In a study involving over 100,000 digital breast tomosynthesis (DBT) screening exams, researchers found the use of ProFoundAI (iCAD), an adjunctive artificial intelligence (AI) software for digital breast tomosynthesis (DBT), increased sensitivity by 5 percent, increased the positive predictive value by 23 percent and increased the breast cancer detection rate by 23 percent for nine fellowship-trained breast radiologists with an average of 22 years of experience.

In a recent interview, Kathy Schilling, M.D., a co-author of the study, conceded a fair amount of surprise with the results.

“We have nine dedicated breast radiologists. That is all we do: mammography and breast imaging studies. We thought we were pretty good. It was really quite interesting and surprising that everyone improved their cancer detection by using the ProFound AI (software),” noted Dr. Schilling, the medical director of the Christine E. Lynn Women’s Health and Wellness Institute at the Boca Raton Regional Hospital in Florida.

Dr. Schilling pointed out these increases were achieved without an increase in recall rate.

“We’re not calling more patients back to find more cancers. We’re actually finding more cancers that we may have missed if we (did not) have the use of the ProFound AI,” added Dr. Schilling, who recently presented the study findings at the European Congress of Radiology.

(Editor’s note: For related content, see “

Large Mammography Study Reaffirms Benefits of Digital Breast Tomosynthesis

Study: AI Improves Cancer Detection Rate for Digital Mammography and Digital Breast Tomosynthesis

FDA Issues Final Rule on National Breast Density Notification for Mammography Reports

In a time of significantly increased imaging volume, Dr. Schilling said the ProFound AI software refocuses radiologists to detecting smaller cancers that may have otherwise gone undetected. In the study, she noted the researchers were diagnosing invasive lobular carcinoma, which commonly presents as architectural distortions, down to three to five millimeters in size.

For more insights from Dr. Schilling, watch the video below.

Articles

In a recent video interview, Kathy Schilling, M.D., discussed findings from a study of ProFound AI, an adjunctive artificial intelligence (AI) software for digital breast tomosynthesis (DBT), that demonstrated a 23 percent increase in breast cancer detection in comparison to DBT alone.

Study: Emerging AI Platform for DBT Shows 23 Percent Increase in Breast Cancer Detection Rate

ChatGPT may have considerable potential for streamlining radiology workflows and reporting but concerns over optimal prompt engineering and hallucination effects may curtail the use of the artificial intelligence (AI) chatbot in radiology for the time being.

In recent video interviews. Nina Kottler, M.D., MS, Tessa Cook, M.D., Ph.D, and Sonia Gupta, M.D., discussed the potential capabilities of ChatGPT for radiologists and cautioned about current drawbacks of the technology.

Dr. Kottler suggested that ChatGPT could help remedy the frustration of receiving incomplete patient histories.

“We’ve never (received) great (clinical) histories. That decreases the quality of our output and causes us to hedge a lot,” notes Dr. Kottler, associate chief medical officer of Clinical AI and vice-president of Clinical Operations at Radiology Partners. “ChatGPT could either summarize information from the EMR (electronic medical record) and provide us with that robust patient history or summarize prior reports for us so we get that history and the pathology that we’re finding.”

Another potential benefit of ChatGPT could be converting patient summaries of radiology reports into different languages and possibly expanding access to care, suggested Dr. Gupta, the chief medical officer of Change Healthcare and a board member of the American Board of Artificial Intelligence.

Emerging Insights on Improving Radiology Workflows

Assessing the Value Proposition of AI in Radiology

Emerging Trends with AI in Radiology in 2023

However, Dr. Kottler noted that the quality of the prompt input has a strong effect on the results one gets with ChatGPT. Drs. Cook, Gupta and Kottler all emphasize having appropriate safeguards in place to ensure the accuracy and verifiable referencing of information provided by ChatGPT. While Dr. Cook noted use cases with ChatGPT that have provided a synthesis of literature on a given topic, she said other cases have demonstrated “hallucination effects” in which there has been rewritten history or the “invention of citations that don’t exist.”

“Just because ChatGPT is far more fluent than any other chatbot we might have encountered in the past doesn’t necessarily mean it is not fallible in some way as well,” cautioned Dr. Cook, an associate professor of radiology, and director of the Center for Practice Transformation in Radiology at the Hospital of the University of Pennsylvania.

