RU2713707C2 - Classification of brain tissues - Google Patents

Abstract

FIELD: physics.

SUBSTANCE: invention relates to the computer equipment. Cerebrum classification system comprises an image data interface for accessing a patient's brain image; a processor configured to apply the method of automatically classifying tissues to an image based on a map of a priori probabilities; user interaction subsystem which enables the user to mark a point on the erroneous classification portion to obtain a user-marked point, wherein the subsystem comprises: i) output to display for display of brain tissue classification map on display, ii) input from a user device for receiving input commands from a user device controlled by a user; and processor is configured to: determining the boundary of the erroneous classification portion based on the point marked by the user, correcting the a priori probability map based on feedback from the user to obtain the corrected map of a priori probabilities and repeated application of method of automatic classification of tissues to image based on corrected map of a priori probabilities.

EFFECT: technical result consists in improvement of quality of brain tissue classification.

13 cl, 10 dwg

Description

FIELD OF TECHNOLOGY

[001] The present invention relates to a system and method for classifying brain tissue. In addition, the present invention relates to a workstation or visualization device comprising such a system. In addition, the present invention relates to a computer program product containing instructions for the execution of the method by a processor system.

BACKGROUND

[002] In recent years, the role of imaging in the detection and differential diagnosis of neurodegenerative diseases has increased. One reason is the emergence of quantitative methods to detect mild changes in the brain that develop in the early phase of the disease or even before the clinical manifestation. For example, T1-weighted magnetic resonance imaging (MRI) scans are widely used to evaluate brain atrophy, which is a key indicator of the debut and progression of many neurodegenerative diseases.

[003] Image analysis techniques help quantify cerebral atrophy by classifying brain tissue voxels into various tissue classes, such as gray matter (CB), white matter (BV), and cerebrospinal fluid (CSF). Classification of brain tissue is particularly useful in assessing brain atrophy, since the volume of gray matter serves as a biomarker of cortical atrophy.

[004] However, automatic tissue classification methods sometimes produce erroneous tissue classification maps due to various reasons, such as:

i. Residual bias field (even after bias field correction)

ii. Noise

iii. Artifacts of movement

iv. Low spatial resolution due to low magnetic field voltage

v. Focal lesions

[005] As a result, as a rule, on the tissue classification map, two types of tissue classification errors can occur, namely, isolated erroneous classifications such as “spots”, as well as excessive or insufficient display of the cortical gray matter near its border with the white matter.

[006] The disadvantage is that such areas of erroneous classification on the tissue classification map may impede further detection and differential diagnosis of neurodegenerative diseases.

[007] US2002 / 0186882 A1 discloses a method and corresponding device for obtaining quantitative data on a 2-dimensional, 3-dimensional image or image of a different dimension, for example, for classifying and counting the number of elements that are contained in an image. Each element contains an element, structure, or some other type of identifiable part of the image having defined characteristics. Elements located within the image may have a different shape, color, texture or other defined characteristic, but at the same time belong to the same classification. In other cases, elements having similar color and texture can be classified as one type, while elements having different color and texture can be classified as another type. An image can contain many elements, while each element can belong to a different class. Thus, the system can determine the amount of image data in accordance with a set of changing criteria and obtain one or more classifications of elements in the image. After classifying the image data, the total number of elements in the image is calculated and given to the user. In embodiments, a method is provided by a computer to determine what kind of element (for example, elements) is present in the image, and to count the total number of elements that can be visually identified in the image. Information used during the training process can be stored and applied to various images.

SUMMARY OF THE INVENTION

[008] It is preferable to provide a system or method for the classification of brain tissue, providing the ability to correct areas of erroneous classification.

[009] In the following aspects of the present invention, it is contemplated that the user interactively provides feedback in the erroneous classification portion of the tissue classification map, wherein feedback from the user is indicative of correcting the erroneous classification. Instead of directly correcting the tissue classification map based on user feedback, user feedback is used to adjust the a priori probability map, which is used as input to the automatic tissue classification method to produce an adjusted a priori probability map. After that, the automatic tissue classification method is re-launched based on an adjusted map of a priori probabilities.

[0010] In a first aspect of the present invention, there is provided a system for segmenting an anatomical structure, comprising:

- an image data interface for accessing an image of a patient’s brain;

- a processor configured to use the automatic tissue classification method to an image based on a map of a priori probabilities,

a priori probability map is recorded in the image and is indicative of the probability of a specific location in the brain belonging to a particular class of brain tissue, and

the output obtained by the automatic tissue classification method is a map of the classification of tissue of the brain of the patient;

- a user interaction subsystem that allows the user to mark a point in the erroneous classification area to obtain a point marked by the user, while this subsystem contains:

i) access to a display for displaying a brain tissue classification map on a display,

ii) input from a user device for receiving input commands from a user device controlled by a user,

moreover, the input commands represent user feedback, which is indicative of a) a section of an erroneous classification on the tissue classification map and b) correction of an erroneous classification; a

the processor is configured to:

j) determining the boundary of the erroneous classification portion based on a point marked by the user,

jj) adjusting the map of a priori probabilities based on user feedback to obtain an adjusted map of a priori probabilities, and

jjj) reapplying the automatic tissue classification method to an image based on an adjusted map of a priori probabilities.

