By Samir Parikh
With October designated as Breast Cancer Awareness Month, now is an especially pertinent time to reflect on the most talked about trend in breast cancer screening – Artificial Intelligence (AI) – and where it’s headed. When you consider the primary functionality of AI in breast imaging, and its initial impact on enhancing workflow and improving accuracy in recent years, the topic is even more relevant in light of the backlog of cases facing many clinicians as screening facilities open their doors for mammograms after the COVID-19 lockdown.
Image reading, in particular, has benefited greatly from innovations in AI technology. Over the last decade, the introduction of Digital Breast Tomosynthesis (DBT) provided radiologists with more image slices and data than ever before, allowing clinicians to have a more holistic view of the breast and detect more invasive cancer – a huge boon since early detection saves lives.
Although this wealth of information has been helpful in identifying cancer, the associated large data file sizes and abundance of images can hinder the efficiency of the image reading process. This issue spurred the development of AI technology that can identify clinically relevant regions of interest and preserve important features during the creation of 6 mm slices from the original high-resolution 3D data. The process reduces the number of images to review and thus the amount of time necessary for review, without compromising image quality, sensitivity or accuracy.
This example is only one recent advancement of many that demonstrates how AI is positively influencing mammography and where it is likely to go in the future. With AI, time is on our side – the longer AI applications are in play, the more cases consumed; and therefore, more insights are revealed. There are still many other aspects of breast imaging that could use refinement in efficiency in the long-term, paving the way for AI technology to continue to grow and make a sizeable impact on the delivery of patient care.
One aspect is the development of new risk models for patients based on their electronic health records in order to have an efficiently compiled and more personalized screening pathway plan before they head in for screening. Currently, women who are at high risk of breast cancer can qualify for an MRI scan, and women with dense breasts (who have a slightly elevated risk of breast cancer) can qualify in some states for an additional test, such as ultrasound. The current risk models use simple metrics such as a woman’s breast density, but machine learning offers the opportunity for improved risk prediction, by, for example, finding patterns in mammograms that are predictive of breast cancer but are not identified by radiologists today.
In fact, beyond imaging, AI may one day similarly have the ability to efficiently determine the best course of treatment after screening and diagnosis.
Additionally, to further streamline the screening process and as AI technology quite literally grows smarter, there exists potential for certain mammograms to be marked almost definitely benign, giving clinicians the confidence to either skip the case entirely or review quickly. This would allow clinicians to move on to the more complex, higher risk cases in their workload.
In conclusion, the need for efficiency and quality care in breast imaging endures, solidifying a critical role for AI in mammography that will only continue to develop. By examining the current inefficiencies with breast screening and health care, trend-followers in the breast imaging technology industry can best anticipate the new needs that AI has the exciting potential to meet. •
Samir Parikh is the global vice president of research and development for Hologic Inc.
