Вештачка интелигенција

The Impact of Artificial Intelligence in Radiology: Lunit’s Groundbreaking Research and Solutions

Summary

Lunit, a leading provider of AI solutions for cancer diagnosis and treatment, is set to showcase eight revolutionary abstracts at the Annual Meeting of the Radiological Society of North America (RSNA) 2023. The meeting will take place in Chicago from […]

The Impact of Artificial Intelligence in Radiology: Lunit’s Groundbreaking Research and Solutions

Lunit, a leading provider of AI solutions for cancer diagnosis and treatment, is set to showcase eight revolutionary abstracts at the Annual Meeting of the Radiological Society of North America (RSNA) 2023. The meeting will take place in Chicago from November 26-30. Three abstracts will be presented orally, while the remaining five will be showcased as electronic posters.

In one of the oral presentations, Lunit explores the efficacy of a newly developed artificial intelligence (AI) model for filtering normal chest X-rays in autonomous reporting. The research evaluates the sensitivity of AI in detecting abnormalities, the percentage of chest X-rays that can be excluded from reporting, and the combined impact of Lunit INSIGHT CXR, a commercially available AI solution for detecting predefined findings. The model for filtering normal chest X-rays demonstrated an average sensitivity of 97.8% when applying a specificity threshold of 50%, resulting in a 22% reduction in the number of chest X-rays for reporting. Additionally, the commercial AI model identified an additional 16.7% of clinically relevant abnormalities missed by human radiologists, positioning it as a valuable safety solution. This research showcases the potential of AI in reducing the workload of radiologists, with Lunit INSIGHT CXR serving as a crucial mechanism for preventing the overlook of possible abnormalities. Lunit intends to elevate its involvement in innovative AI solutions in radiology by launching the model for filtering normal chest X-rays in the near future.

Lunit is also investigating parenchymal breast patterns and longitudinal changes to develop an AI model that predicts the risk of future breast cancer. The AI model, developed based on 16,113 digital mammograms from over 9,000 women, exhibited improved risk prediction with AUC values of 0.75 for 1-year, 0.76 for 2-year, and 0.73 for 3-year results. This research provides insights into the feasibility of an AI predictive model for identifying parenchymal characteristics in mammograms, enhancing risk stratification with longitudinal changes, and expediting personalized breast cancer screening.

Furthermore, a Swedish research team led by Dr. Karin Dembrouer, Chief Physician at Capio S:t Göran Hospital’s Mammography Clinic, presents findings from a sub-study of the ScreenTrustCAD investigation. ScreenTrustCAD is the first prospective study on the application of AI in mass breast cancer screening worldwide, published in The Lancet Digital Health. The study demonstrated that under double reading conditions, Lunit INSIGHT MMG, along with an additional reader, achieved better breast cancer detection than two human readers, without an increase in the recall rate. This sub-study analyzed differences in recall decisions through consensus discussion, depending on whether the mammogram was initially flagged as abnormal by AI or radiologists. A total of 5,515 mammograms were flagged for consensus discussion using AI, reader 1, and reader 2. Among the 2,501 consensus cases not flagged by AI but flagged by one or both radiologists, 25% required further testing, and 0.8% detected cancer, resulting in a positive predictive value of 3.4%. For the 3,014 consensus cases flagged by AI (and zero to two radiologists), 24% required further testing, and 7.2% detected cancer, resulting in a positive predictive value of 29%. It is worth noting that for mammograms flagged by radiologists in consensus discussion for further examination, the positive predictive value for mammograms previously flagged by AI is nearly ten times higher compared to those flagged by radiologists alone. The study concludes that radiologists participating in consensus discussions should be aware of their biases, and the design of consensus discussions may need to be reconsidered to optimize the utilization of AI readings.

“Our research demonstrates how AI can enhance radiology, from reducing the workload of radiologists to predicting the risk of breast cancer. I firmly believe that these studies represent a significant contribution to a new era of radiology with personalized care for cancer patients,” said Brendon So, CEO of Lunit. “Join us at our booth to delve deeper into our work and gain valuable insights through our presentations by renowned speakers we have invited.”

Explore these studies at Lunit’s RSNA 2023 AI Showcase booth (#4165) and attend the eight presentations at the booth by distinguished speakers like Dr. Karin Dembrouer and Dr. David Gruen, a radiologist at Jefferson Radiology and former Chief Medical Officer at IBM Watson Health. Witness firsthand how Lunit is shaping the future of radiology with the Lunit INSIGHT package, implemented in over 3,000 medical institutions worldwide.

FAQ