https://doi.org/10.31449/inf.v45i4.3503 Informatica 45 (2021) 605–616 605
A Blockchain and NLP Based Electronic Health Record System: Indian
Subcontinent Context
Pranab Kumar Bharimalla
School of Computer Engineering, KIIT University, India
E-mail: pranab.bharimalla@gmail.com
Hammad Choudhury
Infosys Ltd., USA
E-mail: hchoudhury99@gmail.com
Shantipriya Parida
Idiap Research Institute, Martigny, Switzerland
E-mail: shantipriya.parida@idiap.ch
Debasish Kumar Mallick and Satya Ranjan Dash (Corresponding Author)
School of Computer Applications, KIIT University, India
E-mail: mdebasishkumar@gmail.com, sdashfca@kiit.ac.in
Keywords: blockchain, electronic health record, ResNet, CNN, natural language processing
Received: April 8, 2021
The healthcare system in the Indian subcontinent is plagued with numerous issues related to the access,
transfer, and storage of patient’s medical records. The lack of infrastructure to properly communicate and
track records between all key participants has allowed the distribution of counterfeit drugs, dependency
on unsafe methods of communication, and lack of trust between patients and providers. During the global
COVID-19 pandemic, the need for a robust communication and record tracking system has been further
emphasized. To facilitate efficient communication and mitigate the mentioned issues, a nationwide EHR
(electronic health record) system must be introduced to bring the healthcare system into digital space. To
further enhance security, efficiency, and cost, the innovation of Blockchain is introduced. Blockchain is
a decentralized data structure that allows secure transactions between untrusted parties without needing
a central authority. In this paper, a Hyperledger fabric-based Blockchain Electronic Healthcare Record
(EHR) system is proposed. The system is integrated with technologies such as NLP (Natural Language
Processing), and Machine Learning to provide users with practical features.
Povzetek: Predstavljen je elektronski zdravstveni zapis na osnovi bloˇ cnih in NLP tehnologij v kontekstu
Indije.
1 Introduction
The SARS-CoV-2 or commonly known as the Coronavirus
pandemic, has challenged healthcare systems worldwide
and has exposed vulnerabilities even among the best pre-
pared due to its uncertainty of transmission, the unavail-
ability of a patient’s proper medical history, and lack of
adequate contact tracing, to name a few examples. Track-
ing a threat such as this pandemic requires dynamic adap-
tation of resource deployment to manage rapidly evolving
care demands, ideally based on real-time data from a large
population sample. Healthcare issues plague all nations re-
gardless of development status due to environmental, eco-
nomic, or societal conditions; according to The Institute
of Medicine (IoM), over one hundred thousand people die
each year from preventable medical errors in the US [2].
While healthcare systems in developed nations are in no
way perfect as stated by IoM and proven by the pandemic,
developing or underdeveloped regions such as the Indian
sub-continent remain more vulnerable. Some common is-
sues that have arisen are due to the lack of communication
and tracking infrastructures, such as the influx of coun-
terfeit drugs in the market, dependability on handwritten
prescriptions, especially in remote areas lacking any com-
puter systems, and almost nonexistent integration between
healthcare and insurance systems. The inadequate infras-
tructure facilitates the lack of accountability for healthcare
providers and further damages relations with patients. Re-
solving the many health care issues faced in the Indian
subcontinent is a formidable challenge but can be signifi-
cantly improved with the implementation of an end-to-end
integrated Electronic Health Record (EHR). In comparison
to paper-based record keeping, a practice still utilized in
the Indian subcontinent, EHR has clear and distinct advan-
tages [7]. It is fair to say that EHR holds a lot of promise:
lower morbidity and mortality rates, better continuity of

606 Informatica 45 (2021) 605–616 P.K. Bharimalla et al.
care, increased efficiencies, fewer adverse drug reactions,
and, most importantly, lower healthcare costs. Paper-based
records are more susceptible to human error due to funda-
mental factors such as legibility or loss of the physical item,
causing a delay in treatment and possible fatality, which
could have been prevented[18]. Also, the impact of EHR
will undoubtedly be felt during the coming months as the
global effort to distribute and administer the Coronavirus
vaccine intensifies.
An ideal healthcare system should be open, afford-
able, innovative, and secure. Care costs should not ham-
per patients from receiving required treatments or buy-
ing medicines, and healthcare should be affordable to
all regardless of wealth[19]. In the current scenario,
an ideal healthcare system might seem far-fetched, but
there is always room for improvement using the latest
innovations[13]. As our digital infrastructure evolves, the
need for robust privacy is increasing. Because of the sensi-
tivity of an individual’s healthcare information, a breach
in the system could jeopardize the identity of a patient
and the reputation of providers. According to [10], the
patient’s data contains information that is highly prized
by cybercriminals. Few noticeable hacks of medical in-
formation such as AMCA Data Security Incident (ap-
prox. 25M records) and Anthem data breaches (approx.
