Frontiers in Computing and Intelligent Systems 
ISSN: 2832-6024 | Vol. 2, No. 3, 2022 

 

94 

Trust Evaluation of Doctor Behavior Based on GCN 
Network in Medical Big-Data Access Control 

Xun Tian 

Yunnan University of Finance and Economics, School of Information, Kunming 650221g, China 

gk700.gjgcs@foxmail.com 

 

Abstract: The establishment of health care big data has brought great convenience to population health and medical research, 

but at the same time a series of privacy protection issues must be considered as a result. In this paper, we propose a graphical 

convolutional neural network to detect the access behavior of doctors in medical big data. In this paper, we propose a graphical 

convolutional neural network to model the access behavior of doctors in medical big data and perform trust evaluation, so as 

restrict such doctors or behaviors. In this paper, by taking the doctor behavior features and the doctor-doctor relationship network 

as input, the GCN network is used to supervise the learning of the department to which the doctor belongs, and the last layer is 

used as the characterization learning result. Finally, the similarity between doctor and department is used as the doctor behavior 

trust evaluation index. The experimental results show that the proposed model in this paper can well identify doctors' behaviors 

with malicious intent and assign a low trust value, laying the foundation for further research. 

Keywords: Trust evaluation; Medical big-data; GCN. 

 

1. Introduction 

With the rapid development of Internet information 

technology, all walks of life have stepped into the era of big 

data. Especially in the medical field, the era of big data has 

created an unprecedented and great convergence of digital 

with health and medical care. This convergence has enabled 

new medical research and new health services to manifest 

great value and potential. In the healthcare big data 

environment, based on data standards and data integration, 

doctors can call up patients' historical information and even 

genetic information during the treatment of patients. In this 

way, doctors can provide better and more targeted specific 

treatment plans to specific patients at specific points in time. 

And big data in healthcare can bring many benefits to medical 

institutions, government management departments and 

related enterprises. However, each stage of collecting, mining, 

analyzing and utilizing big data in healthcare requires a good 

health big data industry ecosystem. One of the basic and 

important features of this industrial ecosystem is "openness". 

However, the "openness" of health care big data will increase 

the risk of data leakage to a certain extent. For example, in 

September 2017, the service information system of a hospital 

in China was hacked, resulting in the leakage and trafficking 

of a large amount of citizens' information; in January 2018, 

hundreds of thousands of information about newborn babies 

and pregnant women due to unauthorized access by staff of a 

community health service center in China was leaked. 

Therefore, preventing the leakage of healthcare big data has 

become an urgent issue now. Currently, the main access 

control model uses a static authorization approach. It uses 

fixed policies and does not consider uncertainty and big data 

environments. It is difficult to apply in scenarios where 

authorization changes frequently in big data environments. 

The main direction of scholars in recent years is dynamic 

access control methods. Among them, risk-based and trust-

based are two more mainstream approaches. And this paper 

finds the doctors with malicious access behavior by modeling 

the trust value of doctors' access behavior. 

2. Related works 

In the context of big data in health care, medical 

information construction is bound to break the traditional data 

silos and move toward sharing and openness. However, in the 

process of data "flow", there are many hidden problems. 

Among them, the privacy leakage of data has become the core 

concern of big data in health care. At present, many scholars 

have conducted research on the privacy leakage of data and 

proposed many solutions in terms of technology. Examples 

include pseudonymization of patient identity, encryption of 

patient data and information, creation of public and private 

clouds to handle sensitive data, privacy-protected data 

distribution, privacy-centric access control and data 

outsourcing, and dynamic reconstruction of data [1-4]. In the 

healthcare domain, Soceanu propose an advanced encryption 

scheme and attribute-based access control authorization 

framework for clinical data privacy and security issues. 

Allowing partial visibility and effective protection of 

authorized parts. Suneetha . proposed a new framework spark 

using K-anonymization and L-diversity to mask patient 

sensitive information. It ensures that the shared data does not 

reveal the original medical data. And isolate sensitive data 

before transferring to the system. Literature [7,8] used 

blockchain technology for effective management of big data 

set access control to solve the data privacy security problem. 

Lee designed a medical big data privacy protection system 

based on Diffie-Hellman protocol to protect patients' private 

information and avoid the leakage of medical data. Wu built 

a medical big data privacy protection platform based on IoT . 

For medical big data transmission sharing security. Medical 

data security is provided. Y Yang proposed a privacy-

preserving electronic health record (EHR). and developed a 

policy update mechanism based on keyword matching. To 

achieve flexible access updates without compromising 

privacy. Mariagrazia proposed an ABAC model-based subject 

and object access control principle based on risk scenarios. It 

achieves dynamic modification of authorized adaptive access 



 

95 

control based on risk values. L Chen proposed a risk-aware 

RBAC model. The model consists of three components: user's 

trust, user's ability to assume roles, and appropriateness 

between roles and permissions. Riaz added trust and risk 

relative to traditional access control considering different 

sensitivity requirements of different applications. Two 

dynamic risk-based access control system decision methods 

are proposed. 

