Developing a prototype for automated analysis of Focused Cardiac Ultrasound (FoCUS) images using artificial intelligence

BACKGROUND

Lack of echocardiography assessment is the major cause of increased risk of mortality and morbidity among patients with suspected heart failure in low-resource clinical settings. Our study in India found that cardiac problems in pregnant women are associated with a case-fatality of 40% and another study from rural United States found lower baseline echocardiographic assessment at out-patient clinics to be a major reason for delays in initiating appropriate treatment for heart failure patients (doi:10.1002/ejhf.3201). 

Democratisation of echocardiography can save the lives of patients with cardiac problems in low-resource settings. Hand-held ultrasound devices can facilitate this, however technological advancement or improved accessibility of the devices has not translated in their clinical use in resource-limited settings where they are needed.

To facilitate prompt diagnosis, we developed and validated a 4-step FoCUS Solutions (doi: 10.1016/j.echo.2022.07.014) for echocardiography of pregnant women in India using low-cost hand-held cardiac ultrasound devices. The 4-step system is based on the principles of ‘task-shifting’ and ‘telemedicine’ and can be used for echocardiography in any patient suspected of a cardiac problem, so its use is not just limited to pregnant women. The scale of its clinical use could be enhanced by Artificial Intelligence (AI) enabled image analysis and decision support software to trigger referrals in real-time. This can make a life-saving difference in resource-limited settings. We have more than 10,000 cardiac images from ~1400 pregnant/ postpartum women obtained using the system from 10 hospitals across India, and we have already developed a proof-of-concept prototype. Our aim is to further develop the prototype for AI-enabled automated cardiac pathology assessment using FoCUS. The research objectives are:

• Development of AI/ML models for cardiac pathology assessment that could support referral decision making through FoCUS Solutions in resource-limited settings.
• Development of methodological advances to deploy such models using ‘low computational resources’ to address inequalities in access to affordable diagnostics in resource-limited settings.

Potential AI methods will include: Explainable AI, Affordable AI, Fairness in AI, AI for low computational resources, etc.

RESEARCH EXPERIENCE, RESEARCH METHODS AND TRAINING

Opportunity to attend research seminars offered at the NDPH, the BDI and the Institute of Biomedical Engineering (IBME) as B. Papiez (co-supervisor) is a member of Imaging Hub at the IBME. The student will be supported to attend relevant training in AI/ ML based on identified needs. There will be opportunities to present research work at relevant international/national conferences, and publish papers from the research.

prospective student

The ideal student would have a degree in computer science, statistics, engineering or a related discipline; strong programming skills (preferable Python, or Matlab/C++ and willing to learn Python), experience in machine learning (Deep Learning) and interest in medical image analysis.