Medical Chatbot for Patients

This repository details the development of a Medical Chatbot designed to provide patients with personalized and immediate access to medical information and services, utilizing AI and NLP techniques.

Introduction
Problems and Solutions
Motivation
Solution Requirements
System Architecture (Framework)
Methodology
Deployment
Project Management
- Task Distribution
Dataset Information
Reference Papers

Introduction

Project Background

Medical chatbots, powered by AI technology, provide personalized and convenient access to medical information and services, acting as virtual assistants for users(patients). They offer immediate responses to inquiries, guidance on health issues and medication management. This chatbot utilizes NLP techniques and vast medical data to enhance precision, empower users to seek accurate medical advice, make informed decisions, and access reliable healthcare information efficiently.

Motivation

By addressing the challenges mentioned, the chatbot aims to achieve the following objectives:

Empower users to make informed decisions about their health and medical conditions, enhancing overall health literacy and promoting proactive healthcare management.
Provide a convenient and efficient platform for users to access medical information anytime, anywhere, facilitating timely decision-making and intervention in healthcare matters. This project will serve as a practical application of the concepts studied in the NLP course and reinforce our learning and showcasing the real-world relevance of the technologies and strategies employed in improving healthcare accessibility and patient outcomes.

Problems and Solutions

Problems

Limited Access to Healthcare Professionals: Long wait times for appointments,overburdened healthcare systems, geographical barriers and financial constraints results in the limited access to healthcare professionals
Delayed Medical Attention: Lack of awareness or understanding of the seriousness of symptoms contributes to delayed medical attention, especially in case of Cancer.
Technological problems: Traditional chatbots were rule-based and had limited understanding of natural languages and inability to learn and adapt which resulted in limited accuracy.

Solutions

The key factors to deploy medical chatbots include scalability, availability, data collection, instant information provision, and cost efficiency.

Scalability: Efficiently handles increased demand without compromising quality
Availability: Provides 24/7 instant support for patients
Data Collection: Analyzes patient data for informed insights and personalized care
Instant Information: Offers rapid and accurate responses to inquiries
Cost Efficiency: Optimizes operations, reduces costs, and streamlines tasks

Related Works

1. Buoy Health

Features: Buoy Health is a conversational AI chatbot designed to provide personalized health assessments and guidance to users. It uses natural language processing to understand users' symptoms, medical history, and concerns, and then offers tailored recommendations such as possible conditions, next steps, and self-care advice. Buoy Health aims to empower users to make informed decisions about their health and navigate the healthcare system more effectively. Objectives: The primary objective of Buoy Health is to improve healthcare access and outcomes by leveraging AI technology to deliver accurate and reliable health information and support to users. It aims to reduce unnecessary healthcare visits, alleviate patient anxiety, and facilitate early detection and management of medical conditions. NLP Architecture: Buoy Health employs a combination of NLP techniques, including rule-based systems, machine learning models, and medical knowledge graphs. It uses advanced algorithms to analyze user input, extract relevant information, and match symptoms to potential conditions. The architecture may include components such as named entity recognition (NER), intent classification, and dialogue management systems.

2. Ada Health

Features: Ada Health is an AI-driven health assessment platform that offers personalized symptom analysis and health advice through conversational interactions. Users can describe their symptoms to Ada, and the chatbot will ask relevant questions to gather more information and provide possible explanations and recommendations. Ada Health aims to empower individuals to manage their health proactively and seek appropriate medical care when needed. Objectives: The main objective of Ada Health is to democratize access to healthcare by leveraging AI technology to deliver comprehensive and reliable health assessments and recommendations to users worldwide. It aims to augment healthcare services, improve health literacy, and promote early intervention and prevention of diseases. NLP Architecture: Ada Health utilizes a sophisticated NLP architecture consisting of deep learning models, probabilistic algorithms, and medical knowledge bases. It employs techniques like natural language understanding (NLU), sentiment analysis, and probabilistic reasoning to interpret user input, generate hypotheses, and provide personalized health insights. The architecture may incorporate pretrained language models, medical ontologies, and domain-specific knowledge graphs to enhance accuracy and relevance.

Gap Between Related Work

The existing landscape of chatbots available on the market predominantly revolves around facilitating interactions based on users' medical history. These chatbots excel in managing and tracking users' existing medical conditions, enabling communication with healthcare providers, and aiding in the coordination of care based on past medical records. While these functionalities are invaluable for personalized healthcare management, they often fall short in addressing broader aspects of healthcare, such as medical health literacy, accurate diagnosis of new symptoms, and providing reliable information for diverse medical queries.

In contrast, our medical chatbot for patients is designed to fill this crucial gap by focusing on enhancing users' medical health literacy, providing accurate diagnoses, and addressing a wide range of medical queries beyond historical data. We have leveraged advanced language understanding and generation models, such as Seq2Seq and GPT-2, to develop a conversational chatbot capable of comprehensively addressing users' medical-related concerns. Through the integration of advanced language models, we strive to deliver a user-friendly and intelligent conversational interface that enhances health literacy, supports accurate diagnosis, and fosters proactive healthcare management.

Solution Requirements

Utilize advanced NLP for accurate comprehension of user queries and responses.
Integrate with a constantly updated medical knowledge base.
Conduct trials to identify the most effective model.
Ensure smooth integration with a web application
Offer a user-friendly interface to increase user satisfaction.

System Architecture (Framework)

Dataset Information

We have used the MedQuad dataset from :

HuggingFace Dataset: https://huggingface.co/datasets/keivalya/MedQuad-MedicalQnADataset

Methodology

1. Data Collection and Preprocessing

Datasets

Sources: Leverage the MedQuad dataset and supplementary datasets from Huggingface and GitHub.

Data Preprocessing

We have used the GPT2TokenizerFast from the transformers library to tokenize text efficiently for processing with the GPT-2 model.
Similarly, for the seq2seq model, we’re using NLTK library for carrying out the tokenization.
Dataset is further splitted into training, validation and testing pairs on 80%, 10%, and 10% respectively

2. Model Development

We have trained 2 models: Seq2Seq and GPT2.
Integrate advanced NLP techniques like validation and perplexity,adaptation validation seq2seq for high quality performance.

3. Experimental Design

We conducted experiments to train and fine-tune two models, namely GPT-2 and Seq2Seq, for medical question-answering (QA) tasks. Below are the details of the experimental design and hyperparameters used for each model:

a. GPT-2 Model:

Training Data: We utilized a dataset consisting of 143,250 samples for training the GPT-2 model.
Fine-tuning: The GPT-2 model was fine-tuned on medical QA data to adapt it specifically for medical-related inquiries. Hyperparameters for Fine-Tuning:
Evaluation Strategy: Epoch-based evaluation
Learning Rate: 2e-5
Weight Decay: 0.01
Number of Training Epochs: 3

b. Seq2Seq Model:

-Training Data: Similar to the GPT-2 model, we used the same dataset containing 143,250 samples for training the Seq2Seq model. Hyperparameters:

Hidden Layers: 512
Number of Iterations: 15,000
Teacher Forcing Ratio: 0.5

MedicalChatBot

Install / Use

README