A deep convolutional neural network (DCNN) can differentiate between males and females on chest X-ray, according to new research.

Based on existing clinical experience, DCNNs have been shown to be biased against men or women if they are trained on datasets that do not have balanced sex representation. Until now, though, is has been unknown if the algorithms can use visual biomarkers that are beyond a radiologist’s perception to accurately predict sex.

Society for Imaging Informatics in Medicine 

(SIIM) 2021 Virtual Annual Meeting, David Li from the University of Ottawa, detailed the predictive efficacy of a DCNN trained on datasets that included equal numbers of males and females.

“DCNNs trained on two large chest X-ray datasets accurately predicted sex on internal and external test data with similar heatmaps across DCNN architectures and datasets,” he said. “These findings support the notion that DCNNs can leverage imaging biomarkers to predict sex and potentially confound the accurate prediction of disease on chest X-rays and contribute to biased models.”

To test DCNN ability to predict sex, as well as evaluate the visual biomarkers used, Li gathered chest X-ray data from the Stanford CheXPert and National Institutes of Health (NIH) Chest XRay-14 data set, including 224,316 and 112,120 scans from two heterogeneous patient populations, respectively. By using random under-sampling the data volume was reduced to 97,560 images that were balanced to 50 percent male and 50 percent female.

Overall, the dataset was split into 70-percent training, 10-percent validation, and 20-percent test sets. Li used multiple DCNN architectures pre-trained on ImageNet – Inception-V3, ResNet-18, ResNet-50, and VGG-19 – for transfer learning. They were also externally validated.

According to his analysis, on the internal test set, DCNNs trained with both datasets reached an area under the cure ranging from 0.98 to 0.99. External validation showed a peak cross-dataset performance of 0.94 for VGG19-Stanford model and 0.95 for InceptionV3-NIH model.

Additionally, heatmaps showed similar attention areas between model architectures and datasets. They were localized to the mediastinal and upper rib regions and to the lower chest and diaphragmatic regions.

For more coverage based on industry expert insights and research, subscribe to the Diagnostic Imaging e-Newsletter 

Articles

SIIM

Study results show DCNN can use imaging biomarkers to predict sex, potentially confounding the accurate prediction of disease.

Deep Convolutional Neural Network Can Predict Sex on Chest X-ray

Putting standard processes in place can make it easier to implement point-of-care ultrasound (POCUS) in a facility, potentially opening the door to millions of dollars in new revenue.

Although it is an easy-to-use modality, POCUS utilization can be slow if institutions are unable to bill for it because their providers aren’t credentialed or if their images are not stored in an enterprise archive.

(SIIM) 2021 Virtual Annual Meeting, Laurie Perry, RN, lead systems analyst at Cincinnati Children’s Hospital, detailed how the hospital successfully turned the corner to properly billing for POCUS – a move that brought in $2.25 million in new services from 5,000 studies.

“POCUS was successfully implemented at our hospital,” she said. “This has enabled the reproducible implementation of a new technology demanded by physicians and has generated new revenue for our institution.”

To reach this goal, the hospital launched an implementation team comprised of imaging informatics professionals and electronic health record (EHR) analysts, as well as a project manager, business director, an educator, and a physician. They focused on five key components to successful implementation: EHR workflow design and creation, procedure codes for billing, provider credentialing, user training, and modality procurement and configuration.

Based on their interactions with leaders from each service, the team pinpointed procedures that could be performed in the division and drafted appropriate billing codes. They also identified providers who should receive customized training and credentialing. In addition, they assessed existing POCUS equipment for Wi-Fi capability and whether it could use a DICOM worklist, as well as create and store DICOM images in an enterprise archive. For divisions without POCUS, the team offered a choice of hospital-approved devices.

Throughout the process, the team overcame two significant challenges. First, they worked with clinical leaders to create a standard POCUS workflow in the EHR that automated as many steps as possible while pulling from both existing workflow and ad-hoc order creation. Second, in order to correctly bill for imaging studies, the team collaborated with the radiology business director to identify and create proper procedure codes.

Based on their success, the hospital has three more go-lives planned for the upcoming year that will focus on each division’s individual needs.

“Creating a standard process has allowed us to implement POCUS in multiple divisions,” Perry said. “Based on our experience, we have made improvements in our process and now provide interested divisions with a list of approved POCUS vendors and involve radiology business directors early in the project.”

Working with clinical and business leaders to create a standard process for point-of-care ultrasound opened the door for greater utilization and revenue capture. 

