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

Study Suggests AI Improves Chest X-Ray Interpretation

Article

Researchers found that stand-alone use of an artificial intelligence (AI) model led to a 24.9 percent increase in sensitivity for diagnosing pulmonary nodules and a 21.4 percent increase in sensitivity for diagnosing pneumonia.

Emerging research suggests artificial intelligence (AI) can significantly enhance sensitivity rates for diagnosing major thoracic findings on chest X-rays.

In a recently published study, researchers reviewed 497 frontal chest radiographs drawn from two institutions. Comparing radiologist assessment versus an AI algorithm (Lunit INSIGHT CXR, Lunit), the study authors examined the diagnosis of pulmonary nodules, pneumonia, pneumothorax, and pleural effusion. The reviewing physicians included two experienced thoracic radiologists, two thoracic imaging fellows and two radiology residents.

The researchers noted abnormal findings in 351 of the 497 cases including 195 cases of pneumonia, 149 diagnoses of pleural effusion, 80 cases of pneumothorax and 114 lung nodules. Stand-alone use of AI resulted in a 21.4 percent increase in sensitivity for pneumonia, a 24.9 percent increase in sensitivity for pulmonary nodules and a 19.6 percent increase in the sensitivity rate for pneumothorax in comparison to unaided radiologist assessment, according to the study.

In comparison to radiologist assessment, the study authors said the combination of AI and radiologist assessment resulted in sensitivity increases of 17.3 percent, 6.2 percent and 4.6 percent for pneumothorax, pulmonary nodules, and pneumonia respectively.

“The use of an AI algorithm was associated with sensitivity gains for all 4 target chest radiograph findings across all readers regardless of their experience and training status,” wrote study co-author Mannudeep K. Kalra, M.D., a professor of radiology at Harvard Medical School, and colleagues.

The study authors also noted that AI-aided radiograph review shortened image interpretation time by 3.9 seconds (36.9 seconds) in comparison to unaided radiologist review (40.8 seconds).

“Specifically, improved interpretation both in terms of finding detection and interpretation time was most notable for 3 of the 4 residents and thoracic imaging fellows,” noted Kalra and colleagues. “Although neither attending radiologist improved their reading efficiency with AI, there was an improvement in detection of target findings with use of AI for both radiologists.”

In regard to study limitations, the researchers noted possible inaccuracies stemming from the fact that ground truths were derived only from frontal chest X-rays. They acknowledged that instances requiring more than three minutes of interpretation time were excluded from the study data but noted this data comprised less than 10 percent of the original study data.

While the study primarily focused on four findings (pneumonia, pulmonary nodules, pneumothorax and pleural effusion), the researchers said the AI algorithm can detect additional findings beyond those in the study. However, they also pointed out that an increased number of findings with AI can lead to increases in false positives and true-positive annotations that may inhibit the efficiency of radiologists.

Recent Videos
Can Fiber Optic RealShape (FORS) Technology Provide a Viable Alternative to X-Rays for Aortic Procedures?
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.