Optimizing Hospital-Patient Interactions through Advanced Machine Learning and NLP Methodologies

International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
Volume: 10 Issue: 11 | Nov 2023 www.irjet.net p-ISSN: 2395-0072
© 2023, IRJET | Impact Factor value: 8.226 | ISO 9001:2008 Certified Journal | Page 895
Optimizing Hospital-Patient Interactions through Advanced Machine
Learning and NLP Methodologies
Ujwal K C 1 , Srividhya Ravichandran 2
1Student, Department of Information Science , Ramaiah Institute of Technology, Bangalore, India
2Student, Department of Information Science , Ramaiah Institute of Technology, Bangalore, India
--------------------------------------------------------------------***---------------------------------------------------------------------
Abstract - This research presents an innovative approach
to streamline the patient journey within the healthcare
ecosystem, leveraging the capabilities of DialogFlow and
Feedforward Neural Network(FNN) model. Users interact
with Google Assistant, detailing their symptoms which
intelligently recommends relevant departments in nearby
hospitals using Google Maps API and classifies them to a
department based on symptoms. The integration of BERT-
based NLP techniques in analyzing and summarizing
patient-doctor conversations reduces diagnostic errors,
aligning the doctor's diagnosis with the model's insights.
This innovation streamlines the hospital experience,
ensuring a more accurate and efficient diagnosis. The future
implications extend towards the evolution of patient-centric,
technology-driven healthcare experiences, marking a
significant advancement in the convergence of artificial
intelligence and medical services.
Key Words: DialogFlow, FNN, Machine Learning,
Google Assistant, Maps API, BERT, NLP
1. INTRODUCTION
The modernization of healthcare systems is imperative for
ensuring optimal patient experiences and efficient clinical
workflows. This paper introduces a comprehensive
intelligent appointment system designed to streamline
various aspects of the healthcare journey. By integrating
cutting-edge technologies such as natural language
processing, machine learning, and QR codes, the system
aims to optimize the appointment booking process,
enhance hospital check-in procedures, and augment
doctor-patient interactions.
It is crucial to acknowledge the existing drawbacks in
some of the current literature. Many studies lack a holistic
approach, often focusing on isolated aspects of healthcare
management. This literature review critically examines
these limitations in current papers and sets the stage for
proposing a more integrated and inclusive intelligent
appointment system and seeks to bridge these gaps by
identifying shortcomings, and proposing avenues for
further exploration and improvement in intelligent
healthcare systems.
1.1 LITERATURE REVIEW
[1] Kyambille and Khamisi (2015) conducted a study on
improving patient appointment scheduling through
mobile technology. The research introduces a mobile
application system that enables patients to conveniently
register and schedule appointments using their phones at
their preferred time. The system, developed using MySQL,
WAMP server, and PHP, facilitates efficient appointment
management. However, a limitation of the system is its
inability to redirect appointment requests to other
hospitals with doctors possessing similar expertise.
[2] This paper emphasizes the importance of prioritizing
health using Artificial Intelligence (AI) to deliver
convenient and affordable healthcare services. It discusses
the role of high-tech gadgets and AI applications,
particularly disease prediction through big data analysis.
The study introduces an online medical services platform
with a chatbot for disease prediction, employing Natural
Language Processing and Decision tree algorithms for
effective communication and accurate predictions.
[3] The literature highlights the positive impact of
technology on healthcare, emphasizing its role in
improving efficiency and accessibility. It advocates for AI-
driven healthcare services, showcasing how technology
enhances medical guidance through tools like disease
prediction chatbots, contributing to overall well-being.
[4] This paper highlights a comprehensive, integrated
information system designed to manage the
administrative, financial, and clinical aspects of a hospital.
Developed as an area of medical informatics, the system
aims to achieve the best possible support of patient care
and administration through electronic data processing. It
provides an intelligent front desk information service for
patients at the hospital entrance and software assistance
for doctors to diagnose patients quickly and accurately.
The system is designed to handle paper-based information
processing as well as data processing machines.
2. SYSTEM DESIGN
The overall workflow of the system is as follows - The user
(patient) interacts with Google Assistant in a language of