Standard Enterprise Imaging Process for POCUS Brings in Millions for Hospital

Given the advancements in CT, each patient scan can include thousands of individual images. It’s nearly impossible to examine them all, and radiologists must keep in mind that they’re not necessarily seeing everything when they read studies.

(SIIM) 2021 Virtual Annual Meeting, Fletcher Moore, M.D., a diagnostic radiologist with the University of Maryland Medical System showed the frequency at which images are skipped during an average interpretation.

“Images are imperceptibly skipped during normal scroll operation on thin slice CT under experimental conditions,” he said. “Skipped images may contain clinically significant information that could impact patient outcomes. Radiologists should be aware of this possibility.”

Based on his analysis, the thinner the CT slice, the more images are skipped.

Many PACS viewers do have a default image stack scroll feature in place, he said. To determine the amount of images skipped during a normal scroll operation and how it relates to monitor refresh rate and slice thickness, he and his colleagues examined screen capture data, slow-motion video, and optical character recognition (OCR).

The team recorded a radiologist scrolling through 5-mm or 1-mm slices of publicly available chest CT scans using the RadiAnt DICOM viewer set to either 120 Hz or 60 Hz. They extracted the current image number from the video on a per-frame basis with OCR and analyzed that dataset to determine the frequency of image skips. They compared the skip rates to the Student t-test.

Based on their analysis, no image skips occurred during 5-mm slice scrolling. But, when they examined 1-mm slice scrolling at a 60 Hz monitor refresh rate, they saw that 1.13 percent of images were skipped. It increased to 1.17 percent for 1-mm slice at 120 Hz. The Student t-test revealed no statistically significant difference between the skip rates. In addition, they determined the OCR error rate on screen-captured data was less than 1 percent, and it was roughly 5 percent on slow-motion video.

Given that the monitor refresh rate did not have a significant impact on results, Moore’s team said the outcome points to a computer hardware or software limitation on frame rate.

Images can be skipped with monitors with refresh rates of both 60 Hz and 120 Hz.

During Normal Scrolling, Clinically Relevant Images Can Be Skipped

Much has happened with enterprise imaging in recent years, but the one constant has been that facilities have individual and unique needs when it comes to finding the right platform. And, to find the right solution, you have to ask the right questions.

(SIIM) 2021 Virtual Annual Meeting, Kim Garriott, chief innovation officer for NetApp Healthcare, walked through 10 questions you should pose to potential vendors – but, she pointed out, these queries are only a starting point.

“These suggestions are basic suggestions. Solutions have changed over time and what is provided varies from vendor to vendor, and the needs that you have, the priorities that you have as an organization will vary, as well,” she said. “Take this information as basic information, and see how you can apply it inside your organization.”

How does the enterprise imaging solution accommodate various orders- and encounters-based workflows?

Be sure your vendor can support both orders- and encounters-based workflows, she said. Focus on the discrete workflow needs of each of your service areas, and ask your potential vendor what consultation they can provide that will help you design these various workflows. It is likely the vendor has encountered a myriad of workflow designs, so draw from their experience to create what works best in your institution.

How does this solution leverage AI? What type of AI inference models are available?

Find out how the vendor can leverage current AI technologies to assist in your workflow and image interpretation. Ask if they have a marketplace with AI inference models that you can use within your organization, as well as what learning models they support.

How does the solution manage DICOM and non-DICOM images?

Whether your organizations uses DICOM or non-DICOM images, be sure you select a vendor that can work with you to retain your images. Be sure you have a good understanding of what their capabilities are.

What interoperability standards does the solution use and support?

Interoperability is critical to being able to provide the highest level of patient care, so when choosing a solution, be sure your choice has the highest degree of standards being applied to it. Can it support cross-enterprise document sharing? Does the company participate in some of the standards organizations as a way of staying abreast with what’s coming with interoperability in the future?

What are implementation and post-implementation support options?

Depending upon your needs and what solution you choose, you will likely need additional support from the company. Ask what training options are available – and at what cost – and whether the vendor can provide at-the-elbow support or if it will be offered by a third party. Your vendor should be able to offer support for the various types of workflow you have in your organization, but you should also ask if they will ever train you to build your own workflow configurations if you are interested in doing so. These details are critical to the success of your implementation and beyond, she said.

How does the solution design mitigate performance latency and workflow disruption?

