Chapter 8. Artificial Intelligence-based Software
124 Regulatory Affairs Professionals Society (RAPS)
21 CFR §870: Cardiovascular Devices
After general radiology, the second highest num-
ber of AI/ML SaMD applications that the FDA
has granted marketing authorization for are
used in cardiology.53 A number of cardiological
applications use AI and have significant overlap
with medical imaging.54 Table 8-3 provides a
few examples of FDA-cleared AI/ML clinical
applications for cardiology use that are related to
medical imaging the majority are programma-
ble diagnostic devices that can compute various
physiologic, or blood flow parameters based on
medical images.55-57
21 CFR §892.1750: CT X-ray system – Image
Reconstruction Algorithms
In addition to standalone SaMD incorporating
AI/ML, these technologies are also frequently
integrated within the medical device itself (i.e.,
SiMD). Examples of the latter exist throughout
the radiological health space and one example
is the use of AI/ML for medical image recon-
struction. Deep learning-based reconstruction
algorithms have recently begun to be used in CT,
MR, and nuclear medicine devices examples
include GE Healthcare’s Deep Learning Image
Reconstruction (DLIR) (K183202)58 for CT and
Precision DL for PET/CT, Canon Medical’s
Precise IQ Engine (PIQE) for CT (K182901),59
and Siemens Deep Resolve for MR (K220939) 60
see Chapters 3, 4, and 6 for more on these sys-
tems). When used in image reconstruction, deep
learning-based algorithms have the potential to
maintain quantitative integrity while providing
denoising capabilities.61
Image reconstruction software is part of
embedded hardware and therefore falls within
the scope of the existing regulations for imaging
hardware for example, CT and MR reconstruc-
tion are regulated under the 21 CFR §892.1750
and 21 CFR §892.1000 regulations respectively.
Image reconstruction is a mathematical pro-
cess that generates images from noisy sample
measurements. For example: Philips Iterative
Reconstruction Technique Software Application
(K113483)62 is cleared to reconstruct raw data
from CT scanners to produce images containing
less or equal noise when compared to images
produced by standard filtered back projection
reconstruction. DLIR (K183202)58 was the first
FDA-cleared technology to use deep neural
network-based CT reconstruction, followed 2
months later by Canon Medical’s Advanced
intelligent Clear-IQ Engine (AiCE)
(K183046).63 According to the 510(k) summary
for K183202, the latter’s FDA clearance was
based on a combination of bench data and a
retrospective study of 60 cases, with images eval-
uated by nine radiologists using a Likert-scale
study. Efficient validation methods for AI/ML
based reconstruction algorithms remain a subject
of ongoing research.61
21 CFR §886.1100: Retinal Diagnostic
Software Device
While ophthalmology is not directly con-
nected to radiological health, developments in
Table 8-3. Examples of FDA-Cleared AI/ML Devices for Cardiology
Device Name Regulation Clearance
Number
Company Short Description
DeepVessel
FFR55
21 CFR §870.1415
(Product code: PJA)
K213657 KeyaMed NA
Inc.
Coronary physiological simulation software
intended for evaluation and assessment of
coronary arteries
HeartFlow
Analysis56
21 CFR §870.1415
(Product code: PJA)
K203329 HeartFlow,
Inc.
Semi-automated tool for extraction of ana-
tomic data for coronary physiologic simulation
to aid in diagnosis of coronary artery disease
Feops
HEARTguide57
21 CFR §870.1405
(Product code: QQI)
K214066 Feops NV Software that performs computer simulation
to predict implant frame deformation to sup-
port the evaluation for left atrial appendage
occlusion device size and placement
124 Regulatory Affairs Professionals Society (RAPS)
21 CFR §870: Cardiovascular Devices
After general radiology, the second highest num-
ber of AI/ML SaMD applications that the FDA
has granted marketing authorization for are
used in cardiology.53 A number of cardiological
applications use AI and have significant overlap
with medical imaging.54 Table 8-3 provides a
few examples of FDA-cleared AI/ML clinical
applications for cardiology use that are related to
medical imaging the majority are programma-
ble diagnostic devices that can compute various
physiologic, or blood flow parameters based on
medical images.55-57
21 CFR §892.1750: CT X-ray system – Image
Reconstruction Algorithms
In addition to standalone SaMD incorporating
AI/ML, these technologies are also frequently
integrated within the medical device itself (i.e.,
SiMD). Examples of the latter exist throughout
the radiological health space and one example
is the use of AI/ML for medical image recon-
struction. Deep learning-based reconstruction
algorithms have recently begun to be used in CT,
MR, and nuclear medicine devices examples
include GE Healthcare’s Deep Learning Image
Reconstruction (DLIR) (K183202)58 for CT and
Precision DL for PET/CT, Canon Medical’s
Precise IQ Engine (PIQE) for CT (K182901),59
and Siemens Deep Resolve for MR (K220939) 60
see Chapters 3, 4, and 6 for more on these sys-
tems). When used in image reconstruction, deep
learning-based algorithms have the potential to
maintain quantitative integrity while providing
denoising capabilities.61
Image reconstruction software is part of
embedded hardware and therefore falls within
the scope of the existing regulations for imaging
hardware for example, CT and MR reconstruc-
tion are regulated under the 21 CFR §892.1750
and 21 CFR §892.1000 regulations respectively.
Image reconstruction is a mathematical pro-
cess that generates images from noisy sample
measurements. For example: Philips Iterative
Reconstruction Technique Software Application
(K113483)62 is cleared to reconstruct raw data
from CT scanners to produce images containing
less or equal noise when compared to images
produced by standard filtered back projection
reconstruction. DLIR (K183202)58 was the first
FDA-cleared technology to use deep neural
network-based CT reconstruction, followed 2
months later by Canon Medical’s Advanced
intelligent Clear-IQ Engine (AiCE)
(K183046).63 According to the 510(k) summary
for K183202, the latter’s FDA clearance was
based on a combination of bench data and a
retrospective study of 60 cases, with images eval-
uated by nine radiologists using a Likert-scale
study. Efficient validation methods for AI/ML
based reconstruction algorithms remain a subject
of ongoing research.61
21 CFR §886.1100: Retinal Diagnostic
Software Device
While ophthalmology is not directly con-
nected to radiological health, developments in
Table 8-3. Examples of FDA-Cleared AI/ML Devices for Cardiology
Device Name Regulation Clearance
Number
Company Short Description
DeepVessel
FFR55
21 CFR §870.1415
(Product code: PJA)
K213657 KeyaMed NA
Inc.
Coronary physiological simulation software
intended for evaluation and assessment of
coronary arteries
HeartFlow
Analysis56
21 CFR §870.1415
(Product code: PJA)
K203329 HeartFlow,
Inc.
Semi-automated tool for extraction of ana-
tomic data for coronary physiologic simulation
to aid in diagnosis of coronary artery disease
Feops
HEARTguide57
21 CFR §870.1405
(Product code: QQI)
K214066 Feops NV Software that performs computer simulation
to predict implant frame deformation to sup-
port the evaluation for left atrial appendage
occlusion device size and placement