Real Life Business Impact of AI/ML on GE CT Scanners Using Glassbeam Clinsights™

Vivek Sundaram
Wednesday, September 25, 2019

We are now in an era where machine learning isn't just hype. In fact, it is absolutely real in its business impact as Glassbeam has recently demonstrated for its growing network of connected medical imaging equipment.

With Glassbeam's extensive experience in data engineering and analytics related to GE CT Scanner machine logs, we now have dozens of powerful use cases where automated anomaly detection via machine learning has been used to detect potentially expensive part failures well before the end-user even noticed an issue.  Here are 3 real-world examples:

Use Case 1: Tube Spit Ratio

Glassbeam continuously monitors critical system health information such as tube spits per exam as well as per 10-second exposure. By applying anomaly detection per tube, we can identify what the normal ratios areas not all tubes have a 0 spit ratio.

Recently, Glassbeam flagged a tube-based on an increasing count of anomalies in spits per exam and 10-second exposure. This was increasing at an alarming rate unbeknownst to the technologist. Glassbeam alerted this to the field service engineer, who requested the hospital to advance the next scheduled PM so that they can do a thorough inspection. They found a burnt high voltage pin that was resulting in the spits. Cleaning up the pin, and placing an order for a replacement pin eliminated unnecessary downtime at a fraction of the cost and dramatically increased the life of the tube by reducing damage.

The root cause for tube spits are manifold and corrective measures can be any of the following:

  • Tube replacement
  • Tube PM, such as degassing, reseating, heat soak, oil check, etc.
  • Identify and replace electrical components
  • HV tank PM or replacement

Use Case 2: Collimator Filter Errors

A typical collimator filter inside GE CT Scanner traps the x-ray beam and adjusts its characteristics to optimize patient dose and increase image quality. When the filter isn't seated properly or the motors don't align correctly, it results in move errors, which over time can result in image artifacts and unexpected dose.

The root cause may be:

  • Control board issues, which may require a replacement
  • Filter issues, which may require reseating or replacement

By applying anomaly detection on these readings, Glassbeam flagged an increasing number of move errors so that the field service engineer could pre-emptively replace the control board and avoid image quality-related complaints, which almost always results in soft downtime, where the machine is technically up, but not used due to the unsatisfactory output quality.

Use Case 3: CAM motor move retries and/or errors

GE CT Scanner uses rotating anode x-ray tubes and in order to maintain tube alignment against the centripetal forces, it's crucial to keep the CAM motors in good condition. With age and/or wear and tear, the CAM motor may miss the target, resulting in move retries and errors. Leaving this unattended can eventually result in scan aborts and downtime. With anomaly detection, Glassbeam identified a cyclical increase in CAM retries that directly related to the number of scans and an increase in scan time. The engineer was advised to replace the CAM motor and it immediately resulted in dramatically reduced retry attempts, and as a result, more efficient scans.

Above three examples relate to GE CT Scanners only.  Overall, Glassbeam cloud-based platform today has over 128 Machine Learning models across GE, Siemens, Phillips models for MRI and CT Scanner modalities.  This knowledge base is getting refined and updated each day as we get more data and use cases from our customers.  It is our vision and strong belief that the service approach in the healthcare industry has been woefully reactive and it's about time we bring healthcare device maintenance to the 21st century. Being at the forefront of this endeavor, we are excited to learn and identify new ways to push our technology forward.

For more information, here are some more pointers: