Autonomous Imaging

Over the past several years, ultrasound technology has rapidly trended toward low-cost and miniaturized implementations, laying the foundation for widespread adoption of the technology in the same way that smartphones or computers became cheap and widely available. However, a skilled sonographer (operator) is still required to perform a successful scan, limiting adoption to where you have a trained expert. If we can enable novice users to perform ultrasound scans (general medical professionals, doctors, nurses, people throughout the developing world, and even someday people in the home), we open the technology to widespread growth. By enabling these users, ultrasound can provide improved quality of care with increased access to basic screening exams, access to ultrasound in the developing world, reduction of unnecessary referrals to high-cost imaging modalities, and reduction in scan times and variability of diagnosis.
 

The team demonstrated a proof-of-concept holistic solution that automates many operator actions and guides the operator through the exam. They explored the fusion of many technologies, such as sensing systems, probe and patient tracking, image processing, atlas-based registration, deep learning, anatomy modeling, AR visualizations, and robotics to realize this vision. Similar themes are present in other medical imaging modalities and industrial applications, and we see clear opportunities to translate and integrate this technology there as well.

Key demonstrations/milestones:

• January 2018: Integrated demonstration on-stage at GE’s Global Leadership Meeting: The Autonomous Imaging mission was selected for highlight on one of GE’s biggest stages: GE's Global Leadership Meeting held in Boston, MA. After a two-week sprint at the end of December, the team was able to pull together an end-to-end demonstration, allowing Kieran Murphy, GM of GE Healthcare and untrained ultrasound operator, to detect the volume of a live subject’s kidney in less than a minute.
• April 2018: Integrated demonstration of a guided fetal scan: The team developed a hands-on demonstration system in FORGE that integrates patient tracking, anatomy modeling, probe guidance, and AI image processing into a system that enables novice users to find the age of the fetus.
• August 2018: Mobile app for probe guidance: The team put together a prototype mobile app that dynamically builds an anatomy model of a patient using only a phone’s camera, and shows an Augmented Reality overlay of target probe position to begin the exam.