- Email: [email protected]
Analysis of parameters affecting blood oxygen saturation and modeling of fuzzy logic system for inspired oxygen prediction
Computer Methods and Programs in Biomedicine 176 (2019) 43–49
Contents lists available at ScienceDirect
Computer Methods and Programs in Biomedicine journal homepage: www.elsevier.com/locate/cmpb
Analysis of parameters affecting blood oxygen saturation and modeling of fuzzy logic system for inspired oxygen prediction Sita Radhakrishnan a,∗, Suresh G Nair b, Johney Isaac a a b
Department of Instrumentation, Cochin University of Science and Technology, Kochi, Kerala, India Anaesthesia and Critical Care, Aster Medcity, Kochi, Kerala, India
a r t i c l e
i n f o
Article history: Received 11 December 2018 Revised 21 March 2019 Accepted 12 April 2019
Keywords: Oxygen saturation Mechanical ventilation Data collection Fuzzy logic
a b s t r a c t Background and Objective: Fraction of Inspired Oxygen is one of the arbitrary set ventilator parameters which has critical influence on the concentration of blood oxygen. Normally mechanical ventilators providing respiratory assistance are tuned manually to supply required inspired oxygen to keep the oxygen saturation at the desired level. Maintaining oxygen saturation in the desired limit is so vital since excess supply of inspired oxygen leads to hypercapnia and respiratory acidosis which lead to increased risk in cell damage and death. On the other side a sudden drop in oxygen saturation will lead to severe cardiac arrest and seizure. Hence intelligent real time control of blood oxygen level saturation is highly significant for patients in intensive care units. Methods: This paper gives statistical pair wise analysis for finding out deeply correlated physiological parameters from clinical data for fixing fuzzy variables. An advisory fuzzy controller using Mamdani model is developed with R programming to predict FiO2 which is to be delivered from the ventilator to maintain SaO2 with in required levels. Results: Fuzzy variables for the fuzzy model is fixed using 75% of the clinical data collected. Remaining 25% of the data is used for checking the system. Compared the predictive output of the system with physicians’ decisions and found to be accurate with less than five percentage error. Conclusions: Based on the comparison the system is proved to be effective and can be used as assist mode for physicians for effective decision making. © 2019 Elsevier B.V. All rights reserved.
1. Introduction Continuous monitoring and control of physiological parameters of patients in intensive care units is a challenge to researchers. Among the various parameters to evaluate the health of a patient, the blood oxygen content or blood oxygen saturation level (SpO2 ) is highly significant. Normal blood oxygen saturation level is 95–100% or normal partial pressure of oxygen in blood (PaO2 ) is 75–100mm Hg. SpO2 levels below 90% or PaO2 < 60 mm Hg is considered as critical. Mechanical ventilators that provide respiratory assistance are tuned to supply controlled quantities of inspired oxygen (FiO2 ) to keep the blood saturation (SpO2 ) at the desired level. Optimal control of FiO2 is necessary since a sudden drop in oxygen saturation level leads to a condition called hypoxemia, which leads to inadequate oxygen content at the cellular levels and subsequently to end organ damage. Alternatively excess
∗
Corresponding author. E-mail addresses: [email protected] (S. Radhakrishnan), [email protected] (S.G. Nair), [email protected] (J. Isaac). https://doi.org/10.1016/j.cmpb.2019.04.014 0169-2607/© 2019 Elsevier B.V. All rights reserved.
quantity of inspired oxygen leads to hyperoxia, which is also harmful to the body. Monitoring and control of oxygen saturation level of patients is very critical in intensive care units. To meet the oxygen requirements of the body FiO2 must be provided in controlled amount to ventilated patients in the intensive care unit. Usually the clinician manually adjusts the ventilator settings twice or thrice a day based on pulse oximetry readings (SaO2 ) and arterial blood gas (ABG).Adequacy of arterial oxygen can is influenced by the settings on the ventilator, which includes the FiO2 , the inspiratory and expiratory times, the inspiratory flow pattern and respiratory rates. In addition, multiple modes or patterns of ventilation are available on modern ventilators. Real time control of inspired oxygen for getting desired SaO2 is an ever interesting area to researchers. Setting FiO2 considering various physiological parameters is very crucial in clinical scenario. Exploring through literature we can see many works that have been there for intelligent control of inspired oxygen [1]. H Luepschen, L Zhu and S Leonhardt [2] implemented a PID controller for FiO2 control which was based on a modified Smith Predictor. M H Giard et al. [3] used a state feedback control scheme to regulate the concentration of oxygen and carbon dioxide in the
44
S. Radhakrishnan, S.G. Nair and J. Isaac / Computer Methods and Programs in Biomedicine 176 (2019) 43–49
respiratory gases by balancing the relative fraction of pure oxygen and rebreathing gases using a first order model which describes the whole system. In first order model the system nonlinearities cannot be accounted. Also the control law developed does not support the rapid changes in physiological characteristics of the patient which effects the robustness of the system. J R Anderson and T D East [4] developed a PID controller to maintain the levels of PEEP and inspired oxygenin Adult Respiratory Distress Syndrome (ARDS). PID controller hereshowed poor performances in large time delay. Since PID controller has to balance all the three gains, it has to compromise the transient response characteristics. PID does not support the uncertainties and disturbances in physiological system showing low robustness. Goode et al. [5] developed a fuzzy rule based advisor using patient model to predict the values of ventilator parameters namely volume control and pressure regulated volume control. In [6] Kretschmer et al. proposed a gas exchange model that predicts oxygen saturation values at different FiO2 s by calculating the shunt fraction. A two compartment lung model was used for gas exchange. The model gave a 10% error range for PaO2 prediction comparing with actual value. Goluck et al. [7] in 2016 designed and constructed a mechanical ventilator which is micro controller based with fuzzy logic control to calculate FiO2 and PEEP from SpO2 .Here a pulse oximeter also designed and developed. The fuzzy decision system here strictly depended on pulse oximeter value only. Pulse oximetry was reliable only in normal conditions. It has limitations in some situations in which patient suffering from hypotension, anemia, restlessness etc. In [8] Goluck, Guler developed another innovative ventilator prototype. Here fuzzy controller was used for calculating the rate for closing and opening the stepper motor controlled valve. In this work the amount of oxygen in terms of liter/minute which was given to the patient, according to the suggested FiO2 by the physician was calculated using fuzzy logic. Here KE-25 series oxygen sensor was used for sensing the percentage of oxygen in air and PIC18F4550 microcontroller was used for driving stepper motor. Also mathematical modeling is somehow impossible in the case of human beings. Most of the systems developed were not patient specific and generalized models were considered [9–11]. This paper shows an innovated method of inspired oxygen prediction using fuzzy logic controller by combining the real time values of pulse oximetry with periodical readings of arterial blood gas analysis. The rest of the sections of the paper is categorised as follows: The section named method is sub divided into two. First section gives statistical analysis for fixing the deeply correlated variables in ABG readings with inspired oxygen which is delivered from ventilator for post operated patients. By correlating periodically collected ABG readings with the continuous values of pulse oximetry, the second phase of the section concentrates on developing an advisory fuzzy model for controlling blood oxygen concentration by appropriate inspired oxygen input. In this paper, using R programming, Mamdani model is applied for developing the fuzzy rule based system for inspired oxygen control [12]. Then follows the sections of result and discussion which analyses the output of the system and conclusion and future scope that discusses the overview and future developments of the system.
