By Samir Parikh
The greatest medical innovations are typically designed to help clinicians do their jobs better in some way, whether they improve efficiency, accuracy, comfort, or another aspect of the job. More importantly, these enhancements can, in turn, produce various benefits for patients. This concept holds true in the radiology field, especially in regard to the use of AI—a technology that continues to gain attention in the radiology community, particularly with the emergence of deep learning.
AI offers incredible potential to transform clinical and administrative workflow tasks. More specifically, AI plays a powerful role in mammography screening for breast cancer detection in a number of ways, starting with risk models and breast density assessments that are completed prior to cancer screening.
Risk Models and AI
Each patient profile is unique, from family history to environmental surroundings, lifestyle habits, and beyond, creating a different level of risk for any given condition, including breast cancer. Through risk stratification—classifying patients by level of risk—clinicians can guide patients on the screening regimen that is optimal for their individual situation, whether that is starting them at an earlier age or encouraging them to be screened with a certain imaging modality.
Although an important process, developing risk models requires a great deal of data about each patient, sourced from various medical encounters they’ve had with different physicians, creating an opportunity for AI integration to make improvements. AI technology uses algorithms to create patterns of significance from large volumes of data. As such, it may one day be possible for AI to gather large amounts of patients’ data from their EHRs, analyze the patterns, and produce new risk models.
Additionally, conventional risk models use more straightforward factors, such as a woman’s breast density, to inform risk, while machine learning can potentially provide enhanced risk prediction by, for example, finding patterns in mammograms that are predictive of breast cancer but not recognized by radiologists today. AI-based risk models can potentially be game-changing for clinicians when guiding patients about which screening options make the most sense for their needs.
Breast Density Assessments and AI
Traditional breast density assessments are completed visually by radiologists. The assessments are subjective; two radiologists may interpret the same images differently and have different views about which breast density indicator to assign. These disparities can have implications when making clinical decisions. As radiologists know, breast density assessments are an important part of mammography screening because a woman’s breast density can potentially influence her level of risk for getting breast cancer as well as which imaging modalities are optimal to use for screening.
From a robust database of breast images spanning a variety of different patients and breast tissues, AI-powered technology exists that uses pattern and texture analysis to automatically classify breast density according to the BI-RADS score system. As a result, clinicians can have more consistent and reliable scoring to support their clinical decisions for each patient’s pathway of care.
The benefits of AI technology here are two-fold: clinicians are able to streamline their manual workflow by transitioning from visual assessments, though there may be some unique cases where their nuanced expertise must supplement the technology’s evaluation, and patients are able to receive an accurate breast density assessment, leading to informed screening plans that can help them to detect cancer, hopefully early on.
Image Reading and AI
Like many physicians, one of the greatest challenges that radiologists face is managing their workloads. For example, digital breast tomosynthesis (DBT) technology, which has been clinically proven to detect more invasive cancers than 2D mammography, also produces many more breast images and larger data files, resulting in additional work for radiologists.
Fortunately, AI can help streamline the data. One existing technology uses AI-powered analytics to generate 6 mm image slices from the original high-resolution 1 mm 3D images. The AI-powered software identifies clinically relevant regions of interest and preserves important features during the creation of the 6 mm slices, and each of those slices overlaps the previous slice by 3 mm, ensuring no loss of 3D data. As a result, clinicians can expedite reading time by reducing the number of images to review, without sacrificing image quality, sensitivity, or accuracy.
Additionally, in high-volume screening centers, radiologists have many patient cases to review. Many are benign and require no further action, but others are more complex, requiring additional follow-up. As AI technology continues to evolve, there is an opportunity to apply deep learning to DBT images to flag subtle suspicious areas and ensure they are not overlooked, with fewer false-positives than previous technologies. As the accuracy of these tools continues to improve, it will become possible for clinicians to spend less time on cases that are determined to be benign or, perhaps, bypass them altogether.
It’s important to note that AI technology does not replace radiologists but, rather, serves to augment their skills, helping to identify suspicious areas that warrant additional, in-depth review. AI has the potential to help radiologists identify more cancers at an earlier stage, ultimately providing enormous value to patients and health care providers.
In summary, there are many aspects of mammography screening where AI technology can potentially be used or is currently being used to support radiologists’ professional needs—which ultimately meet patients’ clinical needs. By examining these mutually beneficial opportunities, health care professionals should feel empowered to explore how the integration of AI technology can make a positive impact on their daily lives and embrace the many innovations that are yet to come.
— Samir Parikh is the global vice president of research and development for Hologic, Inc.