Method and device for segmenting a medical examination object using quantitative MR imaging methods

Abstract

Method for segmenting image data of an object under investigation, with the following steps: - Recording raw data of the object under investigation; - Determining initial quantitative image data of the object under investigation, in which a first physical quantity is quantitatively determined and represented pixel by pixel, - Segmenting the initial quantitative image data, - Determining a first quality parameter for the segmented first image data, which contains information about how reliably the predetermined objects can be identified in the segmented first image data, - Determining second quantitative image data of the object under investigation, in which a second physical quantity is quantitatively determined and displayed pixel by pixel, - Segmenting the second quantitative image data, - Determining a second quality parameter for the segmented second image data, which contains information about how reliably the predetermined objects can be identified in the segmented second image data, - Comparing the segmented first quantitative image data with the segmented second quantitative image data, whereby in an area where the identified objects in the segmented first quantitative image data do not match the identified objects in the segmented second quantitative image data, those quantitative image data with the greater associated quality parameter are used to identify the predetermined objects.

Description

[0001] The invention relates to a method and a device for segmenting a medical examination object by applying quantitative MR imaging methods for medical questions and segmentations.

[0002] In MRI in general, and in orthopedic MRI in particular, the importance of segmentation—that is, the visualization of specific sub-areas of objects under investigation—has increased significantly in recent years. One application is the segmentation of cartilage tissue and the subsequent automated evaluation of thickness, volume, etc., to detect potential cartilage damage as early as possible and to objectively quantify the success of treatment or the progression of diseases (e.g., osteoarthritis). Other areas include, for example, the segmentation of tumors or vascular deposits. Segmentation techniques are also used for bone or fluid visualization, or for MR-PET attenuation correction. Current segmentation algorithms for structures with complex geometries, such as...Joints are often computationally intensive and particularly prone to errors in standard MR images that only generate qualitative contrast.

[0003] Current segmentation methods are based on one or more MR image datasets with a specific contrast weighting. The image contrast is generally chosen to ensure the highest possible contrast between the tissue types to be segmented.

[0004] Segmentation algorithms generally need to learn the contrast used, for example, by adjusting the parameter selection or extracting model parameters from training data, since different protocol parameters can result in different contrast. If the measurement protocol is changed too drastically, segmentation errors can occur. This is further complicated by the fact that the qualitative contrast of common MRI images can vary from MRI scanner to MRI scanner, from coil to coil, and even from day to day. Further image differences can arise from patient positioning and coil selection, scan adjustments, noise levels, manufacturer, software versions, etc. To achieve greater robustness, modern segmentation algorithms often utilize prior knowledge about the shape of the structures to be segmented (atlases, shape models).The results of methods that work with such qualitative data can be poor, however, as soon as a case deviates significantly from the models derived from training data (e.g., rare pathologies). In general, a reproducible measurement result, and therefore reliable segmentation, cannot always be guaranteed.

[0005] German patent DE 10 2006 014 882 A1 describes a method for imaging the myocardium of a heart attack patient.

[0006] WO 2014 / 047 326 A1 describes a method for tissue characterization using MR fingerprinting.

[0007] US 2010 / 0 127 704 A1 describes a method in which MR images that quantitatively represent MR parameters are segmented.

[0008] The present invention therefore aims to provide a method that overcomes the disadvantages of previously known segmentation methods and, in particular, provides a more accurate and reproducible method for segmenting objects under investigation.

[0009] According to the invention, this problem is solved by the features of the independent claims. These include a method for segmenting image data of a medical examination object by at least one quantitatively measured physical quantity, and a device for segmenting image data of a medical examination object by at least one quantitatively measured physical quantity. The dependent claims define preferred and advantageous embodiments of the present invention.

[0010] First, the raw data is acquired through the MR signals of the object under investigation. This makes the unprocessed data available for further analysis and thus contains comprehensive information that can be used for further processing.

[0011] Furthermore, quantitative image data of the object under investigation are determined, whereby at least one physical quantity of the object is quantitatively determined and displayed pixel by pixel. The quantitative determination of the physical quantity has the advantage that, unlike qualitative data, the influence of interfering measurement conditions, such as different coils, scanners, or patient positioning, is much less than when determining qualitative quantities.

[0012] In a further step, the quantitative image data is segmented (e.g., based on atlases, shape models, or other segmentation approaches) to identify predetermined objects within it. These predetermined objects could be, for example, bones or tendons, or any type of organ or tissue structure. This type of segmentation can be used to determine the most accurate possible identification of the object's spatial structure.

