Inventor of the Year at Siemens Healthineers

Inventor of the Year at Siemens Healthineers: Recognizing breast cancer at a very early stage

By: Norbert Aschenbrenner, Senior Communications Expert

Anna Jerebko has a calling: to identify cancers or any abnormalities as early as possible. The Inventor of the Year at Siemens Healthineers won’t rest until she’s made the latest technology usable for diagnostic purposes. That’s why she invented algorithms that can be used in mammography to show breast-tissue structures in high-rotating 3D animations and in single depictions for the first time. The award in the Single Outstanding Invention category recognizes her work in this field.


Non-specialists looking at a mammogram will see nothing but tissue in various shades of black, white, and gray. “Even radiologists sometimes find it difficult to spot every tiny change, especially in highly dense breast tissue, despite their years of experience,” Jerebko says. 

Thanks to a new imaging technique, digital breast tomosynthesis, Siemens Healthineers has achieved a depth of resolution never seen before in mammography images. This high-focus depth for mammography is achieved by rotating the x-ray source at an angle of up to 50 degrees around the target tissue (other manufacturers only work with ranges of 15 to 40 degrees). The tomographic images taken at a 50-degree angle show the composition of the tissue in much greater detail. But these were still tomographic slices through breast tissue, not the 2D images depicting the complete breast that radiologists are accustomed to using for diagnosis. 


Conventional 2D mammography, however, has its limitations, because overlapping dense tissue often obstructs abnormalities in the breast. How could the convenience of a conventional 2D mammogram be combined with the diagnostic advantages 3D imaging? Algorithms that are available to calculate 3D renderings of medical data don’t work for digital breast tomosynthesis like they do for computed or magnetic resonance tomography.


“I was thinking that we had to be able to do a much better job of depicting the breast tissue as well as any abnormal changes and microcalcifications to be able to recognize breast cancer at a very early stage,” Jerebko says.


She had found a new research project: How to turn the data generated during a digital breast tomosynthesis (DBT) scan into a 3D model that would represent the entire breast with the greatest possible depth resolution, depicting even tiniest structures. A model existed in the form of cinematic rendering, which uses raw data from CT and MRI scans to create hyper-realistic representations of internal body structures.

Taking breast-tissue images to the next level

Cinematic rendering works best with organs whose structures are clearly delineated, such as the heart with its chambers, valves, and arteries.


“But if you use this procedure with breast tissue, all you see is a cloudy area where micro-structures and sharpness could be lost,” Jerebko says.


With the algorithm she developed, however, the raw data from the wide-angle DBT images could be used to calculate breast tissue rendering that had never been seen before: The structure was now visible in detail. Areas where the breast tissue is more or less dense can be clearly identified. Microcalcifications can be clearly seen. Accurately identifying these is the most important task for diagnostics because breast cancer can develop from an accumulation of microcalcifications, but not every microcalcification is malignant. The calcifications of interest are tiny—just 100 to 200 microns in size—and the radiologist must determine their morphology to decide whether the cluster is harmless or dangerous.


The Insight 3D procedure developed by Jerebko now makes it possible to view the clusters and all other tissue from different sides. The procedure is now in the process of being registered globally for clinical use. The path registration was a long and arduous one.


“First I and then my team developed countless variations of the algorithm,” says Jerebko.


Images may be sharp and detailed, but they’re worthless if the radiologist can’t get any clinical benefit from them. Multiple variations were assessed by teams of clinicians and medical physicists in Sweden, the Netherlands, Belgium, and Japan. Application Specialists and Product Managers at Siemens Healthineers looked at thousands of images and compared different forms of presentation and versions of the algorithm.


“As computer scientists, we need feedback from the clinicians and specialists, and we got the whole village involved,” Jerebko says.


She finally got a result: She invented a new procedure for representing breast tissue that’s now being used in hospitals and hopefully will help to improve cancer detection. “Although we have done research studies on the diagnostic value of Insight 3D with clinicians, I will keep my fingers crossed as we expect more broad clinical feedback on its usefulness from all over the world,” she says.

Identifying changes before they turn malignant

Jerebko trained as a computer scientist in Russia, but she quickly discovered that “just programming was not very exciting” and specialized at an early stage in neural networks for image processing and then obtained a PhD in this field. She went to England, where she worked on a research project on image restoration in poor visibility and foggy conditions.

“We were looking at images of typical English fog,” she says.


The subject may not sound particularly exciting, but it sparked her interest in recovering barely visible objects in images. At that time, the National Institutes of Health in the U.S. were starting on one of the first projects for computer-assisted colon polyp and cancer detection, and Jerebko was a part of it.


“That’s where I finally found my calling,” she says, “using medical imaging and machine learning to discover changes that could become malignant.”


After two years, she learned that Siemens was setting up a research group for computer-assisted diagnostics and medical-image processing with the final goal of making a medical product.

“I was one of the first to work there, and we were far ahead of our time,” she recalls.


An algorithm developed at the time for early lung-nodule detection, which has been on the market for 13 years, is still considered by experts to be one of the best, according to their publications, Jerebko recounts. She gradually assumed more responsibility in her area of specialization: machine learning and pattern recognition. While still in the U.S., she became a Senior Key Expert, and today she has risen to the position of Principal Expert.


Jerebko spent more than eight years in Germany, first as a research scientist and then as the head of a team developing algorithms for mammography, digital breast tomosynthesis and 3D-radiography systems, including those that now perform the calculations for the innovative breast imaging as part of Insight3D.


Two years ago, she returned to the U.S. with her family and now heads a team of artificial intelligence, medical imaging and text analytics scientists in Malvern, Pennsylvania, and leads AI-focused product development for teams in Bangalore, India. The many changes of location have been a challenge, both for her and her children.


“But we’re always interested in new experiences,” she says about her family, “and it’s great that Siemens Healthineers makes these opportunities available.”