AI & Medical Imaging: Artificial Intelligence Provides Added Support for Patient Diagnosis

Innovation is a core company value at Bumrungrad and key to improving standards of care. The hospital has been an early adopter for a wide range of promising healthcare technologies. Some of the recent additions include:
  • a new electronic medical records (EMR) system, known as TrakCare, that Bumrungrad physicians and medical staff use to provide a more efficient experience each time a patient visits the hospital
  • a new telemedicine service, Bumrungrad Anywhere, that enables a patient to consult with a physician by mobile phone or tablet through the service’s mobile application
  • IBM Watson for Oncology, an artificial intelligence-based system that supports the evaluation and decision-making process for cancer diagnosis and treatment. Bumrungrad was the first hospital in the world to launch the Watson for Oncology AI system.
Medical imaging has been another area of healthcare identified as “high potential” for artificial intelligence (AI) applications. In simple terms, the principle involves teaching a machine how to accurately read CT scans, X-rays, mammograms and other medical images that are an important diagnostic tool for identifying many types of cancer, numerous other diseases and injuries, such as fractures.  
The role we envisioned for the AI system was not to be a substitute or replacement for the physician or the radiologist, whose active role in the process wouldn’t change. But rather, AI would be a support tool for the human decision-making process, providing a “second opinion” for the doctor’s consideration during the diagnostic process. An additional benefit we anticipated for the AI imaging system is its ability to speed up the learning curve for junior doctors with less experience reading images.

Pilot Study

About three years ago, Bumrungrad began a proof-of-concept pilot study of AI-based imaging software systems capable of reading CT scans. Most of the companies in the AI field are based in the U.S. or Europe, and we narrowed our list of candidates to companies whose products had already completed the regulatory approval process.
Similar to the process of obtaining regulatory approval before a new drug can be made available to patients, there are now regulatory bodies that evaluate new AI medical products. The US FDA is considered the leading AI regulator, so we prioritized applications that had received FDA 510(k) approval. This would also ensure that whichever product we selected would be commercial-ready and could be launched into service at the hospital fairly quickly thereafter.

Physician Panel

An important element of the pilot study was the panel of Bumrungrad physicians we recruited as subject matter experts in their respective fields. They represented the views and interests of the doctors who would be the primary users of the chosen system. Physicians on the panel drafted a set of acceptance criteria that contained their key requirements for the system, and they took an active role in the process of evaluating each supplier.
The system we selected was able to satisfy the key physicians’ criteria and could be fully integrated into Bumrungrad’s existing PACS system without requiring any modification to the hospital’s CT scanning processes. As soon as each CT scan is completed, a copy of the scan is transmitted to the AI system, where the machine reads the image and makes its diagnosis — in under a minute, start to finish.

Physician Feedback

When the physician logs in to view the patient’s newest set of CT scans, an additional icon representing the AI system appears, and the physician can click on the icon to read the “second opinion” before or after they do their own review of the patient’s scans. The AI system gives the doctor the opportunity to provide feedback on each individual image reading, so if the doctor believes the AI system is providing an incorrect analysis or overlooking something, their feedback is sent to the project team at the system supplier for follow-up action, which sometimes includes modifications to retrain the relevant algorithm.
We launched the AI system almost two years ago, and the system is now in use for evaluating CT scans for four medical conditions — emphysema, cerebral hemorrhage, fatty liver disease and bone fracture. We continue to evaluate developers’ algorithms for other diseases on an on-going basis, and we expect to add to the current four in the near future, when the new algorithms satisfy our requirements and are able to add value to the patient diagnostic process.

Protecting Patient Information

The entirety of the AI system is installed at the hospital, on premise, which has the significant benefit of eliminating all patient privacy issues. That would not be the case if we had to send each image to an outside vendor or were using cloud-based software. We take great care to ensure we maintain GDPR HIPAA compliance and satisfy all PHI (personal health information) regulations.
By Dr. Teeradache Viangteeravat, Director of Research and Development, Bumrungrad International Hospital
For more information please contact:

Related Health Blogs