Sooraj Shah ’24

The detection of fractures via radiography is one of the most highly used practices in clinical settings such as the emergency room, urgent care, orthopedic and rheumatology offices. The missed fracture diagnosis rate is between 1-3%, accounting for almost 1,200 of every 100,000 patients. A major cause of missed fractures is erroneous initial readings by residents or non-radiologists, which are only corrected after the images are reviewed by board-certified radiologists. A study led by Dr. Daichi Hayashi, an associate professor of clinical radiology at the Renaissance School of Medicine at Stony Brook University, focused on the use of artificial intelligence (AI) in improving the performance and efficiency of fracture diagnosis by physicians on radiographs. 

The AI software, called “AI BoneView,” was trained to detect fractures from a set of 60,170 radiographs of patients seen from 2011 to 2019. The detection specificity included areas such as the foot, ankle, arm, shoulder, elbow, and rib cage, as well as images without fractures. The baseline “gold standard” of the diagnosis of fractures from the radiographic images were conducted by specialized radiologists with 8-12 years of experience. Readers such as radiologists, orthopedic surgeons, physicians, and rheumatologists then read the images with and without the assistance of the AI. The accuracy of detection was then compared with the gold standard. 

Results showed that AI improved fracture recognition by 10.4%, and reduced reading time by 6.3 seconds per patient. In addition to helping detect obvious fractures, AI assistance was also helpful in the detection of non-obvious fractures. While the AI-assisted human readings were 45.4% and 62% accurate for rib and spine fractures, AI detection alone resulted in a 76.7% and 76.9% accuracy for each, respectively. Additionally, in images with more than one fracture, AI improved the accuracy by 30%. 

The use of AI-assisted fracture readings in cases of acute trauma may help to improve the speed and accuracy of reading radiographic images by clinical physicians. In turn, this will greatly help to reduce wait times in emergency rooms and prevent the delays in patient treatment caused by reduced time to analyze the radiographic images. Dr. Hayashi believes this practice can be utilized to detect other diseases such as those in the brain, and this will be the focus of future research.

Works cited: 

[1] A. Guermazi et al. Improving Radiographic Fracture Recognition Performance and Efficiency Using Artificial Intelligence. Radiology  302, no. 3, pp. 627–636(2022). doi: 10.1148/radiol.210937.

[2] Image retrieved from: https://pixabay.com/photos/skeleton-medical-technology-2561573/