ES2387118B1 - METHOD OF SEGMENTATION OF A HEPATIC ORGAN IN A SEQUENCE OF TOMOGRAPHIC IMAGES - Google Patents

Abstract

Segmentation method of a liver organ in a sequence of tomographic images # Segmentation method of a liver organ in a sequence of images [n {sub, 1}? n {sub, j}? n {sub, i}? n {sub, k}? n {sub, N}] tomographic comprising: making a manual selection of the hepatic organ in two reference images (n {sub, j} n {sub, k}); calculating for each image of the sequence an intensity mean, an intensity standard deviation and a centroid of the pixels of each selected region; calculate mean values ** IMAGE-01 **, standard deviation ** IMAGE-02 ** and estimated position (xi, yi); binarize the image by assigning a value to the pixels whose intensity is comprised in an interval ** IMAGE-03 **, where ** IMAGE-04 ** and ** IMAGE-05 ** are proportional to ** IMAGE-02 **; selecting a set of adjacent pixels with said value that comprise the pixel of the position (xi, yi); and assigning said set of pixels as a result. In the case of n {sub, i}, j <i <= k, the mean, standard deviation and estimated position are calculated by interpolating the mean, standard deviation and position values obtained for (n {sub, j} n {sub , k}); for i> k, the values obtained for the previous image are used (n {sub, i-1}); and for i <= j, the values of the image n {sub, j}.

Description

FIELD OF THE INVENTION

The present invention applies to the field of medical imaging, more specifically, to the segmentation of the liver organ in tomographic images.

BACKGROUND OF THE INVENTION

In the field of medical image processing, there are numerous commercial systems that provide means for the management and visualization of the images obtained by any of the technical means available.

(MRI, CT, X-ray, etc.). Thus, for example, WO 2003/043490 Al focuses on a visualization, information management and tool management system for the generic treatment of images. It provides tools to define the Regions of Interest (ROI) and display the results in 2D and 3D, without adapting to any specific organism.

In general, most of the systems aimed at segmenting a generic organ or structure are inscribed in an environment or computer-based system with tools for managing and visualizing both the sequences of

image

original What from the results and measures

obtained

after the operations from segmentation me

definition

from outline. Some from these systems I know

describe as oriented to segment any organ, as is the case with the one described in US 2006/0228009 Al, which

presents an automatic method based on deformable surface algorithms, or in US 2006/0159341 Al, which uses a different method also based on the fitting of deformable models. us 2005/0207628 presents an alternative strategy based on morphological operations such as dilation, erosion, calculation of thresholds and filling of

holes.

All are solutions from character general,

have

the advantage from can be used on a large

rank

from scenarios, but suffer on consequence a

lack

from specialization that supposes a worse result,

so much

on precision What on load computational, to the

face a specific scenario, in this case, the segmentation of the liver organ.

A possible strategy for the specific segmentation of the hepatic organ is the one presented by us 2009/052756 Al, in which it starts from three-dimensional forms of training that try to adjust to the images on which you want to perform the segmentation through rotations and translations. The three-dimensional treatment of the problem, however, supposes a high computational load, in addition to its precision being limited by the set of training forms.

On the other hand, US 2009/097726 Al presents a segmentation method based on multiphase tomographic images, where said phases refer to moments in which contrast agents are concentrated in different areas of the liver (in this case, veins or arteries) . On said images, a processing is subsequently applied that takes into account the intensity histograms for each of the phases. This method places higher requirements on the input images, which translates into greater use and demand for the image capture system.

Likewise, the values and distributions of the pixels differ significantly from one patient to another, so that fully automatic segmentation methods are more susceptible to being altered by

the diversity of properties of images.

There therefore remains a need for a liver organ segmentation method that is tailored to

the

morphology own of the same, doing a

segmentation

precise, robust, and computationally

efficient.

SUMMARY OF THE INVENTION

The present invention solves the above problem by means of a semi-automatic segmentation method that uses as initial information a manual segmentation of the hepatic organ in two reference tomographic images of a sequence of axial sections of a human trunk comprising axial sections of said hepatic organ.

