Maintenance Notice

Due to necessary scheduled maintenance, the JMIR Publications website will be unavailable from Wednesday, July 01, 2020 at 8:00 PM to 10:00 PM EST. We apologize in advance for any inconvenience this may cause you.

Who will be affected?


Citing this Article

Right click to copy or hit: ctrl+c (cmd+c on mac)

Published on 27.07.20 in Vol 8, No 7 (2020): July

This paper is in the following e-collection/theme issue:

    Original Paper

    An Artificial Intelligence Fusion Model for Cardiac Emergency Decision Making: Application and Robustness Analysis

    Hebei North University, School of Information Science and Engineering, Zhangjiakou, China

    Corresponding Author:

    Xiao Zhang, MD

    Hebei North University

    School of Information Science and Engineering

    No. 11 Zuanshi South Road



    Phone: 86 03134029808



    Background: During cardiac emergency medical treatment, reducing the incidence of avoidable adverse events, ensuring the safety of patients, and generally improving the quality and efficiency of medical treatment have been important research topics in theoretical and practical circles.

    Objective: This paper examines the robustness of the decision-making reasoning process from the overall perspective of the cardiac emergency medical system.

    Methods: The principle of robustness was introduced into our study on the quality and efficiency of cardiac emergency decision making. We propose the concept of robustness for complex medical decision making by targeting the problem of low reasoning efficiency and accuracy in cardiac emergency decision making. The key bottlenecks such as anti-interference capability, fault tolerance, and redundancy were studied. The rules of knowledge acquisition and transfer in the decision-making process were systematically analyzed to reveal the core role of knowledge reasoning.

    Results: The robustness threshold method was adopted to construct the robustness criteria group of the system, and the fusion and coordination mechanism was realized through information entropy, information gain, and mutual information methods.

    Conclusions: A set of fusion models and robust threshold methods such as the R2CMIFS (treatment mode of fibroblastic sarcoma) model and the RTCRF (clinical trial observation mode) model were proposed. Our study enriches the theoretical research on robustness in this field.

    JMIR Med Inform 2020;8(7):e19428





    Based on data released by the organizing committee of the 15th annual meeting of the Asian Society of Cardiovascular Surgery, we estimate there are currently 60 million potential heart disease patients in China, of which 8 million patients require cardiovascular surgery. We also surmise that globally two-thirds of heart attack patients can recover, but if ambulance service is not delivered in a timely or improper manner, the proportion of deaths will rise dramatically. In the United States, where medical resources are relatively abundant, the Institute of Medical Research's study of medical institutions in the past 40 years has demonstrated that 7% of patients have suffered as a result of serious medical errors. The World Health Organization reports that in countries such as Canada, New Zealand, and the United Kingdom, 10% of patients suffer from one medical adverse event every year [1]. The accuracy of the doctor's diagnosis, the implementation of ambulance service to patients, and a low error rate must be pursued. However, various errors frequently occur during medical treatment, which pose a threat to the life of patients [2].

    As a country with relatively scarce medical resources, there are few theoretical and practical studies in the fields of medical quality and efficiency in China. The existing research focuses primarily on hospital management and the medical expert system. Current research on improving medical quality and reducing adverse events during medical treatment is typically based on the experiences of countries and regions with advanced medical and health management, such as the United States, Singapore, Taiwan, and Hong Kong. From the perspective of management science, the hospital is regarded as a service industry. From the perspective of patients, it is important to strengthen hospital management, improve the level and quality of physical therapy services, and reduce medical costs [3]. With the robustness of medical emergency decision making as the starting point, research on how to improve the quality and efficiency of the medical system is scarce. Existing research focuses mainly on the attitudes of ambulance personnel in the operating room and the development of personnel training models. From a medical perspective, the current diagnosis and treatment of cardiovascular diseases focus primarily on the development of medical technologies for specific diagnoses and treatments, while neglecting the study of models and systems, with even less research on medical process control.

