I

It* ' catheter

50 100

Real data

Figure 1.39: Real (blue) and simulated (red) gray-level vertical profile (a) of ROIs of Fig. 1.38(b) and data correlation (b).

The maximal difference profiles are localized in the transducer sheath gray-level distribution and the baseline of the transducer sheath inner region. These differences can be smaller, increasing the video and instrumental noise. The high-frequency oscillations in the gray-level profiles come from the concentric arterial structures. We can also observe the gradual reduction of the gray-level magnitude from intima/media interface to adventitia, caused by the ultrasound intensity attenuation.

real in blue and simulated in red real in blue and simulated in red

60 r

50 ■

-1—'

40 ■

E

O

30 ■

u

20 ■

10

n -

-bl

histogram of gray level gray level differences

Figure 1.40: Global projections in the direction 0 (a), from Figs. 1.38(b) and (d) and the corresponding histogram gray-level differences (b).

I 80

I eo

2 40

real in blue and simulated in red real in blue and simulated in red

. ^catheter

\"li "

u

Radius

Radius

histogram of gray level

20 40

gray level differences

Figure 1.41: Global projection in the R direction (c), from Figs. 1.38(b) and (d), the corresponding histogram gray-level differences are shown in (b).

The next step in the validation of the model is to show the significant correspondence between real and simulated gray-level distribution data in the medical zones of interest. For this purpose, 20 validated real IVUS images and their corresponding ROIs were selected. The spatial boundaries of the morphological structures of the real data are kept in the synthetic data. Figure 1.42(a) shows ten real IVUS images and their corresponding simulated (b) synthetic images. The polar images are shown in Fig. 1.43.

Figure 1.44 shows the simulated versus real gray-level correlation for the polar ROIs images selected as shown in Fig. 1.38. The linear correlation coefficients show a good gray-levels correspondence, these being m = 0.90 and b = 1.42. The best correspondence is located by low gray levels (20-40 gray levels), lumen scatterers, lumen/intima transition, and adventitia. The transitions of intima/media and media/adventitia (45-60 gray levels) indicate gradual dispersion. The CNRS average presents significative uniformity values, ¡ = 6.89 and a = 2.88, for all validated frames. The CNRS as figure of merit for each arterial validated region is shown in Fig. 1.45. The CNRS region mean, standard deviation, and the SSE values referring to the 20 image frames are summarized in Table 1.5. The lumen is a good simulated region, with mean ¡x = 0.46 and deviation a = 0.42. The explanation is that the lumen is not a transition zone, the attenuation ultrasound intensity in this region is very poor (1-2%), which determines a simple gray-level profile.

Real images

Simulated images

Real images

Simulated images

Figure 1.42: Ten original IVUS images (a) and the corresponding simulated (b)

images.

The histograms of gray-level differences for each region of interest in the 20 validated frames are displayed in Figs. 1.46 and 1.47. Table 1.6 explains the distribution center ¡x and the standard deviation a for the gray-level difference distribution for each simulated region. The minus sign in the mean values means that the simulated images are brighter than the real images. A symmetric Gaussian can be seen in the lumen gray-level differences distribution, with mean ¡x = -2.44 and deviation a = 15.13. The intima distribution has

Real imaqes

Real imaqes

Figure 1.43: Ten polar real images (a) and the corresponding simulated (b) images.

Figure 1.43: Ten polar real images (a) and the corresponding simulated (b) images.

Table 1.5: CNRS mean, standard deviation (std), and sum square error for different tissues structures

ROI

Mean

Std

SSE

Lumen

0.46

0.42

47.68

Intima

10.0

4.38

12.63

Media

9.91

5.14

15.05

Adventitia

7.21

2.76

Figure 1.44: Simulated versus real gray-level values for 20 ROIs comparing pixel gray-level and the regression line.

real gray level

Figure 1.44: Simulated versus real gray-level values for 20 ROIs comparing pixel gray-level and the regression line.

a mean of ¡x = -18.56 and deviation of a = 24.01, and the media region has a mean of ¡x = -17.82 and a deviation of a = 22.62. The gray-level differences distribution displays a light asymmetry. As a result, the simulated image tends to be brighter than the real image. The adventitia gray-level differences values show a symmetric distribution with a center of ¡x = -13.30 and a deviation of a = 14.27.

Table 1.6: Mean and deviation of the ROIs gray-level differences referred from histograms in Figs. 1.46 and 1.47

ROI

ß

a

Blood

—2.44

15.1B

Intima

— 18.56

24.01

Media

— 17.82

22.62

Adventitia

— 1B.B0

14.27

CNRS Vs for each ROI's in the validated frames

CNRS Vs for each ROI's in the validated frames

Figure 1.45: CNRS values for each ROI of 20 manually segmented image frames.

It is very important to note that the gray-level difference distribution exhibited Gaussian distributions for all regions of interest. Certainly, the synthetic image brightness is an open problem of the image formation model. The simplest approach is to variate it by modifying the original intensity Io of the ultrasound beam, similar to the offset of the image acquisition system. Real and simulated gray-level distributions for each region of interest are shown in Figs. 1.48 and 1.49. We can note the great similarity in the gray-level distributions profile. Figure 1.50 shows the gray-level histogram of the different tissues structures that appear in IVUS images. As expected, it can be seen that the gray-level distributions of different structures overlap and as a result it is not possible to separate the main regions of interest in IVUS images, using only the gray-level distributions as image descriptors.

Figure 1.46: Histogram of gray-level differences for lumen (a) and intima (b).

Figure 1.46: Histogram of gray-level differences for lumen (a) and intima (b).

Figure 1.47: Histogram of gray-level differences for media (a) and adventitia (b).

Figure 1.48: Real (blue) and simulated (red) gray-level distributions for lumen (a) and intima (b).

Figure 1.48: Real (blue) and simulated (red) gray-level distributions for lumen (a) and intima (b).

05 0

05 0

ll ^

1 1 j

11 Í

J M f

J j

Oiiy Jírvels

(b)

Figure 1.49: Real (blue) and simulated (red) gray-level distributions for media (a) and adventitia (b).

104 Simulated gray levels distributions

UBiood :

-

-

-

í/h : : J 1 * 1 '

, -4r— Adventitial i. A iL A i i

f|\! i

1

fjMedia i i v i

i

------

r-^,--

Gray levels

Gray levels

Figure 1.50: Simulated gray-level distributions for blood, intima, media, and adventitia.

Was this article helpful?

0 0

Post a comment