• 专利附录
  • 专利说明
  • 权利要求
  • 类似技术
  • 同族专利
  • 引用文献
  • 法律状态
  • 优先权
    • 专利摘要:
    A method is disclosed for the generation of a background mask from an intensity-value-coded image with a noisy background area. The method includes generating an intensity value histogram of the image; fitting a model curve to a subarea of the intensity value histogram; calculating a threshold value in the intensity value histogram from the fit parameters of the model curve; and generating the background mask starting from at least one defined image area that belongs to the background noise by adding all the pixels whose intensities are below the threshold value to the background mask when connected to the image area via pixels whose intensities are also below the threshold value.
    公开号:NL1032161A1
    申请号:NL1032161
    申请日:2006-07-13
    公开日:2007-01-23
    发明作者:Michael Zwanger
    申请人:Siemens Ag；
    IPC主号:
    专利说明:

    METHOD FOR AUTOMATICALLY MANUFACTURING A BACKGROUND MASK FOR IMAGES WITH NOISE-BACKGROUND AREAS, APPLICATIONS THEREFOR, AND A NUCLEAR SPINTOMOGRAPHY DEVICE FOR PERFORMING THE METHOD
    The present invention relates to a method for automatically producing a background mask for images with noise-backed background areas, in particular for magnetic resonance tomography images (MRT images). Furthermore, the present invention relates to a method for automatically masking noise background areas with images as well as a method for determining a quality measure for images with noise background areas. Furthermore, the present invention relates to a core spin tomography apparatus adapted for performing the method, and to a computer software product that implements the method when it is running on a computer unit connected to the core spin tomography apparatus.
    With MRT images, a different strongly formed background noise occurs depending on the recording sequence. For example, with diffusion-weighted MRT images, which were recorded with an EPI sequence (Echo Planar Imaging), the background is rather strongly noise-treated. Such images are often used for further calculations to better display clinically relevant information. For example, an Apparatus Diffusion Coefficient (ADC) is calculated from multiple diffusion-weighted images of the same intersection with differently strong diffusion weighting, because the image data of at least two images with different diffusion weighting are adjusted by a non-linear fit. In this method, the background noise of the diffusion weighted images is amplified and generates a disturbing noise in the ADC image.
    So far the problem has been solved by producing a background mask that masks the background areas with noise. Here, all pixels whose intensity in the image with the diffusion weight of the strength zero is below a certain threshold value form the background mask. This background mask sets the intensity of the corresponding pixels to zero in ADC images of the same intersection. The threshold value is usually set manually. For example, with the SIEMENS MAGNETOM Harmony MRT device, the threshold value to be used is set manually before the start of a measurement. The success of the method thereby depends on the experience of the user who performs the measurement and determines the threshold value. An incorrectly set threshold value can lead to a loss of medically important information. The measurement may have to be repeated, which is associated with higher costs and a greater time commitment.
    DE 103 56 275 A1 describes a method for automatically segmenting phase-coded current schemes in the MRT. At least one phase image of a blood-flowed area is thereby measured with the aid of the MRT and thereafter flowed-on areas in the phase image are automatically segmented. The segmentation can take place on the basis of a histogram, which is calculated from the phase image, among other things, while recording a threshold value in the histogram. Similarly, a seed growth algorithm can be used for segmenting through-flow areas that, based on a seed in the interior of the flow-through area, scans the present flow areas from within.
    US 2004/0254447 A1 discloses a background suppression method for time-resolved magnet resonance angiographs. In this method, an orthogonality image is calculated from a series of time-resolved MRT images, which is masked in a further process step. To this end, a histogram of the orthogonality image is processed, including a Gaussian curve, and a threshold value is calculated on the basis thereof.
    DE 101 22 874 A1 discloses a method for extracting spider collectives with different chemical shift from phase-coded individual images while taking into account field inhomogeneities. With this method, a threshold value is determined based on an automatic noise level estimation.
    It is the object of the present invention to provide a method which automatically generates a background mask on an image with background noise. Furthermore, it is the object of the present invention to provide methods in which background areas with noise are masked or characterized on the basis of the generated background mask, or on the basis of the generated background mask, a quality measure for an image with noise. provided background areas. It is a further object of the invention to provide a core spin tomography apparatus and a computer software product with which a background mask can be generated automatically in the case of an image with background noise.
    These objects are achieved according to the invention by the features of the independent claims 1, 16, 17, 20, 22 and 23. Advantageous embodiments of the invention are in each case the subject of further claims.
    The method for automatically producing a background mask according to claim 1 to an intensity value coded image with a noise background background has the following steps: - producing an intensity value histogram of the image, - fitting a model curve to a sub-area of the image intensity value histogram, - calculating a threshold value in the intensity value histogram from the fit parameters of the model curve, - manufacturing the background mask by starting from at least one defined image area associated with the background noise, all pixels whose intensities below the threshold value are added to the background mask when they are connected to the image area via pixels, the intensities of which are also below the threshold value.
