Recurrent laryngeal nerve (RLN) monitoring is very important procedure during the neck surgery. For this purpose, special neuro monitors are used. They work based on the principle of surgical wound tissues stimulation and estimation of results of such stimulation [1][2][3][4][5]. The main problem that arises during this process is the proper choice of stimulation methods. In [1] and [4], the latest results of researches related to RLN neuromonitoring are represented.
Other electrophysiological method of RLN stimulation and monitoring requires the alternating current with fixed frequency [7]. The methods and mathematical models for solving this problem are considered in [6][7].
The main problem of the mentioned methods is a high risk of RLN damage. This risk is mainly caused by the accuracy of output information signal (result of surgery wound tissues stimulation) processing. Methods of spectral analysis of this signal [7] give an opportunity to select the main spectral component and to classify the surgery wound tissues. Methods of RLN location visualisation based on building the model of distribution of amplitude of the information signal on the surgery wound allow to detect the high-risky area [7]. The combination of these method in one information technology of processing of signal (reaction on the surgery wound tissues stimulation) is an important task. The paper is dedicated to this task.
The method of RLN monitoring is based on the task of its stimulation as the first sub-task. Other aspect of the task is the processing of reaction on RLN stimulation. After the processing, a conclusion about the RLN location in the surgery area is made.
Let's consider the functionality principles of existing hardware solution for RLN identification. The scheme of the device is shown in Fig. 1 [7].
1 is respiratory tube, 2 is larynx, 3 is sound sensor, 4 are vocal cords, 5 is probe, 6 is surgical wound, 7 is multifunctional block for RLN stimulation and visualize of RLN Flow of air that passes through patient's larynx, is modulated by stretched vocal cords. The result is registered by voice sensor 3. Obtained signal is amplified and is processed by single-board computer 7.
For processing of obtained signal, special software is installed on a single-board computer. The main functions of the software are:
-segmentation of information signal based on analysis of its amplitude; -analysis of amplitude spectrum using Fouriertransform;
-calculation of a spectral component with a maximal amplitude (let's call it "the main spectral component" further); -classification of tissues of surgical environment at the points of stimulation using threshold method.
Software for changing the frequency of RLN stimulation is written in programming language Node.js. Node was created by Ryan Dahl. Now Node.js is a trademark of Joyent, Inc. and is used with its permission and maintained by the Node.js Foundation [8][9]. Node is open-source platform and located on Github.
Block of processing and displaying information is developed based on single-board computer Raspberry PΡ 3 [10].
Raspberry Pi 3 was selected because of two upgrades made to it. The first one is a next generation Quad Core Broadcom BCM2837 64-bit ARMv8 processor making the processor speed increased from 900 MHz in the Pi 2 to up to 1.2GHz in the Pi 3. The second one is addition of the built-in BCM43143 Wifi chip. There is also Bluetooth Low Energy (BLE) module implemented on the board.
At the same time, such an approach does not ensure a significant decrease in the RLN damage risk. For detection of high-risky area of surgery, it is expedient to build a mathematical model for RLN identification.
For these purposes, we use the properties of surgical wound tissues which are characterized by a different reaction on the stimulation by alternating current with a fixed frequency. In Fig. 2, the fragments of amplified information signal obtained from the sound sensor and fragments of their spectral characteristics are shown. We can see the result of stimulation of the muscle tissue at a distance of more than 1 cm from RLN, with a specific blurred spectrum, without a clearly distinguished main spectral component in Fig. 2 a). Fig. 2 b) reflects the result of stimulation of the muscle tissue at a distance of not more than 3 mm, with a specific distinguished main spectral component with a small amplitude value. Finally, the result of RLN stimulation with a specific main spectral component with a sufficiently high normalized amplitude (6 times higher than in the previous case) is illustrated in Fig. 2 c).
The described properties give a possibility to develop the method, mathematical model and to improve the RLN identification technology in general.
Let's represent obtained set of stimulated points in such form: The solution of the obtained system is a vector of unknown parameters of IDDM g ο² . Methods of solving of this system are described in the papers [11][12].
The obtained model in the form of IDDM ( 2) is graphically represented in Fig. 3.