For more insights from Drs. Cook, Gupta and Kottler, watch the video below.

Videos

In recent video interviews, Tessa Cook, MD, PhD, Nina Kottler, MD, MS, and Sonia Gupta, MD shared their insights and perspective on potential benefits and drawbacks of ChatGPT in radiology.

Can ChatGPT Have an Impact in Radiology?

Over one million people in the United States have an abdominal aortic aneurysm (AAA).

 Aortic aneurysms claimed the lives of 9,904 people in the U.S. in 2019.

 Researchers have noted that mortality rates with AAAs could be underestimated as they may cause up to five percent of 200,000 sudden death cases annually in the U.S.

However, Viz AAA (Viz.ai), an emerging artificial intelligence (AI)-powered algorithm, may facilitate earlier detection and treatment for people with AAAs. 

The Food and Drug Administration (FDA) has granted 510(k) clearance to Viz AAA

, which provides automated AI assessments of computed tomography angiography (CTA) to detect suspected AAA, according to Viz.ai.

Philip Batista, M.D., noted the majority of people with AAA are asymptomatic and lack awareness of the disease until they have a rupture. He called Viz AAA a significant advance for this high-risk population.

“This algorithm is a powerful new tool for healthcare professionals to more readily identify and capture individuals with AAA and, importantly, automatically refer those at imminent risk for rupture,” said Dr. Batista, an associate program director of Vascular and Endovascular Surgery Residency and assistant professor of surgery at the Cooper Medical School of Rowan University in New Jersey.

Viz.ai added that Viz AAA can be integrated into existing hospital workflows and will be available through the Viz AORTIC Module, an AI platform geared toward facilitating accelerated treatment of aortic pathology.

AI-Powered Algorithm May Enhance CT Assessment of Aortic Dissection

Could AI Improve Detection of Intracranial Aneurysms on CT Angiography?

1. Kent KC, Zwolak RM, Egorova NN, et al. Analysis of risk factors for abdominal aortic aneurysm in a cohort of more than 3 million individuals. 

2. Centers for Disease Control and Prevention (CDC). Aortic aneurysm. Available at: 

https://www.cdc.gov/heartdisease/aortic_aneurysm.htm 

3. Schermerhorn M. A 66-year-old man with an abdominal aortic aneurysm: review of screening and treatment. 

4. Viz.ai. Viz.ai is first to receive FDA 510(k) clearance for AI algorithm for abdominal aortic aneurysm. Available at: 

https://www.viz.ai/press-release/viz-ai-is-first-to-receive-fda-510-k-clearance-for-ai-algorithm-for-abdominal-aortic-aneurysm 

. Published March 21, 2023. Accessed March 22, 2023.

Viz AAA is reportedly the first artificial intelligence (AI)-enabled algorithm to garner FDA 510(k) clearance for the detection of abdominal aortic aneurysm.

New AI Algorithm for Abdominal Aortic Aneurysm Detection on CTA Gets FDA Nod

The Food and Drug Administration (FDA) has granted 510(k) clearance for the Brainomix 360 e-ASPECTS software (Brainomix), which provides adjunctive artificial intelligence (AI) assessment of computed tomography (CT) scans of the brain and automated scoring with the Alberta Stroke Program Early CT score (ASPECTS) system.

As part of the Brainomix 360 stroke platform, which emphasizes real-time AI-enabled interpretation of brain imaging in stroke patients, the e-ASPECTS software provides 

enhanced heatmap visualization and automated ASPECTS scoring based on non-contrast CT scans

The company said recent studies have demonstrated the stroke imaging platform’s ability to reduce door-in door-out times and reportedly triple the number of patients regaining functional independence after stroke.

“Our e-ASPECTS tool has been shown, in multiple countries and healthcare systems, to improve physicians’ interpretations of ASPECTS scores on non-contrast CT scans … ,” noted Michalis Papadakis, CEO and co-founder of Brainomix. “Our technology supports these physicians who are making time-sensitive, critical decisions around transfer and treatment, strengthening networks and facilitating an improved stroke service.”