Find out the time lag for delivery of images to the devices your radiologists use for image interpretation. In addition, ask about how the vendor pushes out upgrades. Do they support non-disruptive upgrades and maintenance? Will these activities need to take place after hours, and is there a charge for conducting maintenance during off hours? Having these details will help you design your internal support processes and determine budget allocations.

How often are solution enhancements/upgrades published?

Find out how long it takes for your enhancement requests to be integrated into any future design. Time frames on enhancements will vary from vendor-to-vendor and based upon the importance of the enhancement request. Also, ask how frequently your solution will be upgraded so you can plan appropriately.

What image life cycle management capabilities does the solution support?

Getting details about image life cycle management can help you design and coordinate your data and image retention. Will you be able to automatically tier data or age it off it if hasn’t been accessed for a certain period of time, and will you be able to implement rules that will move certain types of images, such as point-of-care ultrasound images used to support billing, to different tiers? As you generate more and more imaging data, you will need to understand your management capabilities.

Data migration is an enormous and costly endeavor, and you need to know if your vendor will help you or if they will train you to do it on your own. Ask how long it will take and what you will be responsible for doing. Your vendor can also tell you what they can do to limit the number of images that must be moved and how they can make the migration as non-disruptive as possible.

How does the solution leverage cloud technology?

With the growing volume of imaging data, the cloud is here to stay, Garriott said. Bring your IT staff into the conversation, and ask a potential vendor how it can leverage the cloud for image retention and disaster recovery. Determine their road map for using the cloud and if they use a private cloud. Or, do they rely on a major provider, such as Google or Amazon, and will you be able to move your data between clouds.

Choosing a vendor for an enterprise imaging solution can be complex. Here’s what you need to inquire about to make the best choice for your organization.

10 Questions to Ask Your Enterprise Imaging Vendor

No accomplished radiologist reaches the peak of performance without guidance from others who have more experience and wisdom. That’s why the mentor-mentee relationship is so critical – but merely asking for or agreeing to be a mentor isn’t enough. Both parties must be invested for the partnership to be fruitful.

(SIIM) 2021 Virtual Annual Meeting, both Eliot Siegel, M.D., professor of diagnostic imaging and nuclear medicine, as well as vice chair of information systems at the University of Maryland School of Medicine, and Omer Awan, M.D., assistant professor of radiology at Temple University Lewis Katz School of Medicine, discussed what it takes to be a successful mentor and a great mentee.

They offered 15 tips that can help set these partnership off on the right foot.

“The most satisfying thing in my career has been serving as mentor to more than 100 mentees from the high school level to the attending physician level,” Siegel said.

Along the way, he said, he has picked up 10 tips that are a critical part of being an effective mentor.

 Have a conversation early about what the mentee is seeking from the relationship. Get both a general and specific idea of his or her goal.

 While it can sometimes be difficult, step back and view a question or situation from the mentee’s shoes. Think about things from their perspective or with their insights.

 Focus on keeping conversations positive and engaging. If both mentor and mentee enjoy the conversation, then the advice and guidance are likely to be more impactful.

 Challenge mentees to think outside-the-box and to identify opportunities they might not have thought of before. For example, Siegel said, he frequently asks mentees to envision themselves five years in the future and to think about what piece or pieces of advice their future self might wish to give to their present self.

 Too often, physicians fervently pursue their next challenge without taking the opportunity to relish their accomplishments. Take the time with mentees to celebrate the small and large wins.

 Most mentors have identified their niche be it private practice or academia. For many mentees, it’s too early to decide on a niche for their career, but encourage them to think about the many options that are available. It’s never too early to begin contemplating it.

 Don’t limit what you offer a mentee to fulfilling their specific request. Always be prepared to contribute more to the relationship – those efforts can lead to expanded, future partnerships and collaborations.

 While mentees are actively searching for guidance and advice, be open to learning from them, as well. Throughout his career, Siegel said, he has learned far more from his mentees in aggregate than they ever learned from him individually.

 Be prepared to recommend, promote, encourage, and become a champion for a mentee, and be sure to mentor by example.

 It doesn’t matter whether a mentee has asked for a week-long, month-long, or year-long mentorship. Invest your time and energies as if you will serve as a mentor ad infinitum.

Having an outstanding mentor is only part of the recipe for a successful mentor relationship. Mentees carry responsibilities, as well, Awan said. He offered these five tips to help mentees maximize their role.

 Take the initiative to start conversations and drive discussion – don’t wait for a mentor to reach out. Be proactive in asking questions and sharing ideas.

