A Critical Care MonitoringSystem for Depth of Anaesthesia Analysis Based on Entropy Analysis and PhysiologicalInformation Database
Qin Wei1, Yang Li2,Shou-Zen Fan3,Quan Liu2, Maysam F. Abbod4,Cheng-Wei Lu5,6, Tzu-Yu Lin5,6,Kuo-Kuang Jen7,Shang-Ju Wu7,
Jiann-ShingShieh6,8,*
1School of Mechanical and Electronic Engineering, Wuhan University of Technology, Wuhan 430070, China; E-Mail: ;
2School of Information Engineering, Wuhan University of Technology, Wuhan 430070,China; E-Mail: ;
3Department of Anesthesiology, College of Medicine, National Taiwan University, Taipei 100, Taiwan; E-Mail: ;
4School of Engineering and Design, Brunel University, London UB8 3PH, UK;
E-Mail: ;
5Dept. of Anesthesiology, Far Eastern Memorial Hospital, Ban-Chiao, Taiwan; ; ;
6Department of Mechanical Engineering, Yuan Ze University, 135, Yuan-Tung Road, Chung-Li32003, Taiwan;
7Chung-Shan Institute of Science and Technology, Taoyuan, Longtan Taiwan; ; ;
8Center for Dynamical Biomarkers and Translational Medicine, National Central University, Chung-Li 32001, Taiwan.
*Author to whom correspondence should be addressed;E-Mail: ;
Tel.: +886-3-4638800 ext. 2470; Fax: +886-3-4558013.
Abstract
Diagnosis of depth of anaesthesia(DoA) plays an important role for treatment and drug usage in operating theatre and intensive care unit.With the flourishing development of analysis methods and monitoring devices for DoA, a small amount of physiological data had been stored and shared for further researches. In this paper, a critical care monitoring (CCM) system for DoA monitoring and analysis were designed and developed, which includes two main components: the DoA analysis subsystem and the physiologic information database. In the analysis of DoA, according to our previous studies on approximate entropy (ApEn), sample entropy (SampEn) and multi-scale entropy (MSE), the SampEn and MSE were integrated into the system for indicating the state of patients underwent surgeries at real time because of their stability. Moreover, a physiologic information database including biologic data and clinical information was constructed through a browser and sever (BS) model so as to provide a safe and open platform for storage, share and further study of clinical anaesthesia information. Therefore, this CCM system not only supplies the original biological data and information collected from operation room, but also shares our studies for improvement and creation on the research of DoA.
Key words:Depth of anaesthesia,approximate entropy,sample entropy,multi-scale entropy,physiologicinformation database
1.Introduction
Depth of anaesthesia (DoA) monitoring during surgery is a kind of protection for patients’ live and recovery, therefore a lot of methods with regard to detect the DoA were presented and developed for clinical application in operating theatre and intensive care unit (ICU), like Bispectral Index (BIS) [1-3], response entropy (RE) and state entropy (SE)[4,5], and so on.Both BIS and RE/SE indexes are acquired through the analysis of the electroencephalograph (EEG)detected from patients undergoing surgeries. In the operation room, apart from EEG, other physiological signals such as electrocardiograph (ECG), respiration (Resp), blood pressure (BP), pulse oxygen saturation (SpO2), there are also the important responses of patient for indication of DoA. However, the BIS Module in MP60 and Entropy Module in Datex-Ohmeda S/5 pays more attention to online analysis of DoA but neglect of the storage and sharing of the all physiologic data. In addition, the module and devices of BIS and Entropy are too expensive to be widely available to most of operation rooms and hospitals.Acritical care monitoring (CCM) system for DoA based on other entropy indexes and a physiologic database is presented in this paper, which include two main components: the DoA analysis subsystem and the physiologic information database (PID).
In our previous studies on DoA analysis, the entropy family plays an important role in measuring the complexity of complex time series. Especially, the approximate entropy (ApEn) is able to be an index of degree of conscious state or DoA of patients [6]. But ApEnis more adaptive to long-term time series compare to the sample entropy (SampEn), and is less sensitive to the change of complexity than SampEn [7]. Furthermore, the distribution of complexity in different time scales is capable of describing the different states of patients during operation by using multi-scale entropy (MSE) [8]. Therefore, the SampEnand MSE were integrated into the DoA analysis subsystem of the service system for DoA in order to detect the status of patients and provide the new indexes. Then, a PID was constructed to store and manage the data and information in operation. The structure of the PID is based on browser and server (BS) model with classicalthree-tier architecture, by which the cost can be greatly reduced in the load of client computer, system update, and routine maintenance. The service system provides the data access via the web browser, so thatthe PID has a small part of the business logic in the front (web browser) and the main business logic in the application server. Due to the nature of the model, it is convenient to manage and maintain the system, and is easy to access and share physiological data and information in PID for authorized users from different placesand through different ways,such as the LAN, the WAN, the Internet / Intranet, etc.
The paper is organized as follows. sectionII introduces simply the complexity analysis methods for DOA monitoring, such as ApEn, SampEn and MSE. The third sectiondescribes theCCM system for DoA in components, functions and examples. The practical application is illustrated in the fourthsection and the last sectionpresents the conclusion and discuss the future work.
