Gestational diabetes mellitus (GDM) is a growing health concern that affects many pregnant women worldwide, including in Uganda. Despite its prevalence, access to accurate and comprehensive information on GDM remains a significant challenge.
Traditional methods of information retrieval can be cumbersome and may not provide the tailored responses needed to manage and mitigate the risks associated with GDM effectively. This is where our innovative project, "Digital GDM Screen," comes into play.
Understanding the Need
GDM is a condition characterized by high blood sugar levels that develop during pregnancy and usually disappear after giving birth. It poses serious health risks to both the mother and the baby if not managed properly. In Uganda, the need for reliable and easily accessible information on GDM is critical, as many women and healthcare providers struggle to find trustworthy resources.
GDM in Uganda: The Statistics
According to recent studies, the prevalence of GDM in Uganda ranges from 2% to 9% of all pregnancies.
A study conducted at Mulago National Referral Hospital in Kampala found that 8.1% of pregnant women screened had GDM.
Despite these significant numbers, many women remain undiagnosed and untreated, leading to increased risks of complications.
Introducing the Digital GDM Screen
The Digital GDM Screen is an intelligent information retrieval system designed to deliver accurate, contextually rich answers to queries related to gestational diabetes. Leveraging the Retrieval-Augmented Generation (RAG) framework, our system combines the strengths of retrieval-based and generative models to provide high-quality responses tailored to individual needs.
Fig 1: RAG Application UI screen with a prompt, response, and chat history.
Objectives of the Digital GDM Screen
The primary objectives of the Digital GDM Screen are to:
Improve the accessibility and accuracy of information on gestational diabetes.
Enhance the user experience by delivering personalized responses.
Integrate state-of-the-art retrieval and generation techniques to ensure coherence and relevance.
Architecture
Fig 2: Architectural Abstraction Overview for the RAG Application
How It Works
Data Collection and Preparation: Our system uses a corpus of scientific articles, clinical guidelines, and educational materials related to GDM. The data is carefully selected based on relevance, credibility, and comprehensive coverage of GDM topics.
Embedding the Corpus: The collected texts are converted into numerical embeddings using Sentence Transformers, facilitating efficient similarity searches.
Retrieval and Augmentation: When a user submits a query, the system retrieves relevant text chunks from the corpus and integrates them with the query to form a comprehensive prompt.
Response Generation: The generative model, OpenAI's GPT-4, produces a coherent and contextually appropriate response, providing the user with high-quality information.
High-level Architecture Diagram
On the repo, you have to install the necessary requirements, have the API, then run embeddings file first before running the python app. Contributions are welcome!
Fig 3: Architectural Abstraction Overview for the RAG Application
📌 GitRepo:https://github.com/Ssemaganda-George/Digital-GDM_Screen
📌 Demo Video: https://tinyurl.com/zmwwb5t6
Tools, Libraries, and Frameworks Used
Sentence Transformers: For generating text embeddings.
OpenAI GPT-4: For generating responses.
Vectorstore: As a database for storing and retrieving embeddings.
Flask: For developing the user interface.
Python: For overall implementation.
Expected Outcomes
Our system is expected to significantly improve the accuracy and relevance of information delivered to users. The Digital GDM Screen aims to:
Achieve a high accuracy rate in answering GDM-related queries.
Ensure the relevance and coherence of the generated responses.
Enhance user satisfaction by providing timely and useful information.
Initial Testing Results
Accuracy: The system demonstrated an accuracy rate of 85% in delivering relevant information.
Relevance: User feedback indicated that 90% of the responses were highly relevant to their queries.
Response Time: The average response time was 10.38 seconds, ensuring quick access to information.
A Step Towards Better Health
By addressing the gaps in traditional information retrieval methods, the Digital GDM Screen represents a significant step forward in managing gestational diabetes. It empowers pregnant women and healthcare providers with reliable and easily accessible information, ultimately contributing to better health outcomes for mothers and their babies.
Join Us in Transforming GDM Management
We invite you to join us in our mission to transform the future of gestational diabetes management in Uganda. Together, we can ensure that every woman has the information she needs to manage her pregnancy healthily and safely.