Believe in yourself, but understand that experiencing failure at some point is inevitable. Embrace it, and use it as a learning experience for the future while maintaining confidence in your capabilities.

 Always be willing to learn something new. Seek out advice from multiple mentors for research, education, or practice. Having a plethora of mentors can only be helpful, Awan, so the more the merrier.

 Now is not the time to be shy, he said, and do not be afraid of being judged. Express your concerns and discuss any problems openly and honestly. Approach all discussion with open eyes, ears, and heart to maximize what a mentor has to offer. And, remember, all communication is a two-way street.

 Express thanks for the time and effort a mentor volunteers. They are freely giving of their time and expertise, so take the time to show appreciation.

For more coverage based on industry expert insights and research, subscribe to the Diagnostic Imaging e-newsletter

15 tips for being an ideal mentor or mentee.



Mentor or Mentee: What It Takes to Be Great

Using a 3D convolutional neural network (CNN) opens the door for automatic segmentation of the pancreas on CT imaging to pinpoint biomarkers for both endocrine and exocrine disease, new research shows. 

(SIIM) 2021 Virtual Annual Meeting, Panagiotis Korfiatis, Ph.D., assistant professor of radiology and senior associate consultant for artificial intelligence in the medical physics division in the Mayo Clinic’s radiology department, revealed how using a 3D CNN automates pancreas segmentation on CT to save time in patients with both normal pancreas or pancreatic ductal adenocarcinoma (PDA).

“A deep-learning model developed on a large dataset of CT scans enabled accurate fully automated pancreas segmentation across a wide-range or CT slice thicknesses in cases with normal pancreas, as well as in those with PDA,” he said. “Furthermore, the model’s performance generalized on an external public dataset for normal pancreas.”

The model, he added, functions as a discovery tool for automated quantitative pancreas imaging in patients with neoplastic and non-neoplastic disease. It is an improvement over manual pancreas segmentation which is difficult to integrate into clinical practice because it can be time-consuming and cumbersome.

To test the CNN’s efficacy, two radiologists completed volumetric pancreas segmentation with a 3D slicer on 2,369 abdominal CT scans performed between 2006 and 2017. Of the scans, 519 showed PDA. In order to develop a two-stage 3D CNN model, Korfiatis’s team randomly divided the dataset into training (1,612 scans), validation (284 scans), and test (473 scans) sets. They compared the model’s performance for pancreas segmentation and differences in model-predicted pancreatic volumes versus ground truth volumes to the test set, and they used an external dataset of 80 CTs from the National Institutes of Health-Pancreas CT 3 for validation.

According to the team’s analysis, there was no difference in the model’s performance between different CT slice thicknesses, ranging from 1.25 mm to 5 mm, based on non-inferiority statistical testing. The team also found no difference in model-predicted and GT pancreatic volumes.

3D Convolutional neural network can effectively replace manual segmentation of the pancreas with or without cancer.



AI Accelerates Fully Automated CT Segmentation of the Pancreas 

Providers can do a better job of predicting how patients with hepatocellular cancer (HCC) will respond to Yttrium-90 radioembolization (Y90-RE) treatment when they use a deep convolutional neural network (DNN).

(SIIM) 2021 Virtual Annual Meeting, William Wagstaff, M.D., MS, a third-year radiology and imaging sciences resident at Emory University School of Medicine, revealed that applying the algorithm improved the ability to determine how a patient will respond to Y90-RE therapy.

“The deep learning algorithm had a 16-percent higher area under the receiver operating characteristic curve (ROC-AUC) for predicting immediate, post-radioembolization treatment response in patients with HCC compared to the current best method,” he explained. “This algorithm has the potential to highlight patients that may benefit from earlier follow-up or additional therapy.

Only a small number of patients with HCC qualify for curative surgery or transplant, so Y90-RE is the alternative treatment method. However, a patient’s response to Y90-RE is assessed via imaging after three-to-six months, potentially giving the tumor time to progress or metastasize. To find out whether using a DNN could help providers predict a patient’s Y90-RE response, Wagstaff’s team conducted a retrospective analysis on HCC patients who received Y90-RE treatment between December 2014 and January 2019 at a single institution.