    The State Space of Cardiac Emergency Decision-Making Information and Its Description

    The CMO-EMPHA First Aid Process

    According to the CMO-EMPHA knowledge framework system, the basic cardiac emergency protocol includes nine major first-aid procedures: (1) rescue procedures and analysis of cardiac arrest, (2) cardioversion procedures and analysis, (3) emergency cardiac procedures and analysis, (4) a mechanical separation processing program and analysis, (5) a cardiac pacing program and analysis, (6) a bradycardia processing program and analysis, (7) an acute myocardial infarction rescue program and analysis, (8) a tachycardia processing program and analysis, and (9) first aid for ventricular (VT) fibrillation and VT procedures and analysis. In order to describe the process of cardiac emergency experts' observations, emergency decision making, and treatment of patients during the emergency decision-making process, Figure 1 shows the workflow description of the nine procedures mentioned above. The main operating nodes of this process include tracheal intubation, opening the airway, oxygen inhalation, suction, carotid pulsation examination, anterior cardiac stroke, cardiopulmonary resuscitation (CPR), defibrillation, pacing, and medication.

    Figure 1. The CMO-EMPHA emergency cardiac rescue process. MSA: multiple system atrophy; MSNA: muscle sympathetic nerve activity; OH: orthostatic hypotension.
    View this figure

    Network Chain of Complex Systems Centered on Cardiac Emergency Decision Making

    Definition of a Complex System Network Chain

    The information chain of heart disease emergency treatment can be described as an information set in the information source that conforms to certain time-dependent data constituting an information transmission chain structure and the information chain of emergency decision making during a heart attack (link i,i = 1,2,…,n) (Figure 2) [4]. The emergency treatment process of the heart disease itself is a feedback system composed of multitasking and time nodes. In this system, each data point forms a dynamic fusion, cooperation, and feedback process of multiple devices and information sources.

    Figure 2. Network structure of a complex system centered on cardiac emergency decision making. VF: ventricular fibrillation, VT: ventricular, ROSC: return of spontaneous circulation, ED: emergency department; CPC: cerebral performance category.
    View this figure
    Network Chain Composition of a Complex System

    The complex system network chain structure centered on heart disease emergency decision making is mainly composed of an emergency information resource system, a first aid rule system, a cardiac emergency rescue system, a medical staff organization structure, a monitoring system (which consists of six subsystems like aldosterone), and a diagnostic system. For example, the organizational structure of medical staff includes attending physicians, associate attending physicians, perfusionists, clinical engineers, etc, and information is connected and shared between the various subjects [5].

    The entire complex system network chain structure is composed of three interactive information chains. Information chain #1 includes an emergency information resource system, clinical path, and a diagnostic system; information chain #2 includes an emergency information resource system, clinical path, and a diagnostic system; and information chain #3 includes a cardiac emergency care system, medical staff organizational structure, and a diagnostic system. Each information chain has a separate subnet chain, and through the transmission channel Ch1, the complex system network chain structure can be shared and called. For example, the vital sign parameters of the patient are obtained from the emergency decision information resource system, including the attributes of the patient's vital signs parameters such as temperature, blood pressure, heart beat (f1,f2,…,fn), etc, to provide emergency decision information for the cardiac emergency specialist. Regarding the information chain of emergency decision making in cardiac emergency, due to the restriction of the resistance coefficient of channel network Ch1, the information transmission process needs to pass as few channels as possible to achieve the most efficient and seamless transmission of information.

    Definition of State Space for Cardiac Emergency Decisions

    The entire first aid process is a large complex system. The various data, information, and knowledge generated and used in it are employed as resources to support first aid decision making. They constitute a multilayer, heterogeneous, and massive complex information space. To represent this state, state space equations can be used, and subspaces can be transformed. A state space equation composed of matrices is a widely used form in subspace identification. Most of the subspace identification algorithms are carried out using subspaces and their mapping methods. In this paper, orthogonal projection is used to obtain the vector space for the state of first aid for heart disease [6].