2. Methods Main challenge of the work is to model the system from real time data. About 75% of the work depends on data collection and analysis. Data includes ABG results, pulse oximetry values, ventilator settings and also the valuable suggestions from physicians. Data samples are collected from the concerned authorities following the relevant laws and institutional guidelines. All the above, the
Data Collection
Data Analysis
Data Collection
Fixing Fuzzy Variables Fuzzy Rule Formation
Modelling of system
Testing Fig. 1. Developing stages of proposed system.
institutional ethical committee had approved the work .The below block diagram shows work flow of the proposed system, see Fig. 1. According to the work flow this section is divided into two- Statistical Analysis for fixing fuzzy variables and Fuzzy Expert System Model for maintaining blood oxygen saturation 2.1. Statistical analysis for fixing fuzzy variables Arterial blood gas analysis readings, pulse oximetry readings and corresponding ventilator settings were collected during the period from December 2017 to November 2018.Real data samples of 40 patients in ventilation, including postoperative and respiratory failure ones were taken. Collected data includes vitals like heart rate, temperature, blood pressure, pulse rate, respiration rate, blood oxygen saturation rate from pulse oximetry (SaO2 ), arterial blood gas readings including pH of blood, partial fraction of oxygen PaO2 in mmHg, partial pressure of carbon dioxide in blood (PaCO2 )(mm Hg), bicarbonate (HCO3 ) (mmol/L) and ventilator settings including positive end-expiratory pressure (PEEP) rate, Minute volumes, FiO2 etc. Among the various parameters collected some had high significance, some had moderate significance and some others had least significance in the control of inspired oxygen. Pulse oximeter readings give real time continuous oxygen saturation values (SaO2 ) which is really significant for the control of inspired oxygen. It can be fixed as one of the fuzzy variable. But there are a number of parameters in ABG analysis. Statistical analysis using scatter plot was done for deciding the critical parameters from among the collected ABG data for fixing fuzzy variables. Scattered plots of each and every parameter against inspired oxygen were generated using R programming [13,14]. From this we could find out PO2 (mm Hg), PCO2 (mm Hg), pH and HCO3 (mmol/L) values of ABG were mostly correlated to FiO2 . Fig. 1. shows the scatter plot of pH verses FiO2 . Normal range of pH is 7.35 to 7.45. Scatter plots shows that when pH goes below 7.3, the patient becomes critically acidotic and FiO2 was increased. Decrease in PO2 is unsafe and necessary increment must be given to FiO2 , see Fig. 2. Scatter plot showing the relationship between PCO2 and FiO2 is
S. Radhakrishnan, S.G. Nair and J. Isaac / Computer Methods and Programs in Biomedicine 176 (2019) 43–49
Fig. 2. Scatter plot of FiO2 verses pH.
45
Fig. 5. Scatter plot of FiO2 verses HCO3.
2.2. Fuzzy expert system model for maintaining blood oxygen saturation
Fig. 3. Scatter plot of FiO2 verses PO2.
Fig. 4. Scatter plot of FiO2 verses PCO2.
shown in Fig. 3. Here we can see that increase in PCO2 is dangerous and hence FiO2 must be increased. Scatter plot of Fig. 4 shows how FiO2 increase when HCO3 decreases (Fig. 5). Main challenge to model a system was the discontinuity of data collection. ABG readings were collected two or three times a day while pulse oximetry readings (SaO2 ) were continuously monitored. FiO2 values were dependent more on ABG readings than pulse oximetry readings which were continuously monitored. During emergency conditions SaO2 readings are very critical. Hence, we had to consider both the readings, pulse oximetry as well as ABG. For this it is difficult to model the system with conventional mathematical modeling and we have to go for another nonlinear model such as Fuzzy Expert System Model [15].
Predictive modeling of human processes turns out to be essentially more refined and boundless nowadays. Fuzzy logic systems used for health care problems have some advantages over classical control systems [16]. Non-linear responses which cannot be modeled mathematically can be analysed by fuzzy logic. Fuzzy logic provides the degree of membership functions to form hybrid systems for real time problems. The ability of fuzzy logic to handle data which was uncertain and vague, enabled its applications in medical field. Fuzzy logic has been implemented in various healthcare problems to describe complex system modeling, which cannot be possible by mathematical modeling [17].There are two methods of fuzzy inference systems – Mamdani and Sugeno. Mamdani inference model gives fuzzy set as output whereas Sugeno gives either a linear mathematical expression or a constant as output [18]. We can apply fuzzy logic to machine learning, since it focuses on the development of computer programs that can access data and make use of the gathered data to learn themselves. Machine learning is a more extensive field of artificial intelligence where we assist PC with learning things and let them do assignments. Artificial intelligence is the superset of fuzzy and machine learning [19]. Since in the case of respiratory control large number of data samples is collected for system study and this work is a small step towards machine learning approach. The goal of this work is to develop a generalized fuzzy inference engine using Mamdani model [20] to control inspired oxygen (FiO2 ) which will be given to the ventilated patients. Since the system considered to be a generalized one, all the categories of post-operative patients of age group between twenty and seventy years were considered [21]. From the statistical analysis in the first phase, physiological parameters that are deeply correlated with inspired oxygen level were calculated. These parameters were used for fixing membership functions to form empirical fuzzy rules [22–24]. The below figure, Fig. 6 shows the block diagram of fuzzy logic inference system. Firstly the crisp set is converted into fuzzy set by fuzzification. The knowledge base creates the membership functions for fuzzy rules. The decision making unit performs the functioning of rules in Data base whereas defuzzification unit converts fuzzy set to crisp sets. Steps used for the development of fuzzy expert system using Mamdani inference system are given below. Step 1:Data Collection For forming fuzzy empirical rules expert opinion regarding patient physiological parameters, ventilator settings and physician’s opinion for FiO2 prediction were needed. As already stated data samples of 40 ventilated patients, both male and female of age
46
S. Radhakrishnan, S.G. Nair and J. Isaac / Computer Methods and Programs in Biomedicine 176 (2019) 43–49 Table 1 Fuzzy regions for membership functions. Parameters
Region1
pH PCO2 (mmHg) PO2 (mmHg) HCO3 (mmol/L) SaO2 (%) FiO2 (%)
Critically acidosis Acidosis Normal pH Respiratory acidosis normalCO2 carbonated Critically hypoxemia Hypoxemia Normal oxemia Metabolic acidosis Normal bicarbonate High bicarbonate Critically hypoxemia Hypoxemia Hyperoxemia RATE1, RATE2,RATE3,RATE4,RATE5, RATE5, RATE7, RATE8
Region2
Crisp Input
Fuzzificaon
Decision Making
Knowledge Base,Fuzzy rules
Defuzzificaon
Crisp Output Fig. 6. Block diagram of fuzzy inference system.
