医学影像质量评估：从ICRU Report 54的视角
需积分: 15 179 浏览量
更新于2024-07-15
1
收藏 8.78MB PDF 举报
"ICRU Report 54：The Assessment of Image Quality"
报告《ICRU Report 54：图像质量评估》详细探讨了医学成像领域在过去的二十年间取得的巨大进步，这些进步主要体现在成像系统能力的提升和系统复杂性的增加。这一变革的主要驱动力是计算技术的发展以及数字技术在数据采集、处理和显示中的广泛应用。尽管医学成像的各个分支都受到了显著的影响，但最为突出的例子是X射线计算机断层扫描（CT）和磁共振成像（MRI）。
报告中指出，对诊断成像技术进行定量测量方法的共识正在逐渐形成。这是基于对不同成像模态之间共同特征的认识，使得我们能够在统计决策分析的框架内理解这些技术的局限性。这种评估方法对于确保医学成像的有效性和可靠性至关重要，因为它可以帮助识别并减少由于系统性能限制导致的误诊可能性。
图像质量的评估涵盖了多个方面，包括空间分辨率、对比度、噪声、信噪比（SNR）、对比度噪声比（CNR）等关键指标。空间分辨率是衡量成像系统能分辨最小细节的能力，而对比度则涉及到图像中不同区域之间亮度差异的识别。噪声是图像中随机变化的部分，它可能掩盖或干扰实际的解剖结构。信噪比和对比度噪声比则是评估图像质量和诊断效能的重要参数，高信噪比和对比度噪声比意味着图像更清晰，更利于医生做出准确的诊断。
报告还强调了质量控制的重要性，这包括定期的系统校准、维护和性能测试，以确保设备在整个工作周期内都能提供一致的图像质量。此外，还讨论了图像处理技术如何通过提高图像的视觉表现来改善图像质量，例如使用滤波、降噪算法以及后处理技术。
报告进一步涵盖了法律和通知方面的内容，强调了ICRU作为国际辐射单位和测量委员会的责任，旨在提供准确、完整且实用的信息。报告的编制者们对其内容的准确性承担责任，同时也提醒用户，尽管努力确保信息无误，但使用者仍需结合具体情况进行判断和应用。
《ICRU Report 54》为医学成像领域的专业人员提供了一个全面的指南，用于评估和优化图像质量，从而提升诊断的准确性和患者护理的质量。报告涵盖了从基本的图像质量指标到复杂的统计决策分析，对理解和改进医学成像系统的性能具有深远的意义。
4
requires
measurements
of
the
intrinsic
contrast
of
the
feature,
such
as
a lesion,
being
detected
(i.e.,
the
signal),
the
contrast
transfer
function
of
the
imaging
system,
and
the
noise
power
spectrum
(NPS)
at
a
given
imaging
condition.
These
fundamental
measure-
ments
and
their
normalization
for
portability
be-
tween
laboratories
will
be
described
in
Section 3.2.
These
measurements
are
the
ingredients
required
to
calculate
the
SNR
that
determines
the
intrinsic
detect-
ability
of
a lesion
in
a simple
phantom,
the
so-called
ideal
observer
signal-to-noise ratio.
1.4.2
Quasi-Ideal
Observers
While
the
strength
of
the
ideal
observer
model is
that
it
indicates
the
best
possible
performance,
its
application is
usually
limited
to
simple
phantoms
and
decision
tasks
because
of
the
difficulty
of
specifying
it
for
more
complicated
tasks.
While sacrificing
the
ideal
nature
of
the
analysis
it
is possible
to
modify
this
observer
to
give model decision
makers
which
are
applicable
to
a
wider
range
of
tasks
and
which
may,
in
principle,
more
closely
approximate
the
performance
of
human
observers.
Several
such
models
are
re-
viewed
in
Section
3.5. A
method
is also
presented
for
directly
measuring
the
SNRs
of
these
observers
with-
out
requiring
the
full
set
of
physical
measurements
needed
to
calculate
the
ideal-observer
performance.
These
observers
are
based
on
matched
filters
and
various
discriminant
functions
(e.g.,
Hotelling/
Fisher).
They
can
be
used
to
assess
the
contributions
to
the
separation
of
image
classes
from
the
combina-
tion
of
random
or
stochastic
noise
and
the
determinis-
tic
artifacts
that
arise
in
the
detection
and
formation
of
images.
Work
is also
presented
to
show
how
well
these
models
correlate
with
the
performance
of
a
human
observer.
1.4.3
Human
Observers
While model
observers
are
useful
for
providing
a
rapid
assessment
of
performance,
they
are
still lim-
ited
to
relatively simple
tasks.
To
incorporate
the
complete
realm
of
clinical complexity
it
is
necessary
to
present
images
to
a
human
observer
and
analyze
performance
by psychophysical
techniques.
The
most
rigorous
of
these,
the
only
one
allowing
performance
to
be
separated
from
the
observer's
bias, is
the
application
of
the
ROC methodology.