80M records) caused enormous damage to the medical
system[20]. Blockchain technology could help secure and
protect sensitive patient information and is emerging as an
alternative to the conventional way to log transactions and
transfer data through a trusted intermediary to provide va-
lidity to the transaction [3].
So far, we have discussed various healthcare issues pre-
vailing in the Indian subcontinent, such as the lack of
complete end-to-end EHR interlinking between individu-
als, stand-alone hospital recording systems, extensive use
of handwritten prescriptions, unsafe hospital databases vul-
nerable to data manipulation by hospital authorities with-
out patient permission and lack of accommodation for
caregivers. We also understood how a Blockchain-based
EHR system can be a perfect solution to these problems.
This article proposes a patient-centered blockchain-based
EHR system that identifies patients using their national ID,
grants individual control over their health records, protects
against unauthorized data manipulation, and allows scal-
ability. Cloud integration stores large CT scans and X-
ray reports. An EHR patient-centered system must accu-
rately identify individuals using unique identifiers. There
is no end-to-end EHR system with national ID cards like
Aadhar or PAN (Permanent Account Number). Integrating
unique identifiers like national IDs is essential to success-
fully identify patients. The proposed scalable system will
enable existing EHR and healthcare databases to be inte-
grated along with other useful features, including a mobile
app-based interface to convert paper prescriptions to text
using Natural Language Processing (NLP) algorithms to
bring old paper-based medical records into the new system.
The paper is organized as follows: Section 2 briefly de-
scribes some related work and their major contributions as
well as research gaps. The proposed system architecture
is presented in Section 3, followed by the proposed algo-
rithms. Section 4 discusses the implementation strategy as
well as the analysis of the results. Finally, Section 5 pro-
vides the conclusion and suggestions for future research.
2 Related work
With the advent of different Blockchain platforms like Hy-
perledger Fabric, Ethereum, and Azure Blockchain Work-
bench, many patients centric, permission-based Blockchain
EHR schemes have been proposed in the literature. [17]
proposed a Permission-based EHR sharing system intend-
ing to enhance security and privacy. They also proposed a
design access control policy algorithm with a smart con-
tract and formulated a performance optimization mech-
anism of the system. However, this work ignores the
reusability of existing healthcare records sitting in individ-
ual hospital databases. [8] proposed a Blockchain-based
permissioned EHR system with the capability of data in-
tegration of local standalone EHR systems that house in
different hospitals or clinics. The framework proposed to
store metadata and access only in Blockchain whereas ac-
tual health records in the cloud. This is a novel concept,
but actual patient-sensitive EHR data resides in the cloud
and does not enjoy the immutability that Blockchain of-
fers. [15] tried to enhance the framework proposed in
[17]; and added few new modules like a chemist, insur-
ance, and doctor’s appointment. Even though the authors
formulated a comprehensive approach,the work ignores the
scalability of data and interoperability of existing EHR and
healthcare databases. [6] proposed a similar Blockchain-
based searchable encryption scheme for EHR that not only
brings convenience to patients, healthcare providers but
also to researchers. In the proposed system only, indexes
are added to the Blockchain, whereas actual patient in-
formation is stored in an encrypted format in the cloud.
There are different ways of encrypting data in the cloud
and achieving privacy preservations. [9] defined a novel
way of splitting EHR records into sub-messages and fi-
nally construct shares of EHR to store in different com-
puter nodes locally and upload the indexes in healthcare
Blockchain. [3, 11] evaluated the performance for common
public and private/consortium Blockchain-based health-
care systems using metrics such as memory consumption,
disk write and read performance, network data utilization,
transaction execution per unit time, and CPU usage with
consortium-based systems yielding the best performance
results. [4] proposed a framework called Blockchain-Based
Deep-Learning as a-Service. The framework shares EHR
records among multiple healthcare users and operates in
two phases to prevent collusion attacks through authenti-
cation and predicts possible future conditions for patients
through deep learning. [5] proposed a Blockchain-based
architecture that allows access to the database based on user

A Blockchain and NLP Based Electronic Health Record System. . . Informatica 45 (2021) 605–616 607
roles and enhances the traditional encryption system em-
ploying the Quantum blind signature to protect the system
from quantum attacks using hyperledger fabric.
On the other hand, [14] proposed a system that uses AES
cryptography to perform the cryptic operation and block
chaining it through the hash keys. In addition to that, their
proposed healthcare ecosystem includes a prediction model
to diagnose the disease of the patient with the deep learn-
ing algorithm. In addition to the related works mentioned
above, we also performed literature surveys on the ANN-
based text extraction model, which could contribute to our
work. [1] have implemented the Artificial Neural Network
(ANN) approach for text extraction from 64 different types
of prescriptions with 98% accuracy. [12] have proposed
a CNN common approach for extracting numeric text us-
ing handwritten numbers commonly found in India, includ-
ing Odiya, Telugu, Devnagari, Bangla, and English. The
Bangla characters were 95% accurate, Devanagari charac-
ters were 98.54% accurate, Odia characters were 97.2% ac-
curate, Telugu characters were 96.5% accurate, and English
characters were 99.10% accurate.
