hamidehkerdegari / ai-echocardiography-for-low-resource-countries

AI-assisted echocardiography for low-resource countries

AI-assisted echocardiography for low-resource countries

This work is 100% Reproducible, lead by https://github.com/mxochicale

Summary

This repository contains documentation and code for AI-assisted echocardiography for low-resource countries. This work presents a Machine Learning (ML) study case to illustrate the challenges of clinical translation for a real-time AI-empowered echocardiography system with data of ICU patients in low income countries (LMICs). Such ML case study includes data preparation, curation and labelling from 2D Ultrasound videos of 31 ICU patients in LMICs and model selection, validation and deployment of three thinner neural networks to classify apical four-chamber view. This work is based on the total product life cycle (TPLC) approach on AI/ML workflow from Good Machine Learning Practices established by the U.S. Food and Drug Administration, Health Canada, and the UK's Medicines and Healthcare products Regulatory Agency (MHRA).

Data collection, validation and management 📂

1. Video data from GE Venue Go GE and Probe 3SC-RS was collected with Portable Video Recorder. See more.
2. Creation and verification of annotations with VGG Image Annotator (VIA) software. See more.
3. Jupyter notebook 📓 for data curation, selection and validation;
   Fig 1. Workflow for data annotation and validation.

Deep learning pipeline 📂

1. Model selection (MobileNetV1, MobileNetV2, SqueezeNet, EfficientNet, SqueezeNet, AlexNet, etc). See more.
2. Model training and tuning (Fig. 2). See more.
3. Model validation (performance evaluation and clinical evaluation).
4. AI-based device modification, and (perhaps) AI-based production model .
   Fig 2. Deep learning pipeline of the AI-empowered echocardiography.

AI-based research system 📂

Figure 3 illustrates the real-time AI-empowered clinical system based on EPIQ 7 ultrasound, X5-1 xMATRIX array transducer and USB framegrabber MiraBox Video Capture. The software of the system is based on Machine learning pipeline and Plug-in based, Real-time Ultrasound. See further details here on laptop hardware, OS and software of the system.

Fig 3. Real-time AI-empowered clinical system.

Licence and Citation

This work is under Creative Commons Attribution-Share Alike license . Hence, you are free to reuse it and modify it as much as you want and as long as you cite this work as original reference and you re-share your work under the same terms.

Clone repository

After generating your SSH keys as suggested here, you can then clone the repository by typing (or copying) the following lines in a terminal:

mkdir -p $HOME/repositories/vital-ultrasound  && cd $HOME/repositories/vital-ultrasound
git clone [e-mail]:vital-ultrasound/ai-echocardiography-for-low-resource-countries.git

BibTeX to cite

@misc{https://doi.org/10.48550/arxiv.2212.14510,
  author = {Xochicale, Miguel and 
	    Thwaites, Louise and 
            Yacoub, Sophie and 
            Pisani, Luigi and 
            Phung, Tran Huy Nhat and 
            Kerdegari, Hamideh and 
            King, Andrew and 
            Gomez, Alberto}, 
  title = {A Machine Learning Case Study for AI-empowered echocardiography of 
           Intensive Care Unit Patients in low- and middle-income countries},
  doi = {10.48550/ARXIV.2212.14510},
  url = {https://arxiv.org/abs/2212.14510},
  keywords = {Medical Physics (physics.med-ph), 
	      Machine Learning (cs.LG), 
	      Image and Video Processing (eess.IV), 
              FOS: Physical sciences, 
	      FOS: Computer and information sciences, 
	      FOS: Electrical engineering, electronic engineering, information engineering},
  publisher = {arXiv},
  year = {2022},
  copyright = {Creative Commons Attribution 4.0 International}
}

Contributors

Thanks goes to all these people (emoji key):

Miguel Xochicale 💻 🔬 🤔 📖

Hamideh Kerdegari 💻

Nhat Phung Tran Huy 💻

Louise Thwaites 🔬 🤔

Sophie Yacoub 🔬 🤔

Andrew King 🔬🤔

Alberto Gomez 💻

This work follows the all-contributors specification. Contributions of any kind welcome!

Contact and issue report

If you have specific questions about the content of this repository, you can contact Miguel Xochicale. If your question might be relevant to other people, please instead open an issue.