[0011] In another aspect of the present invention, there is provided a workstation or visualization apparatus comprising a system.

[0012] In another aspect of the present invention, a method for classifying brain tissue is provided, comprising:

- access to the image of the patient’s brain;

- application of the automatic tissue classification method to the image based on a map of a priori probabilities, while

a priori probability map is recorded in the image and is indicative of the probability of a specific location in the brain belonging to a particular class of brain tissue, and

the output obtained by the automatic tissue classification method is a map of the classification of tissue of the brain of the patient;

- providing the user with the opportunity to mark a point in the area of erroneous classification with getting marked by the user points;

- display of the brain tissue classification map on the display;

- receiving input commands from a user device controlled by the user,

moreover, the input commands represent user feedback, which is indicative of a) a section of an erroneous classification on the tissue classification map and b) correction of an erroneous classification;

- determination of the boundary of the erroneous classification area based on the point marked by the user;

- adjustment of the map of a priori probabilities based on feedback from the user with the receipt of an adjusted map of a priori probabilities and

- re-application of the automatic tissue classification method to the image based on the adjusted map of a priori probabilities.

[0013] In yet another aspect of the present invention, there is provided a computer program product comprising instructions for performing a method on a processor system.

[0014] The above measures include accessing an image of a patient's brain. Thus, the image can be a scanned image of the brain and can be obtained by various imaging methods, including, without limitation, T1-weighted magnetic resonance imaging (MRI). The automatic tissue classification method is applied to an image based on a map of a priori probabilities. Such a priori probability maps are known per se and can describe the probability of a known position in the brain belonging to one of various classes of tissues. Typically, these a priori probability maps were created from a cohort of samples of correctly classified scanned images of the brain.

Automatic tissue classification methods that use a priori probability maps are also known per se, for example, from the field of medical image analysis. A priori probability map is registered with the image compared, for example, by a method that is known per se in the field of medical image registration. As a result of automatic tissue classification, a tissue classification map is obtained that locally classifies the brain according to the type of brain tissue.

[0015] It should be noted that the term “brain tissue classification” is used interchangeably with “brain tissue segmentation” in this document, since the resulting tissue classification map segments the brain into various types of tissues and thus provides segmentation .

[0016] After receiving the tissue classification map, the tissue classification map is displayed and the user can provide feedback from the user that is indicative of the erroneous classification portion of the tissue classification map and which is indicative of correcting the erroneous classification. As a result, the user provides feedback from the user, which is indicative of where the erroneous classification took place and what the correction should be. For example, user feedback may indicate an area to be biased towards the white matter.

[0017] Thereafter, the a priori probability map is adjusted based on feedback from the user, resulting in an adjusted a priori probability map that contains one or more local probability corrections. After that, the automatic tissue classification method is restarted based on the adjusted map of a priori probabilities, resulting in an additional tissue classification map.

[0018] The above measures have the effect that the user can provide feedback from the user that is indicative of where the erroneous classification took place and what the correction should be. Instead of directly correcting the tissue classification map based on user feedback, user feedback is used to adjust the a priori probability map, and then the automatic tissue classification technology is re-launched based on the adjusted a priori probability map.

[0019] The authors of the present invention have revealed that if the automatic tissue classification method produces an erroneous tissue classification map, it may be preferable, oddly enough, to customize the input data of the automatic tissue classification technology, namely, a priori probability maps, instead of directly correcting its output, namely the tissue classification card. The reason for this is that user feedback may not be accurate enough to directly correct tissue classification maps. Namely, such a direct correction may require, for example, an accurate delineation of the area of erroneous classification, an exact indication of the type of brain tissue, etc. Such accurate user feedback may not be available or may unduly burden the user. As a result, the intention of the authors of the present invention is to adjust the map of a priori probabilities and restart the method of automatic classification of tissues in the whole image using a corrected map of a priori probabilities. This provides a degree of indirection in the sense that user feedback is used to adjust probabilities, instead of direct classification.

[0020] Advantageously, the user is not required to indicate the area of the erroneous classification and / or correction of the erroneous classification with the highest degree of accuracy. Instead, it may be sufficient to indicate it approximately. In contrast, by setting the degree of accuracy for user feedback, a more accurate tissue classification map can be obtained compared to directly correcting the tissue classification map. Another advantage is that the classification of tissues anywhere in the brain can also be improved, since the difference between classes of tissues can be better modeled by the presence of more “controlled” evidence.