Portable MRI System Gets FDA Nod for AI-Powered Brain Imaging Software

eraRecon Launches AI-Driven Neuroimaging Platform

Through artificial intelligence (AI) assessment of non-contrast computed tomography (CT) scans, the Brainomix 360 e-ASPECTS software provides an automated ASPECTS score and heatmap to enhance stroke imaging.

FDA Clears AI-Powered Stroke Imaging Tool for Non-Contrast CT

Vaccine-related lymphadenopathy occurs in nearly 19 percent of people who have had COVID-19 booster vaccination, regardless of age or type of vaccine, according to magnetic resonance imaging (MRI) findings from a new study.

For the 128-patient retrospective study, recently published in 

, researchers utilized breast MRI scans taken after COVID-19 booster vaccination and breast MRI scans taken prior to initial COVID-19 vaccination to compare measurements of axillary lymph nodes ipsilateral to the side of COVID-19 booster vaccination. Seventy percent of the patients had the Pfizer-BioNTec COVID-19 booster vaccine and 30 percent of the cohort had the Moderna booster vaccine. There was a median of 31 days between COVID-19 booster vaccine administration and the post-booster vaccine breast MRI, according to the study.

The researchers found that 24 patients (19 percent) had 0.2 cm or greater increases in lymph node cortical thickness after COVID-19 booster vaccine administration. For patients with cortical thickness increases of less than 0.2 cm, the study authors noted a median of 36 days between receiving the booster vaccine and the post-booster vaccine breast MRI exam in comparison to a median of nine days for patients with 0.2 cm or greater increases in cortical thickness. The researchers found no significant associations between age or type of COVID-19 booster vaccine in patients with 

Twenty-three of the 24 patients with cortical thickening 

 0.2 cm had follow-up imaging. According to the researchers, there was no detection of ipsilateral breast cancer or axillary malignancy as of December 31, 2022.

“Given the high frequency of lymphadenopathy following COVID-19 vaccination, further evaluation or follow-up may not be necessary in patients without a suspicious finding within the breast,” wrote study co-author Joao Vicente Horvat, M.D., an instructor of radiology at Memorial Sloan Kettering Cancer Center in New York, N.Y., and colleagues. “ … A similar strategy should be considered for patients following the COVID-19 booster vaccination, as the frequency of lymphadenopathy can be considered high, especially during the first 3 weeks after the booster.”

Post-Vaccine Axillary Lymphadenopathy: What a New Chest MRI Study Reveals

New Study Points to Longer Duration of Axillary Lymphadenopathy After COVID-19 Vaccine

Emphasizing the challenge of diagnosing lymphadenopathy, the researchers noted that breast MRI features associated with lymphadenopathy range from lack of a fatty hilum and irregular contours to a decreased long/short axis ratio. While cortical thickness is the most common morphological feature utilized to diagnose malignant lymphadenopathy, Horvat and colleagues maintained that identifying abnormal lymph nodes is subjective and there is no specific cutoff standard for cortical thickness increase that would be indicative of lymphadenopathy.

For persistent lymphadenopathy after COVID-19 vaccination, the study authors suggest follow-up fine needle aspiration or core needle biopsy. As vaccine-related lymphadenopathy and malignant lymphadenopathy may have some overlapping imaging features, the researchers advocate a higher index of suspicion in patients with elevated risk of malignancy.

“In patients with abnormal lymph nodes and known ipsilateral breast malignancy, lymphadenopathy should not be assumed to represent vaccine-related changes,” emphasized Horvat and colleagues.

In addition to the retrospective nature of the work, the authors said another limitation of the study included variation with the timing of breast MRI after COVID-19 booster vaccination. The researchers also acknowledged a lack of consistent follow-up and noted that some patients did not have follow-up MRI exams. Horvat and colleagues also pointed out that one radiologist made all of the MRI measurements in the study.

In a study of 128 patients who had breast magnetic resonance imaging (MRI) exams after COVID-19 booster vaccination, researchers noted that nearly 19 percent of the cohort had equal to or greater than 0.2 cm of increased lymph node cortical thickness.

What a Breast MRI Study Reveals About Axillary Lymph Nodes and COVID-19 Booster Vaccination

Eric Postal, MD, is a diagnostic radiologist with the University of Pittsburgh Medical Center. 

