# US 8,280,136 B2 · Method and apparatus for detecting intraventricular dyssynchrony

> Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt

- Number: US 8,280,136 B2
- Document ID: US8280136B2
- Status: Granted
- Kind: B2
- Application number: 12/067,093
- Filed: 2006-09-18
- Published: 2008-10-23
- Granted: 2012-10-02
- Earliest claimed: 2005-09-16
- Assignees: The Ohio State University
- Inventors: Paulo F. U. Gotardo; Kim L. Boyer; Subha V. Raman
- Web page: https://patentgrep.com/patent/US8280136B2

## CPC

- G06T 7/0012: {Biomedical image inspection}
- G06T 7/12: Edge-based segmentation
- G06T 7/149: involving deformable models, e.g. active contour models
- G06V 10/755: {Deformable models or variational models, e.g. snakes or active contours}
- G06T 2207/10088: Magnetic resonance imaging [MRI]
- G06T 2207/20116: Active contour; Active surface; Snakes
- G06T 2207/30048: Heart; Cardiac
- G06T: IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06V: IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING

## Abstract

A method and apparatus for detecting intraventricular dyssynchrony is provided. In one embodiment, the method includes: a) receiving information from an observer viewing a source image to define a plurality of points associated with muscle boundaries of a ventricle, b) extracting muscle boundaries for a series of source images to define a corresponding series of contour shape images using a Fourier descriptor model, and c) deforming each contour shape based at least in part on image forces, external forces, and internal forces to track movement of the muscle boundaries over time. In another embodiment, the method includes: a) characterizing the series of source images as a corresponding series of contour shape images representative of movement of muscle boundaries of the ventricle over time, b) classifying the series of contour shape images in a dyssynchronous class or a non-dyssynchronous class, and c) pre-diagnosing the ventricle as dyssynchronous or non-dyssynchronous.

## Claims

### Claim 1

A method of detecting intraventricular dyssynchrony for a ventricle of a beating heart, the method including:
a) receiving initialization information from an observer viewing a first source image of a sequence of source images of a ventricle of a beating heart over at least one complete cardiac cycle to define a single vector constructed by concatenating Fourier shape descriptors extracted from a plurality of points associated with muscle boundaries of the ventricle over time;
b) extracting muscle boundaries for each source image of the sequence of source images to define a corresponding spatio-temporal characterization of a complete source image sequence with contour shapes corresponding to the muscle boundaries using a Fourier descriptor model; and
c) deforming the single vector of Fourier shape descriptors representing each contour shape in the complete source image sequence based at least in part on image forces, external forces, and internal forces to track movement of the muscle boundaries for the ventricle over time in the complete source image sequence.

### Claim 2 (depends on 1)

The method of claim 1 wherein the initialization information includes at least four points.

### Claim 3 (depends on 1)

The method of claim 1 wherein the first source image corresponds to the end of diastole for the cardiac cycle.

### Claim 4 (depends on 1)

The method of claim 1 wherein the muscle boundaries include an endocardial wall and an epicardial wall.

### Claim 5 (depends on 1)

The method of claim 1 wherein the set of source images include short-axis cine magnetic resonance images.

### Claim 6 (depends on 1)

The method of claim 1 wherein the Fourier descriptor model includes:
s
⁡
(
t
)
=
x
⁡
(
t
)
+
jy
⁡
(
t
)
=
∑
k
=
-
N
N
⁢
S
k
⁢
ⅇ
j
⁢
⁢
kt
,
with
⁢
⁢
j
=
-
1
,
where each complex number Sk=ak+jbk, with ak, bkεR, is treated as a rotating phasor with magnitude |Sk|=√{square root over (ak2+bk2)} and phase angle
ϕ
k
=
tan
-
1
⁡
(
b
k
a
k
)
.
where: x(t) and y(t) are projections, with t ε[0,2π]; and
ak and bk are real numbers.

### Claim 7 (depends on 6)

The method of claim 6 wherein each pair of descriptors Sk and S−k describes an ellipse.

### Claim 8 (depends on 1)

The method of claim 1 wherein the image forces include edge-based forces and texture-based forces.

### Claim 9 (depends on 8)

The method of claim 8 wherein the edge-based forces have a limited zone of attraction.