    The static structure (L) of the cardiac emergency decision-making system consists of the emergency information space M, the emergency reasoning space Q, and the emergency decision space D. The cardiac emergency information space is a vector space defined on a real or complex field F as M. It consists of dimension space, and dimension space contains collection space, and collection space contains attribute space. Therefore, M can be represented by the following quaternion:

    M (x)Z, Se, At, Qij (1)

    Among them, Z, Se, and At represent dimension space, set space, and attribute space, respectively, and are composed of the emergency medical information vector. The dimension space constituting the emergency response decision M mainly includes the case dimension space Zc of case-based reasoning (CBR), the rule dimension space Zr of rule-based reasoning (RBR), the resource dimension space Ze of decision reasoning, and the time dimension space Zt of decision reasoning. The dimensional space is multidimensional, and the contained collection space or attribute space is heterogeneous. For example, the information in the vital sign collection Sev in Zc is different from the automatic control information in the medical emergency equipment collection SeQU in Ze. We introduce the definition of emergency reasoning space given in the cardiac emergency decision system below.

    The knowledge structure composed of all the data, information, and knowledge of the subdecision process in the cardiac emergency information space is called the emergency reasoning space, which is represented by Q. Q can be expressed as a quaternion:

    Q = (U, A, V, C) (2)

    Among them, U represents the fusion first aid reasoning space metadata, A is the feature set of the first aid reasoning space; its feature value range is V, , where ui is the feature value vector, and C is the solution of the first aid reasoning space. In Q, an inference information function is formed between C and U:

    f: U x A → C (3)

    According to the properties of the corresponding space vector in the state space, the above relationship shows that Q can be divided into several subspaces. For Q, according to the definition of the state space for cardiac emergency, there are:

    There is an invertible matrix Si,I = 1,2,…,n such that A – BD-1C = SiA-TSi-1.

    Cardiac Emergency Decision Data Space Framework

    In space three, the information space for cardiac emergency decision making is constructed. The dimension is an item-level division of the entire cardiac emergency decision-making system, including case dimension (CBR), rule dimension (RBR), resource dimension, and time window. The concept of subsets is divided under the dimensions, and there is a corresponding subset under each dimension. The next level of attributes describes these subsets and their relationships in detail. Different subsets contain different attributes, and the subset attributes of different dimensions are described in different ways. This vector space system reflects the most important characteristics of cardiac rescue and is the core foundation for the establishment of a cardiac rescue model [7]. The ternary composition of M is a “dimension-subset-attribute” frame (digital subtraction angiography frame), which describes the entire dimension space of the first aid of the heart by constructing a dimensional matrix space. The concept of subset space is used to describe the content of each dimension space, and the attribute space is used to construct a local model frame in each subset space.

    Dimension space (Z) is an 4D solid vector space structure model used to describe cardiac emergency information space. It mainly includes case-dimensional space, regular-dimensional space, resource-dimensional space and time-dimensional space. Z can be expressed as:

    Among them, Zc, Zr, Ze, and Zt refer to the case dimension, rule dimension, resource dimension, and time dimension, respectively.

    Subset space (Se) is a vector space structure model about a subset of features that exists in any dimension space in the 4D cardiac emergency information space. It can be seen that Zc is a dimension space formed by and represents the richness of the emergency decision-making experience of the first-aid heart disease specialists. Attribute space (At) refers to the specific combination of attribute vectors that constitutes the space of each subset, and can be expressed as:

    In the formula, atdi(se) represents the attribute component, subscript di represents the dimension, and se represents the subset component. Take the physical collection space Sev as an example to illustrate the correlation. Let body temperature, blood pressure, heart beat, and other attributes be denoted as a1,a2,a3,…,atCBR (Sev), the attribute vector of the knife dimension, so atCBR (Sev) = {a1,a2,a3,…,an}. The Dimension One Subset One Attribute framework of the cardiology emergency information space is shown in Figure 3. In Figure 3, a knowledge representation model of the state space for cardiac emergency response is constructed using the “dimensional-subset-attribute” framework [8].

