Pioneering the application of deep learning in breast imaging
Imaging Artificial Intelligence (AI), or deep learning, has become a much buzzed-about term, despite the fact that it has been utilised by many of us in the healthcare industry for quite some time.
In breast screening in particular, Hologic has been the pioneer in applying machine learning in breast imaging and developed the first-ever commercial computer-aided cancer detection software product, which was based on conventional machine learning algorithms. More recently, AI has been an extremely influential tool in Hologic’s breast screening product suite because it provides a multitude of benefits for radiologists, from impacting the accuracy of cancer detection, to helping with workflow.
The potential of AI and deep learning
This exciting era of AI research means we are experiencing deep learning technology having a wide-spread and powerful impact on many applications, both inside and outside of the medical imaging space. AI, with deep learning, is rapidly approaching human-level performance in many cases previously considered to be very difficult challenges.
"The AI algorithm learns from the radiologist; while the radiologist, with the help of AI algorithms, can find more cancers."
In breast screening in particular, we see that deep learning algorithms can learn from massive amounts of data and generate impressive results in both cancer detection and diagnosis. At Hologic, we believe any viable future breast screening products and devices will have AI and deep learning as a fully integrated and built-in key component – and we are committed to continuing to lead from the front with the introduction of such products.
For example, AI played a large role in the development of Hologic’s 3D Mammography™ exam, which has been clinically proven to detect breast cancer earlier than traditional mammography. In fact, the 3D Mammography™ exam detects up to 65 per cent more invasive breast cancers compared to 2D alone, with an average increase of 41 per cent.1 In this way, and many others, the application of AI has become more and more a partner in helping radiologists. It is a mutually interactive environment in which the AI algorithm learns from the radiologist; while the radiologist, with the help of AI algorithms, can find more cancers, find them earlier, and most likely become more efficient – without becoming overwhelmed with the large amount of 3D screening data to review.
Delivering unbiased breast density assessment
We’ve seen this first hand with Intelligent 2D™ imaging technology, a feature available on Hologic’s 3Dimensions™ mammography system, introduced in Europe in July 2017. Intelligent 2D™2 is a synthetic image product that has a built-in machine learning algorithm known as smart mapping, enabling radiologists to instantly move from suspicious areas detected on the 2D image to the point of interest on the 3D slice, saving time and optimising workflow.3,4 Yet another example of AI in action is the Quantra™ 2.2 breast density assessment software, developed by matching thousands of mammographic images with corresponding radiologist assigned BI-RADS density categories, training and achieving a fully automated breast density categorisation. With this product, radiologists can deliver unbiased breast density assessment that removes the potential for visual subjectivity; a big feat for clinicians and women with dense breasts.
Breast cancer screening on a whole has – and will continue to have – tremendous benefits from the application of deep learning technology. The participating populations will see improved detection accuracy with reduced false positive findings. The clinicians will see increased productivity and better clinical performance. And, while AI will never take the place of radiologists and the great work that they do, with ongoing innovation, AI has a promising future to positively impact these areas of breast imaging to help radiologists provide the best possible care to patients.
1 Friedewald SM, Rafferty EA, Rose SL, et al. Breast cancer screening using tomosynthesis in combination with digital mammography. JAMA. 2014 Jun 25;311(24):2499-507.
2 CE marked only at this time, pending FDA approval
3 Compared with combo mode ~10 second scan time per view.
4 Feature is used in combination with SecurViewâDX diagnostic review workstation mapping tool in v9.0.1 and above.
Hologic Deutschland GmbH, Otto-von-Guericke-Ring 15, 65205 Wiesbaden-Nordenstadt, Germany
Geschäftsführer:Robert McMahon, Michelangelo Stefani, Amtsgericht Wiesbaden HRB 26790, UST-ID: 813395831