Unless you’ve been living under a rock, you’re aware that there have been changes at Twitter in recent months. Some of the more important have impacted what turns up on a given user’s “feed.” I have happily seen a substantial increase in the number of health-care related items that populate my screen.

It’s not all worthwhile stuff. Plenty of idiots and trolls post about health care, just like anything else. Even their nonsense can be entertaining, but for the most part, I get interesting tidbits about things a teleradiologist might otherwise never know.

Last week, a bunch of posts turned up pertaining to this year’s Match. Evidently, a bunch of emergency medicine (EM) residency spots went unfilled, and people were drawing all sorts of conclusions about it.

This is not a new phenomenon. Every year, all kinds of statistics from the Match are watched, dissected, and subjected to prognostication. Conclusions are drawn. Repercussions are theorized (or proclaimed as facts). One can go pretty deep into the rabbit hole if one is so inclined.

It is, of course, how our brains work. Diagnosticians make a living out of this, but people in general tend to look for patterns and build narratives. Just taking in information and doing nothing with it doesn’t come naturally to us. This is especially the case when the subject at hand is emotionally charged. For instance, topics tied to our economic fortunes (or lack thereof) may ignite some lively discussion.

Tweeting EM docs took last week’s Match news badly. They had the understandable reaction that, with so many residency spots going unfilled, their specialty was falling into disfavor. Some took that ball and ran with it to sound off about all the things going wrong in EM that weren’t being addressed. Of 

, they explained, nobody wants to enter a field like that. Others spoke of how this would mean a future shortage of docs in EM and drew dire conclusions for health-care workers and patients alike. I got the impression that similar stuff was happening elsewhere in primary care.

I don’t claim immunity from such things. One of the reasons I went into radiology to begin with, back in the late 1990s, was that plenty of spots in our residency match had gone unfilled in previous years. I figured that meant I would be able to stroll in the door to a top-flight program with little competitive effort. Wasn’t this a reasonable conclusion?

I thought this may be the case based on the little bit of information I had used to draw my narrative. Popular wisdom at the time was that folks weren’t applying to the specialty on account of a bad rad job market. I figured, by the time I got out of residency, that might turn around.

It did, of course, but a little sooner than expected. Just as I was applying, new stats revealed that the market had come roaring back for rads. Med students all over the place heard the same thing, and suddenly I was surrounded by a sea of other hopefuls. I considered myself lucky to get a spot at all.

Folks like me might not have been surprised if we had based our narratives on more info. For instance, if we had been keeping closer tabs on the job market, we might have seen that the number of unemployed (or employed but unsatisfied) rads was steadily diminishing. The information was out there. But to get it, you would have to first conceive of its possible existence, moving it from your “unknown unknown” to your “known unknown.” Then you would have to come up with ways to track it down and get a handle on the real facts of the situation.

That’s a lot of outside-the-box thinking, not to mention time and effort for someone who is otherwise busy. For me at the time, I was focused on med school. Rads already in the field may have had little time outside of their actual jobs and personal lives to explore the matter. It’s a lot easier to take one or two factoids or overheard scuttlebutt while sitting in your metaphorical mental armchair and imagine what your incomplete info means.

Many moons ago, I picked up a bit of wisdom from a podcaster. If you size up whatever you know about a situation and can only think of one probable explanation, consider it a warning that you haven’t thought about it enough or that there is more relevant information you are not aware of just yet.

, that is especially true when dealing with complex systems. With so many intricate layers of information in such systems, you can’t ever really know all of the pertinent details, especially when more than one complex system intersects, like radiology and economics.

Faced with such an overwhelming amount of unknown data (or the project of gathering it all up), there are a handful of ways we tend to cope. We might take the pittance of info we actually have and assume (or hope or pray) that the sum total of everything we don’t might all just cancel itself out. Alternatively, we may capitulate that there is no way to take all factors into account, and “go with our gut.”

Another option is referring to sources we consider smarter than ourselves. We could take their appraisal as the closest we can get to the gospel truth and turn a willfully blind eye to the notion that they are still fallible, not fully informed, and perhaps biased, whether it is with honest intent or a result of having their own axes to grind.

