May 11, 2021
DCNN recorded 90 percent accuracy
AI may prove a boon to identifying small bone fractures on X-rays that a human specialist might miss, suggests a study from academics at the University of Michigan Medical School and the Taiwanese Center for Artificial Intelligence in Medicine.
The study suggests that around 20 percent of scaphoid fractures — a break in one of the smallest bones of the wrist — cannot be seen in initial radiographs.
There is often only a little swelling which usually goes away in a few days, and a variable amount of pain may be present. Such fractures can often be mistaken for sprained wrists, and if left untreated could lead to arthritis in the wrist.
The academics trained an AI model to detect the so-called “occult fractures” - potential fractures that cannot be identified by an X-ray. The model managed to successfully identify more than 90 percent of such injuries across a two-stage process.
“It achieved high sensitivity and specificity, suggesting that [deep convolutional neural networks] can be trained to reliably detect fractures in small bones,” Alfred Yoon, MD, a plastic surgery resident with Michigan Medicine and one of the study’s co-authors, said.
“In addition, this study found that the deep convolutional neural networks (DCNNs) could detect occult fractures that are not readily visible to physicians. This enhanced diagnostic capacity can help to solve medical problems with high monetary or quality-of-life costs and improve fracture care.”
What’s the diagnosis?
Yoon, Yi-Lun Lee from the Taiwanese Center for Artificial Intelligence in Medicine, and Chang Gung Memorial Hospital’s Chang-Fu Kuo conducted a diagnostic study of 11,838 scaphoid radiographs.
The authors trained the DCNN to distinguish scaphoid fractures from scaphoids without fracture.
The study used a dataset compiled of patients with possible scaphoid fractures at the Chang Gung Memorial Hospital in Taiwan and Michigan Medicine in Ann Arbor between January 2001 and December 2019.
X-ray images were passed through a detection model to crop around the scaphoid and were then used to train the DCNN model based on the EfficientNetB3 architecture to classify apparent and occult scaphoid fractures.
Data analysis was conducted from January to October 2020.
Of the near 12,000 potential fractures, 4917 (41.5 percent) were actual fractures, while 6921 (58.5 percent) weren’t.
Overall, the two-stage model correctly identified 90 percent of occult fractures, with the authors suggesting the model may be able to assist with radiographic detection of fractures that are not visible to human observers.
“In the clinical setting, this two-step process could not only increase the probability of detecting a true occult fracture but also increase the model's sensitivity to rule out scaphoid fractures, especially in cohorts with [a] low prevalence of occult scaphoid fractures, precluding the need for advanced imaging,” the study concluded.
“This enhanced diagnostic capacity can help to solve medical problems with high monetary or quality-of-life costs and improve fracture care.”
About the Author(s)
You May Also Like