3 Proposed model
The proposed system prototype is based on Hyperledger
Fabric, an open-source distributed ledger technology built
to meet enterprise requirements. It is a widely used private
Blockchain option. Being a permissioned platform with
improved configurability and modularity using pluggable
consensus protocols, it is ideal for a range of industries
requiring a level of trust between known participants in a
governance model. For a transaction to occur, the user must
be admitted with the organization’s certificate authority,
received the means for network authentication, the chain
code must be deployed to the channel and installed on the
peers, and both parties must have agreed to the endorse-
ment policy. A transaction proposal using an SDK is con-
structed along with a signature to call a chain code function
with parameters to update the ledger. After approvals from
peers are received, the chain code is executed against the
current database, and the response, read set, and write set
are received; these values and the signature are sent back
to the SDK to be parsed and consumed by the application.
Once the application has validated the responses, the pro-
posal and response are bounded in a transaction message
for the ordering service; the ordering service creates blocks
of transactions per channel. The transactions are sent to
all peers in the channel, the peers complete a final valida-
tion, the ledger is updated, and the peer reports to the client
about transaction validation or invalidation.
3.1 System architecture
In the proposed patient-centric, Blockchain-based health-
care system, there are seven key participants, the patients
or the public, doctors or caregivers, pharmacies, labs, in-
surance company staffs, government institutions, and the
admin user. Assistance from the government is neces-
sary to implement a functional Blockchain-based EHR sys-
tem. The government provides credibility to insurance
companies, labs, pharmacies, doctors, and even patients
through national identification numbers. The proposed sys-
tem provides patients substantial control over their medical
records, including the right to read, write, authorize, and
revoke records in the Hyperledger Fabric Blockchain net-
work. Doctors work closely with patients to diagnose con-
ditions, plan treatments, and prescribe medication. Phar-
macies work in parallel with doctors to distribute medica-
tions to patients. Due to the many factors related to an ac-
curate diagnosis, labs are specialized in detecting distinct
conditions with the help of specific tools and trained pro-
fessionals. Insurance companies help share risk between a
large population, making the cost of healthcare affordable
for the public, especially during unexpected events such as
accidents. Doctors, pharmacies, labs, and insurance com-
panies can read and update the patient’s medical record in
the Hyperledger Fabric Blockchain network if access has
been provided. The admin is critical for system mainte-
nance, and they have unrestricted access to the system, in-
cluding the right to read, write, update, remove and grant
access to participants in the Hyperledger Fabric Blockchain
network. The admin’s enrollment certificate is obtained
from the certification authority. Implementing a national
health portal or EHR system is not possible without the
government’s involvement in India due to the sizable pop-
ulation of 1.4 billion. In the proposed system, the govern-
ment institution acts as a founder organization or a trusted
anchor that can provide credibility to hospitals, pharma-
cies, insurance companies, labs, and other institutions and
provide them with trusted roles to play. The participants
with trusted roles can create and issue credential schema
and definition to the public or patients
The participants could register through a client applica-
tion or SDK and request an enrollment certificate from a
Membership Service Provider (MSP) to the certificate au-
thority. An MSP allows peers to validate incoming trans-
actions and sign off endorsements. After receiving the en-
rollment request, the certificate authority issues the certifi-
cate and private key with a new ID to enroll the participant.
The Hyperledger Fabric Blockchain network distributes
all transactions. Participants such as doctors, pharmacies,
labs, insurance companies, etc., have different roles in the
system and are only granted access when authorized. In
the proposed system, an individual’s identity could be ver-
ified against a national identification number such as Aad-
har and be structured to contain identity information like
names, date of birth, gender, and other identifying infor-
mation which could be fetched from the Aadhar database.
Patients can use the client application to update details like
blood group, allergies, medications, insurance details, etc.
Once the transaction is submitted, it will be broadcasted
to the network. Endorsing peers will verify the transac-
tion and authenticate using their certification and private
key. The transaction will next go to the Orderer through

608 Informatica 45 (2021) 605–616 P.K. Bharimalla et al.
the SDK client. The Orderer creates a block and sorts the
blocks based on different ordering algorithms (viz crash
fault-tolerant) and broadcasts to the network peers. All
the committing peers validate the blocks once again and
check if it is from the correct Orderer and validate con-
flicts before committing. MS is the body that manages the
network identities of organizations and users; however, it
does not have access to medical records on the Blockchain
network. The MS verifies participants based on TIN/PAN
before enrolling them within the network. CouchDB, a
NoSQL database that stores data in JSON-based format,
is a popular database option used alongside Hyperledger
Fabric. A network is comprised of peer groups that hold
ledgers and smart contracts used to encapsulate shared net-
work processes and information. In Hyperledger Fabric,
transactions produced by smart contracts are contained in
a chain code. The key components and processes of the
proposed system are highlighted in the below architecture
Figure 1. The algorithm 1 explains enrollment of patients
whereas algorithm 2 refers to hospital, pharmacy, etc. addi-
tion to the network. Table 1 details the abbreviations used
in both the algorithms.