While social media posts abound about the impact of the residency match in different fields and other topics, this author encourages perspective that is grounded in the facts.

Commentary on the Match, Economics and Radiology: Armchair Spin or Gospel Truth?

Dilated cardiomyopathy (DCM) reportedly affects one of 2,500 people in the United States and sudden cardiac death (SCD) is responsible for 30 percent of the deaths in this patient population.

 However, recent findings with myocardial tissue characterization on cardiac magnetic resonance imaging (MRI) may help identify those at higher risk of SCD and facilitate primary preventive options.

In a prospective study, recently published in 

, researchers reviewed data from 858 people (mean age of 48) with nonischemic DCM who had cardiac MRI scans between June 2012 and August 2020. During a 33-month median follow-up period, the study authors noted that 70 patients (8 percent) had SCD-related events.

After adjustment for factors such as atrial fibrillation, systolic blood pressure and left ventricular ejection fraction, the researchers found that late gadolinium enhancement (LGE) had a 1.87 hazard ratio (HR) for SCD-related events. The study authors also noted a 1.26 HR per 3 percent increases in extracellular volume fraction and a 1.07 HR per 10-msec increase in native T1 value.

In addition to identifying these independent predictors of SCD-related events, the researchers found the combination of LGE and native T1 mapping beneficial in predicting SCD risk. According to the study, they found the highest annual SCD-related events rate (9.3 percent) in patients with a native T1 value four standard deviations (SDs) above the mean (1382 msec) while the lowest annual SCD-events rate occurred for patients with negative LGE and a native T1 value two SDs below the mean (1292 msec).

“A risk category based on LGE and native T1 could be used to stratify participants with different risks of experiencing SCD-related events … and was superior to the LGE and left ventricular ejection fraction … in recognizing SCD-related events,” wrote Yucheng Chen, M.D., FACC, who is affiliated with the Department of Cardiology at Sichuan University in Sichuan, China, and colleagues.

For high-risk patients with DCM, the researchers noted a 3.09 HR for heart failure-related events and a 9.71 HR for SCD-related events.

(Editor's note: For a related article, see "

What a New Study Reveals About MRI Use and Implantable Cardioverter Defibrillators

, Hajime Sakuma, M.D., Ph.D., noted that T1 mapping with cardiac MRI is routinely employed at many hospitals for patients with DCM.

 While he cautioned that different MRI scanners and data acquisition approaches can lead to varied T1 mapping cutoff values employed to differentiate between normal myocardium and diffuse interstitial fibrosis, Dr. Sakuma said the combination of T1 mapping with LGE on cardiac MRI offers “great potential” in identifying appropriate candidates with DCM who may benefit from ICD treatment.

“The improved risk stratification of arrhythmic end points by the combination of LGE and native T1 shown in the study … has important implications for better selection of patients for ICD therapy,” noted Dr. Sakuma, a professor and chairperson of the Department of Radiology at Mie University in Japan, and past president of the Asian Society of Cardiovascular Imaging.

In regard to study limitations, the researchers acknowledged the limitations of a single-center study. They noted that the low proportion of patients on ICD treatment in the study may have led to underestimation of arrhythmic events. Chen and colleagues pointed out the use of different MRI scanners and techniques can impact the values one may see with native T1 scans and findings with extracellular volume fraction.

Emerging research suggests that cardiac MRI findings of late gadolinium enhancement, native T1 values and extracellular volume fraction may predict sudden cardiac death-related events in patients with nonischemic dilated cardiomyopathy. 

Could Cardiac MRI Improve Risk Stratification in Patients with Dilated Cardiomyopathy?

's Weekly Scan is your chance to catch up on the latest radiology content, from new research and device news to the latest case quizzes and conference reports. Review the slideshow below to see this week's top content:

Diagnostic Imaging's Weekly Scan: March 12-March 18

Slideshows

Catch up on the top radiology content of the past week.

Noting a bevy of concerns over small sample sizes, poor reporting completeness and insufficient handling of data complexity, the authors of a new meta-analysis claimed a significant majority of neuroimaging-based artificial intelligence (AI) models for psychiatric disorder diagnosis are inadequate for clinical application.

