News | Artificial Intelligence | April 28, 2021

Researchers Use AI to Detect Wrist Fractures

Overview of the scaphoid fracture detection pipeline, which consisted of a segmentation and detection convolutional neural network (CNN). A class activation map is calculated and visualized as a heatmap for fracture localization. Image courtesy of Radiological Society of North America

Overview of the scaphoid fracture detection pipeline, which consisted of a segmentation and detection convolutional neural network (CNN). A class activation map is calculated and visualized as a heatmap for fracture localization. Image courtesy of the Radiological Society of North America

Scatter plot of the confidence score of the detection convolutional neural network (CNN) versus the average confidence score of the radiologists per fracture case (n = 95) and nonfracture case (n = 95) in Dataset 3. Fracture and nonfracture cases are respectively marked with a red and blue color (ground-truth). The confidence scores range from 0 to 1, where 1 means absolute certainty that a fracture is present. Example radiographs where the CNN and radiologists showed a high disagreement are shown next to t

Scatter plot of the confidence score of the detection convolutional neural network (CNN) versus the average confidence score of the radiologists per fracture case (n = 95) and nonfracture case (n = 95) in Dataset 3. Fracture and nonfracture cases are respectively marked with a red and blue color (ground-truth). The confidence scores range from 0 to 1, where 1 means absolute certainty that a fracture is present. Example radiographs where the CNN and radiologists showed a high disagreement are shown next to the plot: (a) 17-year-old male with a proximal scaphoid fracture; (b) 21-year-old male with a fracture of the distal pole of the scaphoid (volar and radial side); (c) 45-year-old female with an intact scaphoid with a slightly irregular radial side; (d) 13-year-old female with an intact scaphoid with no irregularities. Image courtesy of the Radiological Society of North America

Examples of class activation maps (CAMs) for localizing fractures. The interpretation of the CAMs follows from the color coding of a heat map, in which pixel regions with a warm color signify a greater influence on the final decision of the network than regions with a cold color. The yellow arrows projected on the input images indicate the fracture lines, and they are only shown for reference. Image courtesy of the Radiological Society of North America

Examples of class activation maps (CAMs) for localizing fractures. The interpretation of the CAMs follows from the color coding of a heat map, in which pixel regions with a warm color signify a greater influence on the final decision of the network than regions with a cold color. The yellow arrows projected on the input images indicate the fracture lines, and they are only shown for reference. Image courtesy of the Radiological Society of North America

April 28, 2021 — An automated system that uses artificial intelligence (AI) is effective at detecting a common type of wrist fracture on X-rays, according to a study published in the journal Radiology: Artificial Intelligence. Researchers said the AI-derived algorithm could help speed diagnosis and allow earlier treatment.

Scaphoid fractures are injuries to one of the small bones of the wrist that typically occur when people try to break a fall with their hands. They comprise up to 7% of all skeletal fractures. Prompt diagnosis is important, as the fracture may fail to heal properly if untreated, leading to a host of problems like arthritis and even loss of function.

Conventional X-ray is the imaging technique of choice for diagnosing scaphoid fractures, but it is often limited by overlap of the scaphoid with the surrounding bones of the wrist. Variations in wrist positioning and X-ray technique can also limit the visibility of fractures.

"Consequently, scaphoid fractures can be overlooked during initial X-ray examinations," said study lead author Nils Hendrix, a Ph.D. candidate at the Jeroen Bosch Hospital and Jheronimus Academy of Data Science in the Netherlands.

Hendrix and colleagues recently studied a system that could aid radiologists in detecting these common fractures. The system is based on deep learning with a convolutional neural network, a sophisticated type of AI that is capable of discerning subtle patterns in images beyond the capabilities of the human eye.

While previous research found that a convolutional neural network was inferior to human observers at identifying scaphoid fractures on X-rays, the new study used larger datasets and further algorithm refinements to improve detection. It also employed class activation maps, which are AI tools that help users understand what region of the image is influencing the network's predictions.

The researchers used thousands of conventional X-rays of the hand, wrist and scaphoid to develop the system. They tested it on a dataset of 190 X-rays and compared its performance to that of 11 radiologists.

The system had a sensitivity of 78% for detecting fractures with a positive predictive value of 83%, which refers to the likelihood that patients the AI identifies as having a fracture really do have one. Analysis showed that the system performed comparably to the 11 radiologists.

