Manually reviewing breast images for small
abnormalities in a deluge of hundreds of mammographic images can mean long days
for radiologists, which can lead to reader fatigue and increase the potential
for missed cancers. False positives from older 2D computer-aided detection (CAD)
technologies result in unnecessary biopsies, which are not only stressful for
patients, they have been shown to cost the healthcare system more than $2
billion per year.[i]
Artificial intelligence (AI) presents new
opportunities to improve radiologists’ performance and enhance patient care. Used
in diagnostic technologies, AI can increase radiologists’ diagnostic accuracy,
save time and improve patient satisfaction. However, not all AI vendors are created
equally. Before investing in AI technology to enhance your diagnostic
capabilities, the following 10 factors are important to consider.
1. Clinical Value
The AI product should offer significant
clinical value. It should:
- Improve radiologist sensitivity and
increase the numbers of cancers radiologists detect; - Improve specificity, making the reading
more accurate; and - Reduce recall rates, thereby reducing
the burden of unnecessary supplemental imaging or biopsies, and minimize reading
time for radiologists.
ProFound AI is a high-performing workflow
solution available for 2D and 3D mammography, or digital breast tomosynthesis
(DBT), featuring the latest in deep-learning AI. This leading-edge
technology was the first software of its kind to be FDA-cleared; it is also CE
marked and Health Canada licensed. It is clinically proven to offer an 8
percent increase in sensitivity, a 6.9 percent increase in specificity, a 7.2
percent reduction in the rate of false positives and unnecessary recalls, and a
52.7 percent reduction in reading time.[ii]
2. Algorithm and Data Quality
The AI vendor should offer a true deep-learning
algorithm and provide evidence about the quality and quantity of images used. The
credibility of the images and the credentials of the team members, such as
image annotators and chief investigators, who validated the images are critical.
A trusted vendor will explain the types of images obtained, the clinical expert
protocol design used, whether clinical biomarkers or biopsies were used to
validate the annotated images, and if different image sets were used for
training and testing the algorithm.
ProFound AI uses a true deep-learning algorithm.
iCAD employs biostatisticians and chief investigators from leading academic
medical centers and private clinics to help design clinical protocols detailing
optimal population size, heterogeneity of population and gantry manufacturers. We
follow strict FDA guidelines for
developing a medical product that requires separation of training and
regulatory-validation data sets, and we employ expert truthers (radiologists)
to provide ground truth. Our ground truthing for cancers is always done with
biopsy-proven pathology data by trained radiologists and clinical experts.
3. Workflow Integration
When choosing an AI solution, it is critical
to determine whether the AI results can integrate easily and seamlessly into
your preferred diagnostic viewer. The vendor should explain how quickly its
technology can deliver results, whether this data can integrate into your PACS
and primary viewing interfaces, how many PACS vendors its AI product integrates
with, and whether there is a dedicated expert resource to manage PACS upgrades.
The vendor should explain if its AI product can export results in major
reporting structures, and what happens if you switch to another gantry vendor. End-to-end
complete integration into a voice-dictation, PACS or RIS-driven workflow should
always be an option, allowing radiologists to always keep an eye on images and
not worry about using an additional interface. True deep-learning AI should not
be introducing unnecessary technology complexity into readers’ already rich
workflow. It should be a true concurrent reader assisting radiologists in the
background.
Built on powerful GPU technology and algorithm
optimization techniques, ProFound AI is compatible with multiple vendors, PACS,
and workstation vendors, and delivers results rapidly. Our expert technical
sales and integration consultants provide on-site service and support
customers’ changing needs.
4. Data Collection and Clinical
Validation
In deep-learning AI, data are gold. The quality
of data with which the technology is trained determines the quality of its clinical
outcomes. A reliable AI vendor will have an internal dedicated team of experts
for data collection, curation and annotation, and software updates should be
seamless. It is important for the vendor to explain the technology’s process
for collecting, curating and annotating data, as well as its indications for use (independent reader or
concurrent reader), and how to swiftly re-train its algorithms and introduce
updated software versions. Ask the vendor if the technology is supported by clinical
research (including whether it is a prospective, retrospective or true reader
study), the number of readers that participated in the study, and whether the
study was single-center or multi-center. Furthermore, the most important metric
is the quality of the outcomes and claims, not merely the number of images on
which the algorithm is trained.
