While the pandemic raged, work continued to advance the development of AI tools with breast imaging services.
It is no secret that mammography services faced a significant set-back this year when the COVID-19 pandemic erupted. Eventually, the service line was able to rebound from the catastrophic 92-percent plummet it experienced during the summer months – but, that was not all that happened.
That recovery was, without a doubt, a success, but it was by no means the only positive development with mammography during 2020. This was the year to watch advances in artificial intelligence tools in breast imaging. To pinpoint the ones that will be most impactful, Diagnostic Imaging spoke with Randy Miles, M.D., MPH, assistant professor of radiology at Harvard Medical School. What it all boils down to, he said, is how we can improve workflow efficiency and radiologist interpretative performance.
Diagnostic Imaging: Overall, 2020 has been a difficult year – particularly for mammography – but, what have been the biggest developments?
Miles: We still have a lot of work to do following the pandemic, in terms of clearing the backlog of women who have not returned for their mammogram this year. Breast imaging centers will need to be both resourceful and nimble in encouraging women to return for screening and finding additional slots for them to schedule their appointments. In terms of innovation, the most significant news has definitely been continued advances in artificial intelligence in breast imaging. After using much of the same technology over the past few decades, advances in computing power and data storage have provided us the opportunity to push the field forward through the application of artificial intelligence using deep learning techniques.
Related Content: Additional Mammography 2020 Coverage
Numerous AI projects were highlighted at this year’s RSNA conference. There was a presentation that evaluated potential differences in AI model performance in mammography interpretation between Black and White patients. The group demonstrated no difference in performance between the two racial groups. It is an important question for researchers to ask.We must ensure that all subgroups of women benefit from the application of these new technologies to ensure we do not exacerbate pre-existing healthcare disparities in breast cancer care. This study highlights the importance of training models on diverse patient groups. In the future, it will be important to focus on ensuring availability of these technologies for all women when approved for use.
Other AI projects that were presented focused on workflow efficiency. One study, in particular looked at how an AI model could be used to help reduce reporting turnaround time. As radiologists, our workload increases each year. A shortage of breast imaging specialists across the country has only increased demand for our services. If there are tools that can help us become more efficient by reducing this often monotonous and time-consuming task, it will provide us more time to focus on patient care. The research group reported an average time for reporting mammography studies without the AI tool of six minutes, which decreased to four minutes using their AI algorithm. If you look at that over the course of a week or month, that extra time adds up to a lot of time saved. That same study also showed 15.5 and 10.75 cases reported per hour with and without use of the model, respectively. The volumes we are seeing are not decreasing anytime soon, so efforts to improve efficiency will be very important for radiologists to continue to meet interpretative performance standards.
In addition, there was work presented on using AI models to triage screening mammograms as cancer-free. Work in this area is really interesting because it shows how AI models can directly impact workflow by potentially reducing the number of studies that radiologists interpret based on desired levels of sensitivity and specificity. These issues, while controversial, will be continually discussed, particularly how we can optimize performance thresholds of these models in real-world settings to maximize cancer detection, while reducing overall workload.
Diagnostic Imaging: Outside of artificial intelligence, what advancements stand out in 2020?
Miles: There are emerging technologies outside of AI, where we are seeing a lot of growth currently. Contrast-enhanced mammography, abbreviated breast MRI, and molecular breast imaging show promise in helping to identify cancers that may not be evident on conventional mammography examinations. Women at elevated risk may derive the greatest benefit from these technologies, including those with dense breast tissue, whose tumors tend to develop at an earlier age and grow at a more rapid rate.
Diagnostic Imaging: What do you think 2021 will hold for mammography?
Miles: We are going to see continued advances in artificial intelligence in breast imaging in 2021. Much of the work in AI thus far has been performed using retrospective data.Over the next year, it is important that these models are evaluated in real-world settings in diverse patient groups. This will allow us to truly understand how we can fully apply these tools in the clinical environment to improve radiologists’ performance and workflow efficiency. We’ll also be able to reflect on how artificial intelligence can transform the future of breast imaging care for our patients, who will benefit from radiologists having more time to provide consultations, interpret examinations, and perform procedures.
Study: AI Boosts Ultrasound AUC for Predicting Thyroid Malignancy Risk by 34 Percent Over TI-RADS
February 17th 2025In a study involving assessment of over 1,000 thyroid nodules, researchers found the machine learning model led to substantial increases in sensitivity and specificity for estimating the risk of thyroid malignancy over traditional TI-RADS and guidelines from the American Thyroid Association.
Can CT-Based AI Provide Automated Detection of Colorectal Cancer?
February 14th 2025For the assessment of contrast-enhanced abdominopelvic CT exams, an artificial intelligence model demonstrated equivalent or better sensitivity than radiologist readers, and greater than 90 percent specificity for the diagnosis of colorectal cancer.
Emerging MRI Scoring System May Help Predict Recurrent and Metastatic Hepatocellular Carcinoma
February 12th 2025Preoperative use of the scoring system for gadoxetic acid-enhanced MRI demonstrated an average AUC of 85 percent and average specificity of 89 percent in external validation cohorts for pathologic features of hepatocellular carcinoma.