their choice to book an appointment based on the
symptoms they are facing. Under the hood, this interaction
happens with DialogFlow which has NLP capabilities to
identify the symptoms of the patient depending on the
information that the patient provides to the assistant.
Based on the user’s current location, Google Maps API, is
used to determine the nearest hospitals to the user. A
FNN(Feedforward Neural Network) based machine
learning model was used to analyze the symptoms and
predict the correct department within the nearest hospital
for the user to go to, along with a list of available doctors,
which the user can use to book an appointment.
Subsequently, a unique QR code, containing appointment
details and user information, is generated and emailed to
the user for convenient check-in at the hospital.
Upon arrival at the hospital, the user scans the QR code
using a barcode reader at the check-in counter. The
system retrieves information, displaying real-time wait
times and room details for the selected doctor within the
department This can help to expedite the check-in process,
minimizing wait times and enhancing overall efficiency.
Furthermore,users can actively monitor live waiting times
and the number of patients in the queue based on the
website.
A HealthScribe is used to record the conversations
between the patient and the doctor. This can be useful
when the doctor needs to diagnose the disease and can be
useful to keep the conversation on file.
Diagram -1: Workflow diagram
During the consultation, a health scribe system plays a
pivotal role in supporting the doctor-patient interaction.
Utilizing natural language processing and speech-to-text
technologies, the health scribe transcribes and analyzes
the conversation. This assistance aids the doctor in
diagnosing the patient's condition, mitigating human
error, and providing valuable insights.
Additionally, the health scribe system contributes to
comprehensive documentation, ensuring future reference
and analysis of the conversation for continuous
improvement and patient care enhancement.
2.1 IMPLEMENTATION
- Techstack: Flask, Keras, Google Maps API, AWS Health
Scribe, DialogFlow
- Identifying patients symptoms : DialogFlow was used
in order to identify patients symptoms by defining specific
intents. Intents could be ‘Cough’, ‘Headache’, ‘Fever’ etc.
Each intent is trained with a variety of training phrases
that users might use to describe their symptoms.
DialogFlow uses NLP internally which allows for a more
nuanced understanding as it analyzes the context and
extracts entities from the user's input. Entities represent
specific details within an intent, which could be severity of
a symptom, duration etc.
- Locating Hospitals + department : Google Maps API
was used to locate the hospitals near the user. FNN was
trained to predict medical departments in these hospitals
based on the identified symptoms. FNN is used due to
their ability to learn complex patterns in medical data,
capture nonlinear relationships between symptoms and
diseases, and automatically extract relevant features
through hidden layers, enabling effective and accurate
predictions. The Sequential model, built using Keras,
features enhanced capacity with 128 neurons in the initial
hidden layer, fostering a deeper understanding of intricate
patterns in symptom data. The addition of a second hidden
layer with 32 neurons further refines the model's ability to
discern nuanced relationships. The final layer, with
softmax activation ensures accurate and comprehensive
predictions for disease identification. The Flask webhook
retrieves information about the nearest hospitals with
high ratings for the predicted department based on Google
reviews.
- Generating QR Code : When the user selects a
department, DialogFlow communicates with the Flask to
generate a QR code without sensitive patient data using
the ‘qrcode’ library in Python. This QR code is sent to the
user via an email client. The user then scans the QR code at
the hospital, where there is no human receptionist,
automating the entire workflow. The unique ID in the QR
code is mapped to the hospital database.

