• AI
  • Molecular Imaging
  • CT
  • X-Ray
  • Ultrasound
  • MRI
  • Facility Management
  • Mammography

Fast Labeling: Automated Tool Labels More Than 120,000 MRI images in Under 20 Minutes

Article

Achievement removes significant time barrier to building robust artificial intelligence tools.

It’s now possible to automatically label brain MRI images thanks to work done by imaging experts at King’s College London.

This work is a significant step forward for artificial intelligence tools as it truncates the amount of time needed to annotate the thousands of images required to create strong products.

In a study published in European Radiology, a team from the School of Biomedical Engineering & Imaging Sciences at King’s College London shared they were able to label more than 120,000 images in less than 30 minutes.

“By overcoming this bottleneck, we have massively facilitated future deep learning image recognition tasks, and this will almost certainly accelerate the arrival into the clinic of automated brain MRI readers,” said senior author Tom Booth, Ph.D., senior lecturer in neuroimaging. “The potential for patient benefit through, ultimately, timely diagnosis, is enormous.”

To build their platform, the team used more than 121,556 MRI images captured between 2008 and 2019 at King’s College Hospital NHS Foundation, as well as data pulled from corresponding reports and other institutions. Not only did they use unseen radiology reports to evaluate the performance of their model, but they also assessed it on unseen images.

“While this might seem obvious, this has been challenging to do in medical imaging because it requires an enormous team of expert radiologists,” he said.

According to their assessments, their model achieved accurate classification in all categories when tested against reference-standard report labels. They did detect a slight dip in performance with three categories – atrophy, encephalomalacia, vascular – against the reference-standard image labels. This outcome pointed to discrepancies in the original reports.

But, even though one large hurdle has been removed, there is still more work to be done, the team said. For example, performing deep learning image recognition tasks and ensuring the developed models continue to perform accurately in different hospitals with different scanners is still on the horizon.

For more coverage based on industry expert insights and research, subscribe to the Diagnostic Imaging e-Newsletter here.

Recent Videos
Where the USPSTF Breast Cancer Screening Recommendations Fall Short: An Interview with Stacy Smith-Foley, MD
A Closer Look at MRI-Guided Transurethral Ultrasound Ablation for Intermediate Risk Prostate Cancer
Improving the Quality of Breast MRI Acquisition and Processing
Can Fiber Optic RealShape (FORS) Technology Provide a Viable Alternative to X-Rays for Aortic Procedures?
Can Diffusion Microstructural Imaging Provide Insights into Long Covid Beyond Conventional MRI?
Emerging MRI and PET Research Reveals Link Between Visceral Abdominal Fat and Early Signs of Alzheimer’s Disease
Nina Kottler, MD, MS
Practical Insights on CT and MRI Neuroimaging and Reporting for Stroke Patients
Related Content
© 2024 MJH Life Sciences

All rights reserved.