    Figure 3. The "dimension-subset-attribute" framework of the cardiac emergency information space. OS: overall survival, IP: intraperitoneal, ASN: asparagine, GSM: Global System for Mobile Communications.
    View this figure

    Dynamic Performance of the Cardiology Emergency Reasoning Space

    The process of acquiring and transferring knowledge in the first aid reasoning space is complicated. In order to study the robustness of the first aid decision-making system, the dynamic performance of the first aid reasoning space needs to be studied. From these spaces containing decision-making reasoning information, a knowledge space (knowledge base) for decision-making processes is collected, extracted, and mapped from the cardiac emergency information subspace to the reasoning knowledge space. In the multidimensional cardiac emergency information space, the data sets required for decision-making reasoning are obtained and transmitted, and these data constitute heterogeneous knowledge of decision-making reasoning.

    The neutron space of complex cardiac emergency decision making is recessive, random, or uncontrollable, and the information generated shows system characteristics such as multidimensional, dynamic, and uncertain. In terms of inference knowledge transfer, for all the attributes obtained in the decision-making process of cardiac emergency, using attribute reduction, the initial inference state space S formed, eliminating attribute redundancy and retaining the relevant attribute set of information, reasoning for t temporal. The space S is used for the information characterization of relationship mapping, the role of reasoning, or the influence of uncertain factors to obtain the t + Δt temporal reasoning state Sn. In this way, the Ns step reasoning is performed, and finally, the emergency decision-making knowledge C is obtained, as shown in Figure 4 [9].

    Figure 4. The state chain of the cardiac emergency reasoning space.
    View this figure

    Regarding the state transition chain, for the two information state subspaces S = (U, AT, V, f) and S’ = (U’, AT’, V’, f’) of the state at any t and t + Δt in the emergency reasoning space Q, where AT, AT’A, S and S’ are the two reasoning state subspaces in Q.


    Robustness of Cardiac Emergency Decisions

    Definition of Robustness

    System robustness is a characteristic that maintains some of its original performance under certain internal or external parameter perturbations. The robustness complements the system’s vulnerability to ensure the overall safety of the system, which can reduce the uncertainty impact caused by the errors or parameter errors of the first aid reasoning model of cardiac disease. System fragility and system stability and robustness are two aspects of the same problem. Things that are robust in one aspect may be vulnerable in another aspect, or be robust in one set and vulnerable in another.

    From an application perspective, the objects of robust control have been extended from the initial basic control problems to multiple areas such as optimal control, system identification, decision making, and reasoning. Many scholars have carried out in-depth research work in the theoretical and application stages of robust control, which has extended the robustness research to many specific application levels, such as network theory, ecology, computer theory, mechanical vibration theory, pattern recognition, and reasoning. In the fields of prediction and system modelling, robustness research has achieved different degrees of results and has been applied in physical systems. Most indicators that measure a system comprehensively can be attributed to the system's stability (anti-interference ability), accuracy (steady-state error), and fastness (response speed). Cardiac emergency decision-making robustness can be divided into static performance and dynamic performance according to its state. Static performance refers to the state characteristics of the system when various factors inside the system, system input, and external interference are constant. Specifically, it includes system stability, steady-state error, and anti-interference. Dynamic performance refers to when the system has various internal factors. When factors, system inputs, or external disturbances change, the characteristics of the transition process of the system from one state to another, including the speed of response. These static and dynamic properties are interrelated and affect each other, and together they constitute the robustness of cardiac emergency decision making.

    Uncertainty in Cardiac Emergency Decisions

    Under the condition of information uncertainty or system disturbance, emergency decision making in cases of heart disease shows vulnerability and lacks robustness. Certainty means that a cardiac emergency specialist has sufficient information of adequate quality to accurately describe, reason, or make emergency decisions about the behavior or other characteristics of heart disease in a quantitative manner. What cannot be described by the above definition is called uncertainty. Cardiac emergency decision-making uncertainty refers to the hidden, random, or uncontrollable characteristics of the subspaces involved in emergency response in the decision of cardiac emergency, such as the difference in the medical level of medical personnel, technical limitations, and medical emergency equipment data errors, etc. The uncertainty of cardiac emergency decision making is reflected in Table 1. Turning these uncertain situations into deterministic situations requires more or better information.