### Claim 10 (depends on 1)

The method of claim 1, the deforming in c) associated with image forces further including;
d) defining edges of the contour shapes in the complete contour shape image sequence using an optimal zero crossing operator model to determine edge-strength vectors and gradient vectors for associated pixels of the contour shape images;
e) deforming edges of the contour shapes in the complete contour shape sequence using an averaging mask in conjunction with a range of pixels along, the edges defined by the optimal zero crossing operator model;
f) determining a length of the contour shapes in the complete contour shape sequence using an approximation model; and
g) determining edge-based image forces for the contour shapes the complete contour shape sequence based at least in part on the determined length for the corresponding contour shape.

### Claim 11 (depends on 10)

The method of claim 10 wherein the approximation model includes:
L
2
=
4
⁢
π
2
⁢
∑
k
=
-
N
N
⁢
k
2
⁢

S
k

2
=
4
⁢
π
2
⁢
∑
k
=
-
N
N
⁢
k
2
⁡
(
a
k
2
+
b
k
2
)
,
where: L is the length of the contour shape; and
each complex number Sk =ak +jbk, with ak, bk εR, is treated as a rotating phasor with magnitude |Sk|=√{square root over (ak2+bk2)}.

### Claim 12 (depends on 1)

The method of claim 1, the deforming in c) further including:
d) sampling pixels values for the contour shapes in each contour shape image along a normal direction and at selected locations along a length of the corresponding contour shape to determine texture vectors associated with pixels of the contour shape images.

### Claim 13 (depends on 12)

The method of claim 12 wherein the sampling in d) includes sampling pixel values for the contour shapes using:
s
⊥
(
t
)
=
[
u
⁡
(
t
)
v
⁡
(
t
)
]
=
∑
k
=
-
N
N
⁢
k
⁡
[
a
k
-
b
k
b
k
a
k
]
⁡
[
cos
⁢
⁢
kt
sin
⁢
⁢
kt
]
⁢
⁢
at
⁢
⁢
locations
⁢
⁢
t
i
∈
[
0
,
2
⁢
π
]
,
where 1≦i≦L0 and L0 represents the length of the corresponding contour shape, to determine the texture vectors, and where ak and bk are real numbers.

### Claim 14 (depends on 1)

The method of claim 1 wherein the extracting in b) for the first source image is based at least in part on the initialization information.

### Claim 15 (depends on 14)

The method of claim 14 wherein the extracting in b) for each subsequent source image is based at least in part on the previous contour shape image.

### Claim 16 (depends on 1)

The method of claim 1, the deforming in c) associated with external forces further including:
d) normalizing the scale of each Fourier descriptor associated with the Fourier descriptor model; and
e) determining external forces for each contour shape image to limit contour shape variations in consecutive contour shape images based at least in part on a predetermined value for a variance parameter.

### Claim 17 (depends on 16)

The method of claim 16 wherein the scale-normalizing for each Fourier descriptor includes:
S
k

S
1

.
where: Sk represents components of shape descriptors: and
|S1|represents a magnitude of the largest of the components of the shape descriptors.

### Claim 18 (depends on 1)

The method of claim 1, the deforming in c) associated with internal forces further including:
d) determining a length of the contour shapes in the complete image sequence using an approximation model; and
e) determining internal tension forces to minimize contour shape length and avoid self-crossings as contour shapes are deformed.

### Claim 19 (depends on 18)

The method of claim 18 wherein the internal tension forces are weaker than the image forces.

### Claim 20 (depends on 18)

The method of claim 18 wherein the approximation model includes:
L
2
=
4
⁢
π
2
⁢
∑
k
=
-
N
N
⁢
k
2
⁢

S
k

2
=
4
⁢
π
2
⁢
∑
k
=
-
N
N
⁢
k
2
⁡
(
a
k
2
+
b
k
2
)
,
where: L is the length of the contour shape; and
each complex number Sk =ak +jbk, with ak, bk εR, is treated as a rotating phasor with magnitude |Sk|=√{square root over (ak2+bk2)}.

### Claim 21

A method of detecting intraventricular dyssynchrony, including:
a) providing a spatio-temporal characterization of a complete source image sequence of a ventricle of a beating heart over at least one complete cardiac cycle as a single vector constructed by concatenating Fourier shape descriptors extracted from a corresponding sequence of contour shape images representative of movement, of muscle boundaries of the ventricle over time;
b) classifying the single vector of Fourier shape descriptors representing the complete contour shape image sequence in a dyssynchronous class or a non-dyssynchronous class; and
c) pre-diagnosing the ventricle as dyssynchronous or non-dyssynchmnous based at least in part on the classifying.

