At the recent AuntMinnie.com Spring 2022 Virtual Conference Artificial intelligence (AI), Ralph Highnam, PhD discussed the Volpara philosophy for successful AI implementation for breast imaging. Dr. Highnam focused on significant opportunities for AI to improve quality and workflow in breast imaging and how Volpara’s approach has led to its AI software participating in 35% of all US breast cancer screenings:
As more breast centers evaluate the ability for AI to work in clinical practice, there are a number of standard questions they ask us every day. We call it the Breast Imaging AI Checklist:
- How does it improve patient care?
- What is its regulatory status?
- What’s the return on investment?
- Does the vendor have information security certification?
- How well can it integrate into my workflow?
While the questions in this checklist are important, they are the same type of questions clinical users should be asking about any product they are evaluating, whether it’s a scanner, a reporting package or an AI tool. Beyond the hype that currently surrounds AI, to build trust an AI tool can truly improve breast imaging, they must be appropriately designed and validated through rigorous processes to ensure performance and accuracy.
Volpara Health has taken steps to establish best practices for the creation of our AI, abiding by the following guiding principles:
- Is it explainable?
- Is it generalizable? Does it work across populations and machines?
- Is it validated internally and externally?
Evolving Role of AI in Breast Imaging
Artificial intelligence (AI) is an all-encompassing term, broadly covering how to program a machine to simulate human intelligence. Computer vision is a subset of AI focused on trying to replicate human vision. The history of AI in breast imaging dates back to the 1950 and 60s with computer vision. In fact, the first paper on computer aided detection was published in 1967 paper, looking at the feasibility of automated reading.
Key learning: Humans are very good at identifying signs related to breast cancer. As a result, does it make sense to dedicate resources to replicate something that radiologists can already do well?
In 1976, John Wolfe published a paper about the relationship between the tissue patterns shown in a mammogram and the development of breast cancer. This pioneering work set the stage for the next 40-50 years of research into breast density and AI. This includes the development of Cumulus, a semi-automated density software based on visual area-based density assessment, by Professors Norman Boyd and Martin Yaffe.
Key learning: Humans are not so good at quantification and you can transform patient care by making radiology a more quantitative science.
Building Trust in AI that Works for Clinical Practice
Volpara’s approach to AI for breast imaging is a blend of computer vision and deep learning, which leads to robust and reliable AI, which is generalizable and can be strongly validated. We focus on areas where AI can be used to improve patient care and increase productivity, especially where quantification is needed. This includes:
- Breast density assessment
- Image quality analytics
- Positioning metrics
- Breast compression metrics
- Radiation doses
To build trust AI products must be appropriately designed and validated through rigorous processes to ensure performance and accuracy. Currently, there are no widely accepted standards for design or validation of AI today, which is why many tools come to market having been developed and tested on only a few hundred cases.
These tools often have limited clinical utility because they have not accounted for the tremendous variability in mammography systems, breast composition across patient populations and image processing algorithms.
To ensure that our algorithms are robust and generalizable, we utilize more than 70 datasets with some 50 million images and cases from facilities around the world. We perform more than 35 tests to ensure that it works with every FDA-approved mammography and tomosynthesis system and that it can resolve for implants, pacemakers, image quality issues and other real-world noise that are not generally included in clean training datasets. We also perform tests to compare how it performs on digital 2D mammography versus 3D tomosynthesis, how it scores the same woman when the exam is taken on different mammography systems, and measure how it performs compared to radiologists.
The accuracy of Volpara TruDensity® has been validated by many independent studies, showing high correlation to breast MRI, which is considered the source of ground truth in breast density. This includes evaluating the breast density of women screened in the 10-year DENSE Trial, which demonstrated that offering MRI to women with very dense glandular breast tissue and normal results on screening mammography reduced the number of interval cancers by 50%.1 As a result of the DENSE Trial, the European Society of Breast Imaging (EUSOBI) recommended that women with extremely dense breasts receive breast MRI.
AI to Support, Not Replace, Radiologists
At Volpara, we design and develop AI to support, not replace, the clinical judgment of radiologists and to bring consistency to the very subjective area of breast imaging. Our AI tools have been featured in over 200 peer-reviewed studies as well as 200 scientific works, making it the most independently validated AI tool of its kind.
Ultimately, wider adoption of AI will depend on robust evidence of improved quality of patient care, increased efficiency, and cost-effectiveness, but explainability and external validation are key factors in building trust in AI for breast imaging.
For More Information
- For more information on how to evaluate AI for your practice, click here.
- To learn more about the history of AI in breast imaging and how AI can work successfully in clinical practice, watch the full webinar at the link below:
https://www.youtube.com/watch?v=8quN3gLPMbY&feature=emb_imp_woyt
References:
1. Klinkenbijl JHG, van Leeuwen E, Verkooijen HM. Wat zeggen de uitkomsten van de DENSE-studie? [What do the results of the DENSE-trial tell?]. Ned Tijdschr Geneeskd. 2020 Apr 21;164:D4822. Dutch. PMID: 32395945.