[0021] Furthermore, in one embodiment of the invention

- the user interaction subsystem allows the user to mark a point in the erroneous classification area to obtain a point marked by the user;

- the processor is configured to determine the boundary of the erroneous classification portion based on a point marked by the user.

[0022] Thus, the user can be sufficiently marked with a point that lies within the area of the erroneous classification. Nevertheless, such a point marked by the user can provide the system with the ability to determine (the whole) area of the erroneous classification, namely by using the method of determining the boundary. For example, the processor can consider the user-marked point as a starting point in the method of growing areas, thereby obtaining the boundary of the erroneous classification portion. Alternatively, other boundary determination methods that are known in the field of medical image analysis can be used, including, but not limited to, analysis of related components and methods based on morphological operations.

[0023] In one embodiment of the invention, the user interaction subsystem allows the user to note the correction of an erroneous classification by manually specifying the brain tissue class to obtain the user specified brain tissue class. It may be that the user is able to directly determine the class of brain tissue in the area of erroneous classification. The user has the opportunity to give such feedback from the user, namely by directly indicating the class of brain tissue.

[0024] In one embodiment of the invention, the processor is configured to adjust the map of a priori probabilities by increasing, in the map a priori probabilities, the likelihood that a user-specified class of brain tissue is in the area of erroneous classification. Based on the user's direct indication of the class of brain tissue, the probability of the specified class of brain tissue in the map of a priori probabilities within the area of the erroneous classification can be increased. For example, the probability can be increased to 80% or higher, 90% or higher, 95% or higher, or essentially up to 100%, for example, up to 99% or higher.

[0025] In one embodiment, the user interaction subsystem allows the user to note the correction of an erroneous classification by changing the probability ratio between the gray matter and the white matter. It has been found that changing the probability ratio between the gray matter and the white matter is a particularly advantageous way of providing feedback from the user in case of excessive or insufficient display of the cortical gray matter near its border with the white matter. For example, a user may be able to increment the probability relation, for example, by dragging the mouse up or down with the left mouse button, by controlling specific keys on the keyboard (for example, plus and minus keys), etc.

[0026] In one embodiment, a user interaction subsystem allows a user to indicate an erroneous classification portion on a tissue classification map displayed on a display. Thus, the user can indicate a specific area of the erroneous classification, as it happens on the tissue classification map, on the tissue classification map itself. For example, a user can use the annotation tool to draw a contour on a displayed tissue classification map.

[0027] In one embodiment, the user interaction subsystem is configured to:

- display the image on the display and

- allows the user to mark the area of the erroneous classification on the tissue classification map by indicating the area of interest in the image.

[0028] As an alternative to indicating the area of the erroneous classification, as this happens on the tissue classification map, on the tissue classification map itself, the user can mark the area on the image that can be displayed simultaneously with the tissue classification map. For example, a user can use the annotation tool to outline a path in a displayed image.

[0029] In one embodiment, an automatic tissue classification method is based on maximizing expectations. Recently, it has been demonstrated in academic literature that automatic tissue classification methods based on maximizing expectations, in combination with regularizing Markov random fields, provide the best overall performance. However, other automatic tissue classification methods that use a priori probability maps can also be used.

[0030] Those skilled in the art will appreciate that two or more of the above embodiments, embodiments, and / or optional aspects of the present invention can be combined in any way that seems appropriate.

[0031] Modifications and variations of the imaging device, workstation, method and / or computer program product that correspond to the described modifications and variations of the system can be carried out by a person skilled in the art based on the present description.

[0032] One skilled in the art will understand that the method can be applied to multidimensional image data, for example, two-dimensional (2D), three-dimensional (3D), or four-dimensional (4D) images obtained by various acquisition means, such as , without limitation, standard X-ray imaging, computed tomography (CT), magnetic resonance imaging (MRI), ultrasound (ultrasound), positron emission tomography (PET), single-photon emission computed tomography (SPECT) and nuclear medicine (NM).

BRIEF DESCRIPTION OF THE DRAWINGS

[0033] These and other aspects of the present invention will become apparent and will be explained with reference to the embodiments described by way of example in the description below and with reference to the accompanying drawings, in which:

in FIG. 1 shows a system for classifying brain tissue in which the a priori probability map is adjusted based on user feedback and the adjusted a priori probability map is used to reapply the automatic tissue classification method to the brain image;

in FIG. 2 shows part of an MRI image, and

in FIG. 2B shows a tissue classification map obtained by automatically classifying tissue as an MRI image, wherein the tissue classification map contains an erroneous classification in the form of an area like a “spot” that was erroneously classified as a cortical gray matter, and not as a white matter;

in FIG. 3A shows a portion of another MRI image, and

in FIG. 3B shows a tissue classification map obtained by automatically classifying tissue images of an MRI image, wherein the tissue classification map contains an erroneous classification in the form of a cortical gray substance overexposed to a border with a white substance;

in FIG. 4A depicts a user providing user feedback that is indicative of an erroneous classification portion by indicating a point in the erroneous classification portion;

in FIG. 4B shows the result of an erroneous classification of a site boundary, automatically determined based on a user-specified point;

in FIG. 4C shows the result of a user indicating correcting an erroneous classification by manually specifying the class of brain tissue;

in FIG. 5 shows a method for classifying brain tissue, in which the a priori probability map is adjusted based on user feedback and the adjusted a priori probability map is used to reapply the automatic tissue classification method to the brain image; and

in FIG. 6 depicts a computer-readable storage medium containing instructions for conditioning the execution of a method by a processor system.