### Claim 22 (depends on 21)

The method of claim 21, the classifying in b) further including:
d) normalizing the complete contour Shape image sequence;
e) processing the sequence of contour shape images into a lower dimensional representation of principal and discriminative features of the sequence of contour shape images to form a classifier sample; and
f) classifying the classifier sample in the dyssynchronous class or the non-dyssynchronous class.

### Claim 23 (depends on 22)

The method of claim 22, the classifying in f) further including:
g) projecting the classifier sample in a ventricle classification space defining the dyssynchronous class by a dyssynchronous class mean and defining the non-dyssynchronous class by a non-dyssynchronous class mean, the classification sample being represented in dimensions corresponding to dimensions for the ventricle classification space;
h) determining if the classification sample is closer to the dyssynchronous class mean or the non-dyssynchronous class mean; and
i) classifying the classifier sample in the dyssynchronous class or the non-dyssynchronous class based on which corresponding class mean is closest to the classifier sample.

### Claim 24

A method of detecting intraventricular dyssynchrony, including:
a) observing a spatio-temporal characterization of a complete source image sequence of movement of muscle boundaries of a ventricle of a beating heart over at least one complete cardiac cycle, wherein the spatio-temporal characterization is provided as a single vector constructed by concatenating Fourier shape descriptors extracted from a corresponding, sequence of contour shape images representative of movement of the muscle boundaries of the ventricle over time, and wherein the single vector of Fourier shape descriptors representing the complete contour shape image sequence are classified in a dyssynchronous class;
b) diagnosing the ventricle as dyssynchronous based at least in part on the observing;
c) planning treatment of the dyssynchronous ventricle to include epicardial implantation of a cardiac resynchronization therapy device, using a Fourier descriptor model;
d) identifying a dyssynchronous region of the ventricle based at least in part on the observing; and
e) planning lead placement for the cardiac resynchronization therapy device based at least in part on the dyssynchronous region of the ventricle.

### Claim 25 (depends on 24)

The method of claim 24 wherein the dyssynchronous region of the ventricle includes at least one of an anteroseptal region, an inferoseptal region an anterior region, an anterolateral region, an inferolateral region, and an inferior region.

### Claim 26 (depends on 24)

The method of claim 24 wherein the identifying in d) is based at least in part on identifying a cardiac segment associated with the dyssynchronous region thickening late in relation to overall myocardial thickening.

### Claim 27

A method of detecting intraventricular dyssynchrony, including:
a) spatially-temporally characterizing a complete source image sequence of a ventricle of a beating heart over at least one complete cardiac cycle as a single vector constructed by concatenating Fourier shape descriptors extracted from a corresponding sequence of contour shape images representative of movement of muscle boundaries Of the ventricle over time;
b) classifying the single vector of Fourier shape descriptors representing the complete contour shape image sequence in at least one of a plurality of dyssynchronous subclasses or in a non-dyssynchronous class, each dyssynchronous subclass associated with a different region of the ventricle; and
e) pre-diagnosing the ventricle as having at least one dyssynchronous region or as non-dyssynchronous based at least in part on the classifying.

### Claim 28 (depends on 27)

The method of claim 27, the classifying in b) further including:
d) normalizing the sequence of contour shape images;
e) processing the sequence of contour shape images into a lower dimensional representation of principal and discriminative features of the sequence of contour shape images to form a classifier sample; and
f) classifying the classifier sample in at least one dyssynchronous subclass or the non-dyssynchronous class.

### Claim 29 (depends on 28)

The method of claim 28, the classifying in f) further including:
g) projecting the classifier sample in a ventricle classification space defining each dyssynchronous subclass by a corresponding dyssynchronous subclass mean and defining the non-dyssynchronous class by a non-dyssynchronous class mean, the classification sample being represented in dimensions corresponding to dimensions for the ventricle classification space;
h) determining if the classification sample is closer to any dyssynchronous subclass means or the non-dyssynchronous class mean;
i) classifying the classifier sample in each dyssynchronous subclass where the classification sample is closer to the corresponding dyssynchronous subclass mean than the non-dyssynchronous class mean; and
j) classifying the classifier sample in the non-dyssynchronous class where the classification sample is closer to the non-dyssynchronous class mean than any dyssynchronous subclass mean.