For their study, they divided 77 patients with a total of 103 lesions who had no prior Y90-RE therapy into two groups – 57 with complete response to treatment and 46 with incomplete response. They hand-segmented lesions on pre-treatment arterial phase MRI to pinpoint lesion location, as well as on post-treatment arterial phase MRI to determine slice-wise response, and they used immediate post-Y90-RE SPECT as a proxy for treatment area and dose. Ninety percent of lesions were in a training set, and 10 percent were part of a hold-out test set.

Based on their analysis, the five-fold cross-validation results for the ensemble model were sensitivity of 0.73 +/- 0.29, specificity of 0.70 +/- 0.14, accuracy of 0.75 +/- 0.09, and ROC-AUC of 0.79 +/- 0.12. By comparison, the partition model had a sensitivity of 0.36 +/- 0.34, specificity of 0.80 +/- 0.40, accuracy of 0.63 +/- 0.04, and ROC-AUC of 0.62 +/- 0.09.

In addition, test set results for the ensemble model were sensitivity 0.32, specificity 0.85, accuracy 0.65, and ROC-AUC 0.66, and the results for the partition model were sensitivity 0.03, specificity 0.99, accuracy 0.62, and ROC-AUC 0.57.

Using a deep convolutional neural network can improve providers’ ability to predict how patients with hepatocellular carcinoma will respond to Y90-RE treatment, enabling them to offer additional therapies, if necessary.

Deep Learning Helps Predict Y90-RE Response in HCC Patients

Adding an artificial intelligence (AI) tool that can assess the risk of malignancy in a thyroid nodule can not only improve accuracy, but it can lead to faster diagnoses, according to newly-published research.

(SIIM) 2021 Virtual Annual Meeting, Lev Barinov, Ph.D., vice president of medical applications and applied research at clinical decision support company Koios Medical shared research that showed including AI can lead to fewer unnecessary thyroid biopsies.

Thyroid nodules are common, affecting approximately 68 percent of people, and 95 percent of these findings are benign. Still more than 600,000 fine needle aspirations – a method that only roughly has 30-percent positive predictive value – occur annually. Implementing TI-RADS has reduced the rate of unnecessary biopsies by 19.9 percent to 46.5 percent, Barinov said, but the Koios Medical team wanted to see if it were possible to generate more improvement.

Based on their work, they developed an AI tool that can pre-populate TI-RADS descriptors to create a putative point total that readers can review and, potentially, use to modify their diagnostic decisions.

“Automated AI-based pre-population of the 5 TI-RADS descriptors combined with use of an additional AI-based risk descriptor and point modifier significantly improves reader diagnostic accuracy while simultaneously decreasing interpretation time and inter-reader variability,” he said.

To create the tool, the team had 15 readers (11 radiologists and four endocrinologists) review 650 nodules that were proven by fine needle aspiration twice across two sessions that were separated by four weeks. Of the group, 130 nodules were malignant. The readers randomly evaluated the nodules in one of two conditions – manual scoring of the TI-RADS report and AI-pre-populated TI-RADS report that was augmented by the AI tool and point modifier.

According to their analysis, adding the AI tool improved both sensitivity and specificity, as well as inter-reader variability. It also reduced interpretation time. 

Specifically, the average area under the curve improvement for TI-RADS+AI and TI-RADs alone was 0.083, and TI-RADS+AI produced an 8.4-percent sensitivity increase and a 14-percent specificity boost. Inter-reader variability for TI-RADS and TI-RADS+AI was 0.622 and 0.876, respectively, and the time for interpretation dropped by 23.6 percent when readers used TI-RADS+AI, he said.

Incorporating artificial intelligence with TI-RADS improves sensitivity, specificity, and reduces interpretation time by nearly one-fourth. 

AI Improves and Reduces Time to Thyroid Nodule Diagnosis

Jonathan Kruskal, M.D., Ph.D., chair of radiology and chief of abdominal imaging at Beth Israel Deaconess Medical Center, became the 121st president of the American Roentgen Ray Society (ARRS).

The organization announced his new leadership role during its 2021virtual annual meeting this past week. Kruskal will take over for Alexander Norbash, M.D., an interventional neuroradiologist from the University of California-San Diego Health.

For more ARRS 2021 Virtual Annual Meeting conference coverage, click 

Alongside his new ARRS position, Kruskal also serves as Consultant to the Editor for 

a position he will hold from until 2026. He is also a leading voice on quality improvement initiatives and peer learning efforts in radiology, having authored more than 200 publications on these subjects.