As we see in Fig. 3, the distribution of values of amplitude of the main spectral component of signals as the reaction on RLN stimulation and surgical wound tissues stimulation is marked with grey color. The RLN location is marked by black colored line on the plane.
For RLN location identification in the surgery area during surgery on neck organs, we improved the information technology. This technology consists of 4 main steps, a detailed description of which is given below. The scheme of the information technology is shown in Fig. 4. (as a result of stimulation of surgical wound of tissues) processing.
Step
At this step, the multifunctional block for surgical wound tissues stimulation is used. The result of stimulation is recorded by the sound sensor. The obtained information signal is digitized and processed by the single-board computer Raspberry Pi 3. The stimulation results are processed in a real-time mode. Also, they may be stored on the external data storages.
Step 2. Segmentation of information signal. This step is needed to highlight the patient reaction on tissues stimulation from sound signal taken during inhale or exhale.
In Fig. 4, the information signal is displayed and the segmentation principle is visually represented. As we can see from the Fig. 2, there are two segments. Each of them is result of the stimulation of RLN. The intervals between the segments represent delay of a patient's breath. These intervals are not informative for tissue classification method and named as the noise.
Unlike existing method of segmentation "synchronization with the appearance of the current of stimulation", main approach of automatic segmentation is based on the principle of threshold choice of an informative segment. If this energy exceeds the threshold, then, this is the beginning of the segment: πΈπΈ β₯ πΈπΈ π‘π‘π‘π‘ , then, π’π’ π π π‘π‘π π π‘π‘π‘π‘ = π’π’ ππ .
(5) If the energy of n counts is less than the threshold, then, this is the end of the segment: πΈπΈ β€ πΈπΈ π‘π‘π‘π‘ , then, π’π’ π π π‘π‘πππ π = π’π’ ππ . (6) So, the resulting segment consists of a set of countdowns: ππ = οΏ½π’π’ ππ β οΏ½π’π’ π π π‘π‘π π π‘π‘π‘π‘ ; π’π’ π π π‘π‘πππ π οΏ½οΏ½, (7) where οΏ½π’π’ π π π‘π‘π π π‘π‘π‘π‘ ; π’π’ π π π‘π‘πππ π οΏ½ is interval of countdowns of determined signal.
Step 3. Selecting of the main spectral component.
After segmentation of information signal recorded by sound sensor, it is necessary to select the main spectral component of the information signal. At first, the Fourier transform is used for this.
As it is shown in Fig. 2, the amplitude of the main spectral component depends on the distance from the stimulation point to the RLN. Therefore, the amplitude of the selected spectral component is in inverse proportion to this distance.
The obtained at this stage data array is represented in the form (1).
Step 4. Modeling of distribution of the main spectral component amplitudes.
A mathematical model for RLN identification is considered as an interval discrete dynamic model that is, the difference scheme in form (2). For its identification, the method of structural and parametric identification based on the behavioral model of artificial bee colony [13][14] is used. Behavioral model of artificial bee colony imitates the foraging behavior of the honeybee colony [13].
The application of this IDDM structural identification method involves the implementation of activity phases of all types of bees in the colony: onlooker bees, employed bees and scout bees. Let's consider all stages of implementation of IDDM structural identification method in more details. The result, at this stage, is the model that graphically represented in Fig. 3. Probable RLN location is illustrated by black line.
The test examples with sample of over 1500 stimulation points for different patients showed that using presented technology allow to decrease the RLN damage risk from 20% to 14%. So, the proposed methods and tools for electrophysiological RLN monitoring have a good perspective of development and application.
The proposed methods and tools for electrophysiological RLN monitoring and identification are realized using single-board computer Raspberry Pi 3. Applying of this methods for previously obtained sample of signals (reaction on stimulation of surgical wound tissue) showed good results. In particular, it was shown that for sample of over 1500 stimulation points for different patients using this technology, the risk of RLN damage is decreased from 20% to 14%.
This research was supported by National Grant of Ministry of Education and Science of Ukraine "Mathematical tools and software for classification of tissues in surgical wound during surgery on the neck organs" (0117U000410).