To address the above problems, this paper transforms 

doctors' high-dimensional access features into low-

dimensional section relevance features based on their 

historical access records using graph convolutional neural 

networks, and then uses the Euclidean distance similarity 

between each section doctor and the section center feature as 

the reputation evaluation index based on the reduced-

dimensional features. Therefore, the research in this paper 

provides a new idea for the access control research of 

healthcare big data. 

3. Trust evaluation method 

In this paper, we focus on supervised learning by using the 

graph of relationships between doctor access object features 

and doctors as the input to the GCN network. 

3.1. Doctor Characteristics and Relationship 

Chart 

For an electronic case 𝐸𝑐 , let there be m kinds of data 

related to diagnosis and treatment in the case, such as blood 

and urine laboratory data, ECG, chest X-ray, clinical 

symptoms, etc.. Use the one-hot code to indicate whether 

there is a record of the corresponding data in the case 𝐸𝑐 

table, noted as 𝐸𝑐 = [𝑓1, 𝑓2, … , 𝑓𝑚], where 𝑓𝑖 = 1 means that 

item ⅈ has a record. 

A doctor's access record 𝐷𝑟  should contain the case 

records of all non-primary patients accessed within a certain 

work duration, denoted as 𝑅𝑑 = [𝐷𝑟1, 𝐷𝑟2, … , 𝐷𝑟𝑛] . Since 

any Rp_i is a vector of fixed dimensions, 𝑅𝑑  can also be 

represented as a two-dimensional matrix as follows: 

𝐷𝑟 = [
𝑓11 … 𝑓1𝑚

⋮ ⋱ ⋮
𝑓𝑛1 … 𝑓𝑛𝑚

] 

The data features required for a doctor access, Rd, can be 

characterized based on all historical data features from the 

doctor's access to the non-primary patient's medical record as 

follows: 

𝐷𝑟 = [∑ 𝑓𝑖1

𝑛

𝑖

, ∑ 𝑓𝑖2

𝑛

𝑖

, … , ∑ 𝑓𝑖𝑚

𝑛

𝑖

] 

A doctor access to a patient case creates a relationship 

between the doctor and the case, and the element level of the 

weighted adjacency matrix 𝐴𝑑2𝑐  of the doctor-case 

relationship graph is represented as: 

𝐴𝑑2𝑐(𝑖, 𝑗)
= 𝑘 (𝐷𝑜𝑐𝑡𝑜𝑟 𝑖 𝑎𝑐𝑐𝑒𝑠𝑠 𝑚𝑒𝑑𝑖𝑐𝑎𝑙 𝑟𝑒𝑐𝑜𝑟𝑑 𝑗 𝑘 𝑡𝑖𝑚𝑒𝑠) 

Doctors who access the same case in the diagnostic process 

by correlation, and therefore can be transformed from a 

doctor-case relationship graph to a doctor-doctor relationship 

graph, whose weighted adjacency matrix 𝐴𝑑2𝑑  can be 

calculated as following: 

𝐴𝑑2𝑑 = √𝐴𝑑−𝑐 ⋅ 𝐴𝑑−𝑐
𝑇  

The Laplace matrix is the central object used to study the 

structural properties of the graph. The Laplace matrix is 

defined as follows: 𝐿 = 𝐷 − 𝐴, 𝐷 is the diagonal matrix of 

degrees of 𝐴 . The Laplacian matrix also has a regularized 

representation: 

𝐿𝑠𝑦𝑚 = 𝐷−
1
2𝐿𝐷−

1
2 

In this paper we use a two-layer GCN network with fixed 

graph filters: 

�̃�𝑠𝑦𝑚 = 𝐿𝑠𝑦𝑚 + 𝐼 

𝐻 = 𝜎(�̃�𝑠𝑦𝑚𝑋𝑊1 + 𝐵1) 

𝑌 = 𝑠𝑜𝑓𝑡𝑚𝑎𝑥(�̃�𝑠𝑦𝑚𝐻𝑊2 + 𝐵2) 

Similarly, the central feature points of all doctors in each 

department are found by deep learning, and the cosine 

similarity between each doctor and the central feature point is 

calculated separately as the trust evaluation index: 

𝑇𝑟𝑢𝑠𝑡𝑖 = 𝑠𝑖𝑚(𝑦𝑖 , �̂�) =
𝑦𝑖�̂�

‖𝑦𝑖‖‖�̂�‖
 

3.2. Simulation experiment 

This experiment was conducted by computer simulation to 

generate 10 departments with different medical records of 50 

doctors in each department for access, totaling 200,000 access 

of data. The sequence of 40 doctors with normal access and 

those with malicious access in each department. 