2.Method for DOA Analysis
2.1Approximate Entropy
Entropy family as widely used methods in nonlinear dynamic analysis has already been applied in analysis of physiological time series. ApEnis one of themethods which has been employed in clinical cardiovascular studies [9-11] and irregularity of EEG signals under general anesthesia[12,13]. It is referred to as a measure of the complexity of the sequence and statistical quantitative rules [14], and its computation is as follows:
(1)
where N is the length of time series, a toleranceris the threshold to identify whether a selected sequence matches the given pattern in the same length m or not, and Фm(r)is the matching number of time series within parameters m and r.
But there are two shortcomings [15-17] in application of ApEn:
(1)The length of time series affects ApEn heavily, and it gets lower uniformly than expected for short data;
(2)It lacks of the relative consistency.
Thereby, the SampEnwas proposed to improve the ApEn, which does not count self-matches in the computation.
2.2Sample Entropy
Richman and Moorman [18] developed animprovement of the ApEn algorithm, named SampEn, inorder to remove some of the deficienciesmentioned above.Its parameters are the same as ApEn, so it is also expressed as SampEn(m, r, N). The differences with respect to ApEn are: 1) self-matchingis deleted; 2) the first N-m vectors of the length are only chosen in time series; and 3) the conditional probabilities are not estimated ina template manner. Thus, the probability measure is computeddirectly as the logarithm of conditional probability instead ofthe ratio of the logarithmic sums, which is described as follow:
(2)
Even though the application of ApEn to clinical recordings oftenobtains good results,SampEn is more suitable in real time analysis of physiological signals and acquisition of general information about the regularity and the persistence of those signals. In comparison among SEF95%, RE, SE and SampEn in DoA monitoring, firstly EEG data were collected from twenty patients undergoing urological surgery in Far Eastern Memorial Hospital of Taiwan. Age of patients ranges from 15 to 60 years. Theused anestheticagent in the operation is propofol. The device monitoring EEG is Datex-Ohmeda S/5TM Compact Anaesthesia Monitor, in which response entropy(RE) and state entropy (SE) are displayed as an index of DOA for anesthetists. The one compared result of twenty cases is depicted in Figure 1. On account of the urological surgery, the two stages of pre-operation and recovery werein a short time. In addition, the frontal electromyogram (FEMG) mixed with EEG in the two stages, so the SampEn is disturbed badly in these two stages. But it has the same trend with RE and SE and is better than SEF 95% during intubation and maintenance, shown in Figure 1. In order to decrease impact of noises and describe the state of patient during a short time, MSE is employed to give the pattern of complexity in different scale factors.
2.3Multi-scale Entropy
Costaet al. [19, 20] proposed the MSE to observe the complexity of physiologic time series in different time scales. Its calculation consists of two procedures: at first, a “coarse-graining” process is used to construct the new sequence of the entire time series in different time scales. For a given time series, multiple coarse-grained time series are constructed by averaging the data points within non-overlapping windows of increasing length τ. Each element of the coarse-grained time series,is calculated according to the equation:
(3)
where τ represents the scale factor and 1≤j≤N/τ. The length of each coarse-grained time series is N/τ. Moreover, the coarse-grained time series is original time series if τ= 1. For multiple coarse-grained time series, SampEn is secondly calculated for each of them to indicate the complexity in different scale factor τ, which presents the distribution of complexity from small to big time scale
3.A CCM System forDoA Analysis
A CCM system for DoA Analysis was developed to provide a platform for real time DoA analysis based on entropy family and the collection and share of physiological data and operation information. Its structure is shown as Figure 2. In the operation room, physiological data detected from the patient are transmitted from monitoring device to our DoA analysis subsystem for signal processing and collection. Then, the data and information of operation will be uploaded to the server and stored into PID through the internet for the available services supplied from application server.
3.1 DOA analysis subsystem
In Figure 2, the DoA analysis subsystem includes two functions in its program, one is online analysis using traditional methods, SampEn and MSE; the other one is offline analysis for anesthesiologists to have postoperative studies. Therefore,anesthesiologists could not only use indexes to diagnose the DOA while online analysis of EEG, but also process those recordings of signals and indexes according to their ideas after operation.
3.1.1 Online Analysis
For the online analysis of DoA, EEG signals detected at real time are calculated as complexity index of activity of central nervous system. Thanks to the relationship between N and m, the average sum of minimum 10m and maximum 30m(when m = 2) is chosen. Thereby, whenSampEn and MSE are computed online, the selected N is equal to the scale factorτmultiplied to 500. For example, as scale factorτis 6, the N is 3000. Due to sample rate of EEG in Datex-Ohmeda is 100 samples per second, thus, the computation of MSE from 1 to 6 needs EEG samples in 30 seconds collected at first. Under this condition, it is indispensable to use a sliding window for entropy index at every second.