limit between 20–70 years were collected along with ventilator settings. Expert opinions of eight physicians were considered. Step2:Choosing the input, output variables and fuzzy set R programming is used for forming fuzzy system [25]. Input variables taken are PO2 (mmHg), PCO2 (mmHg), pH, and HCO3 (mmol/L) concentrations from arterial blood gas analysis, pulse oximetry readings that is SaO2 (%) and the output variable is inspired oxygen from ventilator (FiO2 ). Input and output variables are analyzed and generalized to form fuzzy sets [26,27]. There are four fuzzy regions for pH and PCO2 . Five for PO2 , three for HCO3 and SaO2 . For the output variable FiO2 there are eight fuzzy regions. Fuzzy regions created and variables are named suitably for the formation of membership functions, shown in Table 1. Step 3: Defining Fuzzy rules Since we are considering probabilistic and multi-valued logic tnorm concept is considered. R package set is used for creating the fuzzy system [28]. A fuzzy system consisting of six variables and twenty rules are formulated. Fuzzy empirical rules that develop the body of fuzzy controller were constructed by standard declarative form. Information and suggestions from doctors’ were considered for decision making. Sample rule formation logic is given below. Rule: If pH is critically acidosis and pCO2 is hyper carbonated and PO2 critically hypoxemia and HCO3 is metabolic acidosis and SaO2 is critically hypoxemia then FiO2 is Rate1. Step 4: Modeling of system using Membership functions The system has five input parameters and one output parameter. Two of the input parameters (HCO3 , SaO2 ) have three membership functions (MF’s), another two have four MF’s (pH, PCO2 ) and the last one has five (PO2 ). The output parameter has eight membership functions (FiO2 ). All the combinations of rules were not necessary for modeling the system as the parameters have solid relationship between each other defined by pairwise
Region 3
Region 4
Region 5
alkaholisis hypercarbonated High oxemia
hyperoxemia
analysis. Membership graph for input parameters PO2 , PCO2 , pH, HCO3 andSaO2 and the output parameter FiO2 are shown in Fig. 7. Step 5: Testing of System and Defuzzification stage Seventy five percentage of the data collected is used for modeling the fuzzy system and twenty five percent of data collected have been used for testing the system. Defuzzification centroid method is used for getting the crisp value from fuzzy sets and membership functions. The defuzzification graph is shown in Fig. 8 As an example Table 2 shows the comparison of predicted FiO2 as per the data samples from the developed fuzzy inference system with the decision of three physicians A, B and C. The deviations from the predicted FiO2 and physician’s suggested FiO2 values give an average percentage accuracy error value of 0.33%. A similar fuzzy control system using well known Tool Box was also developed. Percentage error of this system also calculated. For checking the relevance of the proposed system it is compared with the output produced by the fuzzy inference system which was developed using the known tool box. In Fig. 9 we can see graphical representation showing the variations between average physicians’ suggested FiO2 , proposed system output and output of the tool box. Percentage error comparison of both the systems can be seen in Fig. 10. This comparison proved the accuracy of the system and it was more précised than the system developed by tool box. 3. Result Here we can first see the result of the proposed system. As already discussed twenty five percentage of data collected was used for testing. Each output was compared with the corresponding physicians output. Comparing with the clinicians’ decision making for fixing FiO2 , the proposed system shows high accuracy and average deviation shown is 0.33%. For sample comparison refer Table 2. Percentage accuracy error was calculated between doctors’ accepted values and the predicated values of the proposed system using R programming. Similar method was used for finding out the error calculation for the system developed by known fuzzy logic tool box. Comparing results it was found that the system has got good training and could be trusted. 4. Discussion Formation of a control system considering all the aspects of human physiological parameters is a very tedious and complicated job. Our aim was to build up a generalised fuzzy logic controller with empirical rules for getting controlled FiO2 . Here we were considering various parameters affecting blood oxygen saturation. By analysing the variation of these parameters physicians are adjusting ventilator settings to get the required inspired oxygen flow. An innovative method of combining real time values of pulse oximetry with periodically collected ABG values was established. In ABG analysis we could see a number of parameters like pH, PCO2 , PO2 , HCO3 , Sodium, Potassium, Hb etc. Seventy five percentage of the work depends on analysis and categorisation of data. Statistical analysis with R programming is done for getting the ranges of the
S. Radhakrishnan, S.G. Nair and J. Isaac / Computer Methods and Programs in Biomedicine 176 (2019) 43–49
47
Table 2 Comparison between predicted FiO2 and physicians’ suggestions. Samples
A
B
C
Proposed FiO2
Known fuzzy tool box output
% error of proposed system
% error of known fuzzy logic tool box
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
65 60 55 40 70 35 30 35 65 80 40 65 40 40 30 30 100 100 100 35
65 60 55 50 65 40 30 35 65 75 45 65 45 35 30 30 95 90 95 35
70 60 50 45 70 30 30 35 60 75 40 60 40 40 30 30 90 90 100 35
67 60 53 45 68 35 30 35 63 77 42 63 42 38 30 30 95 93 99 35
66 59 53 44 68 35 31 36 64 78 41 64 42 38 31 30 95 94 100 34
0.5 0 0.625 0 0.487805 0 0 0 0.526316 0.434783 0.8 0.526316 0.8 1.73913 0 0 0 0.357143 0.677966 0
1 1.666667 0.625 2.222222 0.487805 0 3.333333 2.857143 1.052632 1.73913 1.6 1.052632 0.8 1.73913 3.333333 3.333333 0 0.714286 1.694915 2.857143
Fig. 7. Modeling of system by membership functions.
most correlated parameters effecting blood oxygen saturation from ABG analysis. Information from physicians’ is also collected for fixing the parameters effecting SpO2 . Combining the statistical analysis and physicians’ suggestion PO2 , PCO2 , pH, HCO3 , SaO2 are fixed as input variables and output variable as FiO2 for modeling the fuzzy controller. Consulting with clinicians and using 75% of data
collected fuzzy empirical rules are created. The developed system is tested with the remaining 25% of data. For each reading percentage accuracy error is calculated. It is analysed that the average error is equal to 0.33%. This demonstrates the acceptability of utilizing this fuzzy inference system as a suggestive mode for doctors. Further work to convert the proposed system from predictive
48
S. Radhakrishnan, S.G. Nair and J. Isaac / Computer Methods and Programs in Biomedicine 176 (2019) 43–49
Fig. 8. Defuzzification using centroid method.
Fig. 9. Inspired oxygen variations predicted by physicians’, proposed system and known fuzzy tool box verses data samples.
Fig. 10. Percentage error of proposed system, known fuzzy tool box verses data samples.
