RSNA 2021 Digital X-Ray Preview

Road to RSNA 2021: Digital X-Ray Preview

By Will Morton, staff writer
November 17, 2021

Experts suggest one reason digital x-ray is an attractive modality to researchers developing artificial intelligence (AI) models is that large patient datasets are more available than in other modalities. The chest radiographs used in data analysis for the studies we've highlighted in this Road to RSNA section alone number in the hundreds of thousands.

For instance, given the vast number of chest x-rays performed on patients so far during the COVID-19 pandemic, it's no surprise that digital x-ray is leading the way in the development of AI algorithms for diagnosing the disease. But researchers are now digging even deeper to find out whether AI models can detect racial and ethnic health disparities in COVID-19 chest radiography.

Digital x-ray is also proving central in work on AI applications in lung cancer, foot deformities, and tuberculosis, as well as in ways to improve surveillance following total hip arthroplasty procedures.

In addition, dual-energy x-ray absorptiometry (DEXA) appears primed for the application of AI, with researchers exploring whether deep learning can be used to obtain bone mineral density measurements from imaging. In another session, researchers suggest that a model they've developed for use in total-body DEXA imaging can identify body composition changes over time to predict all‐cause mortality.

From the perspective of medical physicists attending RSNA 2021, there's much that can be done to improve the basic functions of x-ray itself. We've highlighted one study presenting a new prototype of a low-dose, multicontrast chest x-ray system that eventually could be put to use for diagnosing and screening for lung cancer.

Plus, we've highlighted a few presentations in interventional radiology. In one, German researchers offer hard clinical evidence that transarterial chemoembolization with degradable starch microspheres is safe and effective for patients with unresectable hepatocellular carcinoma. In another session covering diversity, equity, and inclusion in radiology, researchers took a look at how many women are working in the vascular and interventional radiology field in the U.S. and found the rate is increasing second only to the rate in vascular surgery.

Ultimately, most people want to know when AI algorithms will be ready for prime time. Researchers appear aware of the issue. In one session we've included below, a team from the Henry Ford Health System in Michigan suggests it may be a while. The group will present evidence that AI algorithms don't hold up when applied to external datasets.

Keep reading for highlights of these and other digital x-ray presentations and posters scheduled for this year's meeting. You can also view the complete 2021 scientific and educational program on the RSNA website.

AI captures bone mineral density data from hip x-rays
Sunday, November 28 | 1:00 p.m.-2:00 p.m. | SSMK03-1 | Room TBA
In this talk, researchers will show how bone mineral density data can be obtained from hip radiographs by an artificial intelligence (AI) algorithm.
CNN model can help diagnose foot deformity
Sunday, November 28 | 1:00 p.m.-2:00 p.m. | SSMK03-2 | Room TBA
South Korean researchers will report in this session on a convolutional neural network (CNN) model for automatically diagnosing foot deformity on standing lateral radiographs of the foot. The model appears to work well enough to help in settings with inexperienced readers, according to the group.
Models predict mortality risk on total-body DEXA exams
Sunday, November 28 | 1:00 p.m.-2:00 p.m. | SSMK03-3 | Room TBA
Artificial intelligence can spot changes in body composition over time on total-body dual-energy x-ray absorptiometry (DEXA) exams and predict mortality risk, according to this presentation.
More women joining the VIR workforce than other specialties
Monday, November 29 | 8:00 a.m.-9:00 a.m. | SSNPM01-5 | Room TBA
In this session on diversity, equity, and inclusion in radiology, researchers from Johns Hopkins Hospital will present evidence showing that the number of women in the vascular and interventional radiology (VIR) workforce in the U.S. is increasing. The trend could help stem the negative effect on patient outcomes due to the overall U.S. physician shortage predicted by 2030, the group believes.
Algorithm enhances survival predictions in lung cancer
Monday, November 29 | 9:30 a.m.-10:00 a.m. | SSCH03-2 | Room TBA
Artificial intelligence-based assessment of biological chest age on chest radiographs can help to predict survival in lung cancer patients, according to this presentation.
Researchers test x-ray prototype
Monday, November 29 | 1:30 p.m.-2:30 p.m. | SSPH04-5 | Room TBA
This talk will cover a prototype of a low-dose, multicontrast chest x-ray system that can provide three mutually complementary images from a single acquisition. The new system is fast and eventually could be put to use diagnosing and screening for lung cancer, researchers suggest.
Deep-learning tool can help in THA surveillance
Tuesday, November 30 | 8:00 a.m.-9:00 a.m. | SSMK05-3 | Room TBA
Research to develop a fully automated deep-learning tool to measure femoral component subsidence on hip x-ray without any user input will be discussed in this Tuesday morning session. A team at the Mayo Clinic in Rochester, MN, believes that its artificial intelligence-enabled algorithm could improve total hip arthroplasty (THA) radiographic surveillance efforts.
DSM-TACE is safe when nothing else works
Tuesday, November 30 | 9:30 a.m.-10:30 a.m. | SSIR02-5 | Room TBA
This interventional radiology session will provide information on a German study that evaluated the safety and efficacy of transarterial chemoembolization with degradable starch microspheres (DSM-TACE) for patients with unresectable hepatocellular carcinoma. The patients had high tumor burden and were ineligible for or were failing other palliative therapies.
Can an AI model improve radiologist scheduling?
Wednesday, December 1 | 9:30 a.m.-10:30 a.m. | SSIN06-5 | Room TBA
In this session, researchers from Cincinnati Children's Hospital Medical Center will discuss progress on an artificial intelligence (AI) algorithm to optimize radiologist scheduling for radiographs. Schedules are usually constructed so that a predetermined, fixed number of radiologists cover a service. But what if the daily study volume goes up or the potential radiologist capacity drops?
AI illuminates health disparities in COVID-19 patients
Wednesday, December 1 | 9:30 a.m.-10:30 a.m. | SSIN06-1 | Room TBA
In this talk, researchers will describe how an artificial intelligence (AI) algorithm can reveal racial or ethnic disparities in COVID-19 patients based on their chest x-rays.
AI models don't hold up when applied to external test datasets
Wednesday, December 1 | 3:00 p.m.-4:00 p.m. | SSPH12-2 | Room TBA
Many artificial intelligence (AI)-based algorithms have shown promise in detecting COVID-19 on chest x-ray; however, a team in this afternoon presentation asks, "Can a deep-learning model trained from one clinical site be applied to other sites?" The case report offers the very first set of scientific data regarding the generalizability of AI models in radiology, according to the group.
How does low-dose CT compare with x-ray for diagnosing lung disease?
Prerecorded, available throughout meeting | SPR-MS-9
In this session, researchers from Amsterdam University Medical Center in the Netherlands will offer insight into the pros and cons of using ultralow-dose CT compared to x-ray to diagnose nontraumatic pulmonary disease in emergency department patients.
Deep-learning model meets WHO guideline for diagnosing tuberculosis
Prerecorded, available throughout meeting | SPR-IN-12
In this informatics session, a team at Google Health in Palo Alto, CA, will present work on a deep-learning model for detecting active pulmonary tuberculosis using chest x-rays from countries across Africa, Asia, and Europe. The model surpassed World Health Organization (WHO) performance targets and could assist cost-effective screening efforts in radiologist-limited settings, according to the researchers.