OR WAIT null SECS
COVID-19 Machine Learning Models Not Ready for Clinical Use; CT Colonography Tops Colorectal Cancer Detection; Radiology's Environmental Impact; Plus, DBT in Academic and Community Settings
Welcome to Diagnostic Imaging’s Weekly Scan. I’m senior editor Whitney Palmer.
Before we get to our featured interview with Dr. Amy Patel about the use and performance of digital breast tomosynthesis in both academic and community settings, here are the top stories of the week.
Since the beginning of the pandemic, radiology has been urgently searching for tools that will enhance providers’ abilities to detect and diagnostic viral infection. Consequently, there was a concerted effort to develop image-based machine learning models that used chest X-rays and chest CTs to identify which patients had COVID-19. Unfortunately, according to a study published the Nature Biomedical Intelligence, none of the more than 300 models are suitable for clinical use. According to investigators from the University of Cambridge, some had methodological short-comings, others had biases, and still others used bad data. By conducting a systematic review of 62 studies, the team found that some models were built with training data sets that used images of adults for COVID-19 data, but relied on pediatric images for non-COVID-19 information – a significant shortcoming because children have not been as impacted by the virus. Other models didn’t specify the origin of their data or they used Frankenstein datasets that have evolved and merged over time, preventing reproducibility. But, the team did say it was possible to salvage these machine learning models. Future efforts should avoid using public datasets; datasets should be diverse and appropriately sized; and manuscripts should provide sufficient documentation to enable independent technical and clinical validation.
When it comes to colorectal cancer screening and detection, CT colonography outperforms stool-based non-invasive screenings, said a group of researchers from the University of Wisconsin in the American Journal of Roentgenology. CT colonography has gained popularity in recent years because it is non-invasive and requires no sedation – but investigators wanted to know how it performed, compared to multi-target stool DNA testing and the fecal immunochemical test. They determined that, with a threshold of at least 10 mm, CT colonography did the best job of detecting advanced neoplasia. For the study, the examined 10 multi-target stool DNA, 27 CT colonography, and 88 fecal immunochemical tests, totaling more than 2.8 million patients. Overall, they said, CT conolography with the 10mm threshold most effectively targets advanced neoplasia, has a low optical colonoscopy referral rate, and a higher positive predictive value than the other non-invasive tests. In addition, it had at least a 90-percent sensitivity rate for large adenomas, outperforming both multi-target stool DNA and fecal immunochemical tests. Ultimately, though, they noted, the most effective tests will likely be the one the patient is most willing to undergo.
Radiology doesn’t pop top of mind in conversation about global climate change, but the industry’s impact is greater than you think. Roughly 10 percent of the nation’s carbon emissions – and 9 percent of harmful non-greenhouse air pollutants – come from the healthcare system. And, radiology can be a significant contributor, said a group of experts in the Journal of the American College of Radiology. In fact, a study from Switzerland revealed the three CT and four MRI scanners at one facility accounted for 4 percent of the institution’s overall energy use. This is a big deal with a study from the United Kingdom showing that 92 percent of patients want their healthcare facilities to be good environmental stewards. So, what can you do? The JACR offered four tactics. First, when you can, opt for ultrasound – it’s cheaper, uses less radiation, and has a lower environmental impact. In addition, truncate your MRI protocols and implement life cycle analyses to quantify the environmental impact of your different modalities. Second, don’t leave your machines in standby mode – it can consume between one-third and two-thirds of the energy your equipment uses. Instead, use a 24-hour operating cycle and make improvements to your HVAC system and imaging techniques. Third, power down at night. Yes, it’s easier to leave your PACS on, but doing so can produce the carbon dioxide emissions equivalent to four passenger cars annually. And, fourth, opt for clean energy. Prices are falling, and several institutions have already achieved carbon neutrality or positivity.
And, finally, this week, Diagnostic Imaging spoke with Dr. Amy Patel, medical director of women’s imaging at Liberty Hospital, about the use and performance differences of digital tomosynthesis in both academic and community setting. She shared details about cancer detection and impact on patients. Here’s what she had to say.
For more coverage based on industry expert insights and research, subscribe to the Diagnostic Imaging e-Newsletter here.