[0013] The advantage of this method is that by measuring quantitative physical values, better reproducibility of the quantitative image data as well as improved and more accurate segmentation of the predetermined objects can be achieved.

[0014] According to the invention, in addition to the method described above, after the acquisition of initial image data of the object under investigation, the pixel-by-pixel quantitative determination of the initial physical quantity, and the segmentation of the initial quantitative image data, a first quality parameter is determined for the segmented initial image data. This parameter contains information about how reliably the predetermined objects can be identified in the segmented initial image data.

[0015] In addition, second image data is determined, a second physical quantity is quantitatively determined pixel by pixel, the second image data is segmented, and a second quality parameter is determined for the segmented second image data.

[0016] Subsequently, the segmented first quantitative image data are compared with the segmented second quantitative image data, whereby in an area where the identified objects in the segmented first quantitative image data do not match the identified objects in the segmented second quantitative image data, those quantitative image data with the higher associated quality parameter are used to identify the predetermined objects.

[0017] Determining at least two different physical quantities has the advantage that a predetermined object can be segmented more accurately, because at least two physical quantities can contain more information about the tissue of the predetermined object than one physical quantity.

[0018] Furthermore, a computationally superimposed quantity can be determined from the first and second physical quantities of the object under investigation. This allows for improved segmentation because a more precise segmentation can be derived from the computational combination of the two physical quantities.

[0019] In a further embodiment of the method, the at least one physical quantity in the quantitative image data of the predetermined object can be compared with a first reference database that comprises a multitude of physical quantities of objects under investigation. If a predetermined deviation of the at least one physical quantity is exceeded, this deviation can be indicated. This advantageously allows for the identification of those predetermined objects in which physical quantities deviate from the reference range.

[0020] Furthermore, it is possible to determine the volume of a predetermined object from at least one physical quantity in the quantitative image data and compare it with a second reference database containing a large number of volume data from objects under investigation. If a predefined volume deviation is exceeded, an identification step can be performed to determine the volume deviation.

[0021] Similarly, after segmenting the objects in the quantitative image data, the objects in the image data of the predetermined object can be further segmented by MR fingerprinting subvoxel quantification, whereby at least one physical pixel dimension is taken from the first reference database. This allows, for example, a more precise determination of the composition of an object under investigation.

[0022] Furthermore, in one embodiment of the method, after selecting a pixel in the quantitative image data, a search can be conducted for neighboring pixels that have the same quantitative value as the selected pixel, up to a defined deviation. The same quantitative values ​​are then assigned to the same tissue. In this process step, a tissue region associated with a specific object can thus be advantageously identified.

[0023] In another embodiment of the method, the predetermined objects can be identified by thresholding, correlation analysis, region growing techniques, image processing methods using active shape models, or statistical tests of at least two physical quantities in the quantitative image data by comparison with the reference database. This advantageously allows pathological or degenerative changes to be detected at an early stage.

[0024] Furthermore, in another embodiment of the method, the quantitative image data can be fed into the first or the second reference database. This advantageously provides additional data beyond that already existing in the databases, which can be used in subsequent investigations to achieve greater accuracy in the evaluation of the objects under investigation.

[0025] Similarly, in another embodiment, the quantitative image data can be assigned to a tissue type and, after their assignment, fed together with this assignment to the first or the second reference database.

[0026] In another embodiment, a segmentation method can be carried out that uses, among other things, a statistical model of the object under investigation, whereby at least one statistical value of the at least one quantitatively determined physical quantity is determined. The statistical model takes this at least one statistical value into account. This has the advantage of improving the accuracy of the statistical model.

[0027] Furthermore, in one embodiment of the method, a quantitative ratio between at least two different physical quantities of the predetermined object can be determined and compared with a quantitative ratio of the two different physical quantities from the first reference database. The deviation of the quantitative ratio from the reference database can be displayed as a graphic, a color map, or as color overlays on medical images. This advantageously allows the deviation with respect to a reference model to be visually represented.

[0028] Similarly, the pixel-wise determination of at least one physical quantity can be achieved by means of MR fingerprinting.

[0029] In one embodiment of the proposed method, the at least one physical quantity can comprise one of the quantities T1, T2, T2* or offresonance.

[0030] In another embodiment of the method, after segmenting the image data of the object under investigation, healthy and pathological tissue can be manually identified and assigned to the image data. This assignment can then be added to the first and / or the second database.