### Claim 30

A method of detecting intraventricular dyssynchrony, including:
a) capturing movement of a ventricle of a beating heart over at least one complete cardiac cycle over time in a sequence of source images for a subject;
b) extracting muscle boundaries for each source image of the sequence of source images to define a corresponding spatio-temporal characterization of a complete source image sequence as a single vector constructed by concatenating Fourier shape descriptors extracted from a corresponding sequence of contour shape images representative of movement of the muscle boundaries of the ventricle over time;
c) deforming the single vector of Fourier shape descriptors representing each subject contour shape in the complete source image sequence to track movement of the muscle boundaries for the ventricle over time in the complete subject source image sequence;
d) observing the complete subject source image sequence;
e) comparing the complete subject source image sequence to other complete source image sequences from other cases; and
f) pre-diagnosing the subject ventricle as dyssynchronous or non-dyssynchronous based at least in part on the observing and comparing.

### Claim 31 (depends on 30)

The method of claim 30 wherein the process is used in conjunction with teaching diagnosis of intraventricular dyssynchrony.

### Claim 32 (depends on 30)

The method of claim 30 wherein the process is used as a diagnostic aid by a practitioner having limited experience in diagnosis of intraventricular dyssynchrony.

### Claim 33 (depends on 30)

The method of claim 30 wherein the process is used as a diagnostic aid by a more general practitioner to determine whether or not referral to a specialist is advisable.

### Claim 34 (depends on 30)

The method of claim 30 wherein at least the extracting in b) and the deforming in c) are performed at a different location than the capturing in a), further including:
g) communicating the sequence of source images for the subject from a remote location to a central location via a communication network.

### Claim 35 (depends on 34)

The method of claim 34 wherein the communication network includes the Internet.

### Claim 36 (depends on 34)

The method of claim 34, further including:
h) communicating the subject sequence of contour shape images and the other sequences of contour shape images from the central location to the remote location via the communication network.

### Claim 37

An apparatus for detecting intraventricular dyssynchrony, including:
a contour image processor extracting muscle boundaries for each source image of a sequence of source images of a ventricle of a beating heart over at least one complete cardiac cycle to define a single vector constructed by concatenating Fourier shape descriptors from the extracted muscle boundaries corresponding to a spatio-temporal characterization of the complete source image sequence, images and deforming the single vector of Fourier shape descriptors representing each contour shape in the complete source image sequence based at least in part on image forces, external forces, and internal forces to track movement of the muscle boundaries for the ventricle over time in the sequence of contour shape images; and
an image classifier, in operative communication with the contour image processor, classifying the sequence of contour shape images in a dyssynchronous class or a non-dyssynchronous class.

### Claim 38 (depends on 37)

The apparatus of claim 37, farther including:
a source image processor, in operative communication with the contour image processor, receiving initialization information from an observer viewing, a first source image of the sequence of source images to define a plurality of points associated with the muscle boundaries.

### Claim 39 (depends on 37)

The apparatus of claim 37 wherein the contour image processor defines edges of the contour shapes in each contour shape image using an optimal zero crossing operator model to determine edge-strength vectors and gradient vectors for associated pixels of the contour shape images, deforms edges of the contour shapes in each contour shape image using an averaging mask in conjunction with a range of pixels along the edges defined by the optimal zero crossing operator model, determines a length of the contour shapes in each contour shape image using an approximation model, and determines edge-based image forces for the contour shapes in each contour shape image based at least in part on the determined length for the corresponding contour shape.

### Claim 40 (depends on 37)

The apparatus of claim 37 wherein the contour image processor normalizes the scale of each Fourier descriptor associated with the Fourier descriptor model and determines external forces for each contour shape image to limit contour shape variations in consecutive contour shape images based at least in part on a predetermined value for a variance parameter.

### Claim 41 (depends on 37)

The apparatus of claim 37 wherein the contour image processor determines a length of the contour shapes in each contour shape image using an approximation model and determines internal tension forces to minimize contour shape length and avoid self-crossings as contour shapes are deformed.

### Claim 42

An apparatus for detecting intraventricular dyssynchrony, including:
an image contour processor spatially-temporally characterizing a sequence of source images of a ventricle of a beating heart over at least one complete cardiac cycle as a single vector constructed by concatenating Fourier shape descriptors extracted from a corresponding sequence of contour shape images representative of movement of muscle boundaries of the ventricle over time; and
an image classifier, operative communication with the contour image processor, classifying the concatenated sequence of contour shape images in a dyssynchronous class or a non-dyssynchronous class.