In the segmented area of each reference image, the mean and standard deviation of the pixels contained in said segmented area are calculated, as well as the position of the centroid thereof. These mean, standard deviation, and position values are interpolated to obtain an estimated mean value, an estimated standard deviation, and an estimated position for each image in the sequence between the two reference images.

Preferably, for images whose index is lower than that of the reference image with the lowest index, the estimated mean, the estimated standard deviation and the estimated position take the calculated values of mean, standard deviation and position in said reference image of lower index.

Also preferably, for the images whose index is higher than that of the reference image with the highest index, the estimated mean, the estimated standard deviation and the estimated position take values of mean, standard deviation and position calculated for the image segmented with a index one smaller unit. Said selection of mean, standard deviation and position

Estimates allow the method to adapt to typical intensity changes of the upper and lower extremities of the liver.

For each image in the sequence, a range of intensities is then defined from the estimated mean and the estimated standard deviation for that image, using said range to binarize the image. That is, pixels whose intensity falls within that range are assigned a first value, and pixels whose intensity falls outside the range are assigned a second value.

Finally, a set of adjacent pixels that have said first value and that include the pixel that is in the position estimated for said image is selected as the area corresponding to the hepatic organ.

I know

get So a method computationally

efficient

that I know adapts to the variations on the

parameters

from the images from the cuts of the organ

hepatic

according the height from sayings cuts. Too I know

adapts to patient-to-patient variations and measurement conditions by extracting initial data from manual segmentation.

Preferably, the precision and robustness of the method are improved by incorporating morphological processing (erosion, gap filling, and dilation).

Preferably, the images on which manual segmentation is performed correspond to the maximum surface section of the hepatic organ in the case of the reference image with the highest index, and to a section corresponding to the lower region of the hepatic organ in the case of the image of lower index reference, in both cases through a visual inspection of the images by the user.

In order to increase the robustness of the method, it preferably comprises a results verification step, in which it verifies that the segmented area is

not null and has not undergone a variation with respect to the segmented area of the previous image greater than a predefined threshold. If any of the checks are not met, the method calculates a new segmented area.

In another aspect of the present invention, a computer program is presented comprising computer program code adapted to perform the steps of the described method when said program is executed on a computer, a digital processor of the

sign,

a circuit specific integrated from the app,

a

microprocessor, a microcontroller or any other

shape

from hardware programmable.

BRIEF DESCRIPTION OF THE FIGURES

In order to help a better understanding of the characteristics of the invention according to a preferred example of practical embodiment thereof and to complement this description, the following figures are attached as an integral part thereof, the character of which is illustrative and not limiting:

Figure 1 illustrates the segmentation steps for an example image.

Figure 2 shows the result of applying the method of the invention to a sequence of tomographic images, showing decimated images of the complete sequence.

Figure 3 shows a precision measure of the method according to a particular embodiment thereof.

DETAILED DESCRIPTION OF THE INVENTION

In this text, the term 11 comprises11 and its derivations (such as 11 comprising 11, etc.) should not be understood in an exclusive sense, that is, these terms should not be interpreted as excluding the possibility that what is described and defined may include more elements, stages, etc.

The method starts from a sequence of tomographic images [n1 nJ ni nk nNl that correspond to sections

000 000

axial section of a human trunk in which the hepatic organ is located. n1 corresponds to the lower cut (that is, the one closest to the feet), while nN corresponds to the upper cut (the one closest to the head).

In Figures 1 and 2, each example image is part of a sequence in which each image has been taken with a resolution of 256x256 pixels, with a distance of 3 millimeters between two consecutive cuts, although the method is perfectly valid for other settings of the input images.

To start the method, the user selects a liver region for each of two nJ and nk reference images. Said manual segmentation does not require great precision, since it is only used to obtain the reference parameters of the method, the final result of the segmentation for said reference images being precisely determined automatically by the method itself in subsequent steps.

For each of the segmented regions, the following is calculated:

-The average intensity of the pixels in said region (mJ and mk).

-

The standard deviation of the intensity of the

pixels of that region (oJ Y Ok).