mode to assist mode is going on with reference to the concept of machine learning [29]. 5. Conclusion and future work In this work, a primary fuzzy model is developed for inspired oxygen prediction for ventilator systems. Pairwise analysis of physiological parameters was done to identify parameters which were correlated with FiO2. Fuzzy rule based system developed in consultation with clinicians was tested for checking the efficiency. Comparing the system output with clinicians’ decision making gave a
result with less than 5% error and hence is acceptable. Further work is going on to convert the assist mode of fuzzy system to control mode. For these additional parameters in ventilator settings like modes, PEEP, Rate, Tidal Volumes are under consideration .Here we can apply fuzzy logic to machine learning, since it focuses on the development of computer programs that can access data and make use of the gathered data to learn themselves. Machine learning is a more extensive field of artificial intelligence where we assist computer with learning things. Also similar researches with big data analysis is possible for developing a multi input multi output system [30,31].Deep learning method using adaptive
S. Radhakrishnan, S.G. Nair and J. Isaac / Computer Methods and Programs in Biomedicine 176 (2019) 43–49
neuro fuzzy system to adjust multiple ventilator settings for keeping blood oxygen saturation will be a definitive point of further research. Statements of ethical approval By following the relevant laws and institutional guidelines only data samples for the work were collected and the institutional ethical committee had approved the work. Competing interests The authors have no competing interest to declare. Supplementary materials Supplementary material associated with this article can be found, in the online version, at doi:10.1016/j.cmpb.2019.04.014. References [1] C.S. Wang, D. Shaw, K.S. Jih, An intelligent control system for ventilators, Med. Eng. Phys. 20 (7) (1998) 534–542. [2] H. Luepschen, L. Zhu, S. Leonhardt, Robust closed-loop control of the inspired fraction of oxygen for the online assessment of recruitment maneuvers, in: ConfProc IEEE Eng. Med. Biol. Soc., 2007, pp. 495–498. [3] M.H. Giard, F.O. Bertrand, D. Robert, J. Pernier, An algorithm for automatic control of O2 and CO2 in artificial ventilation, IEEE Trans. Biomed. Eng. (1985) 658–667. [4] J.R. Anderson, T.D. East, A closed-loop controller for mechanical ventilation of patients with ARDS, Biomed. Sci. Instrum. (2002) 289–294. [5] K.M. Goode., D.A. Linkens, P.R. Bourne, J.G. Cundill, Development of a fuzzy rule-based advisor for the maintenance of mechanically ventilated patients in ICU: a model-based approach, Appl. Basis Commun. 10 (4) (1998) 236–246 ISSN 1016-2372. [6] J. Kretschmer, T. Becher, A. Riedlinger, D. Schadler, N. Weiler, K. Moller, A simple gas exchange model predicting arterial oxygen content for various FiO2 levels, in: ConfProc IEEE Eng. Med. Biol. Soc., 2013, pp. 465–468. [7] A. Golcuk, H. Isik, I. Guler, Design and Construction of a Microcontroller-Based Ventilator Synchronized with Pulse Oximeter, J. Med. Syst. 40 (7) (2016) 180 Jul, doi:10.1007/s10916-016-0538-x. [8] A. Gölcük, I.˙ &Güler, The use of stepper motor-controlled proportional valve for fio 2 calculation in the ventilator and its control with fuzzy logic, J. Med. Syst. 41 (1) (2017) 1. [9] M. Bonfanti, A. Cammi, P. Bagnoli, Gas transfer model to design a ventilator for neonatal total liquid ventilation, Med. Eng. Phys. 37 (12) (2015) 1133–1140. [10] S. Cecchini, I. Sardellitti, S. Silvestri, A new methodology for intra-breath control of mechanical ventilation, Med. Eng. Phys. 34 (2) (2012) 256–260. [11] P.A. Dargaville, O. Sadeghi Fathabadi, G.K. Plottier, K. Lim, K.I. Wheeler, R. Jayakar, T.J. Gale, Development and preclinical testing of an adaptive algorithm for automated control of inspired oxygen in the preterm infant, Arch. Dis. Childhood-Fetal Neonatal, Ed., 102 (1) (2017) F31–F36.
49
[12] D. Mindrila, P. Balentyne, Scatter Plots and correlation, The Basic Practice of Statistics, 6th ed., Chapter4, https://www.westga.edu/academics/research/vrc/ assets/docs/scatterplots_and_correlation_notes.pdf, Accessed on: July 7 (2018). [13] L.S. Riza, C.N. Bergmeir, F. Herrera, J.M. Benítez Sánchez, Frbs: Fuzzy Rule-Based Systems For Classification and Regression in R, American Statistical Association, 2015. [14] D.G. Rossiter, An example of statistical data analysis using the R environment for statistical computing, Version 1.4 May 6 (2017) http://www.css.cornell.edu/ faculty/dgr2/teach/R/R_corregr.pdf, Accessed on: May 10 (2018). [15] E.P. Martinoni, ChA Pfister, K.S. Stadler, S. PM, D. Leibundgut, T. Bouillon, T. Bohlen, A.M. Zbinden, Model-based control of mechanical ventilation: design and clinical validation, Br. J. Anaesth. (2004) 800–807. [16] M. Susmitha, M. Prakash, Study of fuzzy logic in medical data analytics, Int. J. Pure Appl. Math. 119 (12) (2018) 16321–16342. [17] Y. Sun, I. Kohane, A.R. Stark, Fuzzy logic assisted control of inspired oxygen in ventilated newborn infants, in: Proceedings of the Annual Symposium on Computer Application in Medical Care, American Medical Informatics Association, 1994, p. 757. [18] MrsSusmita Prakash, M. Mishraand Dr., Study of fuzzy logic in medical data analytics, Int. J. Pure Appl. Math. 12 (2018) 16321–16342 119. [19] E. Hullermeier, Fuzzy methods in machine learning and data mining: status and prospects, Fuzzy Sets Syst. 156 (3) (2005) 387–406. [20] Ion Iancu, A Mamdani type fuzzy logic controller, In Fuzzy Logic-Controls, Concepts, Theories and Applications, InTech, 2012. [21] J. Schaublin, M. Derighetti, P. Feigenwinter, S. Petersen-Felix, A.M. Zbinden, Fuzzy logic control of mechanical ventilation during anaesthesia, Br. J. Anaesth. 77 (5) (1996) 636–641. [22] T. Nemoto, G.E. Hatzakis, C. William Thorpe, R. Olivenstein, S. Dial, J.H.T. Bates, Automatic control of pressure support mechanical ventilation using fuzzy logic, Am. J. Respir. Crit. Care Med. 160 (2) (1999) 550–556. [23] K M Goode, D.A Linkens, P.R. Bourne, J.G. Cundill, Development of a fuzzy rule-based advisor for the maintenance of mechanically ventilated patients in ICU: a model-based approach, Appl. Basis Commun. 10 (4) (1998) 236–246 ISSN 1016-2372. [24] J. Kretschmer, T. Becher, A. Riedlinger, D. Schadler, N. Weiler, K. Moller, A simple gas exchange model predicting arterial oxygen content for various FiO2 levels, ConfProc IEEE Eng. Med. Biol. Soc (2013) 465–468, doi:10.1109/EMBC. 2013.6609537. [25] C. Wagner, S. Miller, J.M. Garibaldi, A fuzzy toolbox for the R programming language, in: In Fuzzy Systems (FUZZ), IEEE International Conference, 2011, pp. 1185–1192. [26] Y. Sun, I. Kohane, A.R. Stark, in: Fuzzy logic assisted control of inspired oxygen in ventilated newborn infants, AMIA, 1994, pp. 757–761. [27] P. Chanyagorn, P. Kiratiwudhikul, Automatic control of fraction of inspired oxygen in neonatal oxygen therapy., The Institute of Electronics Engineers of Korea 5 (2) (2016) 107–116. [28] https://cran.r-project.org/web/packages/sets/sets.pdf. [29] E. Hullermeier, Fuzzy methods in machine learning and data mining: status and prospects, Fuzzy Sets Syst. 156 (3) (2005) 387–406. [30] A. Wang, M. Mahfouf, G.H. Mills, G. Panoutsos, Da Linkens, K. Goode, H.F. Kwok, M. Denai, Intelligent model-based advisory system for the management of ventilated intensive care patients. Hybrid blood gas patient model, ELSEVIER, Comput. Methods Programs Biomed. 99 (2010) 195–207. [31] L.A. Pastur-Romay, F. Cedrón, A. Pazos, A.B. Porto-Pazos, Deep artificial neural networks and neuromorphic chips for big data analysis: pharmaceutical and bioinformatics applications, Int. J. Mol. Sci. (2016) 1313.