This
is de-
scribed
in
detail
in
Section
4.
In
such
studies,
the
observer
is
asked
to
perform
some
visual
task,
and,
through
training,
develops a decision
strategy
for
doing
so.
An
important
aspect
of
psychophysical
studies
of
image
quality
is
that
they
can
use
real
clinical
images
of
patients,
incorporating
the
natu-
rally
occurring
biological variability
to
arrive
at
a
realistic
task.
1.5
Image
Quality
and
the
Diagnostic
Process
Image
assessment
is defined
here
in
terms
of
the
performance
of
well-defined signal
detection
tasks.
Physical
performance
assessment
is
often
specified
in
terms
of
the
detection
or
discrimination
of
simulated
lesions while clinical
performance
assessment
is speci-
fied
in
terms
of
the
detection
or
discrimination
of
clinical lesions
or
other
disease
states.
These
tasks
are
only
part
of
the
larger
clinical
task,
or
process,
that
involves
the
management
of
the
patient
and
the
determination
of
the
patient
outcome
and
its
effect
on
society.
These
issues
come
under
the
broader
heading
referred
to
as
efficacy.
Fryback
and
Thornbury
(1991)
have
proposed
a
six-level model
of
efficacy
and
this
is
summarized
in
Table
1.1.
The
two lowest levels
are
the
concern
of
this
Report,
namely
technical
efficacy (i.e., physical
image
performance
assessment)
and
diagnostic accu-
racy
(i.e., clinical
image
performance
assessment).
A
major
purpose
of
this
Report
is
to
present
consensus
methodology so
that
these
levels
of
efficacy
may
be
characterized
quantitatively.
It
is
to
be
expected,
furthermore,
that
the
quantification
at
the
first level
will serve
as
a specification
of
the
state
of
technology
that
can
be
used
to
label
or
normalize
studies
at
the
second level,
and
ultimately
to
understand
their
ranking
of
imaging
systems.
Similarly,
quantification
at
both
of
these
lower levels
can
serve
to
label
or
normalize
studies
at
higher
levels
and
contribute
toward
the
understanding
of
their
results.
Work
at
all
of
these
levels,
then,
represents
major
efforts
towards
a
quantitative
science
of
medical decision
making.
1.6
Outline
of
the
Report
The
intention
of
this
Report
is
to
present
a frame-
work
within
which
the
diagnostic
quality
of
images
produced
by
a
variety
of
clinical
imaging
devices
can
be
evaluated.
It
is
not
proposed
to
deal
with
the
detail
of
how
this
framework
can
be
applied
to
each
modal-
ity;
although
examples
are
given
in
Appendix D,
this
task
will
be
tackled
in
subsequent
reports.
TABLE
1.1-Six-tier
model
of
efficacy
Level
Technical
efficacy
Diagnostic
accuracy
efficacy
Diagnostic
thinking
efficacy
Therapeutic
efficacy
Patient
outcome
efficacy
Societal efficacy
Typical
Output
Measures
MTF;
Gray
scale
steps
Sensitivity,
specificity, ROC
curve
analysis
Entropy
change;
Change
in
clini-
cian's
diagnostic
probability
Percentage
of
times
therapy
changed
Change
in
quality
adjusted
life
years
Summed
quality
adjusted
years;
Positive
change
in
national
product
Downloaded from https://academic.oup.com/jicru/article-abstract/os28/1/NP/2924011 by Oxford University Press USA user on 30 October 2018
剩余99页未读，继续阅读
2024-11-27 上传
2024-11-27 上传
2024-11-27 上传
2024-11-27 上传
2024-11-27 上传
2024-11-27 上传
2024-11-27 上传
2024-11-27 上传
2024-11-27 上传
haz2yy
- 粉丝: 5
- 资源: 24
我的内容管理
展开
最新资源
- MATLAB新功能：Multi-frame ViewRGB制作彩色图阴影
- XKCD Substitutions 3-crx插件：创新的网页文字替换工具
- Python实现8位等离子效果开源项目plasma.py解读
- 维护商店移动应用：基于PhoneGap的移动API应用
- Laravel-Admin的Redis Manager扩展使用教程
- Jekyll代理主题使用指南及文件结构解析
- cPanel中PHP多版本插件的安装与配置指南
- 深入探讨React和Typescript在Alias kopio游戏中的应用
- node.js OSC服务器实现：Gibber消息转换技术解析
- 体验最新升级版的mdbootstrap pro 6.1.0组件库
- 超市盘点过机系统实现与delphi应用
- Boogle: 探索 Python 编程的 Boggle 仿制品
- C++实现的Physics2D简易2D物理模拟
- 傅里叶级数在分数阶微分积分计算中的应用与实现
- Windows Phone与PhoneGap应用隔离存储文件访问方法
- iso8601-interval-recurrence：掌握ISO8601日期范围与重复间隔检查
