From X-Rays to AI: Navigating US Regulations in Radiological Health
125
ophthalmic AI devices are likely to have poten-
tial future implications for the regulation of
radiological health. This is due to the first FDA
marketing authorizations of autonomous AI/ML
devices being granted within the specialty.
Ophthalmology is an image-based and data-
rich specialty with a trend toward the increasing
use of AI/ML algorithms.64 To date only a few
SaMD AI/ML devices are FDA-cleared for
use in ophthalmology, for diagnostic screening
purposes. These can help identify retinal diseases
or conditions early enough (e.g., mild diabetic
retinopathy) to enable treatment to prevent
vision loss.65
The first SaMD, IDx-DR (Idx, LLC,
Coralville IA) received FDA marketing
authorization via the De Novo pathway
(DEN180001).28 Special controls established by
21 CFR §886.1100 included a requirement for
clinical performance data evaluating sensitivity,
specificity, negative predictive value, and positive
predictive value, in addition to software verifi-
cation and validation tests and usability testing
information.66
Key Design Considerations for
Successful Development of AI/ML
SaMDs
When developing AI/ML medical devices, devel-
opers should always consider the Total Product
Life Cycle (TPLC) aspects of the product (i.e.,
from initial design to post-marketing moni-
toring). The first aspect is understanding the
steps involved in the development of an AI/ML
product (Figure 8-5). In 2021, the FDA, Health
Canada, and the UK’s Medicines and Healthcare
products Regulatory Agency (MHRA) jointly
identified 10 guiding principles that could inform
the development of good machine learning prac-
tice.67The key principles are a useful resource for
developers and are outlined in Table 8-4.68
The development of SaMD requires enlist-
ing multiple internal and external stakeholders to
ensure robust development. Developers should
have a systematic process for engaging key stake-
holders, including physicians and other end users,
to participate in device design and continuously
test and validate assumptions and ideas during
device development. Table 8-5 contains some
key considerations for successful development of
a SaMD.
Figure 8-5. Schematic Representation of Various Development Stages in SaMD
Market Research
• Identify user
needs
• Specify target
users
• Understand
clinical need/
value
• Develop claims
matrix
Discovery
• Identify poten-
tial AI/ML
algorithm
• Define product
scope (indi-
cations) and
performance
criteria
• Develop claims
matrix
• Identify data
sources to
meet the
scope.
Development
• Finalize the
algorithm
• Integrate the
algorithm into
the clinical
workflow
• Verify and lock
the algorithm
Verification &
Validation
• Complete
standalone
bench perfor-
mance testing
• Clinical testing
(as needed)
• User interface
testing
• Submit for
regulatory
clearance/
approval
Deployment
• Move to pro-
duction phase
to serve user
requests
• User site
testing
Maintenance
• Postmarketing
surveillance to
monitor for any
adverse events
or performance
drifts
AI, Artificial intelligence ML, machine learning SaMD, software as a medical device
Created by Gopal Abbineni and Hortense Allison
125
ophthalmic AI devices are likely to have poten-
tial future implications for the regulation of
radiological health. This is due to the first FDA
marketing authorizations of autonomous AI/ML
devices being granted within the specialty.
Ophthalmology is an image-based and data-
rich specialty with a trend toward the increasing
use of AI/ML algorithms.64 To date only a few
SaMD AI/ML devices are FDA-cleared for
use in ophthalmology, for diagnostic screening
purposes. These can help identify retinal diseases
or conditions early enough (e.g., mild diabetic
retinopathy) to enable treatment to prevent
vision loss.65
The first SaMD, IDx-DR (Idx, LLC,
Coralville IA) received FDA marketing
authorization via the De Novo pathway
(DEN180001).28 Special controls established by
21 CFR §886.1100 included a requirement for
clinical performance data evaluating sensitivity,
specificity, negative predictive value, and positive
predictive value, in addition to software verifi-
cation and validation tests and usability testing
information.66
Key Design Considerations for
Successful Development of AI/ML
SaMDs
When developing AI/ML medical devices, devel-
opers should always consider the Total Product
Life Cycle (TPLC) aspects of the product (i.e.,
from initial design to post-marketing moni-
toring). The first aspect is understanding the
steps involved in the development of an AI/ML
product (Figure 8-5). In 2021, the FDA, Health
Canada, and the UK’s Medicines and Healthcare
products Regulatory Agency (MHRA) jointly
identified 10 guiding principles that could inform
the development of good machine learning prac-
tice.67The key principles are a useful resource for
developers and are outlined in Table 8-4.68
The development of SaMD requires enlist-
ing multiple internal and external stakeholders to
ensure robust development. Developers should
have a systematic process for engaging key stake-
holders, including physicians and other end users,
to participate in device design and continuously
test and validate assumptions and ideas during
device development. Table 8-5 contains some
key considerations for successful development of
a SaMD.
Figure 8-5. Schematic Representation of Various Development Stages in SaMD
Market Research
• Identify user
needs
• Specify target
users
• Understand
clinical need/
value
• Develop claims
matrix
Discovery
• Identify poten-
tial AI/ML
algorithm
• Define product
scope (indi-
cations) and
performance
criteria
• Develop claims
matrix
• Identify data
sources to
meet the
scope.
Development
• Finalize the
algorithm
• Integrate the
algorithm into
the clinical
workflow
• Verify and lock
the algorithm
Verification &
Validation
• Complete
standalone
bench perfor-
mance testing
• Clinical testing
(as needed)
• User interface
testing
• Submit for
regulatory
clearance/
approval
Deployment
• Move to pro-
duction phase
to serve user
requests
• User site
testing
Maintenance
• Postmarketing
surveillance to
monitor for any
adverse events
or performance
drifts
AI, Artificial intelligence ML, machine learning SaMD, software as a medical device
Created by Gopal Abbineni and Hortense Allison