3.2 Data pulling and sharing
In a permissioned blockchain system, the patients have the
authority to decide who can read or update their records;
however, the admin reserves the right to grant access to
an institution in case of an emergency. Individual hos-
pitals will integrate their EHR system with a Blockchain
node and a web API that has full access to their local
EHR to convert any existing SQL records to No SQL for-
mat for storage within Blockchain. A hybrid data manage-
ment approach is utilized to facilitate EHR data scalability
where all key patient information, including demograph-
ics, allergies, medications, and access controls, are stored
in Blockchain, and sensitive medical files such as x-ray and
scanning reports are stored in private cloud storage using
encryption in Figure2.
In the proposed system, a patient can provide access to
institutions and participants, including hospitals, doctors,
insurance companies, and pharmacies, through the user in-
terface using the web or mobile app. The patient will need
to identify the participant requiring access, the category of
data to be shared, and the period until the data is accessi-
ble. If a patient has visited a particular institution in the past
and there are medical records contained in the local EHR
system not found on the Blockchain network, they can ini-
tiate a pull request using web apps. Once the pull request
is approved by the institution’s admin user, the web apps
will connect to the local EHR system to fetch relevant data,
insert patient information into the Blockchain network and
upload large files to the private cloud in an encrypted for-
mat.
3.3 Patient data management at hospital
During a routine or emergency hospital visit, the patient
provides the hospital access to their medical records on
the Blockchain network to check and amend their records
based on the latest assessment. The Hospital must have
a valid node on the Blockchain network and request keys
from the network admin to permit the login. The patient
will select the category of data to be shared and how long
the Hospital will have access to the records. Once the Hos-
pital has been provided access, they can read and update the
records for an individual using their Aadhar identification
card described in Figure3.
3.4 Patient data management at pharmacy,
patho lab and insurance firm
During a pharmacy visit to fulfill a prescription, the pa-
tient provides the pharmacy access to their medical records
on the Blockchain network, assuming the pharmacy has a
valid node on the network and has requested keys from the
network admin to enable login. The patient can provide
itemized permissions where the pharmacy will only have
access to select prescriptions through private datasets and
control how long the pharmacy has access to the records.
Once the pharmacy has been provided access, they can read
and update the records for an individual using their Aadhar
identification card.
During a lab visit at a specialized facility to assess spe-
cific conditions, the patient provides the lab access to their
medical records on the Blockchain network. The lab re-
quires a valid node on the network and must have requested
keys from the network admin to allow login. The patient
controls what data is to be shared and how long the lab
will have access. Once the lab has been provided access,
they can read and update the records for an individual us-
ing their Aadhar identification card. Large lab files such
as x-ray reports can be encrypted and uploaded to private
cloud storage. Similarly, Insurance firms need to update a
patient’s insurance and policy information for policy pro-
curement or medical expense claims. The patient provides
the insurance firm access to their medical records on the
Blockchain network if the institution has a valid node on
the network and has requested keys from the network ad-
min to authorize a login. The patient will select the type
of data to be shared and how long the insurance firm will
have access to the record. Once access has been provided,
they can read and update the records for an individual using
their Aadhar identification card.
3.5 Patient uploads old handwritten/
printed prescriptions and bills
A core proposal of this paper is to convert paper prescrip-
tions to text using Natural Language Processing (NLP) al-
gorithms to bring old paper-based medical records into the
new system through a mobile app-based interface. In the

A Blockchain and NLP Based Electronic Health Record System. . . Informatica 45 (2021) 605–616 609
Figure 1: Proposed architectural Framework for Blockchain-based Healthcare System. It depicts key participants, com-
ponents and transaction processes involved.
Figure 2: Secured Hospital data management and data
pulling process from local EHR.
below module in Figure 4, the system flow for NLP-based
data extraction is presented, highlighting the key compo-
nents.
3.5.1 Handwritten prescription data extraction
Convolutional Neural Network (CNN) CNN is a Neu-
ral Network (NN) that performs convolutional operations
instead of simple matrix multiplication operations; it is one
of the layers. The structure of CNN consists of a Convolu-
tion layer, pooling layer, and fully connected layer. Feature
extraction operations are done in the convolution layer, and
the output of it is passed to the activation function. The size
of output reduces by pooling layers and gives robust learn-
ing results for input data. By performing the convolution
layer and pooling layer multiple times, global features can
be obtained. In the end, extracted features are passed to the
fully connected layer for regression and classification.