-

The position of said region, calculated as the centroid of said region ((xJ, y J) and (xk, Yk))

5 The nk image is chosen manually by the user as the image in which, approximately, the section of the hepatic organ has a maximum surface area, while the nJ image is chosen, also manually, of the lower region of the liver, verifying by so much

10 k> j.

Then for each of the N images, a mean, a standard deviation, and a position of the pixels corresponding to the liver organ are estimated. This estimate depends on the position of each image with respect to

15 to relationship images:

m. =

l

ff. =

l

i-j

(mk -m.) - + m.

1 k. 1

-

]

mean (AH)

(J "j

i-j

((J "k - (J".) - + (J ".

1 k. 1

-

]

std (AH)

reference, following the following

isJ J <isk k <i

isJ J <is k k <i

where the std () operator refers to the calculation of the

typical deviation.

(x,., y ,.) =

is J

i-j

(xk -x., yk-y.) - + (x., y.) J <isk

1 1 1 1

k-j

k <i

centroid (AH)

That is, in the area below the first image of

reference (ni, i: -s; j), the values obtained for the area selected manually in said first reference image are used. For ni, with j <i: -s; k, an interpolation of the values obtained for the reference images is performed. Finally, for the images ni, with k <i, values calculated on the segmented area in the previous image are used (Ai-d.

Once the estimated parameters described are available, a thresholding is carried out, that is, a

binarized

the image assigning to each pixel a value or other

depending

from Yes its intensity is bought endida on a

determined interval:

oia <ni (x, y) <ml + oib

Bi (x, y) - {lo

rest

where

hey aloi

and

i <k

{b, ci,

oib =

Do you

b2oi i k To optimize the performance of the method, a1 <b1 is checked. In this way, it is possible to exclude the thoracic wall from the segmentation, separated from the hepatic organ by pixels of less intensity than said hepatic organ. In the images corresponding to the upper slices, the range is restricted by imposing b 2 <b1, to avoid the inclusion in the segmentation of organs such as the heart, present in these images. Specifically, according to a preferred embodiment, the variables a1, b1 and b2 are optimized by assigning them the following values: a1 = 2; b1 = 4; b2 = 3. The method is also valid by varying these values

of the preferred embodiment, being especially advisable to keep them within a 10% variation with respect to said values.

Next, on the binarized image Bi, erosion operators are applied to break the connections between the pixels corresponding to the hepatic organ and other nearby tissues, and hole-filling operators are applied to unify the interior of said hepatic organ.

On the resulting image, the group of adjacent pixels of value 1 containing the pixel of the estimated position for the image being segmented is selected. A dilation operator is applied to said group to compensate for the effect of the erosion operator, segmenting the resulting group of pixels as a liver organ for said image (Ai).

This first segmentation is verified by verifying that the number of segmented pixels is not null, and that its variation with respect to the number of segmented pixels for the previous image is less than a certain threshold, for example, a relative variation of 70% between Ai and Ai

l.

If any of the conditions to be verified are not met, the group of pixels Ai is discarded and an alternative group of pixels A'i is obtained by means of a new segmentation, repeating the process described for the first segmentation with the modifications described below. . First, the value of Bib is reduced, using a new constant b3 less than b 1 • That is, Bib = b3 Bi, where the use of b 3 = 2 allows to achieve an optimal behavior of the method, although the method remains valid varying said value, especially within a 10% margin.

Additionally, the condition that the selected group of pixels must contain the pixel of the position (xi, yd, is relaxed by replacing it with the

condition

from that the group from pixels selected has to

contain

a pixel whose distance to the pixel from the position

(xi,

Yi) be less that a determined threshold.

Yes

the group from pixels resulting from the second

segmentation

Ai verify that contains a number from

pixels

no null and that its variation respect to the number from

Segmented pixels for the above image is less than a certain threshold, it is accepted as segmentation of the liver organ. Otherwise, A'i is discarded, and the same area obtained for the previous image is taken as the segmented area (Ai-d.

When the method has all the segmented areas, it calculates the total volume of the liver by adding the pixel volume associated with each of the segmented pixels, said pixel volume being the result of multiplying the length represented by said pixel in each of the Cartesian axes. These lengths are determined by the resolution of the image along its two axes, and by the distance between two consecutive images.