With ProFound AI, our customers have access to
internal experts dedicated to data collection agreements, the secure transfer
of data, anonymizing, ingesting data and sending it for multiple-expert
annotation. It is supported by a multi-center, global (US and Europe), retrospective
reader study––with HIPAA-compliant IRB protocol for data collection––that evaluated
the concurrent use of AI to shorten DBT reading time, while maintaining or
improving radiologist sensitivity and specificity. Our reader study was
conducted with 24 certified MQSA radiologists (13 expert breast imagers and 11
general radiologists, with 34 years or less experience in clinical practice).[ii]
5. Product Regulation
The AI vendor should explain which US and
non-US regulatory approvals it has, if any. For FDA approval, find out the type
of clearance the vendor received.
In December 2018, ProFound AI for DBT became
the first artificial intelligence software for DBT to be cleared by the U.S.
Food and Drug Administration; it was also CE marked and Health Canada licensed
that same year. ProFound AI for 2D Mammography was CE marked in July 2019.
6. IT Development and Support
The AI solution must be cost-effective, secure
and work quickly. Ask the vendor to explain its IT consultancy and whether its
AI maximizes hardware resources for optimal performance and cost. A vendor
should also explain the extent and duration of its global support as well as
who will provide technical support. Having support within the same time zone is
an important consideration for critical issues involving product and field
integrations.
iCAD’s expert team of integration specialists
helps to optimize IT workflow and provides a tailored solution using a variety
of modern-architecture deployments, which minimizes the hardware footprint and
reduces demand on internet bandwidth.
7. Operational Efficiency
The AI vendor should explain how to prioritize
high-complexity patients to provide immediate supplemental screening or biopsy for
some women, and how to distribute workload across multiple readers to ensure a uniquely
tailored workload for each radiologist (i.e. specialist versus general
radiologist).
ProFound AI provides Certainty of Finding
lesion and Case Scores, which helps radiologists assess caseloads and assists
with clinical decision-making. By using ProFound AI’s Case Scores, you can easily
distribute workload from your workflow list in PACS or your reading station.
Vendors such as GE and ThreePalm have implemented this successfully today for the
assessment of cases, regardless of complexity.
8. Purchasing and Contracting
An AI vendor should outline what purchasing
models it offers––perceptual Cap-EX or operational Op-Ex licenses. The vendor
should work with you to find the best solution for your financial needs
especially in times when funds are tight, and prioritization of financial
resources is needed. iCAD offers a flexible suite of financing models that can
help facilities afford our solution and help to ensure a sound return on their
investment.
9. ROI, Value Proposition and Product
Vision
When choosing an AI vendor for 2D and 3D
mammography, knowing the product’s financial impact is crucial. Ask the vendor
to explain how your patients and your facility will benefit financially, and
what its near-mid-term product vision is. Select a vendor whose vision aligns
with your longer-term growth goals. The AI solution should increase the
accuracy of diagnosis, lower time of reads and total cost of patient spend over
time. Vendors should satisfy a quadruple aim: demonstrate improved clinical
experience, better outcomes, improved patient experience and lower costs. Furthermore,
the vendor should always have a comprehensive near- and long-term product
roadmap that is aligned with customer’s goals.
ProFound AI is clinically proven to cut reading
time by more than half,[ii]
which frees up work time to spend with patients, or perform an additional
biopsy or additional reads.
10. Capital Funding and Leadership
Being able to raise capital through multiple
funding rounds reflects a vendor’s potential long-term success. The AI vendor
should explain how much capital funding it has raised; the number of funding
rounds, how long the typical funding round lasts and when the most recent round
was completed; and disclose who its investors are. The vendor leadership team’s
credibility and experience and its medical advisory board are integral to the
vendor’s success. Ask the vendor about the leadership team’s experience in taking
healthcare technology products to market, and whether its executive team has
prior similar experience working for successful, revenue-generating companies.
iCAD’s senior
executive team offers a legacy of leadership and expertise in women’s
imaging and the AI market. Our medical team comprises renowned global experts
from leading academic medical centers and hospitals, including the University
of Pennsylvania, NYU Langone Health, Elizabeth Wende Breast Care and Boca Raton
Regional Hospital. We aim to serve millions of patients and physicians with clinically
proven AI tools that save lives and enhance patient care.
[i] Vlahiotis
A, Griffin B, Stavros AT, Margolis J. Analysis of utilization patterns and
associated costs of the breast imaging and diagnostic procedures after
screening mammography. Clinicoecon
Outcomes Res. 2018;10:157-167
https://doi.org/10.2147/CEOR.S150260
[ii] Conant, E. et al. (2019).
Improving Accuracy and Efficiency with Concurrent Use of Artificial
Intelligence for Digital Breast Tomosynthesis. Radiology: Artificial
Intelligence. 1 (4). Accessed via
https://pubs.rsna.org/doi/10.1148/ryai.2019180096