Using the doctor's history of time taken for consultation,
and the room where the doctor is located, the current wait
time for the patient is displayed on the hospital screen.
The user also receives an email to click on the link to the
website which has real time updates regarding the wait
time.
This website is designed to provide seamless updates and
transparency for both healthcare providers and patients.
The frontend of the platform is developed using React, and
the backend using Flask. Patient data and queue
information are stored securely in Firebase along with
current status, room numbers, and waiting times. The
system also integrates with electronic health records
(EHR) through APIs, ensuring synchronization with the
latest patient data.
The system calculates and updates the waiting times
dynamically, allowing both healthcare providers and
patients to stay informed about the current status,
optimizing resource allocation and enhancing overall
patient experience.
- Healthscribe: The doctor-patient conversation is
analyzed using pyAudioAnalysis library to identify
whether the speaker is doctor or patient. This is sent to a
BERT based text summarizer that summarizes the
conversation. BERT's contextual embeddings enable the
generation of a concise summary of the entire
conversation. This summarized dialogue acts as a
comprehensive reference for the doctor during future
interactions with the patient, contributing to efficient and
accurate diagnoses. The seamless integration of audio
analysis and text summarization enhances the overall
documentation process in medical practice.
3. RESULTS
Diagram -2 : Model accuracy over epochs for predicting
departments based on symptoms
Diagram -3: User’s conversation with Google Assistant in
regional language (Hindi)
Diagram -4: Website for real-time updates of patient wait
time

Diagram -5: Speaker Identification and Text
Summarization Results
Diagram 4 showcases the evolving accuracy of the FNN
based ML model to predict the department based on the
patients symptoms throughout training, reaching a peak
accuracy of 98.6%. This illustrates the model's proficiency
in accurately predicting patient departments based on
provided symptoms, promising improved healthcare
diagnostics. Our speaker identification system achieved
90% accuracy for doctors and 85% for patients. Integrated
with a BERT-based text summarizer, our system yielded a
remarkable 75% accuracy in generating concise
summaries. These results underscore the effectiveness of
our approach in enhancing medical documentation and
doctor-patient interactions.
4. CONCLUSION
This holistic solution orchestrates a seamless patient
journey from symptom-based appointment booking using
Google Assistant to accurate hospital and department
predictions via Google Maps API and FNN models.
Automated QR code generation and real-time updates
reduce wait times, while the health scribe system ensures
thorough documentation and aids in precise diagnosis.
The tech stack, including Flask, TensorFlow, and
DialogFlow, forms a robust foundation for this patient-
centric approach. The system's adaptability to multiple
languages and its integration of cutting-edge technologies
represents a significant improvement in healthcare
efficiency and patient care.
5. FUTURE SCOPE
1. Integration of the system with IVR (Interactive Voice
Response) to help patients book appointments even
without a smartphone.
2. Support preliminary assessments by enabling
teleconsultations so that patients can get certain lab tests
done prior to the consultation.
3. Integrate ML to predict hospital resource requirements
based on historical wait times and patient data.
6. REFERENCES
[1] Kyambille, G. G., & Khamisi, K. “ Enhancing Patient
Appointments Scheduling that Uses Mobile
Technology” ,pp. 21-27
[2] Athulya N, Jeeshna K, S J Aadithyan, U Sreelakshmi,
Hairunizha Alias Nisha Rose, ”HEALTHCARE
CHATBOT”.
[3] Jian Huang, Jing Li, Zheming Li, Zhu Zhu, Chen Shen,
Guoqiang Qi, and Gang Yu Detection of Diseases Using
Machine Learning Image Recognition Technology in
Artificial Intelligence. Comput. Intell. Neurosci.
[4] B. Koyuncu and H. Koyuncu, "Intelligent Hospital
Management System (IHMS)," 2015 International
Conference on Computational Intelligence and
Communication Networks (CICN), Jabalpur, India,
2015, pp. 1602-1604, doi: 10.1109/CICN.2015.305.

Optimizing Hospital-Patient Interactions through Advanced Machine Learning and NLP Methodologies

More Related Content

Similar to Optimizing Hospital-Patient Interactions through Advanced Machine Learning and NLP Methodologies (20)

More from IRJET Journal (20)

Recently uploaded (20)

Optimizing Hospital-Patient Interactions through Advanced Machine Learning and NLP Methodologies