For the meta-analysis, recently published in 

, the authors reviewed 517 studies that presented a total of 555 neuroimaging-based AI models for use in detecting psychiatric disorders. For systematic assessment of the AI models for risk of bias and reporting quality, the researchers utilized the Prediction Model Risk of Bias Assessment Tool (PROBAST) and the modified Checklist for Evaluation of Image-Based Artificial Intelligence Reports (CLEAR) systems, according to the study.

Out of 555 AI models, the researchers found that 83.1 percent (461 models) had a high risk of bias (ROB). The meta-analysis authors also noted inadequate sample sizes in 71.7 percent (398 models) and insufficient handling of data complexity in 99.1 percent (550 models) of the AL models. 

While sample sizes of more than 200 participants have become acceptable for AI models, the study authors noted this was not the case for the majority of the models reviewed in the meta-analysis.

“ … About 80% of AI models for psychiatric diagnosis were trained on far smaller samples, leading to a high (ROB) and resulting in poor generalizability,” wrote study co-author Hu Chuan-Peng, Ph.D., who is affiliated with the School of Psychology at Nanjing Normal University in Nanjing, China, and colleagues. “On the other hand, model configurations, including data leakage, performance optimization, and absence of handling data complexities, represented key challenges to increasing model ROB … On balance, a high ROB in the analysis domain should be addressed to prompt clinical applications in future AI models.”

The meta-analysis authors also noted that 38.8 percent of the AI models were plagued by incomplete reporting and 60.1 percent had incomplete technical assessment.

“Going forward, complete and transparent reporting quality is imperative for applicable AI diagnostic models in clinical practice,” emphasized Chuan-Peng and colleagues.

In regard to study limitations, the authors noted they did not test the clinical performance of the AI models. They also acknowledged a rating heterogeneity with the benchmarks employed in the meta-analysis and suggested that the lack of a suitable benchmark may have led to a possible overestimation of ROB. Chuan-Peng and colleagues posited that the development of a psychiatry-specific, image-based AI model may bolster the reporting quality associated with these models.

In a review of 555 neuroimaging-based artificial intelligence (AI) models for the diagnosis of psychiatric disorders, researchers found that nearly 72 percent of the AI models had an inadequate sample size and over 99 percent were insufficient at handling data complexity.

Meta-Analysis Finds High Risk of Bias in 83 Percent of AI Neuroimaging Models for Psychiatric Diagnosis

Amy Patel, MD has seen the disparities with state laws on breast density notification. In a recent video interview, Dr. Patel, who practices in Missouri, said the neighboring state of Kansas does not currently require breast density notification. While Missouri has had a state law on breast density notification since 2015, Dr. Patel said the state does have high breast cancer mortality rates and a lack of awareness of breast density can contribute to a diagnosis of advanced breast cancer.

“We see it from all geographic locations, urban, rural. They may say ‘I didn’t know that I had dense breasts. My mammogram was negative one year and now you’re telling me that I have metastatic disease.’ It’s really heartbreaking,” noted Dr. Patel, the medical director of the Liberty Hospital Breast Care Center in Kansas City and an Assistant Professor of Radiology at the University of Missouri-Kansas City.

advent of national breast density notification, which will be required as of September 10, 2024, “really levels the playing field

 in terms of education and making the patients aware of their breast density.” However, she maintained that ongoing education for patients as well as primary care providers will be key to improving outcomes.

“We cannot assume that primary care providers have all the knowledge on breast density,” emphasized Dr. Patel, the President of the American Association for Women in Radiology. “As breast imagers, I fervently feel … that we’re going to need to take more of an active approach to ensuring that our primary care providers we work in concert with to serve these patients have the tools and knowledge so they can educate their patients. If they’re not able to at times, we need to step in and take more of an active role to educate our patients.”

Current Insights on National Breast Density Notification for Mammography Reports

Mammography Study Shows 22 Percent Higher Incidence of Dense Breasts in Women with a Family History of Breast Cancer.

For more insights from Dr. Patel, watch the video below.

In a recent video interview, Amy Patel, M.D., shared her perspective on forthcoming national beast density notification in mammography reporting, emphasized the importance of educating primary care providers on breast density risks, and discussed the ongoing need for coverage of supplemental options for breast cancer screening.  