    Table 1. Uncertainty in cardiac emergency decision making.
    View this table
    Robustness Constraints

    Robust stability criteria and sensitivity functions are the basis for further research on the robustness of cardiac emergency decision making. To this end, it is necessary to determine the robust constraints of cardiac emergency reasoning. Let v be the value of all deterministic variables in the model, such as the age attribute value of heart disease cases, etc; M (v) is a robust random mixed model; M0 (v) is a function matrix determined solely by deterministic variables, and M1 (v) and M2 (v) are uncertainty weight function matrix for random terms; θ is a random term for uncertainty; and I is an identity matrix.


    Modeling of CBR / RBR Fusion Inference Based on Robustness Threshold

    Robust Solution Set for the Fusion Inference Space

    The complex relationships in the knowledge source are mainly reflected in the three perspectives of similarity discrimination relationship, rule discrimination relationship, and inference reliability. The singular value of the fusion space is an intrinsic characteristic quantity calibration in the decision-making process of cardiac emergency, and a 3D robust threshold vector is constructed with the singular value.

    The ternary vector constructed by the case attribute value matrix P, the regular membership matrix Q, and the normalized singular value is called the inference robustness threshold vector.

    The singular value is used as the robustness threshold, and the inherent characteristics of the quantity matrix from case knowledge and rule knowledge are considered, which not only represents the correlation between the metadata of the knowledge source and the solution set of the decision target, but also reflects the threshold in the fusion space The internal differences overcome the strong subjectivity in the previous fusion methods, have strong robustness, and effectively improve the reliability of reasoning.

    Fusion Reasoning Strategies and Steps

    Considering the similar accumulation and reuse of knowledge sources, the simple CBR consists of main five steps, including the following: problem representation, case retrieval, solution transfer, feature mapping, and adjustment of noncorresponding solutions, and problem identification with RBR, generation of proposed solutions, recommendation evaluation, and screening. Based on a four-step method, including scheme modification, a robust threshold CBR is proposed. For the RBR fusion inference method, the specific implementation steps are as follows:

    • Step 1: Knowledge representation and problem identification. A case base CB and a rule base RB are generated on the knowledge source (N,Q) of the unitary space E. The target problem is identified, and the target problem set Cq is established, and the case matrix R, the rule matrix S, the robust solution x* of fusion inference, and the inference operator matrix T(x) are left blank [10].
    • Step 2: Fusion spatial mapping and normalization processing. Normalize the case data and rule knowledge Vij in the knowledge source (N,Q). The orthogonal discriminant method is used to obtain R for the case set, and the decision tree rules and S are obtained to form the fusion unitary space E of fusion reasoning.
    • Step 3: Determine robustness threshold and fusion inference strategy. Using singular value decomposition, the robustness threshold vector is obtained, and the relationship between the target problem Cq and the meta-knowledge in the knowledge base is defined in E, and then the fusion inference strategy is formulated.
    • Step 4: Integrate reasoning in knowledge space E and iterate repeatedly until a robust solution is obtained.
    Experimental Research and Application Effect Analysis

    The application effect of this thesis includes both theory and application. In terms of theory, based on mainstream research such as magnetic induction tomography and this hotspot of medical intelligent reasoning, a set of mathematical models such as the R2MIFS (represents the treatment mode of fibroblastic sarcoma) model and the RTCRF (clinical trial observation mode) model are established to solve the problem of robustness of cardiac emergency decision making. The R2MIFS model can effectively eliminate the attribute feature redundancy of patient characteristic data; the RTCI (the antimicrobial peptide Rana temporaria chensinensis) model can effectively implement resource strategy separation and conflict resolution mechanisms, implement inference knowledge fusion, and have obvious advantages in efficiency and accuracy compared with foreign literature; the BN (Bayesian network)-CBR / RBR model enhances the accuracy and sensitivity of inference decision making. In clinical trials, it has been experimentally verified that it has a good decision-supporting effect on the diagnosis of heart disease.


    From the perspective of the robustness of complex medical decision-making systems, our paper summarizes the evolution of related research content in the field over the years and the research results obtained, and proposes a three-level perspective on the research process. They are as follows: (1) research at the “behavioral level” based on industrial engineering theory (ie, medical industrial engineering); (2) research on the robustness of complex systems based on cybernetics (ie, medical system engineering); (3) “logical level” research based on information theory and artificial intelligence (ie, medical artificial intelligence).


