytical form and quadratic at least relative to the coefficients , . Thus, the task of model's parametric identification for a fixed structure is the search for the optimization problem solution: (13) To calculate the parameters of the interval model based on the optimization problem (13) and the given structure should be used nonlinear optimization methods, such as the Gradient Descent method, the Newton method or a combination of theirs [17, 18, 19]. The implementation of structural identification consists in selecting the structure of the interval model by reducing or increasing structural elements [20]. 3. The interval non-linear model of the information signal characteristics distribution for the detection of the recurrent laryngeal nerve during thyroid surgery. There were carried out the construction of an interval non-linear model of the characteristics of the information signal distribution in the area thyroid surgery based on the developed method. Experimental measurements on a sterile grid in the area of surgical intervention we’ve carried out based on of two coordinates: (14) The data were obtained in interval form based on information signal processing taking into account measurement errors and noises are presented in Table 1. Detailed analysis of the data in the Table 1 showed that the structure of an adequate model of the maximum amplitudes of the information signal distribution in the thyroid surgery area should be to search with the inclusion of trigonometric basic functions. To reduce the number of such structural elements we’ve added for the parameters in the power function form . Accordingly, a set of potential structural elements we’ve formed in this form: (15) In the process of selecting and increasing the model structure with elements from the set F, we've obtained a model structure based on the convolution of the following form: (16) Since the value of the objective function of the optimization problem (13) for this model’s structure is close to zero (please, check Figure 1) and the condition (12) is satisfied, therefore, we obtained the optimal solution based on it. Table 1 Results of experimental measurements of information signal characteristics Measurement Measurement coordinates, Interval value of the maximum signal number amplitude 1 1 1 8,0974 11,5326 2 1 2 9,5576 13,6124 3 1 3 13,0391 18,5709 4 1 4 15,2955 21,7845 5 1 5 19,8619 28,2881 6 1 6 25,6492 36,5308 7 2 1 9,0626 12,9074 8 2 2 14,421 20,539 9 2 3 21,7099 30,9201 10 2 4 28,5656 40,6844 11 2 5 33,8498 48,2103 12 2 6 43,032 61,288 13 3 1 10,1516 14,4584 14 3 2 18,1335 25,8265 15 3 3 31,5892 44,9907 16 3 4 37,8716 53,9384 17 3 5 47,8912 68,2087 18 3 6 50,7004 72,2096 19 4 1 13,1134 18,6766 20 4 2 23,3475 33,2525 21 4 3 43,1516 61,4584 22 4 4 47,4375 67,5625 23 4 5 49,5001 70,4985 24 4 6 43,3125 61,6875 25 5 1 13,2021 18,8001 26 5 2 32,1750 45,825 27 5 3 43,3125 61,6875 28 5 4 47,0250 66,975 29 5 5 44,5501 63,4495 30 5 6 26,8125 38,1875 31 6 1 16,5001 23,4996 32 6 2 37,5375 53,4625 33 6 3 46,4063 66,0938 34 6 4 42,0750 59,925 35 6 5 24,3375 34,6625 36 6 6 16,9125 24,0875 Based on calculated parameter estimates the model of the information signal amplitude distribution on the thyroid surgery area in this form was constructed: . (17) Figure 1: The objective function value in the process of calculating parameter estimates and coefficients for the resulting model There’re given the predictive values of the information signal amplitude that it's obtained based on the constructed model and, accordingly, calculated coefficients in process solving optimization problem (13) for parameter estimates of the resulting model in Table 2. Table 2 The results of the interval model constructing of the maximum information signal amplitude distribution Measurement Predictive Coefficient Measurement Predictive Coefficient number, amplitude estimates, number, amplitude estimates, value, value, 1 10,0373 0,4354 19 14,2191 0,41 2 12,8713 0,1975 20 29,1968 0,7892 3 16,3145 0,4089 21 47,0182 0,3897 4 20,3638 0,2261 22 59,7198 0,4161 5 24,9817 0,3916 23 61,7681 0,464 6 30,1031 0,5902 24 53,1629 0,3642 7 10,823 0,5447 25 16,7887 0,3171 8 16,2673 0,6984 26 37,2303 0,3632 9 23,9583 0,7539 27 55,8709 0,8519 10 33,2932 0,609 28 59,7198 0,7099 11 43,3751 0,3362 29 47,348 0,5194 12 53,1629 0,4447 30 30,1031 0,5332 13 12,218 0,5094 31 19,8618 0,353 14 21,9101 0,7275 32 44,9699 0,6086 15 35,2319 0,3041 33 59,1501 0,6397 16 49,0664 0,4059 34 49,0664 0,5615 17 59,9674 0,3207 35 28,0548 0,3171 18 65,2948 0,7971 36 20,0517 0,3632 Figure 2 shows the graphs of the experimental interval corridor of the information signal amplitude and the predictive values that are obtained based on the model. The presented graphs demonstrate the inclusion of predictive values in the experimental corridor at each measurement point that satisfy condition (11) and indicates the adequacy of the constructed model. So, constructed the interval non-linear model based on of real experimental data that were obtained during thyroid surgery will be able used to detect the placement of the laryngeal nerve in the thyroid surgery area and, accordingly, reduce the risk damage of its. Figure 3 shows the 2- and 3-dimensional visualization of the maximum amplitude distribution over the surgical area, which demonstrates the possible placement of the laryngeal nerve. Figure 2: Interval values of experimental measurements and model-based predictive