Kruskal’s storied career began when he began his MB and ChB degrees from the University of Cape Town (UCT) in South African in 1982. After completing a medical internship at Groote Schuur Hospital, he joined the South African Liver Research Center at UCT Medical School in 1987 as a research scholar, earning his doctorate degree the same year in molecular biology and liver tumor physiology.

After arriving in the United States, in 1987, he completed a post-doctoral research fellowship at the Study Center for Anesthesia Toxicology at Vanderbilt University. He, ultimately, completed his radiology training in 1994 when he finished up a residency at Harvard Medical School’s New England Deaconess Hospital.

His contributions to the field of radiology have been myriad. Not only did he, as an RSNA Scholar, develop a mouse liver tumor model that can be used with optical and molecular imaging technology to assess tumor growth and metastases, but, as an attending provider, he has also been deeply involved in practice and physician quality improvement efforts.

In addition, Kruskal has also help several leadership positions. Within the ARRS, he has also served on the Executive Council and as Vice President (2019) and President-Elect (2020). He served as president of the Society of Abdominal Radiology in 2019, as well as chairman of the Quality Management and Inter-Society Committees and Board of Chancellors of the American College of Radiology. 

ARRS

Abdominal imaging specialist Jonathan Kruskal, M.D., Ph.D., assumed the role during the 2021 ARRS Virtual Annual Meeting.



ARRS Names 121st President

Older patients and those who have prior mammogram screenings are more likely to receive a cancer diagnosis from an initial “probably benign” BI-RADS 3 assessment, according to new research.

In a poster presented during the American Roentgen Ray Society (ARRS) 2021 Virtual Annual Meeting, a team led by Cindy Lee, M.D., a radiologist and clinical assistant professor with New York University Langone Health, examined the results of the National Mammography Database (NMD) and determined that both age and the existence of prior mammogram screenings should be considered with BI-RADS 3 assessments.

“We sought to determine the effect of patient age and availability of prior mammograms on cancer yield and cancer staging of BI-RADS 3 findings in the National Mammography Database,” the team said. “[Based on our results,] discussions with patients should include the estimated risk of cancer.”

Overall, the NMD includes 31,905,228 mammograms collected from 690 facilities throughout 45 states. For their retrospective study, Lee’s team analyzed data gathered between 2009 and 2018 from 471 NMD facilities, including 19,675,062 mammograms from 8,296,728 patients. Their patient population included women who were recalled from screening mammograms for a follow-up diagnostic evaluation that yielded a BI-RADS 3 assessment who either had a biopsy or an imaging follow-up after two or more years. The team stratified their data into 10-year intervals.

For their evaluations, they concentrated on age, the existence of prior mammograms, and lesion type, looking at cancer yield, PPV3, biopsy rate, and the percentage of invasive cancers.

Of the study population, they identified 43,628 women – with an average age of 55 – who had BI-RADS 3 assessments. In the group, the overall cancer yield was 1.86 percent, the PPV3 was 16.6 percent, and the biopsy rate was 11.2 percent. The team did determine the cancer yield was higher among the 25,160 women who had prior mammograms – 2.24 percent. 

All total, they pinpointed 810 cancers, 545 of which (67.3 percent) were invasive. From their analysis, they saw the rate of ductal carcinoma in situ was much higher for women in their 40s than older women, 41.3 percent and 31.1 percent, respectively.

When the team drilled down on the data, they made several additional discoveries. Information on lesion type was available for 10,804 women (24.8 percent). Based on that data, the team said, breast cancer mass is the most common type of lesion (1.11 percent), but it also has the lowest cancer yield among women of all ages (0.47 percent). Architectural distortions are rarer, but they have a much higher cancer yield (3.21 percent). However, the team noted, when prior mammograms were available, the cancer yield for masses jumped significantly for older women – up to 5.1 percent.

While there were few asymmetries, the team said, those that were presumed to be developing consistently had a cancer yield rate of more than 2 percent for women between ages 60 and 90.

Ultimately, the team said, they found no clear relationship between cancer staging and age or between prior mammogram availability and cancer staging or cancer type. But, still women with prior mammograms who had BI-RADS 3 findings still had significantly higher cancer yield. In fact, they said, cancer yield exceeded 2 percent at age 60 for baseline and actually reached 4.6 percent for women ages 80-to-89.

Based on these outcomes, they said, surveillance may still be a reasonable option for findings with a slightly higher cancer yield, such as 3 percent instead of 1 percent or 2 percent.

Research findings show patient age and existing mammogram studies play a role in assessment.

Latest Conference Articles