Residual Network (ResNet) According to the image
processing research, the number of layers or depth of a
network is crucial for the performance of a model, but the
greater number of layers is responsible for the degradation.
Much research shows those types of degradation are not
caused by overfitting but due to the matter of optimization.
The ResNet can solve the degradation problem by intro-
ducing a residual framework.
Long Short-Term Memory (LSTM) RNN is also a
Neural Network that is specially designed for the process-
ing of sequential data. In time-series data, the output t - 1-
time step affects the decision of future time step t. So, RNN
is not able to solve long sequence data. i.e., is called a van-
ishing or exploding problem. LSTM was designed to re-
solve this issue of vanishing or exploding problems. LSTM
has an internal memory cell called a cell state. This gets the
previous output and determines which element should be
updated, erased, and maintained in the internal state vector.
These processes are handled by four gates, forget gate ft,
output gate ot, input gate gt. which are shown in Figure 5.
The proposed model consists of Resnet-50 and three
LSTM layers. To do no-linearity, Rectified Linear Unit
(RELU) activation function is used in every convolutional
layer. The images are divided into 28 sub-windows, so the
image height is equal to the height of text-line images. The
vector map will be produced by the last layer of convolu-
tion. The output of the last convolution fed into the first

610 Informatica 45 (2021) 605–616 P.K. Bharimalla et al.
Figure 3: High-level access flow diagram: The patient can add own records to the Blockchain or grants access to other
participants like Hospital, Laboratory, Pharmacy or Insurance company to add/update records.
Figure 4: NLP based data extraction flow showing the com-
ponents.
LSTM layer. By doing LSTM operations, the weights are
optimized. Initially, 0.001 is used as the rate of learning.
3.5.2 Printed prescription data extraction
For extracting printed prescription data, we have used the
Google tesseract model,[16] a pre-trained character made
by Google. For the extraction process, we have done some
preprocesses. For a more efficient model, first, we convert
the image into grayscale. After the greyscale operation, the
Otsu thresholding is next, where pixels are converted into
zeros and ones. During thresholding, some of the pixels
Algorithm 1: Patient Enrollment.
Input: Enrollment Certificate (E
C
) from
Certification Authority (C
A
)
Output: Successful registration of patient
Initialization:N
Admin
should be valid node.
N
Admin
can Write/Read/Remove/Update patients;
while (true) do
if P
AId
is validand FetchAadharRecord(P
AId
)
not null then
P
Rec
 FetchAadharRecord(P
AId
)
Add Patient (B
NT
,P
AId
)
Grant Access(P
AId
)
Create Record (P
AId
,P
Rec
,B
NT
)
else
Invalid(P
AId
)
end
bool chk (0 :malicious; 1 :genuine)
if !(behaviour(chk)) then
Remove Update (P
AId
)
else
Add Update (P
AId
)
end
end
may be lost. To restore those pixels, Erosion and Dilation
operations are performed where Erosion expands some pix-
els, and Dilation shrinks some pixels as shown in Figure 6a,
Figure 6b and Figure 7.
After these images complete preprocessing, the image
will be sent to Google-Tesseract. The tesseract will extract
all text from the image and send it to the network.

A Blockchain and NLP Based Electronic Health Record System. . . Informatica 45 (2021) 605–616 611
Figure 5: Resnet-LSTM approch for Text Extractions.
(a) Erosion Operation.
(b) Dilation Operation.
Figure 6: Pixel restoration using Erosion and Dilation op-
erations.
4 Implementation and result
analysis
4.1 Blockchain network setup
To realize the proposed architecture, hyperledger fabric
and Sandbox are utilized. Hyperledger is authentication
and distributed ledger-based platform. It is an open-source
technology used to implement different smart contracts
with constraints and logic over the network for applica-
tions. The smart contracts are implemented over the net-
work using the sandbox module. In Sandbox, the partic-
ipants are known, and the Blockchain is in the permis-
sioned consortium mode, making it a secure and trusted
Blockchain. The proposed architecture is not limited to
Figure 7: Some pixels as missed, and after the Erosion and
Dilation operation.
the healthcare domain. Programming languages such as
Node.js, Java, Go, etc., are used for contract and busi-
ness network development. Docker is used for setting
up and initialization when working with hyperledger fab-
ric and composer. Docker is an operating system-level
container used by developers, system administrators, etc.,
for creating, deploying, and running business networks or
hyperledger-based applications in a container, enabling the
dependencies and functionalities to be packaged together.
The hyperledger fabric and composer network can run in-
side a container using Docker.
In our simulation phase, we used a network model of
3 organizations with 2 peers each and one Orderer. The
experiment is carried out with basic writing transactions at
various rates, with 1000 transactions per round at 50, 100,
150, 200, and 250 transactions per second. The experiment
is done for 1 Org 1 Peer, 2 Org 2 Peer, and 3 Org 3 Peer
with different performances of transactions. The results are
calculated over five rounds, with each round consisting of
1000 transactions at various transaction rates per second

612 Informatica 45 (2021) 605–616 P.K. Bharimalla et al.
Algorithm 2: Hospital, Pharmacy, Patho Lab and
Insurance firm Enrollment.