In Figure 3, the main interface of this program is displayed, in which the communication, the signals transmitted from physiological monitor, the operation event and some indexes are integrated together. The RS232 communication function, selection of analyzed physiologic signal, input of operation events are located at upside of waveform display of detected signals and analyzed results. There are totally nine figures lined in three rows. On the first line, Figures 3(a-c)are the waveforms of EEG, ECG and plethysmography from left to right.Physiological indexes, such as RE and SE, heart rate (HR) and SpO2 are displayed on the left side of each panel. EEG signals are collected for 30 seconds at first for the MSE calculation, and then its analysis and computation carry out at every secondwith sliding window. The MSE from scale factor τ =1 to 6 is shown in Figure 3(d), its sum of each scale factor is plotted as a blue line in Figure 3(g), and its ratio of scale factor 6 to 1 represented as a percentage is displayed with red line. This ratio indicates the change of complexity of time series in big time scale compared to that of original time series, which is sensitive to the slow rhythm of EEG, like deltawave in EEG [7]. Figure 3(e) shows indexes of RE and SE from monitor device to compare with SampEn in Figure 3(h). The adjusted value of sample entropy is the product of sample entropy and 10 for clarity. Moreover, the spectrum of EEG computed by fast Fourier transform (FFT)is shown in Figure 3(f), and Figure 3(i) provides the SEF 95% and percentage of delta, theta, alpha, beta bands in frequency domain. All the results are stored in text files at every half an hour for offline analysis.
Those traditional indexes like SEF 95% and percentage of four bands are used to compare with our entropy indexes SampEn, MSE and itscomplexity at the same time in online analysis.Furthermore, as entropy was proved to be a valid indicator of the hypnotic effect of some anesthetic drugs [21], it is more optimal than spectrum in describing the irregularity or unpredictability of a signal.So the online analysis of EEG based on entorpy is significant in measure DOA.
3.1.2 Offline Analysis
There have been many commercial programs for clinician to analyze physiological signals in time and frequency domains and statistics.For validation of our own program, MSE results are compared to the commercial software Visual Signal, and the same results are shown in Figures 4 and 5. The average sum and ratio of MSE are computed as well. Parameters m and r are able to be different for other trials. Figure 6 shows the result of fast Fourier transform of EEG signals, accuracy of which has already been proved by Matlab.
Even though the EEG signals are the direct responses of cerebral cortex, however, some drugs affect brain stem internally, so that it is insufficient to detect DOA only through EEG. At this time, ECG, SpO2and respiration of patients should be integrated into analysis and measurement of DOA. Therefore, the ECG signals are involved in our offline analysis firstly. The R peak can be extracted from ECG waveform displayed in Figure 7. In next step, the interval of R peak is calculated and shown in the list below. SampEn of the HR is shown on the right side if quantity of data is adequate. As m is taken as 2, the minimum for calculation is fixed at 120, which is bigger than 102 [15, 22].
3.2. The Physiological InformationDatabase
The PID was established to store, manage and share the collected and analyzed data and operation information, which has classic three-tier architecture shown in Figure 2. The database is in charge of data and information storage, the application server provides authority, services, and file management, and all services and data access are on the basis of web browser and server model. Thus, a variety of devices can obtain the service in different and remote access. The realization of this platform depends on the popular web server software Apacha[23],the relational database management system MySQL and the hypertext preprocesserPHP, all of which is widely used in server architecture [24].
3.2.1 The functions of PID
Figure 8 displays the main functionsin the PID. First of all, all members should be authorized so as to have further data access and services safely. The authority includes member ID check, random code verification, service acceptance and permission. Then members are permitted to login, upload and download data, view and print, and further self-analysis. Theprinciplefunction of the PID system is a file management system [25]. Due to different member assigned by the different authorities, the application server must identify the identification and authority of member who requires the particular services, and then give him the corresponding treatment and service. In order to manage the data in PID easily, each dataset has its own describing information.Furthermore, the database will give the new data list of database for members after new dataset is uploaded.
3.2.2 The Services in PID
1. Type of physiological information
There are three types of physiologic information about the operation information uploaded. First one is the physiological data such as ECG, EEG, BP, and SpO2 and so on. The second type is event recording during operation, like drug injection, movement, and other general operation during surgeries. The last one is the patient anesthesia record that is a form filled by anesthetist. The three types of information should be converted into uniform format respectively for recycled use and storage. The physiological data and online and offline analysis results are transformed into *.MAT file using Matlab. Because the MAT format has the advantage of less storage space, and Matlab supplies simulation and processing platform for a further research of DoA. The operation events in Chinese and English are combined with the patient anesthesia record into a *. JPG file. For the patient privacy, the picture hides the name of patient.
When all conversion and format verification has been done, the physiological data and information can be uploaded into PID by the authorized members. There are three kinds of authorized members :administrator, member and guest, they have their own service and permission in our PID, shown in Table 1. Administrator has all permission to the services in PID, members in our corporation group are only restricted to delete data and information, and however, the guests can just upload their own physiological data and information and look over in the PID.
2. Login
Figure 9 is the home page of PID website [26], on which the sponsors and creators of PID for DoA analysis namethe PID as YZU (Yuan Ze University) and NUTH (National University of Taiwan Hospital) databank. Members can login and logout through it, the login interface is shown in Figure 10. After entering the account, password and the random verification code, members are able to enjoy the service of PID.