 
Fig 1. Similarity of the between doctors and the center of the 

department 

Through simulation experiments we get the results in Fig 1. 

According to Figure 1 we can easily find the low similarity 

between malicious doctors and section averages. It shows that 

our method effectively distinguishes doctors with malicious 

accessing behavior. 

4. Conclusion 

In this paper, we propose a trust evaluation model for 

medical environment, which uses graph neural network 

algorithm to reduce the high-dimensional access data of all 

doctors into low-dimensional section relevance data based on 

the historical access records of doctors, and then learns the 

average features of the section based on the reduced features 

of doctors in the same section, and cosine similarity to 

evaluate the trust value of doctors within the section. The trust 

model proposed in this paper is different from the traditional 

trust model, this model can not only be based on whether the 

data of doctors' access to patient's medical records are related 

to diagnosis and treatment, but also based on the network 

topology relationship between doctors at the same time, this 

model is more conducive to give credibility judgment on 

doctors' lines, which brings great reference value for the 



 

96 

subsequent access control research. The model is proved to be 

effective in the simulation experiments at the end of the paper. 

References 

[1] AGGARWAL C C, PHILIP S Y . A general survey of privacy-
preserving data mining models and algorithms [M]. Privacy-
preserving data mining. Springer. 2008: 11-52.  

[2] FERNáNDEZ-ALEMáN J L, SEñOR I C, LOZOYA P Á O, et 
al. Security and privacy in electronic health records: A 
systematic literature review [J]. Journal of biomedical 
informatics, 2013, 46(3): 541-62.  

[3] ABBAS A, KHAN S U. A review on the state-of-the-art 
privacy-preserving approaches in the e-health clouds [J]. IEEE 
Journal of Biomedical and Health Informatics, 2014, 18(4): 
1431-41.  

[4] CAMARA C, PERIS-LOPEZ P , TAPIADOR J E. Security 
and privacy issues in implantable medical devices: A 
comprehensive survey [J]. Journal of biomedical informatics, 
2015, 55(272-89.  

[5] SOCEANU A, VASYLENKO M, EGNER A, et al. Managing 
the privacy and security of ehealth data; proceedings of the 
2015 20th International Conference on Control Systems and 
Computer Science, F, 2015 [C]. IEEE.  

[6] SUNEETHA V, SURESH S, JHANANIE V. A Novel 
Framework using Apache Spark for Privacy Preservation of 
Healthcare Big Data; proceedings of the 2020 2nd International 
Conference on Innovative Mechanisms for Industry 
Applications (ICIMIA), F, 2020 [C]. IEEE.  

[7] CHEN X, ZHU H, GENG D, et al. Merging RFID and 
Blockchain Technologies to Accelerate Big Data Medical 
Research Based on Physiological Signals [J]. Journal of 
Healthcare Engineering, 2020.  

[8] UCHIBEKE U U, SCHNEIDER K A, KASSANI S H, et al. 
Blockchain access control Ecosystem for Big Data security; 
proceedings of the 2018 IEEE International Conference on 
Internet of Things (iThings) and IEEE Green Computing and 
Communications (GreenCom) and IEEE Cyber, Physical and 
Social Computing (CPSCom) and IEEE Smart Data 
(SmartData), F, 2018 [C]. IEEE.  

[9] LEE N-Y, WU B-H. Privacy Protection Technology and 
Access Control Mechanism for Medical Big Data; proceedings 
of the 2017 6th IIAI International Congress on Advanced 
Applied Informatics (IIAI-AAI), F, 2017 [C]. IEEE.  

[10] WU X, ZHANG Y , WANG A, et al. MNSSp3: Medical big 
data privacy protection platform based on Internet of things [J]. 
NEURAL COMPUTING & APPLICATIONS, 2020,   

[11]  YANG Y, ZHENG X, GUO W , et al. Privacy-preserving 
fusion of IoT and big data for e-health [J]. Future Generation 
Computer Systems, 2018, 86(1437-55.  

[12] TEIMOURIKIA M F A M. Access Control Privileges 
Management for Risk Areas [J]. Information Systems  for 
Crisis Response and Management in Mediterranean Countries, 
2014,   

[13] CHEN L, CRAMPTON J. Risk-aware role-based access 
control; proceedings of the International Workshop on Security 
and Trust Management, F, 2011 [C]. Springer.  

[14] SHAIKH R A, ADI K, LOGRIPPO L. Dynamic risk-based 
decision methods for access control systems [J]. computers & 
security, 2012, 31(4): 447-64.