values of the information signal amplitude 4. Conclusions There were proposed based on the known method and technical means of the recurrent laryngeal nerve detecting during thyroid surgery the method and the non-linear model for predicting the information signal characteristics in order to detect the laryngeal nerve. The method of building the non-linear models was created to detect the laryngeal nerve location in the area of surgical intervention based on the characteristics of the information signal. Signal was gotten from the previous excitation of the tissues of the surgical wound with an alternating current of a fixed frequency and the subsequent construction of the distribution function of the response to excitation. The proposed method simplifies and, accordingly, reduces the time spent on building a non-linear model based on interval data, in particular, due to the analytical representation of the objective function of the optimization problem of identification model. Based on real experimental data of thyroid surgery the interval non-linear model was built it made possible to identify the placement of the laryngeal nerve in the thyroid surgery area, accordingly, reducing the risk damage of its. a) b) Figure 3: Visualization of the information signal amplitude distribution based on the constructed model: a) two-dimensional image, b) 3d image. 5. References [1] V. O. Shidlovskyi, O. V. Shidlovskyi and A. M. Dyvak, Safety of surgical interventions on the thyroid gland, Clinical Endocrinology and Endocrine Surgery 3 (2019) 79–80. [2] V. O. Shidlovskyi, O. V. Shidlovskyi, I. M. Deikalo and V. M. Lipskyi, Standard and non- standard situations in thyroid surgery, Hospital Surgery 21–24. [3] V. O. Shidlovskyi and M. I. Sheremet, Modern approaches to surgical treatment of autoimmune thyroiditis, Endocrinology 4 (2014) 365–365. [4] F. Y. Chiang, I. C. Lu, H. C. Chen, H. Y. Chen, C. J. Tsai, K. W. Lee, P. J. Hsiao and C. W. Wu, Intraoperative neuromonitoring for early localization and identification of recurrent laryngeal nerve during thyroid surgery, Kaohsiung J Med Sci 26(12) (2010) 633–9. doi: 10.1016/S1607- 551X(10)70097-8. [5] V. I. Tymets, Methods, tools and information technologies for identifying the recurrent laryngeal nerve, Ph.D. thesis, West Ukrainian National University, Ternopil, 2020. [6] Yu. A. Gordievich, N. I. Padletska, A. V. Pukas and I. F. Voytiuk, Interface of the software system for experimental research of surgical wound tissues on neck organs, Proceedings of the VI Ukrainian school young scientists and students on Advanced Computer Information Technologies, ACIT’2016, Ternopil:TNEU, 2016, pp.116–118. [7] N. Ya. Savka and I. V. Gural, Improved information technology for monitoring the recurrent laryngeal nerve, Bulletin of the Vinnytsia Polytechnic Institute 6 (2019) 45–53. [8] G. W. Randolph, H. Dralle, Electrophysiologic recurrent laryngeal nerve monitoring during thyroid and parathyroid surgery: international standards guideline statement, Laryngoscope Suppl 1 (2011) 121S1-16. doi: 10.1002/lary.21119. PMID: 21181860. [9] M. P. Dyvak, A. V. Pukas, N. P. Parplitsia and A. M. Melnyk, Applied problems of structural and parametric identification of interval models of complex objects, Ternopil: Universitetska dumka, 2021. [10] M. Dyvak, A. Pukas and O. Kozak, Tolerance estimation of parameters set of models created on experimental data, Proceedings of the International Conference on Modern Problems of Radio Engineering, Telecommunications and Computer Science, TCSET’08, 2008, pp. 24–26. [11] M. P. Dyvak, Problems of mathematical modeling of static systems with interval data, Economichna dumka, 2011. [12] A. Abraham, R.K. Jatoth, A. Rajasekhar, Hybrid differential artificial bee colony algorithm, J. Comput. Theor. Nanosci 9(2) (2012) 249–257. [13] O. G. Moroz, V. S. Stepashko, Combinatorial algorithm of MGUA with genetic search of the model of optimal complexity, Proceedings of the International Conference on Intellectual Systems for Decision Making and Problems of Computational Intelligence, 2016, pp. 297–299. [14] B. Akay, D. Karaboga, A modified artificial bee colony algorithm for real-parameter optimization, Inf. Sci. 192 (2012) 120–142. https://doi.org/10.1016/j.ins.2010.07.015. [15] N. P. Porplitsia, Comparative analysis of the effectiveness of genetic and bee algorithms in the problem of structural identification of the interval difference operator, Information technologies and computer engineering 1 (2015) 55-67. [16] I. S. Oliynyk, Identification of parameters of interval models of static systems by methods of optimal planning of saturated experiments, Ph.D. thesis, Lviv Polytechnic National University, Lviv, 2018. [17] P. Venkataraman, Applied optimization with MATLAB programming, John Wiley & Sons, 2009. [18] Anders Forsgren, Philip E. Gill, Margaret H. Wright, Interior methods for nonlinear optimization, SIAM review 44.4 (2002) 525-597. doi: 10.1137/S0036144502414942. [19] A. BECK, Introduction to nonlinear optimization: Theory, algorithms, and applications with MATLAB, Society for Industrial and Applied Mathematics, 2014. doi: 10.1137/1.9781611973655. [20] O. G. Moroz and V. S. Stepashko, Testing the combinatorial genetic algorithm in the task of finding the optimal nonlinear model, Proceedings of the International Conference on Intellectual Systems for Decision Making and Problems of Computational Intelligence, 2018, p 256.