Feature | Artificial Intelligence | April 03, 2019 | Greg Freiherr

Artificial Intelligence Helps Detect Breast Cancer and Saves Time

Study of smart algorithm with tomosynthesis is “triple win”

A smart algorithm developed by iCAD

A smart algorithm developed by iCAD outlines and scores a suspicious lesion seen in an image created using digital breast tomography. Displayed is the probability calculated by the algorithm that the outlined area includes a cancerous lesion. Image courtesy of Emily Conant, M.D.

 

Editor's Note: This article was updated following the SBI symposium.

 

In a study evaluating the use of artificial intelligence (AI) as a tool for breast cancer screening, increased sensitivity and specificity were achieved with AI when reading digital breast tomosynthesis (DBT) while reading time was reduced. Study leader, Emily Conant, M.D., framed the results as a triple win.

“We detected more cancers; decreased false positives; and decreased reading time,” said Conant, radiology professor in the Perelman School of Medicine and division chief of breast imaging at the Hospital of the University of Pennsylvania (HUP). She is scheduled to present details of the study April 4 at the Society for Breast Imaging (SBI)/American College of Radiology Breast Imaging Symposium in a talk entitled “Concurrent Use of Deep Learning Based Artificial Intelligence Improves Detection of Breast Cancer and Reading Time with Digital Breast Tomosynthesis in Women with Dense and Non-Dense Breasts.”

In an interview with ITN before the symposium, Conant described details that underscored the substantial impact that the algorithm can have on breast imagers. In regard to a particular case (see image above), without AI, 12 of 24 readers in the study detected the cancer with an average reading time of 72.7 seconds.  With AI, 22 of 24 readers detected the cancer with an average reading time of 45.6 seconds.

“So ten more readers detected the cancer while also reducing their reading time on average 37% (27.1 seconds),” she said.

The test data set was “enriched” with more images of pathology than would occur in a typical screening population. Radiologists served as their own controls. They read the same images either with or without AI, waiting four weeks between readings so they would not have an advantage that might have been gained from previous interpretations.

“They go back to work and see lots of other cases. (After four weeks), they don’t remember how they read the earlier cases,” she said.

The results, she said, were provided to iCAD, developer of the AI algorithm, to support its FDA application for product marketing of the software.

 

New AI Better Than Earlier CAD Software

The great advantage of using the AI algorithm, she said, is that not only did radiologists find more cancers but they did so without increasing the number of false positives. And they did the work substantially faster. This is much different than what happened with CAD (computer-assisted detection) software developed for full field digital mammography (FFDM)

A major problem with CAD software was the high number of false positives that came from its use. CAD software also tended to slow rather than accelerate interpretations.

In contrast to CAD software, the AI algorithm tested at HUP was applied to tomosynthesis rather than FFDM; was used at the same time radiologists’ interpreted images; and calculated probabilities that the suspicious lesions were cancerous, rather than simply marking suspected lesions for radiologists to consider after they had already interpreted the image.

“The concept behind this AI is that you will have it from the ‘get-go.’ It can help you rapidly navigate to slices that have concerning lesions,” Conant said.

 

AI Efficiency Varies With Breast Density

One unexpected outcome from the HUP research was that the algorithm shortened reading times more for dense than non-dense breast.  Average reading times dropped from 62.5 seconds to 32.8 seconds when reading images of non-dense breast – and from 65.8 seconds to 28 seconds for images of dense breasts. 

Conant attributes the extra time for reading non-dense images to the greater complexity associated with images of this type. Although dense breasts can obscure pathology, they may also contain fewer structures that need to be analyzed. 

“When it comes to reading time, sometimes there is not a lot to look at with a dense breast.  If you have a breast that is less dense and has a lot of little nodules and calcifications, it can be a more complex texture and, therefore, require a longer reading time,” she said.

The differences in reading time between images of different breast types, however, were relatively small, averaging just a few seconds.  Similarly sensitivity increased with AI assistance slightly more for dense than nondense breasts as did specificity.  Again the differences were slight.