Input: Enrollment Certificate (E
C
) from
Certification Authority (C
A
)
Output: Access to all nodesH
N
,P
N
,L
N
,I
N
Initialization:N
Admin
should be valid node.
N
Admin
can Read/Write/Update/Remove
participantsH
N
,P
N
,L
N
,I
N
;
while (true) do
if H
N
is validandH
TIN
is valid then
Add Node (B
NT
,H
N
)
Grant Access(H
N
)
else
Invalid(H
N
)
end
if P
N
is validandP
TIN
is valid then
Add Node (B
NT
,P
N
)
Grant Access(P
N
)
else
Invalid(H
N
)
end
if L
N
is validandL
TIN
is valid then
Add Node (B
NT
,L
N
)
Grant Access(L
N
)
else
Invalid(L
N
)
end
if I
N
is validandI
TIN
is valid then
Add Node (B
NT
,I
N
)
Grant Access(I
N
)
else
Invalid(I
N
)
end
end
bool chk (0 :malicious; 1 :genuine)
if !(behaviour(chk)) then
Remove Update (H
N
,P
N
,L
N
,I
N
)
else
Add Update (H
N
,P
N
,L
N
,I
N
)
end
(tps).
The graphs in figure 8a and 8b highlight the average la-
tency and throughput for varying transaction rates along
with the number of transactions completed per minute per
three network models. The network model using 1 Org
and 1 Peer had the lowest average latency and the highest
throughput per transaction rate while completing the high-
est number of tasks per minute. In contrast, the network
model using 3 Org and 3 Peer had the highest average la-
tency and the lowest throughput per transaction rate while
completing the lowest number of tasks per minute. The
network model using 2 Org and 2 Peer fell in between the
above results. (i.e 3org3peer > 2org2peer > 1org1peer).
So, we can conclude from figure 8a, that latency increased
as the system scaled up with more organizations and more
peers. Throughput of 1 org 1 peer is measured to be high-
est of 190 whereas it keeps decreasing with the number of
organization and peer increases. For 2 org 2 peer, it was
found to be 182, and the same for 3 org 3 peer was 180 in
Figure 8b. Figure 8c shows successfully completed trans-
actions per minute. 1 org 1 peer completed 5000 transac-
tions in around 4 minutes whereas the 2 org 2 peers com-
pleted 4500 transactions and 3 org 3 peers completed 4000
transactions at the same time. As a result, transaction time
has been observed to increase in perfect sync with the orga-
nization’s and peers’ growth. Figure 8d on the other hand,
highlights the CPU consumption per network model. For
different rates of transactions, the network models resulted
in varying average CPU usage. We observed that among all
peers, peer1.org1.example.com touched the highest CPU
utilization at a transaction rate of 200 per sec. whereas
peer0.org1.example.com recorded the lowest CPU utiliza-
tion at a transaction rate of 100 per sec. Table 2, details
other resource consumption parameters. With these exper-
iment results, we move forward to our next set of experi-
ments related to data extraction from prescriptions.
4.2 Handwritten prescription data
extraction- training and validation
In regions dominated by paper prescription usage, provid-
ing the ability to transfer vast amounts of existing data into
the digital space is essential. Allowing users to upload pre-
scriptions simplifies the transition to an electronic system
and populates the system with useful patient data given in
Figure 9a, 9b and 10. The following experiments are re-
lated to training and validation for data extraction from
handwritten prescriptions. The first step of training re-
quires the number of inputs, hidden layers, and output lay-
ers. Twenty handwritten prescriptions of different classes
are taken, including numeric characters, alphabetical char-
acters, spaces, and punctuation. To improve image accu-
racy or legibility, the prescription image may be taken by
section by users, causing the model to be confused and less
accurate. To overcome this issue, images are converted
into small segments. For analysis of images 64x64 pix-
els in size, 16 feature vectors are extracted from the feature

A Blockchain and NLP Based Electronic Health Record System. . . Informatica 45 (2021) 605–616 613
Abbreviation Explanation
P
Rec
Patient’s Record
B
NT
Blockchain Network
N
Admin
Admin Node
P
N
,H
N
,L
N
,I
N
Pharmacy, Hospital, Patho lab, Insurance firm nodes respectively
P
TIN
,H
TIN
,L
TIN
,I
TIN
Pharmacy, Hospital, Patho lab, Insurance firm tax Id’s respectively
Table 1: Abbreviation used in the algorithm and its explanations.
(a) Average Latency with Varying Transaction Rate.
(b) Throughput with Varying Transaction Rate.
(c) No. of Completed Transactions with Time.
(d) Resource consumption.
