From X-Rays to AI: Navigating US Regulations in Radiological Health
113
Introduction
Artificial intelligence (AI) has great potential to
address clinical challenges in healthcare.1 With
the recent advancement in data science and
information technology, software incorporating
AI, and its subdivision machine learning (ML)
technology, is becoming an integral part of an
increasing number of medical devices. AI/ML
technologies have revolutionized every aspect
of our lives and are poised to reshape clinical
practice by improving the experiences of physi-
cians and patients.2 The growth in AI research
is evident from the number of peer-reviewed
articles published between 1998 and 2018 which
has grown by over 300 percent.3 The exponential
increase in AI research has led to a corresponding
increase in the number of resulting patents. For
example, the number of AI patents filed in 2021
has increased by more than 30 times compared to
patents filed in previous years.4,5 The increasing
trend in AI research and resulting patents also
correlates well with the increasing rate of mar-
keting authorizations from the Food and Drug
Administration (FDA) in the US.6
AI-based software in the context of med-
ical devices includes standalone and embedded
software – software as a medical device (SaMD)
and software in a medical device (SiMD),
respectively. The use of algorithms containing AI/
ML in SaMD and SiMD has been proven to aid
clinical users in image acquisition and process-
ing, early disease detection and triage, accurate
diagnosis, prognosis, and risk assessment. The
first software incorporating AI/ML technolo-
gies was approved via a premarketing approval
process (P970058) for mammography by the
FDA in 1998. For the next 18 years, marketing
authorizations trickled through the Agency with
fewer than 30 SaMD reaching the US market.
However, the pace of clearances and approvals
has increased substantially since 2016.7
The FDA maintains a list of AI software
that they have granted US marketing authoriza-
tion on their website as of June 2023, over 500
AI/ML algorithms are cleared or approved for
clinical use.6 As shown in Figure 8-1, the FDA
has granted an increasing number of marketing
authorizations every year, with the majority of
applications being in radiology, followed by car-
diology. With an increasing number of validated
AI/ML models along with efforts from health
authorities to streamline regulatory pathways
to bring them to market, a growing number of
manufacturers are developing AI/ML based soft-
ware tools to support clinical decision-making in
medical imaging. Despite the promise of AI/ML
Artificial Intelligence-based Software
Gopal Abbineni, MS, PhD Hortense Allison, MS, MBA
8
113
Introduction
Artificial intelligence (AI) has great potential to
address clinical challenges in healthcare.1 With
the recent advancement in data science and
information technology, software incorporating
AI, and its subdivision machine learning (ML)
technology, is becoming an integral part of an
increasing number of medical devices. AI/ML
technologies have revolutionized every aspect
of our lives and are poised to reshape clinical
practice by improving the experiences of physi-
cians and patients.2 The growth in AI research
is evident from the number of peer-reviewed
articles published between 1998 and 2018 which
has grown by over 300 percent.3 The exponential
increase in AI research has led to a corresponding
increase in the number of resulting patents. For
example, the number of AI patents filed in 2021
has increased by more than 30 times compared to
patents filed in previous years.4,5 The increasing
trend in AI research and resulting patents also
correlates well with the increasing rate of mar-
keting authorizations from the Food and Drug
Administration (FDA) in the US.6
AI-based software in the context of med-
ical devices includes standalone and embedded
software – software as a medical device (SaMD)
and software in a medical device (SiMD),
respectively. The use of algorithms containing AI/
ML in SaMD and SiMD has been proven to aid
clinical users in image acquisition and process-
ing, early disease detection and triage, accurate
diagnosis, prognosis, and risk assessment. The
first software incorporating AI/ML technolo-
gies was approved via a premarketing approval
process (P970058) for mammography by the
FDA in 1998. For the next 18 years, marketing
authorizations trickled through the Agency with
fewer than 30 SaMD reaching the US market.
However, the pace of clearances and approvals
has increased substantially since 2016.7
The FDA maintains a list of AI software
that they have granted US marketing authoriza-
tion on their website as of June 2023, over 500
AI/ML algorithms are cleared or approved for
clinical use.6 As shown in Figure 8-1, the FDA
has granted an increasing number of marketing
authorizations every year, with the majority of
applications being in radiology, followed by car-
diology. With an increasing number of validated
AI/ML models along with efforts from health
authorities to streamline regulatory pathways
to bring them to market, a growing number of
manufacturers are developing AI/ML based soft-
ware tools to support clinical decision-making in
medical imaging. Despite the promise of AI/ML
Artificial Intelligence-based Software
Gopal Abbineni, MS, PhD Hortense Allison, MS, MBA
8