Automated segmentation of organ chambers using deep learning methods from medical imaging

1. A system for automatically segmenting a body chamber from medical images of a patient, the system comprising one or more hardware processors configured to:

obtain medical images comprising at least a representation of the body chamber inside the patient;

obtain a region of interest corresponding to the body chamber from each of the medical images;

organize the obtained region of interest using an input vector;

apply the input vector through a trained graph having at least one hidden layer; and

obtain an output vector representing a refined region of interest corresponding to the body chamber based on the application of the input vector through the trained graph,

wherein the system is configured to utilize a training data set to initialize filters randomly to train the convolutional network, and wherein the filters are convolved with the input medical images to obtain k convolved feature maps of size m 1 ×m 1 , computed as:

Z

l

⁡

[

i

,

j

]

=

∑

k

1

=

1

a

⁢

∑

k

2

=

1

a

⁢

F

l

⁡

[

k

1

,

k

2

]

⁢

I

⁡

[

i

+

k

1

-

1

,

j

+

k

2

-

1

]

+

b

0

⁡

[

l

]

,

(

1

)

for 1≤i,j≤m 1 , l=1, . . . , k, and m 1 =m−a+1; and

wherein p×p non-overlapping regions in the convolved feature maps are computed as:

P

l

⁡

[

i

1

,

j

1

]

=

1

p

⁢

∑

i

=

1

+

(

i

1

-

1

)

⁢

p

i

1

⁢

p

⁢

⁢

∑

j

=

1

+

(

j

1

-

1

)

⁢

p

j

1

⁢

p

⁢

⁢

C

l

⁡

[

i

,

j

]

,

(

2

)

for 1≤i 1 , j 1 ≤m 2 , wherein m 2 =m 1 /p and p is chosen such that m 2 is an integer value.

2. The system according to claim 1 , wherein the one or more hardware processors is/are configured to apply a deformable model on the obtained output vector representing the region of interest.

3. The system according to claim 1 , wherein the one or more hardware processors is/are configured to identify and segment a chamber of a heart.

4. The system according to claim 3 , wherein the chamber of a heart is selected from the group consisting of a left ventricle, a right ventricle, a left atrium and a right atrium.

5. The system according to claim 1 , wherein the medical images comprise magnetic resonance imaging (MRI) images, ultrasound images, or CT scan data.

6. The system according to claim 1 , wherein the system comprises one or more elements selected from the group consisting of an operating system, a data input port or reader, an output display device, a keyboard, a transceiver unit, a speaker, and an antenna for wireless communication with other with other devices.

7. The system according to claim 1 , wherein the system is linked to an additional computing device and/or a network.

8. The system according to claim 1 , wherein the memory comprises one or more of a dynamic random access memory (DRAM), a static random access memory (SRAM), a flash memory, a hard disk drive, a floppy disk drive and a disk-based memory.

9. The system according to claim 1 , wherein the system comprises a graphics processing unit (GPU) configured to accelerate computing.

10. A method for automatically segmenting a body chamber from medical images of a patient, the method comprising using the system according to claim 1 to:

obtain image data representing the stack of medical images comprising at least a representation of a body chamber inside the body;

obtain a region of interest using a convolutional network trained to locate the body chamber, wherein the region of interest corresponds to the body chamber from each of the medical images; and

infer a shape of the body chamber using a stacked auto-encoder (AE) network trained to delineate the body chamber, wherein the AE network segments the body chamber.

11. The system according to claim 1 , wherein the one or more hardware processors is/are configured to identify a segment of a body chamber from an output of the trained graph.

12. The system according to claim 1 , wherein the system is configured to obtain filters using a sparse autoencoder (AE), which acts as a pre-training step.

13. The system according to claim 1 , wherein the trained graph has two or more hidden layers.

14. A system for automatically segmenting a body chamber from medical images of a patient, the system comprising one or more hardware processors configured to:

obtain medical images comprising at least a representation of the body chamber inside the patient;

obtain a region of interest corresponding to the body chamber from each of the medical images;

organize the obtained region of interest using an input vector;

apply the input vector through a trained graph having at least one hidden layer; and

obtain an output vector representing a refined region of interest corresponding to the body chamber based on the application of the input vector through the trained graph,

wherein the one or more hardware processors is/are configured to align the medical images of the body chamber by performing contour alignment to reduce misalignment between the short-axis images, and wherein center coordinates of the images are estimated using the following quadratic assumptions for curvature:

x i =a 1 i 2 +b 1 i+c 1 ,  (Equation 17),

y i =a 2 i 2 +b 2 i+c 2   (Equation 18),

wherein a 1 , b 1 , c 1 , a 2 , b 2 c 2 are unknown parameters estimated based on minimizing the mean squared error as:

a

^

1

,

b

^

1

,

c

^

1

=

arg

⁢

⁢

min

a

1

,

b

1

,

c

1

⁢

∑

i

=

1

n

⁢

⁢

(

x

~

i

-

a

1

⁢

i

2

-

b

1

⁢

i

-

c

1

)

2

,

⁢

a

^

2

,

b

^

2

,

c

^

2

=

arg

⁢

⁢

min

a

2

,

b

2

,

c

2

⁢

∑

i

=

1

n

⁢

⁢

(

y

~

i

-

a

2

⁢

i

2

-

b

2

⁢

i

-

c

2

)

2

.

,

and

wherein, after estimating the unknown parameters, the actual center coordinates are estimated from equations (17) and (18), and wherein the contours are registered, using an affine transformation with linear interpolation, according to the estimated center values to obtain an aligned stack of contours.