Contents lists available at ScienceDirect
Computer Methods and Programs in Biomedicine journal homepage: www.elsevier.com/locate/cmpb
Analysis of parameters affecting blood oxygen saturation and modeling of fuzzy logic system for inspired oxygen prediction Sita Radhakrishnan a,∗, Suresh G Nair b, Johney Isaac a a b
Department of Instrumentation, Cochin University of Science and Technology, Kochi, Kerala, India Anaesthesia and Critical Care, Aster Medcity, Kochi, Kerala, India
a r t i c l e
i n f o
Article history: Received 11 December 2018 Revised 21 March 2019 Accepted 12 April 2019
Keywords: Oxygen saturation Mechanical ventilation Data collection Fuzzy logic
a b s t r a c t Background and Objective: Fraction of Inspired Oxygen is one of the arbitrary set ventilator parameters which has critical influence on the concentration of blood oxygen. Normally mechanical ventilators providing respiratory assistance are tuned manually to supply required inspired oxygen to keep the oxygen saturation at the desired level. Maintaining oxygen saturation in the desired limit is so vital since excess supply of inspired oxygen leads to hypercapnia and respiratory acidosis which lead to increased risk in cell damage and death. On the other side a sudden drop in oxygen saturation will lead to severe cardiac arrest and seizure. Hence intelligent real time control of blood oxygen level saturation is highly significant for patients in intensive care units. Methods: This paper gives statistical pair wise analysis for finding out deeply correlated physiological parameters from clinical data for fixing fuzzy variables. An advisory fuzzy controller using Mamdani model is developed with R programming to predict FiO2 which is to be delivered from the ventilator to maintain SaO2 with in required levels. Results: Fuzzy variables for the fuzzy model is fixed using 75% of the clinical data collected. Remaining 25% of the data is used for checking the system. Compared the predictive output of the system with physicians’ decisions and found to be accurate with less than five percentage error. Conclusions: Based on the comparison the system is proved to be effective and can be used as assist mode for physicians for effective decision making. © 2019 Elsevier B.V. All rights reserved.
1. Introduction Continuous monitoring and control of physiological parameters of patients in intensive care units is a challenge to researchers. Among the various parameters to evaluate the health of a patient, the blood oxygen content or blood oxygen saturation level (SpO2 ) is highly significant. Normal blood oxygen saturation level is 95–100% or normal partial pressure of oxygen in blood (PaO2 ) is 75–100mm Hg. SpO2 levels below 90% or PaO2 < 60 mm Hg is considered as critical. Mechanical ventilators that provide respiratory assistance are tuned to supply controlled quantities of inspired oxygen (FiO2 ) to keep the blood saturation (SpO2 ) at the desired level. Optimal control of FiO2 is necessary since a sudden drop in oxygen saturation level leads to a condition called hypoxemia, which leads to inadequate oxygen content at the cellular levels and subsequently to end organ damage. Alternatively excess
∗
Corresponding author. E-mail addresses: [email protected] (S. Radhakrishnan), [email protected] (S.G. Nair), [email protected] (J. Isaac). https://doi.org/10.1016/j.cmpb.2019.04.014 0169-2607/© 2019 Elsevier B.V. All rights reserved.
quantity of inspired oxygen leads to hyperoxia, which is also harmful to the body. Monitoring and control of oxygen saturation level of patients is very critical in intensive care units. To meet the oxygen requirements of the body FiO2 must be provided in controlled amount to ventilated patients in the intensive care unit. Usually the clinician manually adjusts the ventilator settings twice or thrice a day based on pulse oximetry readings (SaO2 ) and arterial blood gas (ABG).Adequacy of arterial oxygen can is influenced by the settings on the ventilator, which includes the FiO2 , the inspiratory and expiratory times, the inspiratory flow pattern and respiratory rates. In addition, multiple modes or patterns of ventilation are available on modern ventilators. Real time control of inspired oxygen for getting desired SaO2 is an ever interesting area to researchers. Setting FiO2 considering various physiological parameters is very crucial in clinical scenario. Exploring through literature we can see many works that have been there for intelligent control of inspired oxygen [1]. H Luepschen, L Zhu and S Leonhardt [2] implemented a PID controller for FiO2 control which was based on a modified Smith Predictor. M H Giard et al. [3] used a state feedback control scheme to regulate the concentration of oxygen and carbon dioxide in the
44
S. Radhakrishnan, S.G. Nair and J. Isaac / Computer Methods and Programs in Biomedicine 176 (2019) 43–49
respiratory gases by balancing the relative fraction of pure oxygen and rebreathing gases using a first order model which describes the whole system. In first order model the system nonlinearities cannot be accounted. Also the control law developed does not support the rapid changes in physiological characteristics of the patient which effects the robustness of the system. J R Anderson and T D East [4] developed a PID controller to maintain the levels of PEEP and inspired oxygenin Adult Respiratory Distress Syndrome (ARDS). PID controller hereshowed poor performances in large time delay. Since PID controller has to balance all the three gains, it has to compromise the transient response characteristics. PID does not support the uncertainties and disturbances in physiological system showing low robustness. Goode et al. [5] developed a fuzzy rule based advisor using patient model to predict the values of ventilator parameters namely volume control and pressure regulated volume control. In [6] Kretschmer et al. proposed a gas exchange model that predicts oxygen saturation values at different FiO2 s by calculating the shunt fraction. A two compartment lung model was used for gas exchange. The model gave a 10% error range for PaO2 prediction comparing with actual value. Goluck et al. [7] in 2016 designed and constructed a mechanical ventilator which is micro controller based with fuzzy logic control to calculate FiO2 and PEEP from SpO2 .Here a pulse oximeter also designed and developed. The fuzzy decision system here strictly depended on pulse oximeter value only. Pulse oximetry was reliable only in normal conditions. It has limitations in some situations in which patient suffering from hypotension, anemia, restlessness etc. In [8] Goluck, Guler developed another innovative ventilator prototype. Here fuzzy controller was used for calculating the rate for closing and opening the stepper motor controlled valve. In this work the amount of oxygen in terms of liter/minute which was given to the patient, according to the suggested FiO2 by the physician was calculated using fuzzy logic. Here KE-25 series oxygen sensor was used for sensing the percentage of oxygen in air and PIC18F4550 microcontroller was used for driving stepper motor. Also mathematical modeling is somehow impossible in the case of human beings. Most of the systems developed were not patient specific and generalized models were considered [9–11]. This paper shows an innovated method of inspired oxygen prediction using fuzzy logic controller by combining the real time values of pulse oximetry with periodical readings of arterial blood gas analysis. The rest of the sections of the paper is categorised as follows: The section named method is sub divided into two. First section gives statistical analysis for fixing the deeply correlated variables in ABG readings with inspired oxygen which is delivered from ventilator for post operated patients. By correlating periodically collected ABG readings with the continuous values of pulse oximetry, the second phase of the section concentrates on developing an advisory fuzzy model for controlling blood oxygen concentration by appropriate inspired oxygen input. In this paper, using R programming, Mamdani model is applied for developing the fuzzy rule based system for inspired oxygen control [12]. Then follows the sections of result and discussion which analyses the output of the system and conclusion and future scope that discusses the overview and future developments of the system.