[0031] The proposed MRI system for segmenting image data of a subject comprises a measurement unit for acquiring raw data. This measurement unit is connected to a processing unit designed to determine quantitative image data of the subject and to segment this quantitative data. Furthermore, the processing unit is connected to a display unit for presenting the parameters in a segmented representation. The MRI system is thus configured to execute a method for segmenting image data of a subject as described above.

[0032] The user of the inventive method and the associated device is thus advantageously able to perform segmentations of examination objects based on quantitative MR images and thus avoid the disadvantages of qualitative methods.

[0033] The invention is explained in more detail below with reference to the accompanying drawings and exemplary embodiments. The drawings show: Fig. 1 a flowchart for carrying out the inventive method for segmenting image data of an object under investigation, Fig. 2 a flowchart of a further embodiment of the method for segmenting image data of an object under investigation, Fig. 3 A representation of the MR system.

[0034] Fig. Figure 1 shows the process steps for segmenting image data of a subject. In the first step, S1, the raw data of the subject is acquired. This raw data comprises digitized MR signals generated during the measurement sequence of the subject. These signals can then be stored in a matrix (k-space). For example, the stored data can be transformed by a Fourier transform so that it is available for further processing.

[0035] The raw data described above can contain a wide variety of information about the tissue. After the transformation described above, this information is available in the form of quantitative tissue parameters that are spatially and / or temporally resolved and encompass specific areas of the specimen or even the entire specimen.

[0036] These tissue parameters are aggregated into quantitative image data in step S2, which may include, for example, T1, T2, or spin density. T1 represents the relaxation time of the longitudinal magnetization, i.e., the time after which the longitudinal magnetization has recovered to 63% of the initial displacement caused by the excitation pulse. T1 is specifically dependent on the type of hydrogen or water binding in the tissue; therefore, this parameter can be used, for example, to differentiate healthy from diseased tissue.

[0037] T2 denotes the relaxation time of the transverse magnetization, i.e., the time it takes for the transverse magnetization to reach a specific spread. T2 is based solely on spin-spin interactions and is not dependent on interactions with the environment. Therefore, this parameter is also tissue-specific.

[0038] To determine quantitative image data from MR signals, a method incorporating Magnetic Resonance Fingerprinting (MRF) can be used. This involves applying a pseudorandomized signal profile in an MR imaging procedure, where different materials or tissues exhibit a distinct signal development that simultaneously represents a function of their different material properties. After acquiring the signal profile, a pattern recognition algorithm is applied to the data and converted into a quantitative image representation. This is described, for example, in the article "Magnetic Resonance Fingerprinting" from March 14, 2013, 495 (7440): 187-92.

[0039] In the following step S3, the quantitative image data are segmented with respect to at least one quantitative physical quantity. Various methods can be used to identify spatially related sub-areas of the image, thereby defining specific regions within the entire object of study and distinguishing these regions from the rest of the object, i.e., the background.

[0040] For example, there are multi-stage segmentation methods that follow a model-based segmentation approach, which is extended by higher-dimensional input data. A segmentation hierarchy is applied, segmenting structures that are relatively easy to segment first. For example, bones are easy to segment because bone tissue is usually better defined than, for example, cartilage. So-called Active Shape Models (ASMs) are used for this purpose.

[0041] The ASM can be expanded by using quantitative values ​​(e.g., T1, T2) and the quality of the segmentation improved. Based on this, the transition surface between bone and cartilage is identified using a statistical model of cartilage tissue.

[0042] Using a cartilage thickness model and local intensity edges in the image, cartilage pixels can be identified in an iterative process. Training data is used for the bone models from the ASM and the statistical cartilage models (mean thickness and variance).

[0043] In a preferred embodiment, the statistical model of the tissue to be segmented (for example, cartilage) is extended by statistical quantities (e.g., mean and variance) of the physical values ​​relevant to the tissue to be segmented (in particular T2).

[0044] The inventive additional or exclusive use of input data with multiple quantitative physical quantities helps to improve the accuracy of the models. On the one hand, the intensity distribution of the images to be segmented will correspond better to the models due to improved reproducibility. On the other hand, the use of several different physical quantities increases the information content of the models and thus also the robustness of the classification. This allows, for example, the weight of the cartilage thickness model of the object under investigation, or other assumptions that extend beyond individual pixels, to be reduced in favor of the measured values ​​of the individual pixels when classifying pixels. This, in turn, improves the robustness of the segmentation for cases with abnormal anatomy and, in particular, with pathologies.