It should be noted that the figures are only schematic and not scalable. Elements in the figures that correspond to the elements already described have the same reference designations.

Reference List

The following list of reference signs is provided to simplify the interpretation of the drawings and should not be construed as limiting the claims.

020 image storage

022 medical image data

024, 026 medical image

030, 032, 034A, 034B tissue classification map

036, 038 classification error indication

040 database

042 data representing a priori probability map

060 display

062 display data

064 screen cursor

070 user specified point

072 segmentation

080 user device

082 input commands

100 system for the classification of brain tissue

120 image data interface

140 a priori probability data interface

160 processor

162 data transfer to or from a user interaction subsystem

180 user interaction subsystem

182 display output

184 user device login

200 way to classify brain tissue

210 access to the image of the brain

220 application of automatic tissue classification

230 map display tissue classification

240 receiving user feedback

250 correction of a priori probability map

260 reuse of automatic tissue classification

270 computer readable storage medium

280 instructions stored as short-term data

DETAILED DESCRIPTION OF THE INVENTION

[0034] FIG. 1 shows a system 100 for classifying brain tissue. Such a system can be used in various medical applications, including, but without limitation, in the detection and differential diagnosis of neurodegenerative diseases. During operation, system 100 may, in short, apply the automatic tissue classification method to the brain image based on a priori probability map to obtain a brain tissue classification map. After that, the user has the opportunity, with the help of the user interaction subsystem, to issue feedback signals that are indicative of a) an area of erroneous classification on the tissue classification map and b) correction of erroneous classification. After that, the system can adjust the map of a priori probabilities based on user feedback to obtain a corrected map of a priori probabilities, and reapply the technology of automatic tissue classification to the image based on the adjusted map of a priori probabilities.

[0035] The system 100 comprises an image data interface 120 for accessing a patient's brain image 022, hereinafter referred to as a brain image 022 for simplicity. In the example shown in FIG. 1, an image data interface 120 is shown connected to an external image storage 020 that contains a brain image 022. For example, image storage 020 may be or may be part of an image archiving and transmission system (SAPI) of a hospital information system (LSI) with which system 100 may be connected or in which it may be contained. Therefore, system 100 can access brain image 022. Alternatively, access to a brain image 022 may be obtained from an internal data store contained in system 100. In general, the image data interface 120 may take various forms, such as a network interface for a local or global computer network, for example, a network Internet, storage interface for internal or external data storage, etc. It should also be noted that, where applicable, a link to image 022 of the brain should be understood as a reference to image data of the brain.

[0036] In addition, the system 100 includes a processor 160 configured to apply an automatic tissue classification method to a brain image 022 based on a prior probability map 042. To this end, it has been shown that the processor 160 receives a brain image 022 from the image data interface 120, as well as an a priori probability map 040 from the a priori data interface 140. The a priori probability data interface 140 may provide the system 100 with the ability to access the a priori probability map 042 in an external database 040, such as SAPI. Alternatively, system 100 may access the prior probability map 042 internally or from another source. As with the image data interface 120, the a priori probability data interface 140 can take various forms, including but not limited to a network interface for a local or global computer network, for example, the Internet, a storage interface for internal or external storage data etc. After that, as an output, a map 162 of the classification of tissue of the brain of the patient is obtained.

[0037] In addition, the system 100 includes a display output 182 for displaying a visual output of the system 100 on a display 060 with visual output including at least a tissue classification map. To display the tissue classification map, the output 182 on the display is shown communicating internally with the processor 162, for example, to obtain data visualizing the tissue classification map, as well as to provide display data 062 to the display 060.

[0038] In addition, the system 100 comprises an input 184 for a user device for receiving input commands 082 from a user device 080 that the user controls. User device 080 can take various forms, including, but not limited to, the shape of a computer mouse 080, touch screen, keyboard, etc. The input 184 from the user device can be any type of device corresponding to that of the user device 080. The output 182 to the display and the input 184 from the user device together can form a user interaction subsystem 180 that enables the user to interactively issue feedback signals from the user to the system 100 In particular, feedback from the user may be indicative of a) a section of the erroneous classification on the tissue classification map, and b) a systematic way of misclassification. As a non-limiting example, a user can click on a portion of a tissue classification map that has not been correctly classified and select the correct classification from the on-screen menu. After that, user feedback may be available to the system 100 in the form of user feedback information indicating, for example, the coordinate (s) of the erroneous classification in the coordinate system associated with the tissue classification map, and data indicative of correction.