2. Methods Main challenge of the work is to model the system from real time data. About 75% of the work depends on data collection and analysis. Data includes ABG results, pulse oximetry values, ventilator settings and also the valuable suggestions from physicians. Data samples are collected from the concerned authorities following the relevant laws and institutional guidelines. All the above, the
Data Collection
Data Analysis
Data Collection
Fixing Fuzzy Variables Fuzzy Rule Formation
Modelling of system
Testing Fig. 1. Developing stages of proposed system.
institutional ethical committee had approved the work .The below block diagram shows work flow of the proposed system, see Fig. 1. According to the work flow this section is divided into two- Statistical Analysis for fixing fuzzy variables and Fuzzy Expert System Model for maintaining blood oxygen saturation 2.1. Statistical analysis for fixing fuzzy variables Arterial blood gas analysis readings, pulse oximetry readings and corresponding ventilator settings were collected during the period from December 2017 to November 2018.Real data samples of 40 patients in ventilation, including postoperative and respiratory failure ones were taken. Collected data includes vitals like heart rate, temperature, blood pressure, pulse rate, respiration rate, blood oxygen saturation rate from pulse oximetry (SaO2 ), arterial blood gas readings including pH of blood, partial fraction of oxygen PaO2 in mmHg, partial pressure of carbon dioxide in blood (PaCO2 )(mm Hg), bicarbonate (HCO3 ) (mmol/L) and ventilator settings including positive end-expiratory pressure (PEEP) rate, Minute volumes, FiO2 etc. Among the various parameters collected some had high significance, some had moderate significance and some others had least significance in the control of inspired oxygen. Pulse oximeter readings give real time continuous oxygen saturation values (SaO2 ) which is really significant for the control of inspired oxygen. It can be fixed as one of the fuzzy variable. But there are a number of parameters in ABG analysis. Statistical analysis using scatter plot was done for deciding the critical parameters from among the collected ABG data for fixing fuzzy variables. Scattered plots of each and every parameter against inspired oxygen were generated using R programming [13,14]. From this we could find out PO2 (mm Hg), PCO2 (mm Hg), pH and HCO3 (mmol/L) values of ABG were mostly correlated to FiO2 . Fig. 1. shows the scatter plot of pH verses FiO2 . Normal range of pH is 7.35 to 7.45. Scatter plots shows that when pH goes below 7.3, the patient becomes critically acidotic and FiO2 was increased. Decrease in PO2 is unsafe and necessary increment must be given to FiO2 , see Fig. 2. Scatter plot showing the relationship between PCO2 and FiO2 is
S. Radhakrishnan, S.G. Nair and J. Isaac / Computer Methods and Programs in Biomedicine 176 (2019) 43–49
Fig. 2. Scatter plot of FiO2 verses pH.
45
Fig. 5. Scatter plot of FiO2 verses HCO3.
2.2. Fuzzy expert system model for maintaining blood oxygen saturation
Fig. 3. Scatter plot of FiO2 verses PO2.
Fig. 4. Scatter plot of FiO2 verses PCO2.
shown in Fig. 3. Here we can see that increase in PCO2 is dangerous and hence FiO2 must be increased. Scatter plot of Fig. 4 shows how FiO2 increase when HCO3 decreases (Fig. 5). Main challenge to model a system was the discontinuity of data collection. ABG readings were collected two or three times a day while pulse oximetry readings (SaO2 ) were continuously monitored. FiO2 values were dependent more on ABG readings than pulse oximetry readings which were continuously monitored. During emergency conditions SaO2 readings are very critical. Hence, we had to consider both the readings, pulse oximetry as well as ABG. For this it is difficult to model the system with conventional mathematical modeling and we have to go for another nonlinear model such as Fuzzy Expert System Model [15].
Predictive modeling of human processes turns out to be essentially more refined and boundless nowadays. Fuzzy logic systems used for health care problems have some advantages over classical control systems [16]. Non-linear responses which cannot be modeled mathematically can be analysed by fuzzy logic. Fuzzy logic provides the degree of membership functions to form hybrid systems for real time problems. The ability of fuzzy logic to handle data which was uncertain and vague, enabled its applications in medical field. Fuzzy logic has been implemented in various healthcare problems to describe complex system modeling, which cannot be possible by mathematical modeling [17].There are two methods of fuzzy inference systems – Mamdani and Sugeno. Mamdani inference model gives fuzzy set as output whereas Sugeno gives either a linear mathematical expression or a constant as output [18]. We can apply fuzzy logic to machine learning, since it focuses on the development of computer programs that can access data and make use of the gathered data to learn themselves. Machine learning is a more extensive field of artificial intelligence where we assist PC with learning things and let them do assignments. Artificial intelligence is the superset of fuzzy and machine learning [19]. Since in the case of respiratory control large number of data samples is collected for system study and this work is a small step towards machine learning approach. The goal of this work is to develop a generalized fuzzy inference engine using Mamdani model [20] to control inspired oxygen (FiO2 ) which will be given to the ventilated patients. Since the system considered to be a generalized one, all the categories of post-operative patients of age group between twenty and seventy years were considered [21]. From the statistical analysis in the first phase, physiological parameters that are deeply correlated with inspired oxygen level were calculated. These parameters were used for fixing membership functions to form empirical fuzzy rules [22–24]. The below figure, Fig. 6 shows the block diagram of fuzzy logic inference system. Firstly the crisp set is converted into fuzzy set by fuzzification. The knowledge base creates the membership functions for fuzzy rules. The decision making unit performs the functioning of rules in Data base whereas defuzzification unit converts fuzzy set to crisp sets. Steps used for the development of fuzzy expert system using Mamdani inference system are given below. Step 1:Data Collection For forming fuzzy empirical rules expert opinion regarding patient physiological parameters, ventilator settings and physician’s opinion for FiO2 prediction were needed. As already stated data samples of 40 ventilated patients, both male and female of age
46
S. Radhakrishnan, S.G. Nair and J. Isaac / Computer Methods and Programs in Biomedicine 176 (2019) 43–49 Table 1 Fuzzy regions for membership functions. Parameters
Region1
pH PCO2 (mmHg) PO2 (mmHg) HCO3 (mmol/L) SaO2 (%) FiO2 (%)
Critically acidosis Acidosis Normal pH Respiratory acidosis normalCO2 carbonated Critically hypoxemia Hypoxemia Normal oxemia Metabolic acidosis Normal bicarbonate High bicarbonate Critically hypoxemia Hypoxemia Hyperoxemia RATE1, RATE2,RATE3,RATE4,RATE5, RATE5, RATE7, RATE8
Region2
Crisp Input
Fuzzificaon
Decision Making
Knowledge Base,Fuzzy rules
Defuzzificaon
Crisp Output Fig. 6. Block diagram of fuzzy inference system.