[0045] In a further embodiment of the invention, the physical values ​​(e.g., T1, T2) are compared after segmentation with a database containing normal ranges of physical values ​​for a large patient group. This allows for an assessment of whether the quantitative values ​​of the patient under examination indicate disease, damage, or a normal state. Furthermore, specific tissue types, such as water, fat, muscle tissue, etc., can be assigned to the measured regions based on normal values ​​from databases.

[0046] In another embodiment, the volumes determined from the segmentation are compared with normative values ​​from a database of numerous individuals in order to detect pathological changes, such as a decrease in cartilage volume or an increase in synovial fluid in osteoarthritis, at an early stage. The visualization can be as a graph with a measurement point and a normative range, or as a color overlay on the anatomical image.

[0047] The methods and implementation examples can be used in step S4, after segmentation, to identify predetermined objects. A user, for example a physician, can then display and evaluate the identified object, recognize pathological structures within it, and distinguish them from healthy tissue.

[0048] In a further preferred embodiment of the method, after segmentation of the quantitative image data and identification of the predetermined objects, e.g., bone and muscle tissue, the quantitative image data and the signal profiles in the muscle tissue can be re-examined and further subdivided into components using a method for MR fingerprinting subvoxel quantification. The described method assumes that the signal in certain previously segmented areas consists entirely of these components (e.g., bone and muscle tissue), with the required quantitative tissue properties (T1, T2, ...) of the compartments being obtained from a database or the literature.

[0049] In a further embodiment of the method according to the invention, an extended segmentation method is performed based on at least one quantitative physical quantity. The user can select a specific pixel or area of ​​the object, for example, a chord, whereby an algorithm, for example using a region-growing method, searches the surrounding pixels for identical or similar quantitative values ​​and assigns them to the same compartment if a match is found. Starting with the matching pixels, the surrounding pixels are examined again. In this way, for example, a chord can be traced further to capture it completely. In an advantageous display format, the three-dimensional profiles of desired bands can be calculated for the user.By incorporating the above-mentioned segmentation into different tissue types, the user can also obtain an atlas-like anatomical representation of an image in which certain tissue types and regions are marked in different colors and displayed three-dimensionally.

[0050] In another aspect of the invention, a reference database can be generated through calibration. Pathological and healthy cases are measured, the image data is acquired, retrospectively assigned to healthy or pathological tissue by a physician, and then stored in a database along with the measured and assigned signal profiles. In this way, tissue types can also be distinguished whose differences are not yet represented in the models used for simulation and whose influence may not even have been recognized. A particularly interesting application in orthopedics is the early detection of cartilage tissue pathologies even before, for example, a change in volume occurs in the course of osteoarthritis.In this context, it is also possible to evaluate longitudinal studies and retrospectively correlate disease progression with changes in quantitative measurements or signal patterns, so that threshold values ​​can be derived that indicate early signs of degenerative change.

[0051] Furthermore, in another embodiment of the method, the segmentation of certain tissue regions (for example, fat, water, bone, cartilage, muscle, tendons, ligaments, menisci or edema) can be carried out based on at least one quantitative physical value, whereby the segmentation can be carried out by means of thresholding, correlation analysis or statistical test procedures.

[0052] In a further embodiment of the method according to the invention, the described quantitative image data can also be used to be incorporated into the reference database of physical values ​​and the reference database of volume data of the objects under investigation. This can be done in particular with the described calibrated data for different tissue types. This expands the existing databases with additional information and, in conjunction with the embodiments of the method according to the invention described above, can provide segmentation results with improved accuracy.

[0053] Furthermore, a segmentation method is proposed that, in addition to prior knowledge from the reference database about the geometry of the tissue to be segmented, also uses a statistical model for at least one physical quantity of the object under investigation. This method calculates a probability that a pixel belongs to the tissue to be segmented, depending on the physical quantities quantitatively determined by MRF.

[0054] In a further embodiment of the invention, a quantitative ratio between at least two different physical properties of the predetermined object, for example fat / muscle or fat / water, can be determined and this ratio compared with normative data from the reference database, which contains physical properties of a large number of test objects. Deviations can be displayed, for example, in the form of graphs or as a color map or color overlay on medical images or anatomical illustrations. This allows the user to easily identify altered or pathological tissue visually.

[0055] Fig. Figure 2 presents a further embodiment of the method, in which at least two different quantitative physical quantities are used to more accurately identify predetermined objects that can only be imprecisely identified using a single quantitative physical quantity. In the first step, S10, an initial set of quantitative image data is acquired, and a first physical quantity is determined and displayed. This can be, for example, T1; other parameters can also be used. In the second step, S11, the initial image data is segmented. In the subsequent step, S12, a first quality parameter is determined, which describes how reliably the predetermined objects can be identified in the segmented initial image data.

