System and method for automatic detection of end of diastole and end of systole image frames in X-ray ventricular angiography
End-diastolic and end-systolic image frames are automatically selected on a real-time basis from a sequence of X-ray ventricular angiogram images by modeling the angiogram images by a dynamic graphical model and estimating a posterior probability density of the ventricular area in each angiogram image frame using Bayesian probability density propagation and adaptive background modeling. Then, a variation curve plot of expectation values of the posterior probability density of the ventricular area of each angiogram image frame is generated in which peaks and valleys in the variation curve correspond to end-diastolic and end-systolic angiogram image frames, respectively.
1. A computer implemented method of identifying and selecting end-diastolic and end-systolic X-ray image frames from a sequence of X-ray ventricular angiogram image frames, the method comprising:
employing at least one computer for,
modeling said angiogram image frames by a dynamic graphical model, said dynamic graphical model formulating areas of ventricular region in the angiogram image frames as a sequence of foreground states, formulating backgrounds in the angiogram image frames as a sequence of background states, wherein the changes of the foreground states and the background states from one image frame to a subsequent image frame is formulated as a Markov chain and the foreground states and the background states generating corresponding observations that are overlapped in each angiogram image frames;
estimating a posterior probability density of the foreground state of each angiogram image frame based on both foreground and background observations using Bayesian probability density propagation and adaptive background modeling;
calculating an expectation value of said posterior probability density of the foreground state of each angiogram image frame, said expectation value representing an estimation of the area of the ventricular region represented in each angiogram image frame; and
generating a variation curve of the expectation values, said variation curve containing a plurality of peaks and valleys whereby said peaks correspond to end-diastolic angiogram image frames and said valleys correspond to end-systolic angiogram image frames.
2. The method of claim 1 further comprising determining optimal end-diastolic and end-systolic angiogram image frame pairs from the variation curve of the expectation values.
3. The method of claim 1 , wherein estimating a posterior probability density of the foreground state of each angiogram image frame based on both foreground and background observations using Bayesian probability density propagation and adaptive background modeling includes using sequential Monte Carlo simulation method to estimate the probability density propagation.
4. The method of claim 3 , wherein the posterior probability density of the foreground state is expressed as p(χ 1:t |z 1:t , G 1:t ), where χ t represents the area of the ventricular region in X-ray angiogram image frame at time index t, z t represents said foreground observation associated with the foreground state χ t at time index t, and G t represents the background observation associated with the foreground state χ t at time index t, and p(χ 1:t |z 1:t , G 1:t ) is approximated as
p
(
x
1
:
t
❘
z
1
:
t
,
G
1
:
t
)
≈
∑
n
=
1
N
w
t
n
δ
(
x
1
:
t
-
x
1
:
t
n
)
,
where {x 1:t n , n=1 . . . N} is a set of said samples of the foreground states, {w t n , n=1 . . . N} are the associated weights, and δ is the Dirac delta function, and based on importance sampling theory, when the samples are obtained from an importance density q(·), the weights are given by
w
t
n
∝
p
(
x
1
:
t
n
❘
z
1
:
t
,
G
1
:
t
)
q
(
x
1
:
t
n
❘
z
1
:
t
,
G
1
:
t
)
.
5. A non-transitory computer readable program storage device, tangibly embodying a program of instructions executable by the computer to perform a method of identifying and selecting end-diastolic and end-systolic X-ray image frames from a sequence of X-ray ventricular angiogram image frames, wherein said method comprising:
modeling said angiogram image frames by a dynamic graphical model, said dynamic graphical model formulating areas of ventricular region in the angiogram image frames as a sequence of foreground states, formulating backgrounds in the angiogram image frames as a sequence of background states, wherein the changes of the foreground states and the background states from one image frame to a subsequent image frame is formulated as a Markov chain and the foreground states and the background states generating corresponding observations that are overlapped in each angiogram image frames;
estimating a posterior probability density of the foreground state of each angiogram image frame based on both foreground and background observations using Bayesian probability density propagation and adaptive background modeling;
calculating an expectation value of said posterior probability density of the foreground state of each angiogram image frame, said expectation value representing an estimation of the area of the ventricular region represented in each angiogram image frame; and
generating a variation curve of the expectation values, said variation curve containing a plurality of peaks and valleys whereby said peaks correspond to end-diastolic angiogram image frames and said valleys correspond to end-systolic angiogram image frames.
6. The device of claim 5 , wherein the method further comprising determining optimal end-diastolic and end-systolic angiogram image frame pairs from the variation curve of the expectation values.
7. The device of claim 5 , wherein the estimating a posterior probability density of the foreground state of each angiogram image frame based on both foreground and background observations using Bayesian probability density propagation and adaptive background modeling step of the method further includes using sequential Monte Carlo simulation method to estimate the probability density propagation.
8. The device of claim 5 , wherein the posterior probability density of the foreground state is expressed as p(χ 1:t |z 1:t , G 1:t ), where χ t represents the area of the ventricular region in X-ray angiogram image frame at time index t, z t represents said foreground observation associated with the foreground state χ t at time index t, and G t represents the background observation associated with the foreground state χ t at time index t, and p(χ 1:t |z 1:t , G 1:t ) is approximated as
p
(
x
1
:
t
❘
z
1
:
t
,
G
1
:
t
)
≈
∑
n
=
1
N
w
t
n
δ
(
x
1
:
t
-
x
1
:
t
n
)
,
where {x 1:t n , n=1 . . . N} is a set of said samples of the foreground states, {w t n , n=1 . . . N} are the associated weights, and δ is the Dirac delta function, and based on importance sampling theory, when the samples are obtained from an importance density q(·), the weights are given by
w
t
n
∝
p
(
x
1
:
t
n
❘
z
1
:
t
,
G
1
:
t
)
q
(
x
1
:
t
n
❘
z
1
:
t
,
G
1
:
t
)
.