[0039] After receiving feedback from the user from input 184 from the user device, processor 160 may adjust the a priori probability map based on user feedback to obtain an adjusted a priori probability map, and then reapply the automatic tissue classification method to the image based on the adjusted map a priori probabilities.

[0040] It should be noted that various options for the operation of the system 100, including its various operational aspects, will be explained in more detail with reference to FIG. 2-4C.

[0041] The system 100 may be implemented as, or in, a stand-alone device or apparatus, such as a workstation or imaging device. A device or apparatus may contain one or more microprocessors that run the corresponding software. The software may be downloaded and / or stored in appropriate memory, for example, a volatile memory device, such as RAM, or a non-volatile memory device, such as a flash drive. Alternatively, the functional blocks of the system can be implemented in a device or apparatus in the form of a programmable logic circuit, for example, as a user-programmable gate array (PPM). In general, each functional block of the system can be implemented in the form of a circuit. It should be noted that the system 100 can also be implemented in a common manner, for example, involving various devices or apparatuses. For example, distribution may be performed according to a client-server model.

[0042] FIG. 2A-3B depict two general types of tissue classification errors, namely, isolated erroneous classifications such as “spots” (Fig. 2A-2B), as well as insufficient or excessive display of cortical gray matter near its border with white matter (Fig. 3A-3B )

[0043] FIG. 2A shows part of the 024 MR image, and FIG. 2B shows a tissue classification map 030 obtained by automatic tissue classification of MRI images. In FIG. 2B, blue (black in reproduction in gray space) indicates cerebrospinal fluid (CSF), green (dark gray in reproduction in gray space) indicates gray matter (CB), and pink (light gray in reproduction in gray space) denotes white matter (BV). As indicated in FIG. 2B at 036, tissue classification map 030 contains an erroneous classification in the form of a spot-like region that was incorrectly classified as a cortical gray matter, and not as a white matter.

[0044] FIG. 3A shows a portion of another MRI image 026, and FIG. 3B shows a tissue classification map 032 obtained by automatic tissue classification of MRI images. The same color coding (or gray space coding) is used as in FIG. 2B. As indicated in FIG. 3B at 038, tissue classification map 032 depicted in FIG. 3B contains an erroneous classification in the form of a cortical gray matter that was not sufficiently displayed at the border with the white matter.

[0045] The above classification errors can occur frequently, for example, for reasons indicated in the Background section, thereby giving erroneous tissue classification maps.

[0046] As indicated with reference to FIG. 1, the user may provide the system with user feedback that is indicative of a) a section of the erroneous classification on the tissue classification map and b) correction of the erroneous classification. Such user feedback can take many forms.

[0047] FIG. 4A depicts a specific example of such user feedback, namely, the user indicating the point 070 in the erroneous classification portion. An example is based on an erroneous classification like the spot shown in FIG. 2B, wherein in FIG. 4A-4B show an enlarged portion of the corresponding tissue classification card 034A. The user can mark point 070 in various ways. For example, the user can click on the position on the displayed tissue classification map 034A using the on-screen cursor 064. After that, as a result of the point 070 marked by the user in the erroneous classification area, the boundary of the erroneous classification area can be determined, for example, using the border determination method. For example, a user-marked point 070 can be considered as a starting point in the method of growing areas, thereby obtaining the boundary of the area of erroneous classification. Alternatively, other boundary determination methods that are known per se in the field of medical image analysis can be used, including, but not limited to, analysis of related components and methods based on morphological operations. The resulting boundary 072 is indicated in FIG. 4B with a dashed line outlining an erroneous classification like a spot on tissue classification map 034A from its surrounding parts.

[0048] FIG. 4C shows the result of an additional indication by the user of the correction of an erroneous classification. Namely, the user can specify the class of brain tissue to be applied to the erroneously classified region 072 manually. As a non-limiting example, the user can select the correct class of brain tissue from the on-screen menu (the menu is not shown in Fig. 4C). The correction can be shown to the user, namely, by means of an erroneously classified area 072 corrected on the displayed tissue classification map 034A according to the user specified tissue class of the brain. As shown in FIG. 4C, the site can be classified as white matter, and not as gray matter.

[0049] After providing feedback from the user regarding a) an erroneous classification on a tissue classification map and b) correcting an erroneous classification, an a priori probability map may be configured to obtain an adjusted a priori probability map. It should be noted that the initial and corrected a priori probability maps are not explicitly shown because such a priori probability maps are difficult to properly visualize, since each location on the map usually has probability values for each class of brain tissue, making it difficult, therefore gray-scale visualization or even color-based visualization. However, in essence, the adjustment may take a form similar to that shown for the tissue classification card 034A, 034B in FIG. 4B and FIG. 4C, in that the corresponding region on the map of a priori probabilities can be adjusted by increasing the probability of a user specified class of brain tissue in the area of erroneous classification to essentially 100%, effectively “overwriting” existing values of probabilities. After adjusting the map of a priori probabilities, the method of automatic tissue classification can be reapplied to the image, resulting in another, usually (much more) improved tissue classification map.