Figure 8: Measurement of different performance parameters.The results are calculated over five rounds, with each round
consisting of 1000 transactions at various transaction rates per second (tps).
Type Name CPU(avg) Memory(avg) Traffic-In Traffic-out disc Write
Docker peer0.org1.example.com 36.6 284.5MB 10.4MB 4.5MB 4.2MB
Docker peer0.org2.example.com 28.4 280.0MB 10.5MB 5.6MB 4.2MB
Docker peer0.org3.example.com 25.1 275.5MB 9.8MB 9.8MB 4.2MB
Docker Orderer.example.com 2.34 50.0MB 2.5MB 1.2MB 1.2MB
Table 2: Resource consumption of various parameters.
map. The feature vectors are produced by the convolution layer and are extracted from each sliding. The model is

614 Informatica 45 (2021) 605–616 P.K. Bharimalla et al.
trained by 11,000 samples, with the epoch starting from 0
to 10000. The learning rate is 0.0005 when the size of the
batch is 10. For this experiment, 2000 images are validated.
(a) Sample handwritten Prescription.
(b) Sample output.
Figure 9: Handwritten prescription and sample output.
CNN-LSTM ResNet-LSTM
Train 89.1 90.3
Test 81.6 88.3
Table 3: Model Accuracy for Handwritten Data extraction
approach.
According to the above results, the training accuracy was
improved after the 30th epoch. Finally, an 88.3% accuracy
result was accomplished using Test Dataset given in table
3. The generate model will take an input and generate a text
output which is pre-processed by string operations. After
the string operations are completed, the produced output
will be sent to the network.
Figure 10: Training graph sample of Resnet-LSTM with
output.
4.3 Printed prescription data extraction-
training and validation
The Tesseract-OCR is a pre-trained model created by
Google.
For improved image processing dilation, erosion meth-
ods are applied to Otsu’s thresholding. This approach
also provides better accuracy as mentioned by the Google
tesseract research. [16]. In our case, the test data is result-
ing in 99% accuracy. The illustration of the image is shown
in 11.
(a) Printed Prescription.
(b) Sample output.
Figure 11: Output of Tesseract OCR text extraction.
5 Conclusion and future work
EHR systems will be of considerable importance to ad-
vance the digital medical space of developing regions such
as the Indian subcontinent. With the advancement of
Blockchain technology, its potential has been recognized to
significantly impact the future of EHR systems due to the
superiority of Blockchain-based systems over traditional
systems and paper-based record keeping. Blockchain-
based EHR systems improve security, efficiency, and cost,
making it an excellent option for the Indian subconti-
nent. In this paper, we have highlighted some common

A Blockchain and NLP Based Electronic Health Record System. . . Informatica 45 (2021) 605–616 615
issues that have arisen due to the lack of communication
and tracking infrastructure, such as the influx of counter-
feit drugs, dependability on handwritten prescriptions, and
lack of integration between healthcare and insurance sys-
tems. We have discussed solutions to these issues using
Blockchain, NLP, Hyperledger Fabric and Docker Contain-
ers, etc. The proposed scalable system will allow inte-
gration of existing EHR and healthcare databases, national
identification, cloud technology to store large files with en-
cryption, and a mobile app-based interface to convert paper
prescriptions to text using OCR and deep learning tech-
niques, then to bring old paper-based medical records into
the new system. Our future research will include address-
ing implementation challenges at the grassroots level and
also collect more samples for training, to increase the ac-
curacy during converting the handwritten prescriptions into
text.
References
[1] R. Achkar, K. Ghayad, R. Haidar, S. Saleh, and
R. Al Hajj. Medical handwritten prescription recogni-
tion using crnn. In 2019 International Conference on
Computer, Information and Telecommunication Sys-
tems (CITS), pages 1–5. IEEE, 2019. https://
doi.org/10.1109/CITS.2019.8862004.
[2] A. Baker. Crossing the quality chasm: a new health
system for the 21st century, volume 323. British
Medical Journal Publishing Group, 2001. https:
//doi.org/10.1136/bmj.323.7322.1192.
[3] P. K. Bharimalla, S. Praharaj, and S. R. Dash.
Ann based block chain security threat mecha-
nism. International Journal of Innovative Tech-
nology and Exploring Engineering (IJITEE), 8
(10), 2019. https://doi.org/10.35940/
ijitee.J9442.0881019.
[4] P. Bhattacharya, S. Tanwar, U. Bodke, S. Tyagi,
and N. Kumar. Bindaas: Blockchain-based deep-
learning as-a-service in healthcare 4.0 applications.
IEEE Transactions on Network Science and Engi-
neering, 2019. https://doi.org/10.1109/
TNSE.2019.2961932.
[5] M. Bhavin, S. Tanwar, N. Sharma, S. Tyagi, and
N. Kumar. Blockchain and quantum blind signature-
based hybrid scheme for healthcare 5.0 applications.
