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

AI Tool Helps Reduce Re-Take Knee X-rays


Using a deep convolutional neural network tool, radiographers can correct image errors and reduce repeated imaging.

Re-takes for knee X-rays are common, leading to unnecessary radiation exposures and over-use of radiology personnel. But, until now, little has been done to help alleviate this situation.

That’s where investigators from Osaka University in Japan have stepped in. Using a deep convolutional neural network they developed, they can potentially help technologists identify and classify tilt direction errors and guide them to the correct positioning for lateral knee X-rays.

The team, led by T. Ishida, from the medical physics and engineering department at Osaka University, published their findings recently in Radiography.

“Our research is a novel attempt to create a re-taking support system,” said Ishida’s team. “This may reduce the inconvenience of patients and improve the work efficiency of radiological technologists, and we also assume that the proposed system may be used for the assistance and training for inexperienced radiological technologists and students.”

Specifically, their tool aims to reduce the number of and time needed for re-takes by correcting the mistakes found in the rejected images. Their initial results were encouraging, the team said.

For their study, the team used 11,520 synthetic Raysum images from a 3D CT exam set to train their tool. They also added in a flexed knee joint phantom case, as well as 14 knee joint X-rays that had been rejected.

Based on their analysis, their tool was able to categorize tilting directions into four groups: adduction-internal rotation, adduction-external rotate, abduction-internal rotation and abduction-external rotation. Classifications were based on a heat map that showed the medial and lateral femoral condyles and the patella.

According to their analysis, the larger the tilting degree of the knee joint, the more accurate the classification was. Overall, the team said, accuracy of each test dataset was: 88.5 percent +/- 7 percent for the extended-knee Raysum images, 81.4 percent +/- 11.2 percent for the flexed-knee Raysum images, and 73.3 percent +/- 9.2 percent for the rejected images.

At this point, the team said, they do not think knees with a small tilt always require re-imaging, but additional research with more cases is necessary.

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

Related Videos
Nina Kottler, MD, MS
The Executive Order on AI: Promising Development for Radiology or ‘HIPAA for AI’?
Expediting the Management of Incidental Pulmonary Emboli on CT
Related Content
© 2024 MJH Life Sciences

All rights reserved.