[0056] In a further step S13, a second set of quantitative image data is acquired, and a second physical quantity is determined and displayed. This could be, for example, T2, but the physical quantity is not limited to this. Then, in a second step S14, the second set of image data is segmented. Finally, in a further step S15, a second quality parameter is determined, which describes how reliably the predetermined objects can be identified in the segmented second set of image data. This quality parameter can include information about certainty, specificity, probability, or significance.

[0057] In step S16, the first segmented quantitative image data is compared with the second segmented quantitative image data. If the identified objects in the first segmented quantitative image data and the second segmented quantitative image data do not match in a specific area, this area is identified in step S17. Then, in step S18, the image data with the higher quality parameter is used to identify the predetermined objects.

[0058] The described procedure can, for example, employ the method of Magnetic Resonance Fingerprinting (MRF), preferably measuring at least two quantitative physical quantities with spatial resolution in multiple image series. In this process, a specific signal profile is generated voxel-wise by pseudorandomized profiles of certain parameters in the MR imaging over a large number of images (N = 1000-5000). This means that predefined MRF protocols with pseudorandomized profiles of specific parameters are used. This signal profile is called a fingerprint and can be uniquely assigned to a specific n-tuple of physical parameters, such as T1, T2, off-resonance, or M0, using a database. The fingerprint thus represents a unique function of the material properties of the object under investigation.

[0059] For the automated evaluation of multiple image series, registration is usually necessary because patient movements can occur during or between measurements. Registration algorithms for the often complex anatomies and non-rigid movements encountered in practice are frequently error-prone and computationally intensive. The use of multidimensional, simultaneously acquired image series (e.g., with MRF) avoids this otherwise necessary step, provided the acquisition method is sufficiently motion-insensitive.

[0060] In one embodiment, the spatially resolved determination of at least two quantitative physical quantities can be used to computationally superimpose them. For example, the quantities T1 and T2 can be weighted according to the formula: Superimposed Image = n*T1Map + m*T2Map, for example, with m = 0.3 and n = 10. This allows a significantly more accurate representation of the object to be achieved from quantitative image data where only one physical quantity would produce an insufficiently precise depiction of the predetermined object, by combining and weighting two different values.

[0061] Furthermore, in areas where the segmented first and second image data do not match with sufficient accuracy, improved segmentation with mutual correction can be performed by comparing the at least two segmentation results. For example, for a predetermined object where one of the two physical quantities has a higher specificity, if the image data for both physical parameters do not match, the physical quantity with the higher specificity can be used.

[0062] In another embodiment, segmentation can be used to adapt individual implants and thus to plan surgeries. For this purpose, the exact geometry of the bones to be replaced and that of the surrounding bone structure are derived from the data. The preferably at least two quantitative values ​​should be selected such that the individual bone structure differs as advantageously as possible from its surroundings.

[0063] Fig. Figure 3 represents an MRI system for segmenting image data of an object under investigation. The system includes a measurement unit 1, with which the object under investigation can be measured and the raw data recorded. Various methods can be used, such as magnetic resonance fingerprinting.

[0064] The raw data is then fed to a processing unit 3, which is connected to a memory 2. The raw data is stored in memory 2 and is thus available to the processing unit 3 for further processing. The processing unit 3 is designed to extract quantitative image data from the raw data, which can contain a multitude of different physical parameters of the object under investigation. The processing unit 3 can then apply various segmentation methods to the quantitative image data in order to identify segmented objects. When using the segmentation methods, a reference database 4 is used, which contains physical parameters of a multitude of objects under investigation. This database is connected to the processing unit. Furthermore, the processing unit 3 is connected to another reference database 5, which contains volumetric data of a multitude of objects under investigation.Both reference databases are used to provide the algorithms in the segmentation procedures with comparative reference data.

[0065] To enable the visual representation of the segmented objects, the segmentation algorithms can include an image recognition algorithm that can be coupled with at least one of the reference databases 4, 5, which contain known signal waveforms and physical quantities of the object under investigation. The image recognition algorithm can compare the measured signal waveforms with the data from the database and thus generates segmented image representations of the desired objects. The objects segmented in this way are then displayed in a display unit 6.

