AI: Raising Efficiency in Healthcare

AI development makes waves in some areas of medical field, especially in the readings of patients’ electronic records generated by digital equipment.  In this article, we look at two cases, one is the collaborative project between NY University and Facebook and the other refers to an automated system which claims to detect heart murmurs better than the cardiologists.

The first project involves MRI (Magnetic Resonance Imaging) and CT (Computerized Tomography) scans in a huge database that would help medical staff to spot quickly and efficiently lesions and tissue characteristics. A serious present downside in accurately detecting such conditions refers to accessing a relatively small database.

In the joint collaboration between the Medical School of NY University, Facebook using the innovative fastMRI dataset releases a giant database which can be read using AI software to analyze various MRI and CT scans rapidly. It creates opportunities for researchers to read this data and work with many characteristics on medical imaging scans in regards to illnesses and conditions. Besides their own AI software built, these organizations are creating other AI tools to help companies develop applications to read efficiently and accurately imaging scans. NYU gives some information related to the database:

  • 10,000 medical imaging scans
  • Over 1.5 million images
  • There is anatomical data in 1,600 scans
  • Fully HIPAA compliant (Health Insurance Portability and Accountability Act)
  • Absolutely Zero data from Facebook

The goals expressed by its initiators:

  • Improve the technology of diagnostic imaging
  • Capacity to produce detailed images aimed to accurately detect abnormalities from generated imaging scans produced from relatively less compared to the current measurement data.
  • fastMRI supports a AI standardized software which would bring together medical staff to improve imaging scans reading efficiency versus proprietary versions which could be less effective.

The second project presents an AI algorithm on neural network proposed by Eko, the designers of the well-known digital stethoscopes.

This comes as a solution to misdiagnosed heart murmurs by general and family physicians where the access to specialty cardiologist doctors is limited.

At the 2018 scientific sessions of the American Heart Association, Eko showed that a computer based on machine learning software precisely detected abnormal heart murmurs. The computer received numerous sound recordings previously diagnosed by the medical staff. Upon analyzing the records for uniqueness, it found some distinct murmurs which identified them as belonging to a sample auscultation. The AI software results were compared in parallel with those given by five pediatric cardiologists who independently listened to the heart sound recordings.

Thanks to the successful demo of the AI computer, Eko will seek implementation of this powerful heart murmur detection software following the final FDA (Food and Drugs Administration) approval.


Cory Popescu