From the July 2020 issue of HealthCare Business News magazine
By Samir Parikh
The breast radiology field has taken many positive strides over the years to more accurately detect breast cancer. Arguably the most significant innovation in breast health in recent history is the introduction of digital breast tomosynthesis (DBT) more than 10 years ago, which is proven to detect more invasive breast cancers compared to traditional 2D mammography alone.
Most recently, the integration of artificial intelligence (AI) in breast cancer screening technology has generated serious attention, and for good reason when considering the impact it’s already having on clinicians and patients, and the even greater impact it could have in the future. To assess future potential, I like to look back on the past and review the goals of applying AI technology in breast imaging, which tend to fall into three main categories: improving cancer detection, workflow optimization and clinical decision support.
While I mentioned that DBT has been one of the most profound innovations in breast health — if not the most — it does create large amounts of data and files that can impact workflow.
Over the years since DBT’s introduction, there have been several iterations of technology that use algorithms to create synthesized 2D images from tomosynthesis slices to address this and help clinicians navigate to important areas of interest in the tomosynthesis stack. For example, recent technology generates 6mm tomosynthesis slices (referred to as SmartSlices), instead of the 1mm standard tomosynthesis slices, reducing the number of images that need to be interpreted. The technology’s algorithm identifies suspicious areas and makes them more conspicuous, allowing radiologists to more quickly and efficiently review each case.
When thinking about the future of AI, healthcare professionals must ask: what processes can still use more precision and be expedited?
First, the image review process has potential for further improvement. Despite advances in cancer detection as a result of DBT adoption, there’s still wide variability in performance among individual radiologists, and cancers may be missed even with the most current imaging technology. One emerging technology in the field of AI is the application of Deep Learning to identify potential abnormalities in tomosynthesis images and highlight these areas for radiologists. The performance of Deep Learning algorithms in identifying subtle cancers with a very low false positive rate is much better than previous technologies, so these tools have the potential to significantly improve radiologists’ ability to detect breast cancers. Furthermore, the more these algorithms evolve, the more potential they have to reduce radiologists’ workload by identifying patients whose mammograms are almost definitely benign — it’s possible one day these particular mammograms may only need a quick review, or no review at all, so clinicians can focus their time on cases with clear areas of concern.
Personalized screening through creation of new risk models based on patients’ electronic health records could also be game-changing. Every individual is unique, and as a result, customized patient plans are growing in importance. AI will most certainly play an important role in doing so efficiently.
As we look beyond screening, AI can — and will — continue to play a role along the continuum of breast health care. For example, AI may be able to help inform surgical and treatment pathways more effectively and efficiently.
The breast cancer screening patient journey is a closed-loop system of steps, from screening to treatment and back to imaging again. The important thing is to develop technology that keeps the full continuum of care in mind, from start to finish — that’s where I see AI having the biggest impact on breast health and women’s health overall. This holistic view will help realize the full potential of AI and bring even greater value to facilities and patients served.
About the author: Samir Parikh is the global vice president of research and development for Hologic.