[0050] In general, in addition to specifying a point, other options may be available to the user to indicate the area of the erroneous classification on the tissue classification map. For example, a user can directly outline an erroneously classified region on a tissue classification map. Another example is that a user may mark a point in an erroneously classified area, the system then accepting the erroneously classified area as a predetermined area around the point.

[0051] It should also be noted that the erroneously classified area that occurs on the tissue classification map can be marked by the user not on the tissue classification map, but elsewhere. For example, the user can study the area of the erroneous classification after reviewing the displayed tissue classification map, but after that it can mark the indicated area to the system in the image itself, for example, by drawing a rectangle, thereby marking the area on the image.

[0052] In addition, there are various options for the user to indicate correction of an erroneous classification. The first example is the above direct reference to the correct class of brain tissue. Another example is that instead, the user can change the probability ratio between classes of brain tissue, such as the probability ratio between gray tissue and white matter. As a result, a correction is not a binary correction of a class, but a correction of probability. This ratio of probabilities can be changed gradually, for example, in stages. The system may reapply automatic tissue classification technology to the image after each gradual change, or it may not. Another example is that a system can automatically propagate the probabilities surrounding an erroneously classified site to an erroneously classified site. As a result, an indication of an erroneously classified portion is thus also effectively used to indicate correction of an erroneous classification, since it is assumed that it follows from the parts surrounding it.

[0053] In accordance with the above, an example use case may be as follows. Here, the user can point and click on a location on an erroneous classified area like a “spot” and assign the area the correct label, for example, through a pop-up window. This action can activate the following operations:

1) the boundary of the erroneously classified region can be automatically determined based on local analysis of voxels in the labeled image, for example, by analyzing related components, morphological operations or building up areas using a segmented image label,

2) the assignment of a fixed label to voxels in the region in the form of a 100% probability of voxels in the region belonging to the selected tissue class, and

3) automatic optimization of the overall tissue classification result by restarting the automatic tissue classification algorithm, but now using the map of a priori probabilities with local modifications.

[0054] Another example use case may be as follows. Here, the user can mark the studied area and continuously change the probability of its belonging to a particular class of brain tissue in an interactive way. Immediately after satisfaction, the automatic tissue classification algorithm can be re-run. This example use case may include the following operations:

1) The user can mark the studied area with the help of a tool for constructing the image contour. For example, a user can draw an outline or use an annotation tool with a specific shape, such as a circle or rectangle. Since the scanned image of the brain can be a 3D scanned image and the user can be shown a 2D slice of a 3D scanned image, the area can be automatically expanded / extended to 2D slices of images in front of and behind the displayed slice of the image using known technologies, such as a 3D tool spherical annotations instead of a circle.

2) Immediately after marking the study area, the user can start an interactive operation that increases / decreases the values of a priori probability for gray matter compared to white matter. Changes in the probability value can be displayed using, for example, specific keys on the keyboard (+/-), by holding down the left mouse button and moving the mouse up / down, etc.

3) The restart of the automatic tissue classification algorithm, but now using the map of a priori probabilities with local modifications.

[0055] It should be understood that there are a number of approaches for the automatic classification of tissues. At the time of the creation of the present invention, it was demonstrated that algorithms based on maximizing expectations (MO) and regularizing Markov random fields give the best overall performance (see http://mrbrains13.isi.uu.nl/results.php for results and related articles ) However, other automatic tissue classification methods that use a priori probability maps can also be used. Also referring to approaches based on MO: maximizing expectations uses a priori probability maps that describe the probability of a known position in the brain belonging to one of the possible classes of brain tissue. Typically, these a priori probability maps are created from a cohort of samples of correctly segmented scanned images of the brain that are aligned in space, i.e. registered. For a new subject, a priori probability maps can be recorded in a scanned image of the brain at the pre-processing stage. After that, the algorithm for maximizing expectations can iteratively perform two stages:

(I) the so-called M-stage, in which, for given probabilities of a tissue class for each voxel, an intensity model for a particular class of brain tissue can be determined / updated, and

(ii) the so-called E-stage, in which, with an (updated) intensity model for a particular class of brain tissue, peroxel probabilities can be detailed.

