From X-Rays to AI: Navigating US Regulations in Radiological Health
121
understanding the role of human readers (i.e.,
the physicians interpreting the images) and both
inter- and intra-reader variability are vital during
the reader study design. As reader variability
poses challenges to the evaluation and compari-
son of imaging devices, health authorities often
expect to see fully crossed multi-reader, multi-
case (MRMC) reader studies.12
It is recommended that developers and
manufacturers refer to the FDA guidance doc-
ument Computer-Assisted Detection Devices
Applied to Radiology Images and Radiology
Device Data-Premarket Notification 510(k)
Submissions, for more information.12 Note that
the majority of FDA-authorized CAD devices
are intended to be decision support tools rather
than primary diagnostic tools.46
CADx (21 CFR §892.2060): Radiological
Computer-Assisted Diagnostic Software for
Lesions Suspicious of Cancer
CADx systems are intended to aid in the char-
acterization of lesions identified on acquired
medical images, such as MR, mammography,
radiography, or CT images. These software
devices perform functions that may include auto-
matically registering images, segmenting images,
or analyzing user-selected ROI. In July 2017,
the FDA granted a De Novo reclassification
request for QuantX27 as the first CADx intended
to be used as a second reader for breast MRI
(see Chapter 4).23 The device presents an index
indicating a QI score for the lesion site the score
is calculated via an AI algorithm from radiomic
features. CADx software devices such as QuantX
aid in providing diagnostic and patient manage-
ment decisions to clinical users. Other examples
of CADx include Brainomix 360 e-ASPECTS
(K221564),47 a software device intended to assist
the physician in the assessment and character-
ization of brain tissue abnormalities using CT
image data.
The performance study requirements for
CADx are comparable to CADe requirements.
The manufacturer should perform both stand-
alone performance testing and a reader study.
Figure 8-4. Overview of Computer-Assisted Diagnosis Applications
Clinical Decision Support (AI-assisted)
Computer-assisted diagnosis (CADx)
Risk
prediction
Computer-
assisted
detection
(CADe)
Disease
detection
Computer-
assisted
triage (CADt)
Disease
Character-
ization
Rule out
Staging/
planning
Treatment
response
assessment
Prognosis
prediction
Recurrence
predication &
monitoring
AI, Artificial intelligence
Source: Adapted from Hadjiiski L, et al.41
121
understanding the role of human readers (i.e.,
the physicians interpreting the images) and both
inter- and intra-reader variability are vital during
the reader study design. As reader variability
poses challenges to the evaluation and compari-
son of imaging devices, health authorities often
expect to see fully crossed multi-reader, multi-
case (MRMC) reader studies.12
It is recommended that developers and
manufacturers refer to the FDA guidance doc-
ument Computer-Assisted Detection Devices
Applied to Radiology Images and Radiology
Device Data-Premarket Notification 510(k)
Submissions, for more information.12 Note that
the majority of FDA-authorized CAD devices
are intended to be decision support tools rather
than primary diagnostic tools.46
CADx (21 CFR §892.2060): Radiological
Computer-Assisted Diagnostic Software for
Lesions Suspicious of Cancer
CADx systems are intended to aid in the char-
acterization of lesions identified on acquired
medical images, such as MR, mammography,
radiography, or CT images. These software
devices perform functions that may include auto-
matically registering images, segmenting images,
or analyzing user-selected ROI. In July 2017,
the FDA granted a De Novo reclassification
request for QuantX27 as the first CADx intended
to be used as a second reader for breast MRI
(see Chapter 4).23 The device presents an index
indicating a QI score for the lesion site the score
is calculated via an AI algorithm from radiomic
features. CADx software devices such as QuantX
aid in providing diagnostic and patient manage-
ment decisions to clinical users. Other examples
of CADx include Brainomix 360 e-ASPECTS
(K221564),47 a software device intended to assist
the physician in the assessment and character-
ization of brain tissue abnormalities using CT
image data.
The performance study requirements for
CADx are comparable to CADe requirements.
The manufacturer should perform both stand-
alone performance testing and a reader study.
Figure 8-4. Overview of Computer-Assisted Diagnosis Applications
Clinical Decision Support (AI-assisted)
Computer-assisted diagnosis (CADx)
Risk
prediction
Computer-
assisted
detection
(CADe)
Disease
detection
Computer-
assisted
triage (CADt)
Disease
Character-
ization
Rule out
Staging/
planning
Treatment
response
assessment
Prognosis
prediction
Recurrence
predication &
monitoring
AI, Artificial intelligence
Source: Adapted from Hadjiiski L, et al.41