The system has significant potential in clinical use, Hendrix said. It could reduce the incidence and costs of additional imaging exams and unnecessary therapy, speed up diagnosis and allow earlier treatment.

"The system may be able to assist residents, radiologists or other physicians by acting either as a first or second reader, or as a triage tool that helps prioritize worklists, potentially reducing the risk of missing a fracture," Hendrix said.

Such assistance could prevent delayed therapy and reduce complications that may lead to a subpar clinical outcome, according to Hendrix.

"The convolutional neural network may also reduce unnecessary wrist immobilization, performed out of precaution, in more than half of the patients with clinical suspicion for having a scaphoid fracture," he said.

The class activation maps were found to overlap with fracture lines in the scaphoid, suggesting they could be used for localizing potential fractures.

Hendrix and colleagues plan to expand the scaphoid fracture detection system so that it can combine multiple X-ray views for its predictions. They are also conducting an experimental study in which radiologists are asked to identify scaphoid fractures on X-rays with and without the aid of the fracture detection system.

The researchers hope to extend the system to fracture detection in other bone structures.

For more information: www.rsna.org

Related Content

After May 30, 2021, the U.S. Food and Drug Administration (FDA) will no longer accept Quality in 510(k) Review Files.  Companies should consider using the electronic Submission Template and Resource (eSTAR).

Getty Images

News | Radiology Imaging | April 26, 2021
April 26, 2021 — After May 30, 2021, the U.S.
ACR releases 13 new topics and five revised topics to support referring physicians and other providers in making the most appropriate imaging or treatment decisions
News | Radiology Imaging | April 26, 2021
April 26, 2021 — The American College of Radiology (ACR) released an update to its...
Change Healthcare unveiled InterQual 2021, the latest edition of the company’s flagship clinical decision support solution.
News | Clinical Decision Support | April 22, 2021
April 22, 2021 — Change Healthcare unveiled...
Women at high risk of breast cancer face cost-associated barriers to care even when they have health insurance, a new study has found.

Getty Images

News | Women's Health | April 22, 2021
April 22, 2021 — Women at high risk of breast cancer face cost-associated barriers to care even when they have health
Examples of axial FLAIR sequences from studies within dataset A. From left to right: a patient with a 'likely normal' brain; a patient presenting an intraparenchymal hemorrhage within the right temporal lobe; a patient presenting an acute infarct of the inferior division of the right middle cerebral artery; and a patient with known neurocysticercosis presenting a rounded cystic lesion in the left middle frontal gyrus. Image courtesy of Radiological Society of North America

Examples of axial FLAIR sequences from studies within dataset A. From left to right: a patient with a 'likely normal' brain; a patient presenting an intraparenchymal hemorrhage within the right temporal lobe; a patient presenting an acute infarct of the inferior division of the right middle cerebral artery; and a patient with known neurocysticercosis presenting a rounded cystic lesion in the left middle frontal gyrus. Image courtesy of Radiological Society of North America

News | Artificial Intelligence | April 22, 2021
April 22, 2021 — An artificial intellige...
Spectral DLR enables improved assessment of lumen stenosis in the presence of calcified plaque. Interactive monochromatic image display enables improved opacification of the injected contrast with low keV images and reduced calcium blooming artifacts with high keV images

Spectral DLR enables improved assessment of lumen stenosis in the presence of calcified plaque. Interactive monochromatic image display enables improved opacification of the injected contrast with low keV images and reduced calcium blooming artifacts with high keV images. The range of monochromatic energy levels (35-135 keV) can be visualized in real time through an image slider in the application that can be integrated in to a PACS.

News | Cardiac Imaging | April 21, 2021
April 21, 2021 — Meeting the growing cardiovascular needs of healthcare providers today, ...
The American College of #Radiology (#ACR) has brought together the full power of its world-class #informatics and #clinical #research resources to launch the innovative #ACR National Clinical Imaging Research Registry (#ANCIRR).

Getty Images

News | Radiology Business | April 21, 2021
April 21, 2021 — The American College of Radiology (ACR) has brought toget
Low-dose #CT #lung #scans are used to #screen for #lung_cancer in high-risk people such as heavy #smokers

Getty Images

News | Lung Imaging | April 16, 2021
April 17, 2021 — A deep learning algorithm accurately predicts the risk of death from...