Machine Learning-Aided Platform for Point-of-Care Pregnancy Risk Assessment from 2D Ultrasound

Overview

This project aims to develop a machine learning (ML)-aided platform for point-of-care pregnancy risk assessment using 2D ultrasound images. The platform utilizes state-of-the-art ML models to analyze ultrasound images and provide risk assessments for pregnancy-related complications.

Features

• Automated analysis of 2D ultrasound images.
• Prediction of pregnancy-related complications such as fetal abnormalities, placental issues, and maternal health risks.
• User-friendly interface for healthcare professionals to input ultrasound images and receive risk assessments.
• Integration with existing healthcare systems for seamless adoption.

Installation

1. Install dependencies in requirements.txt

Usage

1. Run the training script train.py to use MLFlow for model logging & tracking.
2. train.py also uses MLFlow to register the best trained model and transition it production stage.
3. Run the CMD: !mlflow models serve --model-uri models:/{model_name}/production -p 7777 --no-conda to create a model serving endpoint.
4. Access the model endpoint through your web browser at http://localhost:7777/invocations.
5. Run the dashboard using streamlit run streamlit_dashboard.py
6. Upload 2D ultrasound images for analysis.
7. Receive risk assessments and recommendations based on ML analysis.

Contributors

Marconi Lab@MAK
Makerere AI Lab

Acknowledgements

Special thanks to HASH for their support and collaboration on this project.