Journal of Information Security and Applications, 56:
102673, 2021. https://doi.org/10.1016/
j.jisa.2020.102673.
[6] L. Chen, W.-K. Lee, C.-C. Chang, K.-K. R.
Choo, and N. Zhang. Blockchain based search-
able encryption for electronic health record shar-
ing. Future Generation Computer Systems, 95:420–
429, 2019. https://doi.org/10.1016/j.
future.2019.01.018.
[7] B. Devkota and A. Devkota. Electronic health
records: advantages of use and barriers to adoption.
Health Renaissance, 11(3):181–184, 2013. https:
//doi.org/10.3126/hren.v11i3.9629.
[8] A. Dubovitskaya, F. Baig, Z. Xu, R. Shukla,
P. S. Zambani, A. Swaminathan, M. M. Jahangir,
K. Chowdhry, R. Lachhani, and N. Idnani. Action-
ehr: Patient-centric blockchain-based electronic
health record data management for cancer care. Jour-
nal of medical Internet research, 22(8):e13598, 2020.
https://www.jmir.org/2020/8/e13598/.
[9] J. Fu, N. Wang, and Y . Cai. Privacy-preserving in
healthcare blockchain systems based on lightweight
message sharing. Sensors, 20(7):1898, 2020.
https://doi.org/10.3390/s20071898.
[10] G. Gavrilov, O. Simov, and V . Trajkovik. Blockchain-
based model for authentication, authorization, and
immutability of healthcare data in the referrals pro-
cess. 2020. http://hdl.handle.net/20.
500.12188/8179.
[11] C. Kombe, A. Sam, M. Ally, and A. Finne.
Blockchain technology in sub-saharan africa: Where
does it fit in healthcare systems: A case of tanzania.
Journal of Health Informatics in Developing Coun-
tries, 13(2), 2019. https://www.jhidc.org/
index.php/jhidc/article/view/24.
[12] D. S. Maitra, U. Bhattacharya, and S. K. Parui.
Cnn based common approach to handwritten char-
acter recognition of multiple scripts. In 2015
13th International Conference on Document Analysis
and Recognition (ICDAR), pages 1021–1025. IEEE,
2015. https://doi.org/10.1109/ICDAR.
2015.7333916.
[13] B. Mounia and C. Habiba. Big data privacy in
healthcare moroccan context. Procedia Computer
Science, 63:575–580, 2015. https://doi.org/
10.1016/j.procs.2015.08.387.
[14] R. Shanthapriya and V . Vaithianathan. Block-
healthnet: security based healthcare system using
block-chain technology. Security Journal, pages
1–19, 2020. https://doi.org/10.1057/
s41284-020-00265-z.
[15] A. P. Singh, N. R. Pradhan, S. Agnihotri, N. Jhanjhi,
S. Verma, U. Ghosh, D. Roy, et al. A novel patient-
centric architectural framework for blockchain-
enabled healthcare applications. IEEE Transactions
on Industrial Informatics, 2020. https://doi.
org/10.1109/TII.2020.3037889.
[16] A. P. Tafti, A. Baghaie, M. Assefi, H. R. Arab-
nia, Z. Yu, and P. Peissig. Ocr as a service: an
experimental evaluation of google docs ocr, tesser-
act, abbyy finereader, and transym. In International

616 Informatica 45 (2021) 605–616 P.K. Bharimalla et al.
Symposium on Visual Computing, pages 735–746.
Springer, 2016. https://doi.org/10.1007/
978-3-319-50835-1_66.
[17] S. Tanwar, K. Parekh, and R. Evans. Blockchain-
based electronic healthcare record system for health-
care 4.0 applications. Journal of Information Secu-
rity and Applications, 50:102407, 2020. https://
doi.org/10.1016/j.jisa.2019.102407.
[18] M. Thakkar and D. C. Davis. Risks, barriers, and
benefits of ehr systems: a comparative study based
on size of hospital. Perspectives in Health Infor-
mation Management/AHIMA, American Health Infor-
mation Management Association, 3, 2006. https:
//pubmed.ncbi.nlm.nih.gov/18066363/.
[19] D. Thompson, F. Velasco, D. Classen, and R. J.
Raddemann. Reducing clinical costs with an ehr: in-
vestments in performance management are essential
to realizing the full benefits of an ehr system–
including reduced costs and improved quality of care.
Healthcare Financial Management, 64(10):106–112,
2010. https://link.gale.com/apps/doc/
A243277528/AONE?u=anon~b7bb66fd&
sid=googleScholar&xid=4b596be9.
[20] V . Varadharajan, D. Bansal, S. J. Nair, et al.
Blockchain reinventing the healthcare industry: Use
cases and applications. In Industry Use Cases
on Blockchain Technology Applications in IoT
and the Financial Sector, pages 309–328. IGI
Global, 2021. https://doi.org/10.4018/
978-1-7998-6650-3.ch013.