limit between 20–70 years were collected along with ventilator settings. Expert opinions of eight physicians were considered. Step2:Choosing the input, output variables and fuzzy set R programming is used for forming fuzzy system [25]. Input variables taken are PO2 (mmHg), PCO2 (mmHg), pH, and HCO3 (mmol/L) concentrations from arterial blood gas analysis, pulse oximetry readings that is SaO2 (%) and the output variable is inspired oxygen from ventilator (FiO2 ). Input and output variables are analyzed and generalized to form fuzzy sets [26,27]. There are four fuzzy regions for pH and PCO2 . Five for PO2 , three for HCO3 and SaO2 . For the output variable FiO2 there are eight fuzzy regions. Fuzzy regions created and variables are named suitably for the formation of membership functions, shown in Table 1. Step 3: Defining Fuzzy rules Since we are considering probabilistic and multi-valued logic tnorm concept is considered. R package set is used for creating the fuzzy system [28]. A fuzzy system consisting of six variables and twenty rules are formulated. Fuzzy empirical rules that develop the body of fuzzy controller were constructed by standard declarative form. Information and suggestions from doctors’ were considered for decision making. Sample rule formation logic is given below. Rule: If pH is critically acidosis and pCO2 is hyper carbonated and PO2 critically hypoxemia and HCO3 is metabolic acidosis and SaO2 is critically hypoxemia then FiO2 is Rate1. Step 4: Modeling of system using Membership functions The system has five input parameters and one output parameter. Two of the input parameters (HCO3 , SaO2 ) have three membership functions (MF’s), another two have four MF’s (pH, PCO2 ) and the last one has five (PO2 ). The output parameter has eight membership functions (FiO2 ). All the combinations of rules were not necessary for modeling the system as the parameters have solid relationship between each other defined by pairwise
Region 3
Region 4
Region 5
alkaholisis hypercarbonated High oxemia
hyperoxemia
analysis. Membership graph for input parameters PO2 , PCO2 , pH, HCO3 andSaO2 and the output parameter FiO2 are shown in Fig. 7. Step 5: Testing of System and Defuzzification stage Seventy five percentage of the data collected is used for modeling the fuzzy system and twenty five percent of data collected have been used for testing the system. Defuzzification centroid method is used for getting the crisp value from fuzzy sets and membership functions. The defuzzification graph is shown in Fig. 8 As an example Table 2 shows the comparison of predicted FiO2 as per the data samples from the developed fuzzy inference system with the decision of three physicians A, B and C. The deviations from the predicted FiO2 and physician’s suggested FiO2 values give an average percentage accuracy error value of 0.33%. A similar fuzzy control system using well known Tool Box was also developed. Percentage error of this system also calculated. For checking the relevance of the proposed system it is compared with the output produced by the fuzzy inference system which was developed using the known tool box. In Fig. 9 we can see graphical representation showing the variations between average physicians’ suggested FiO2 , proposed system output and output of the tool box. Percentage error comparison of both the systems can be seen in Fig. 10. This comparison proved the accuracy of the system and it was more précised than the system developed by tool box. 3. Result Here we can first see the result of the proposed system. As already discussed twenty five percentage of data collected was used for testing. Each output was compared with the corresponding physicians output. Comparing with the clinicians’ decision making for fixing FiO2 , the proposed system shows high accuracy and average deviation shown is 0.33%. For sample comparison refer Table 2. Percentage accuracy error was calculated between doctors’ accepted values and the predicated values of the proposed system using R programming. Similar method was used for finding out the error calculation for the system developed by known fuzzy logic tool box. Comparing results it was found that the system has got good training and could be trusted. 4. Discussion Formation of a control system considering all the aspects of human physiological parameters is a very tedious and complicated job. Our aim was to build up a generalised fuzzy logic controller with empirical rules for getting controlled FiO2 . Here we were considering various parameters affecting blood oxygen saturation. By analysing the variation of these parameters physicians are adjusting ventilator settings to get the required inspired oxygen flow. An innovative method of combining real time values of pulse oximetry with periodically collected ABG values was established. In ABG analysis we could see a number of parameters like pH, PCO2 , PO2 , HCO3 , Sodium, Potassium, Hb etc. Seventy five percentage of the work depends on analysis and categorisation of data. Statistical analysis with R programming is done for getting the ranges of the
S. Radhakrishnan, S.G. Nair and J. Isaac / Computer Methods and Programs in Biomedicine 176 (2019) 43–49
47
Table 2 Comparison between predicted FiO2 and physicians’ suggestions. Samples
A
B
C
Proposed FiO2
Known fuzzy tool box output
% error of proposed system
% error of known fuzzy logic tool box
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
65 60 55 40 70 35 30 35 65 80 40 65 40 40 30 30 100 100 100 35
65 60 55 50 65 40 30 35 65 75 45 65 45 35 30 30 95 90 95 35
70 60 50 45 70 30 30 35 60 75 40 60 40 40 30 30 90 90 100 35
67 60 53 45 68 35 30 35 63 77 42 63 42 38 30 30 95 93 99 35
66 59 53 44 68 35 31 36 64 78 41 64 42 38 31 30 95 94 100 34
0.5 0 0.625 0 0.487805 0 0 0 0.526316 0.434783 0.8 0.526316 0.8 1.73913 0 0 0 0.357143 0.677966 0
1 1.666667 0.625 2.222222 0.487805 0 3.333333 2.857143 1.052632 1.73913 1.6 1.052632 0.8 1.73913 3.333333 3.333333 0 0.714286 1.694915 2.857143
Fig. 7. Modeling of system by membership functions.
most correlated parameters effecting blood oxygen saturation from ABG analysis. Information from physicians’ is also collected for fixing the parameters effecting SpO2 . Combining the statistical analysis and physicians’ suggestion PO2 , PCO2 , pH, HCO3 , SaO2 are fixed as input variables and output variable as FiO2 for modeling the fuzzy controller. Consulting with clinicians and using 75% of data
collected fuzzy empirical rules are created. The developed system is tested with the remaining 25% of data. For each reading percentage accuracy error is calculated. It is analysed that the average error is equal to 0.33%. This demonstrates the acceptability of utilizing this fuzzy inference system as a suggestive mode for doctors. Further work to convert the proposed system from predictive
48
S. Radhakrishnan, S.G. Nair and J. Isaac / Computer Methods and Programs in Biomedicine 176 (2019) 43–49
Fig. 8. Defuzzification using centroid method.
Fig. 9. Inspired oxygen variations predicted by physicians’, proposed system and known fuzzy tool box verses data samples.
Fig. 10. Percentage error of proposed system, known fuzzy tool box verses data samples.