In Figure 1, the initial, intermediate and final stages of an image subjected to the steps of the method are observed. Image 1 shows the original image, to which the described thresholding is applied, obtaining image 2. Image 3 shows the result of applying erosion to separate the liver organ, and image 4 shows the result of applying the filling of holes. In image 5, the group of pixels corresponding to the liver organ is selected based on its estimated position. This liver organ appears isolated in image 6, in which a dilation is applied, reversing the effect of erosion. The final result appears superimposed on the original image in image 7.

Figure 2 shows an example of the result of applying the preferred embodiment of the method of the invention to a sequence of sample tomographic images. 24 decimated images out of a complete sequence of 52 images are represented. These images show the differences in the sections of the hepatic organ depending on their height, as well as the organs present in each case during the segmentation process, the method being applied to all images in the sequence, not only to those represented here.

Figure 3 shows the precision of the described method by comparing it with a precise segmentation performed manually by two experts in liver radiology. To do this, part of ten sequences of images used to measure said precision, on which the method of the invention is executed. Subsequently, an expert performs the segmentation manually and it is stored in digital format to measure the differences. The figure shows the percentage of the total area represented by the areas that the expert adds with respect to the automatic segmentation (indicated by the bars 8), and the percentage of the total area represented by the areas that the expert removes with respect to the automatic segmentation (indicated through points 9). The difference between the added area and the removed area represents the total difference between the surface measured by the method of the invention and by manual selection.

In view of this description and figures, the person skilled in the art will be able to understand that the invention has been described according to some preferred embodiments thereof, but that multiple variations can be introduced in said preferred embodiments, without departing from the object of the invention such and how it has been claimed.

Claims (11)

l. Segmentation method of a liver organ in a sequence of tomographic images [n1 nJ ni nk nNl 000 000 000 000 of successive axial sections of a human trunk that include axial sections of said hepatic organ, where n1 is the image that includes the axial section corresponding to a lower end of the hepatic organ and nN is the image that includes the axial section corresponding to an upper end of the organ. liver, and each of said images being formed by a set of pixels with an intensity value, obtaining for each image a set of resulting pixels, characterized by comprising the following steps:

-

for a first nJ reference image:

-

manually select a first region within the axial section of the organ

liver in said first reference image;

-

calculating an average mJ of the intensity of the pixels of said first selected region;

-

calculating a standard deviation oJ of the intensity of the pixels of said first selected region;

-

calculate a position (xJ, yJ) of a centroid of said first selected region.

-

for a second reference image nk, where k> j:

-

manually selecting a second region comprised within the axial section of the hepatic organ in said second reference image;

-

calculating an average mk of the intensity of the pixels of said second selected region;

-

calculating a standard deviation ok of the intensity of the pixels of said second selected region; said second selected region.

-

for each image ni, with 1 ~ i ~ N,

-

calculate an estimated mean mi, an estimated standard deviation Bi, an estimated position (xi, yd, a first interval length Bia and a second interval length Bib 'where if j <i ~ k:

-

the estimated mean is calculated by interpolation of the mean calculated for the first reference image (mJ) and the mean calculated for the second reference image (mk);

-

the estimated standard deviation Bi is calculated by interpolation of the calculated standard deviation for the first reference image (oJ) and the calculated standard deviation for the second reference image (ok);

-

the estimated position (xi, yd is calculated by interpolating the position calculated for the first reference image

(xJ, yJ) and the position calculated for the second reference image (xk, Yk)

-

the first wave interval length is calculated by multiplying the estimated standard deviation Bi by a first positive number predefined to 1 •

-

the second interval length Bili is calculated by multiplying the estimated standard deviation Bl by a second predefined positive number b 1 •

- generating a binarized image Bi by assigning a first binary value to the pixels whose intensity is and a second binary value to the pixels whose intensity is outside said interval;

-

selecting in the binarized image Bi a set of adjacent pixels Ai and with said first assigned binary value that comprise the pixel of the estimated position (xi, yd;

-

assign as result pixels for the image nor the set of pixels Ai.

2.