The key takeaway from the research, according to Conant, is that – regardless of breast density – the use of artificial intelligence, when applied to screening for breast cancer, can reduce the time needed to read DBT while increasing the detection of cancer.

 

Greg Freiherr is a contributing editor to Imaging Technology News (ITN). Over the past three decades, he has served as business and technology editor for publications in medical imaging, as well as consulted for vendors, professional organizations, academia and financial institutions.

 

Related Content

The Evolution of Digital Breast Tomosynthesis

VIDEO: How iCAD Uses AI to Speed Breast Tomosynthesis

3-D Tomosynthesis: Where Are We Now?

 

 

Related Content

Owkin, a startup that deploys artificial intelligence (AI) and Federated Learning technologies to augment medical research and enable scientific discoveries, presented findings in Hepatocellular Carcinoma (HCC) with Cleveland Clinic at the 2021 European Society of Medical Oncology (ESMO) conference.

Illustration courtesy of Cleveland Clinic

News | Artificial Intelligence | September 24, 2021
September 24, 2021 — Owkin, a startup that deploys...
Paige Prostate, is the first artificial intelligence (AI)-based software designed to identify an area of interest on the prostate biopsy image with the highest likelihood of harboring cancer so it can be reviewed further by the pathologist if the area of concern has not been identified on initial review.
News | Digital Pathology | September 22, 2021
September 22, 2021 — The U.S.
This study shows that thanks to deep learning analysis applied to digitized pathology slides, artificial intelligence can classify patients with localized breast cancer between high risk and low risk of metastatic relapse in the next five years.

Getty Images

News | Artificial Intelligence | September 22, 2021
September 22, 2021 — The RACE AI study conducted by Gustave...
HealthMyne, a pioneer in applied radiomics, announced today that peer-reviewed research recently published in the journal Cancers has demonstrated the ability of its radiomics technology to identify biomarkers that predict whether patients with lung adenocarcinoma would benefit from immunotherapy.

Semi-automatic lesion identification: (A) Manual ROI indication. In blue, it is possible to observe the axes that cross the lesion manually delineated by the radiologist on a plane of the MPR. The intensity of the lesion boundary (estimated) is represented with a red outline. (B) Additional axes can be dragged onto other orthogonal MPR views. From left to right, it is possible to observe the initial long axis outlined by the radiologist and the 2D contours on the axial, coronal and sagittal views of the lesion used as a starting point for the RPM algorithms. (C) Resulting 3D contour of the lesion (in blue).

News | Radiomics | September 21, 2021
September 21, 2021 —  HealthMyne, a pioneer in applied radiomics, announced today that peer-reviewed ...
News | Breast Imaging | September 20, 2021
September 20, 2021 — ImageCare Centers is unveiling its new “PINK Better Mammo” service with the addition of...
This is an example of 3-D ultrasound imaging on a breast, designed to help increase efficiency and diagnostic accuracy in any practice. Image courtesy of Hologic.

This is an example of TriVu ultrasound imaging on a breast, designed to help increase efficiency and diagnostic accuracy in any practice. Image courtesy of Hologic.

Feature | Breast Imaging | September 15, 2021 | By Jennifer Meade
The...
While the Mammography Quality Standards Act (MQSA) and the introduction of EQUIP (Enhancing Quality Using the Inspection Program) have been successful in standardizing and enhancing mammographic imaging quality, inadequate breast positioning can dramatically impact the ability of radiologists and technicians to quickly and accurately detect breast cancer and potentially malignant lesions in their patients

Getty Images

Feature | Mammography | September 15, 2021 | By Christopher Austin, M.D. and Randy D. Hicks, M.D., MBA
Cloud services have been utilized within healthcare organizations for more than a decade. Now with the growth of artificial intelligence (AI) it is very common to see organizations adopting cloud services.

Getty Images

Feature | Information Technology | September 14, 2021 | By Jef Williams