    This project was supported partially by the Population Health Informatization in Hebei Province Engineering Technology Research Center, Hebei North University (#JYT2019016) and the Zhangjiakou Municipal Science and Technology and Seismological Bureau (#1911016C-9).

    Conflicts of Interest

    None declared.


    1. World Health Organization. 2004. World Alliance For Patient Safety Forward Programme 2005   URL: [accessed 2020-06-09]
    2. Jeganathan J, Knio Z, Amador Y, Hai T, Khamooshian A, Matyal R, et al. Artificial intelligence in mitral valve analysis. Ann Card Anaesth 2017;20(2):129-134 [FREE Full text] [CrossRef] [Medline]
    3. Vancini-Campanharo CR, Vancini RL, de Lira CAB, Andrade MDS, Lopes MCBT, Okuno MFP, et al. Characterization of cardiac arrest in the emergency department of a Brazilian University Reference Hospital: A prospective study. Indian J Med Res 2016 Oct;144(4):552-559 [FREE Full text] [CrossRef] [Medline]
    4. Bronzi W, Frank R, Castignani G, Engel T. Bluetooth Low Energy performance and robustness analysis for Inter-Vehicular Communications. Ad Hoc Networks 2016 Feb;37(P1):76-86. [CrossRef]
    5. Li Y, Wang X, Li H, Wang J, Zhu X, Li Y, et al. SU-F-T-192: Study of Robustness Analysis Method of Multiple Field Optimized IMPT Plans for Head & Neck Patients. Med. Phys 2016 Jun 07;43(6Part15):3506-3506. [CrossRef]
    6. Lee H, Troschel FM, Tajmir S, Fuchs G, Mario J, Fintelmann FJ, et al. Pixel-Level Deep Segmentation: Artificial Intelligence Quantifies Muscle on Computed Tomography for Body Morphometric Analysis. J Digit Imaging 2017 Aug;30(4):487-498 [FREE Full text] [CrossRef] [Medline]
    7. Kharin AY. An Approach to Asymptotic Robustness Analysis of Sequential Tests for Composite Parametric Hypotheses. J Math Sci 2017 Oct 11;227(2):196-203. [CrossRef]
    8. Iglesias López S, Llopis García G, Yañez-Palma MC, Rodríguez Adrada E. [Detection of palliative patients with acute cardiac insufficiency in the emergency department]. An Sist Sanit Navar 2016;39(2):323-324 [FREE Full text] [CrossRef] [Medline]
    9. Bhoi S, Mishra PR, Soni KD, Baitha U, Sinha TP. Epidemiology of traumatic cardiac arrest in patients presenting to emergency department at a level 1 trauma center. Indian J Crit Care Med 2016 Aug;20(8):469-472 [FREE Full text] [CrossRef] [Medline]
    10. Aguiar PR, Da Silva RB, Gerônimo TM, Franchin MN, Bianchi EC. Estimating high precision hole diameters of aerospace alloys using artificial intelligence systems: a comparative analysis of different techniques. J Braz. Soc. Mech. Sci. Eng 2016 Mar 19;39(1):127-153. [CrossRef]


    BN: Bayesian network
    CBR: case-based reasoning
    CPR: cardiopulmonary resuscitation
    RBR: rule-based reasoning
    RTCI: Rana temporaria chensinensis
    VF: ventricular fibrillation
    VT: ventricular

    Edited by K Kalemaki, Z Du, H Li; submitted 16.04.20; peer-reviewed by J You, S Yao, H Zhang; comments to author 30.04.20; revised version received 03.05.20; accepted 11.05.20; published 27.07.20

    ©Liheng Gong, Xiao Zhang, Ling Li. Originally published in JMIR Medical Informatics (, 27.07.2020.

    This is an open-access article distributed under the terms of the Creative Commons Attribution License (, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work, first published in JMIR Medical Informatics, is properly cited. The complete bibliographic information, a link to the original publication on, as well as this copyright and license information must be included.