Method according to claim 1 characterized in that it verifies a1 <b1.

3.

Method according to any of the preceding claims characterized in that, in the step of calculating

an estimated mean mi, an estimated standard deviation oi, an estimated position (xi, yd, a first interval length oia and a second interval length oib, if 1 <i :: s; j:

-

the estimated mean ml is assigned the mean calculated for the first reference image

(mJ);

-to

the deviation typical Dear I heard I know assigns

the standard deviation reference image

calculated (oJ); for the first

-a the position estimated position calculated for reference (xJ, y J);

(xi, yd be the first assign image that of

-the first interval length oia is calculated _the second interval length Bili is calculated by multiplying the estimated standard deviation Bl by the second predefined positive number b1. Four . Method according to any of the preceding claims, characterized in that the second reference image nk is an image in which the largest area of the axial section of the hepatic organ is estimated manually from among the axial sections of the hepatic organ of all the images [n1 nJ ni nk nNl. 000 000 000 000 5. Method according to any of the preceding claims, characterized in that it also comprises, for each image ni, with 2 ~ i ~ N, a first verification step to verify: a condition i) if the number of segmented pixels Ai is null; a condition ii) if the absolute value of the difference between the number of segmented pixels Ai

and the predefined previous number;

of is larger pixels result or equal for what the one threshold image

and if I know both:

meets cu rent go from the condition ones i) and ii) or

calculate a new value of second interval length B'ili by multiplying the estimated standard deviation by a third positive number -get a binarized Bi image by assigning a primer within an interval: [mi -a ia, mi + 6'ib J and a second binary value to the pixels whose intensity is outside said range;

-

selecting in Bi a set of adjacent pixels A'i with said first binary value comprising a pixel whose distance to the pixel from the position (xi, yil is less than a predefined threshold;

-

assign as result pixels for the image nor the set of pixels A'i

6. Method according to claim 11 characterized in that, if any of the conditions i) and ii) of the first verification step are met, it also comprises, for each image ni, with 2 ~ i ~ N, a second verification step to verify : a condition iii) if the number of segmented pixels A'i is null; a condition iv) if the absolute value of the difference between the number of segmented pixels A'i and the number of resulting pixels for the previous image ni-1 is greater than or equal to a predefined threshold; and if you meet any of the conditions iii) and iv) or

-

assign as result pixels for the image neither the pixels resulting from the previous image ni-1;

7. Method according to any of the preceding claims, characterized in that it further comprises: -before the step of selecting in the binarized image Bi

a

set from pixels Ai with the first value binary,

a

He passed

from Apply to the image binarized Bi a process from

erosion and

filled in from gaps;

-after

of the He passed from to select on the image binarized

Bi a set of pixels Ai with the first binary value, a step of applying a dilation process to Ai. 8. Method according to any of the preceding claims characterized in that, in the step of calculating

a

half Dear me, a deviation typical Dear I heard,

a

position Dear (xi, and d, a first length from

interval

hey and a second Lenght of interval oib, Yes

k <i :: s; N:

-

the estimated mean ml is calculated as the mean of the intensities of the resulting pixels for the image the previous image ni-l;

-

the estimated standard deviation 6i is calculated as the standard deviation of the resulting pixel intensities for the previous image ni-l image;

-

the estimated position (xi, yd is calculated as the centroid of the resulting pixels for the image the previous image ni-l;

the first interval length oia is calculated by multiplying the estimated standard deviation ol by the first predefined positive number a1; the second interval length calculated by multiplying the estimated standard deviation by a fourth predefined positive number b 2;

9.

Method according to claim 6 characterized in that it verifies b2 <b1.

10.

Method according to any of the claims

5 above characterized in that it also comprises a step of calculating a volume of the hepatic organ by adding a volume associated with each segmented pixel, where said volume associated with a pixel is calculated by multiplying a size of a surface represented by said pixel by 10 a distance between two consecutive axial cuts. 11. Computer program comprising computer program code means adapted to perform the steps of the method according to any of the Claims 1 to 15, when said program is run on a computer, a digital signal processor, an application-specific integrated circuit, a microprocessor, a microcontroller, or any other form of programmable hardware.