[0056] It should also be noted that the adjustment of the probability map may not be necessary or preferred. Instead, a parameter or configuration of an automatic tissue classification method may be configured. A specific example is that the erroneously classified portion can be corrected by the user, for example, in an interactive manner, as indicated by reference to FIG. 4A-4C. After that, the method of automatic tissue classification can be re-launched in the study area, which does not include the user specified erroneously classified area. For example, a user-specified erroneously classified region may be excluded from the binary voxel mask that defines the region of interest. After that, the final tissue classification map can be compiled from the results of the automatic tissue classification method in the study area, together with the erroneously classified area, which is interactively corrected by the user. It should be understood that in addition to the above example of eliminating a manually corrected area from the study area, other adjustments to the parameters or configurations of the automatic tissue classification method are available to a person skilled in the art.

[0057] FIG. 5 shows a method 200 for classifying brain tissue. In an operation called “GETTING ACCESS TO THE BRAIN IMAGE”, method 200 includes obtaining 210 access to an image of a patient’s brain. In addition, for an operation called “APPLICATION OF AUTOMATIC CLASSIFICATION OF TISSUES”, method 200 includes applying 220 methods of automatic classification of tissues to an image based on a map of a priori probabilities, while a map of a priori probabilities is recorded in the image and is indicative of the probability of a specific location in the brain, belonging to a specific class of brain tissue, while the output from the automatic tissue classification method is artifact for the classification of brain tissue of the patient. In addition, in an operation called “DISPLAYING TISSUE CLASSIFICATION CARD”, the method 200 includes displaying a 230 tissue classification map on a display. In addition, in an operation called “USER FEEDBACK”, method 200 includes receiving 240 input commands from a user device controlled by the user, the input commands representing feedback from the user, which is indicative of a) an erroneous classification portion of tissue classification map; and b) correction of erroneous classification. In addition, in an operation called “priori probability map correction”, method 200 includes adjusting 250 a priori probability map based on user feedback to obtain an adjusted a priori probability map. In addition, for an operation called “RE-APPLYING AUTOMATIC CLASSIFICATION OF TISSUES”, method 200 involves reapplying the 260 automatic tissue classification method to an image based on an adjusted map of a priori probabilities.

[0058] It should be understood that the above operations can be performed in any suitable order, for example, sequentially, simultaneously or in combination, observing, where applicable, the specific necessary order, for example, using input-output ratios.

[0059] As a non-limiting example, in FIG. 5, an iterative method 200 is shown, namely, the operations of applying 220 and reapplying 260, indicated by arrow 255, are essentially similar, albeit performed at a later stage and using other input data, namely, other a priori probability maps.

[0060] Method 200 may be implemented on a computer as a computer-implemented method, in the form of special-purpose hardware, or in a combination of the two. As also shown in FIG. 6, instructions for a computer, for example, executable code, may be stored on a computer-readable storage medium 270, for example, in the form of a series of 280 machine-readable physical marks and / or in the form of a series of elements having various electrical, e.g. magnetic, or optical properties or values. The executable code may be stored in a short-term or non-short-term manner. Examples of computer-readable storage media include storage devices, optical storage devices, integrated circuits, servers, online software, etc. In FIG. 6 shows an optical disc 270.

[0061] It should be understood that although the claimed invention has been described with reference to the classification of brain tissue, the claimed invention can equally be applied to another type of classification in which a priori probability maps are used as input.

[0062] Examples, embodiments, or optional features, whether they are indicated as non-limiting or not, should not be construed as limiting the claimed invention.

[0063] It should be understood that the present invention is also applicable to computer programs, in particular computer programs on or in a medium configured to put the present invention into practice. The program may be in the form of source code, object code, an intermediate source of code and object code, as in a partially compiled form, or in any other form suitable for use in implementing the method in accordance with the present invention. It should also be understood that such a program can have many different architectural designs. For example, program code that implements the functionality of a method or system in accordance with the present invention may be subdivided into one or more subprograms. The person skilled in the art will understand many different ways of distributing functionality for these routines. Subprograms can be stored together in one executable file to form an independent program. Such an executable file may comprise computer-executable instructions, for example, processor instructions and / or interpreter instructions (for example, Java interpreter instructions). Alternatively, one or more or all of the subprograms may be stored in at least one file of the external library and are associated with the main program in a static or dynamic manner, for example, by counting time. The main program contains at least one call of at least one of the subprograms. In addition, routines may contain each other's function calls. An embodiment related to a computer program product comprises computer-executable instructions corresponding to each processing step of at least one of the methods described herein. These instructions may be subdivided into subroutines and / or may be stored in one or more files, which may be linked in a static or dynamic manner. Another embodiment related to a computer program product comprises computer-executable instructions corresponding to each means of at least one of the systems and / or products described herein. These instructions may be subdivided into subroutines and / or may be stored in one or more files, which may be linked in a static or dynamic manner.

[0064] A computer program medium may be an object or device configured to store a program. For example, the medium may include a data storage medium, such as a ROM, for example, a compact disc, or a semiconductor ROM, or a magnetic storage medium, for example, a hard disk. In addition, the medium may be a transmission medium, such as an electrical or optical signal, which can be transmitted by electric or optical cable or by radio or other means. When implementing the program in the form of such a signal, the medium may be such a cable or other device or means. Alternatively, the medium may be an integrated circuit in which the program is embedded, wherein the integrated circuit is configured to implement the relevant method or used in its execution.