### Claim 43 (depends on 42)

The apparatus of claim 42 wherein the image classifier normalizes the sequence of contour shape images, processes the sequence of contour shape images into a lower dimensional representation of principal and discriminative features of the sequence of contour shape images to form a classifier sample, and classifies the classifier sample in the dyssynchronous class or the non-dyssynchronous class.

### Claim 44 (depends on 43)

The apparatus of claim 43 wherein the image classifier projects the classifier sample in a ventricle classification space defining the dyssynchronous class by a dyssynchronous class mean and defining the non-dyssynchronous class by a non-dyssynchronous class mean, the classification sample being represented in dimensions corresponding to dimensions for the ventricle classification space, determines if the classification sample is closer to the dyssynchronous class mean or the non-dyssynchronous class mean, and classifies the classifier sample in the dyssynchronous class or the non-dyssynchronous class based on which corresponding class mean is closest to the classifier sample.

### Claim 45

An apparatus for detecting intraventricular dyssynchrony, including:
a contour image processor spatially-temporally characterizing a complete source image sequence of a ventricle of a beating heart over at least one complete cardiac cycle as a single vector constructed by concatenating Fourier shape descriptors extracted from a corresponding sequence of contour shape images representative of movement of muscle boundaries of the ventricle over time; and
an image classifier classifying the concatenated contour shape images in at least one of a plurality of dyssynchronous subclasses or in a non-dyssynchronous class, each dyssynchronous subclass associated with a different region of the ventricle.

### Claim 46 (depends on 45)

The apparatus of claim 45 wherein the contour image processor normalizes the sequence of contour shape images, processes the sequence of contour shape images into a lower dimensional representation of principal and discriminative features of the sequence of contour shape images to form a classifier sample.

### Claim 47 (depends on 46)

The apparatus of claim 46 wherein the image classifier classifies the classifier sample in at least one dyssynchronous subclass or the non-dyssynchronous class.

### Claim 48 (depends on 45)

The apparatus of claim 45 wherein the image classifier projects the classifier sample in a ventricle classification space defining each dyssynchronous subclass by a corresponding dyssynchronous subclass mean and defining the non-dyssynchronous class by a non-dyssynchronous class mean, the classification sample being represented in dimensions corresponding to dimensions for the ventricle classification space, determines if the classification sample is closer to any dyssynchronous subclass means or the non-dyssynchronous class mean, classifies the classifier sample in each dyssynchronous subclass where the classification sample is closer to the corresponding dyssynchronous subclass mean than the non-dyssynchronous class mean, and classifies the classifier sample in the non-dyssynchronous class where the classification sample is closer to the non-dyssynchronous class mean than any dyssynchronous subclass mean.

### Claim 49

A method for monitoring an internal area of a subject, including:
a) receiving initialization information from an observer viewing a first source image of a sequence of source images of the internal area over a predetermined time to define a single vector constructed by concatenating Fourier shape descriptors extracted from a plurality of points associated with boundaries within the internal area;
b) extracting boundaries for each source image of the sequence of source images to define a corresponding spatio-temporal characterization of a complete source image sequence with contour shapes corresponding to the boundaries using a Fourier descriptor model; and
c) deforming the single vector of Fourier shape descriptors representing each contour shape in the complete source image sequence based at least in part on image forces, external forces, and internal forces to track movement of the boundaries within the internal area over time in the complete source image sequence.

### Claim 50

An apparatus for monitoring an internal area of a subject, including:
a contour image processor extracting boundaries for each source image of a sequence of source images of the internal area over a predetermined time to define a single vector constructed by concatenating Fourier shape descriptors from the extracted muscle boundaries corresponding to a spatio-temporal characterization of the complete source image sequence, images and deforming the single vector of Fourier shape descriptors representing each contour shape in the complete source image sequence based at least in part on image forces, external forces, and internal forces to track movement of the boundaries within the internal area over time in the sequence of contour shape images; and
an image classifier, in operative communication with the contour image processor, classifying the sequence of contour shape images in a first class or a second class based at least in part on detection of predetermined discriminating characteristics for the first and second classes in the contour shape images.