mode to assist mode is going on with reference to the concept of machine learning [29]. 5. Conclusion and future work In this work, a primary fuzzy model is developed for inspired oxygen prediction for ventilator systems. Pairwise analysis of physiological parameters was done to identify parameters which were correlated with FiO2. Fuzzy rule based system developed in consultation with clinicians was tested for checking the efficiency. Comparing the system output with clinicians’ decision making gave a
result with less than 5% error and hence is acceptable. Further work is going on to convert the assist mode of fuzzy system to control mode. For these additional parameters in ventilator settings like modes, PEEP, Rate, Tidal Volumes are under consideration .Here we can apply fuzzy logic to machine learning, since it focuses on the development of computer programs that can access data and make use of the gathered data to learn themselves. Machine learning is a more extensive field of artificial intelligence where we assist computer with learning things. Also similar researches with big data analysis is possible for developing a multi input multi output system [30,31].Deep learning method using adaptive
S. Radhakrishnan, S.G. Nair and J. Isaac / Computer Methods and Programs in Biomedicine 176 (2019) 43–49
neuro fuzzy system to adjust multiple ventilator settings for keeping blood oxygen saturation will be a definitive point of further research. Statements of ethical approval By following the relevant laws and institutional guidelines only data samples for the work were collected and the institutional ethical committee had approved the work. Competing interests The authors have no competing interest to declare. Supplementary materials Supplementary material associated with this article can be found, in the online version, at doi:10.1016/j.cmpb.2019.04.014. References [1] C.S. Wang, D. Shaw, K.S. Jih, An intelligent control system for ventilators, Med. Eng. Phys. 20 (7) (1998) 534–542. [2] H. Luepschen, L. Zhu, S. Leonhardt, Robust closed-loop control of the inspired fraction of oxygen for the online assessment of recruitment maneuvers, in: ConfProc IEEE Eng. Med. Biol. Soc., 2007, pp. 495–498. [3] M.H. Giard, F.O. Bertrand, D. Robert, J. Pernier, An algorithm for automatic control of O2 and CO2 in artificial ventilation, IEEE Trans. Biomed. Eng. (1985) 658–667. [4] J.R. Anderson, T.D. East, A closed-loop controller for mechanical ventilation of patients with ARDS, Biomed. Sci. Instrum. (2002) 289–294. [5] K.M. Goode., D.A. Linkens, P.R. Bourne, J.G. Cundill, Development of a fuzzy rule-based advisor for the maintenance of mechanically ventilated patients in ICU: a model-based approach, Appl. Basis Commun. 10 (4) (1998) 236–246 ISSN 1016-2372. [6] J. Kretschmer, T. Becher, A. Riedlinger, D. Schadler, N. Weiler, K. Moller, A simple gas exchange model predicting arterial oxygen content for various FiO2 levels, in: ConfProc IEEE Eng. Med. Biol. Soc., 2013, pp. 465–468. [7] A. Golcuk, H. Isik, I. Guler, Design and Construction of a Microcontroller-Based Ventilator Synchronized with Pulse Oximeter, J. Med. Syst. 40 (7) (2016) 180 Jul, doi:10.1007/s10916-016-0538-x. [8] A. Gölcük, I.˙ &Güler, The use of stepper motor-controlled proportional valve for fio 2 calculation in the ventilator and its control with fuzzy logic, J. Med. Syst. 41 (1) (2017) 1. [9] M. Bonfanti, A. Cammi, P. Bagnoli, Gas transfer model to design a ventilator for neonatal total liquid ventilation, Med. Eng. Phys. 37 (12) (2015) 1133–1140. [10] S. Cecchini, I. Sardellitti, S. Silvestri, A new methodology for intra-breath control of mechanical ventilation, Med. Eng. Phys. 34 (2) (2012) 256–260. [11] P.A. Dargaville, O. Sadeghi Fathabadi, G.K. Plottier, K. Lim, K.I. Wheeler, R. Jayakar, T.J. Gale, Development and preclinical testing of an adaptive algorithm for automated control of inspired oxygen in the preterm infant, Arch. Dis. Childhood-Fetal Neonatal, Ed., 102 (1) (2017) F31–F36.
49
[12] D. Mindrila, P. Balentyne, Scatter Plots and correlation, The Basic Practice of Statistics, 6th ed., Chapter4, https://www.westga.edu/academics/research/vrc/ assets/docs/scatterplots_and_correlation_notes.pdf, Accessed on: July 7 (2018). [13] L.S. Riza, C.N. Bergmeir, F. Herrera, J.M. Benítez Sánchez, Frbs: Fuzzy Rule-Based Systems For Classification and Regression in R, American Statistical Association, 2015. [14] D.G. Rossiter, An example of statistical data analysis using the R environment for statistical computing, Version 1.4 May 6 (2017) http://www.css.cornell.edu/ faculty/dgr2/teach/R/R_corregr.pdf, Accessed on: May 10 (2018). [15] E.P. Martinoni, ChA Pfister, K.S. Stadler, S. PM, D. Leibundgut, T. Bouillon, T. Bohlen, A.M. Zbinden, Model-based control of mechanical ventilation: design and clinical validation, Br. J. Anaesth. (2004) 800–807. [16] M. Susmitha, M. Prakash, Study of fuzzy logic in medical data analytics, Int. J. Pure Appl. Math. 119 (12) (2018) 16321–16342. [17] Y. Sun, I. Kohane, A.R. Stark, Fuzzy logic assisted control of inspired oxygen in ventilated newborn infants, in: Proceedings of the Annual Symposium on Computer Application in Medical Care, American Medical Informatics Association, 1994, p. 757. [18] MrsSusmita Prakash, M. Mishraand Dr., Study of fuzzy logic in medical data analytics, Int. J. Pure Appl. Math. 12 (2018) 16321–16342 119. [19] E. Hullermeier, Fuzzy methods in machine learning and data mining: status and prospects, Fuzzy Sets Syst. 156 (3) (2005) 387–406. [20] Ion Iancu, A Mamdani type fuzzy logic controller, In Fuzzy Logic-Controls, Concepts, Theories and Applications, InTech, 2012. [21] J. Schaublin, M. Derighetti, P. Feigenwinter, S. Petersen-Felix, A.M. Zbinden, Fuzzy logic control of mechanical ventilation during anaesthesia, Br. J. Anaesth. 77 (5) (1996) 636–641. [22] T. Nemoto, G.E. Hatzakis, C. William Thorpe, R. Olivenstein, S. Dial, J.H.T. Bates, Automatic control of pressure support mechanical ventilation using fuzzy logic, Am. J. Respir. Crit. Care Med. 160 (2) (1999) 550–556. [23] K M Goode, D.A Linkens, P.R. Bourne, J.G. Cundill, Development of a fuzzy rule-based advisor for the maintenance of mechanically ventilated patients in ICU: a model-based approach, Appl. Basis Commun. 10 (4) (1998) 236–246 ISSN 1016-2372. [24] J. Kretschmer, T. Becher, A. Riedlinger, D. Schadler, N. Weiler, K. Moller, A simple gas exchange model predicting arterial oxygen content for various FiO2 levels, ConfProc IEEE Eng. Med. Biol. Soc (2013) 465–468, doi:10.1109/EMBC. 2013.6609537. [25] C. Wagner, S. Miller, J.M. Garibaldi, A fuzzy toolbox for the R programming language, in: In Fuzzy Systems (FUZZ), IEEE International Conference, 2011, pp. 1185–1192. [26] Y. Sun, I. Kohane, A.R. Stark, in: Fuzzy logic assisted control of inspired oxygen in ventilated newborn infants, AMIA, 1994, pp. 757–761. [27] P. Chanyagorn, P. Kiratiwudhikul, Automatic control of fraction of inspired oxygen in neonatal oxygen therapy., The Institute of Electronics Engineers of Korea 5 (2) (2016) 107–116. [28] https://cran.r-project.org/web/packages/sets/sets.pdf. [29] E. Hullermeier, Fuzzy methods in machine learning and data mining: status and prospects, Fuzzy Sets Syst. 156 (3) (2005) 387–406. [30] A. Wang, M. Mahfouf, G.H. Mills, G. Panoutsos, Da Linkens, K. Goode, H.F. Kwok, M. Denai, Intelligent model-based advisory system for the management of ventilated intensive care patients. Hybrid blood gas patient model, ELSEVIER, Comput. Methods Programs Biomed. 99 (2010) 195–207. [31] L.A. Pastur-Romay, F. Cedrón, A. Pazos, A.B. Porto-Pazos, Deep artificial neural networks and neuromorphic chips for big data analysis: pharmaceutical and bioinformatics applications, Int. J. Mol. Sci. (2016) 1313.