[0065] It should be noted that the above embodiments illustrate and do not limit the present invention, and one skilled in the art will be able to implement a wide range of alternative embodiments without departing from the scope of the attached claims. In the claims, any reference signs enclosed in parentheses should not be considered a limitation of the claim. The use of the verb “contains” and its conjugations does not exclude the presence of elements or stages other than those specified in the claims. The grammatical exponent of the singular in front of an element does not exclude the presence of many such elements. The present invention can be implemented through hardware containing several different elements, as well as through a suitably programmed computer. In a claim on a device in which several means are listed, several of these means may be implemented by the same hardware element. The mere fact that some measures are listed in mutually different dependent clauses does not mean that a combination of these measures cannot be successfully used.

Claims (35)

1. The system (100) for the classification of brain tissue, containing: - an image data interface (120) for accessing an image (022) of a patient’s brain; - a processor (160) configured to apply an automatic tissue classification method to an image based on a map of (042) a priori probabilities, wherein a map of a priori probabilities is registered in the image and is indicative of the probability of a specific location in the brain belonging to a particular class of brain tissue, and the output obtained by the automatic tissue classification method is a map of the classification of brain tissue of the patient; - a subsystem (180) for interacting with the user, allowing the user to mark a point in the area of erroneous classification to obtain a point marked by the user, while the specified subsystem contains: i) an output (182) to a display for displaying a brain tissue classification map on a display (060), ii) an input (184) from a user device for receiving input commands (082) from a user device (080) controlled by a user, moreover, the input commands represent user feedback, which is indicative of a) a section of an erroneous classification on the tissue classification map and b) correction of an erroneous classification; and the processor (160) is configured to: - determining the boundary of the erroneous classification portion based on a point marked by the user, - adjusting the map of a priori probabilities on the basis of feedback from the user to obtain an adjusted map of a priori probabilities and - reapplying the automatic tissue classification method to the image based on the adjusted map of a priori probabilities. 2. The system (100) according to claim 1, wherein the user interaction subsystem (180) allows the user to note the correction of the erroneous classification by indicating the class of brain tissue manually to obtain the user specified class of brain tissue. 3. The system (100) according to claim 2, in which the processor (160) is configured to adjust the map of a priori probabilities by increasing, in the map a priori probabilities, the probability that the user specified class of brain tissue is in the area of erroneous classification. 4. The system (100) according to claim 3, in which the processor (160) is configured to increase, in the map of a priori probabilities, the probability that the user specified class of brain tissue is in the area of erroneous classification, essentially up to 100%. 5. The system (100) according to claim 1, in which the subsystem (180) of the user interface allows the user to note the correction of the erroneous classification by changing the probability ratio between the gray matter and the white matter. 6. System (100) according to claim 5, in which the subsystem (180) of the user interface provides the user with the ability to change the probability ratio in increments. 7. System (100) according to any one of paragraphs. 1-6, in which the subsystem (180) of the user interface allows the user to mark the area of the erroneous classification on the tissue classification map displayed on the display (060). 8. System (100) according to any one of paragraphs. 1-6, in which the subsystem (180) of the user interface is configured to: - display the image on the display (060) and - allows the user to mark the area of the erroneous classification on the tissue classification map by indicating the area of interest in the image. 9. System (100) according to any one of the above paragraphs, in which the automatic tissue classification method is based on maximizing expectations. 10. A workstation containing a system according to any one of the above items. 11. Device for visualization, comprising a system according to any one of paragraphs. 1-9. 12. A method (200) for classifying brain tissue, including: - obtaining (210) access to the image of the patient’s brain; - application (220) of the automatic tissue classification method to an image based on a map of a priori probabilities, while a map of a priori probabilities is registered in the image and is indicative of the probability of a specific location in the brain belonging to a particular class of brain tissue, and the output obtained by the automatic tissue classification method is a map of the classification of brain tissue of the patient; - providing the user with the opportunity to mark a point in the area of erroneous classification with getting marked by the user points; - display (230) maps of the classification of brain tissue on the display; - receiving (240) input commands from a user device controlled by a user, wherein the input commands are user feedback, which is indicative of i) an erroneous classification portion of the tissue classification map; and ii) correction of the erroneous classification; - determination of the boundary of the erroneous classification area based on the point marked by the user; - adjustment (250) of the map of a priori probabilities based on feedback from the user with the receipt of a corrected map of a priori probabilities and - repeated application (260) of the automatic tissue classification method to the image based on the adjusted map of a priori probabilities. 13. A computer-readable medium (270) of information containing instructions for the execution of the method according to claim 12 by the processor system.

Images