Claims

[1] Method for segmenting image data of an object under investigation, comprising the following steps: - Recording raw data of the object under investigation; - Determining initial quantitative image data of the object under investigation, in which a first physical quantity is quantitatively determined and represented pixel by pixel, - Segmenting the initial quantitative image data, - Determining a first quality parameter for the segmented first image data, which contains information about how reliably the predetermined objects can be identified in the segmented first image data, - Determining second quantitative image data of the object under investigation, in which a second physical quantity is quantitatively determined and displayed pixel by pixel, - Segmenting the second quantitative image data, - Determining a second quality parameter for the segmented second image data, which contains information about how reliably the predetermined objects can be identified in the segmented second image data, - Comparing the segmented first quantitative image data with the segmented second quantitative image data, whereby in an area where the identified objects in the segmented first quantitative image data do not match the identified objects in the segmented second quantitative image data, those quantitative image data with the greater associated quality parameter are used to identify the predetermined objects. [2] Method according to claim 1, comprising determining a computationally superimposed quantity from the first and the second physical quantities of the object under investigation. [3] Method according to one of the preceding claims, wherein the at least one physical quantity in the quantitative image data of the predetermined object is compared with a first reference database comprising a plurality of reference data of objects under investigation, and when a predetermined deviation of the at least one physical quantity is exceeded, this exceedance is indicated. [4] Method according to one of the preceding claims, wherein the volume of the predetermined object is determined from the at least one physical quantity in the quantitative image data and compared with a second reference database comprising a plurality of volume data of test objects, and if a predetermined deviation of the volume is exceeded, an identification step of the volume deviation is carried out. [5] Method according to claim 3, wherein after segmenting the objects in the quantitative image data the objects in the image data of the predetermined object are further segmented by MR fingerprinting subvoxel quantification, wherein the at least one physical size of the pixels is taken from the first reference database. [6] Method according to one of the preceding claims, wherein after selecting a pixel in the quantitative image data, neighboring pixels are searched for which, up to a defined deviation, have the same quantitative value and the same quantitative values ​​are assigned to the same tissue. [7] Method according to claim 1, wherein the predetermined objects are determined by thresholding, correlation analysis, region growing methods, image processing methods with respect to active objects or statistical test methods of the at least two physical quantities in the quantitative image data by comparison with the reference database. [8] Method according to claim 3 or 4, wherein the quantitative image data are supplied to the first or the second reference database. [9] Method according to claim 3 or 4, wherein the quantitative image data are assigned to a tissue type and, after their assignment, are supplied together with this assignment to the first or the second reference database. [10] Method according to one of the preceding claims, wherein a segmentation method is carried out which, among other things, uses a statistical model of the object under investigation, wherein at least one statistical quantity of the at least one quantitatively determined physical quantity is determined, wherein the statistical model takes into account the at least one statistical quantity. [11] Method according to claim 3 or 4, wherein a quantitative ratio between at least two different physical quantities of the predetermined object is determined and compared with a quantitative ratio of the two different physical quantities from the first reference database, and the representation of the deviation of the quantitative ratio from the reference database is in the form of a graphic or as a color map or as color overlays on medical images. [12] Method according to one of the preceding claims, wherein the pixel-wise determination of the at least one physical quantity is carried out by MR fingerprinting. [13] Method according to any of the preceding claims, wherein the at least one physical quantity comprises one of the following quantities: T1, T2, T2* offresonance. [14] MR system for segmenting image data of an object under investigation, comprising: - a measuring unit (1) for recording raw data of an object under investigation, - a computing unit (3) for determining quantitative image data of the object under investigation and for segmenting the quantitative image data, - a representation unit (6) for displaying the parameters in a segmented representation, the computing unit is designed to perform the following steps: - Determining initial quantitative image data of the object under investigation, in which a first physical quantity is quantitatively determined and represented pixel by pixel, - Segmenting the initial quantitative image data, - Determining a first quality parameter for the segmented first image data, which contains information about how reliably the predetermined objects can be identified in the segmented first image data, - Determining second quantitative image data of the object under investigation, in which a second physical quantity is quantitatively determined and displayed pixel by pixel, - Segmenting the second quantitative image data, - Determining a second quality parameter for the segmented second image data, which contains information about how reliably the predetermined objects can be identified in the segmented second image data, - Comparing the segmented first quantitative image data with the segmented second quantitative image data, whereby in an area where the identified objects in the segmented first quantitative image data do not match the identified objects in the segmented second quantitative image data, those quantitative image data with the greater associated quality parameter are used to identify the predetermined objects.

Images