    With the invention, a background mask is automatically generated in an image with background noise areas, without the need for interaction with the user. As a result, an error-loaded and delay-sensitive step in image processing is automated. The invention is particularly suitable for diffusion-weighted MRT images, in which background masks are created, in order to further characterize background areas that are particularly disturbing in further processing of these images.
    Advantageously, the sub-area of the intensity value histogram, which is fitted with the model curve, is the sub-area around a local maximum in the intensity-value histogram, which is associated with the background noise, since this sub-area is also clearly in the intensity value for misplaced images. histogram can be seen.
    This area is preferably adjusted with a Gaussian curve, since the intensity distribution of the mass can be properly optimized with this curve shape and since the fit of the Gaussian curve is a numerically stable method. However, other model curves, for example a Rice distribution or polynomial functions, can be fitted to the area when appropriate fit algorithms are used that adjust the model curve to the intensity distribution in the sub-region in a numerically stable manner.
    A threshold value is calculated from fit parameters of the model curve, which is characterized in that the function value of the model curve is always clearly lost on the threshold value. If the model curve to be fitted is a Gaussian curve, the standard deviation σ and the average value μ are those fit parameters from which the threshold value is calculated. In a possible embodiment, the threshold value S is based on the formula
    calculated since at the threshold value the function value of the Gaussian curve has strongly disappeared.
    After calculating the threshold value, starting from at least one defined image area associated with the background noise, all pixels whose intensities are below the threshold value are merged into a background mask when they are associated with the image area via pixels, the intensity of which is also below the threshold value. It is advantageous here to start from parts of the image edge, since this image area usually belongs to the background with the images to be processed. By this method, pixels which have an intensity below the threshold value, but which are located in the interior of an anatomical structure, are not added to the background mask, since they are located in the interior of structures with high signal intensity and not via pixels, whose intensity is below the threshold value are connected to an angle of view. In addition to image angles, other image areas can also be taken as starting points if, due to the special spatial shape of the included anatomical structure, it is known that the image area belongs to the background. For example, with a frontal cross-section of both legs, the center belongs to the background, so that the background mask can be produced from this area.
    In a further embodiment, before or after measurement of the images, the user can characterize an area associated with the background noise, so that the background mask can then be established starting from this area. As a result, the method can be adapted in a simple manner to special events of the included anatomical structure.
    In a preferred embodiment, a modified background mask is created from the background mask by topographically coherent pixels not being detected by the background mask being merged into at least one cluster, and by a cluster thus formed depending on its size and / or its average intensity value is added to the modified background mask. This embodiment achieves that individual pixels which, although associated with the background noise, but whose intensity is above the threshold value, so that they are not included in the background mask, are now added to the modified background mask. These pixels generally form clusters, the size of which is small compared to the clusters formed by anatomical structures, and as a rule have an average intensity which is only slightly above the threshold value, while the clusters of anatomical structures are clearly above the threshold value. This method improves the quality of the background mask.
    In a favorable embodiment, the decision as to whether or not a cluster is added to the modified background mask can be achieved by evaluating the size and the average intensity of the cluster independently of each other with a numerical value Ps and Pj, respectively. If the Ps · Pi product is above a certain limit value, the cluster is added to the modified background mask.
    In a possible embodiment, the size of the cluster Sc is evaluated with the number value Ps, which is according to the formula
    is calculated. The average intensity Ioem_c is evaluated with the number value Pj, that according to the formula
    is calculated, where luax is the maximum intensity value of the image and S is the threshold value. The cluster is now added to the modified background mask when the product of the two numerical values Ps Pi is above the value 10. This advantageous version is thereby adapted to the special requirements of diffusion-weighted MRT images. Other types of images may require different weights of the size and / or of the average intensity value of the cluster and these must be adapted to the special properties of the recordings.
    The method can reach its limits in the case of particularly highly noisy images with a very weak signal from anatomical structures, since in the intensity value histogram the area assigned to the background noise can no longer be clearly separated from the area that actually image information. By applying the method to such images, image areas can be counted as the background mask, which includes clinically relevant information.
    Such a picture is due to a high quotient
    characterized in that S is the threshold value and Imox is the maximum intensity value of the image. In an advantageous embodiment of the invention, this quotient is calculated as a measure of the quality of the background mask. On the basis of this measure, it can be decided whether the background mask should be used for further processing steps. For example, if the background mask is used to mask a background noise area in images, the masking may be rejected if the quotient exceeds a certain value - e.g., 0.3 - that is, if the quality of the mask is too poor. is. This prevents the background mask from masking areas that carry clinically relevant image information.
    Particularly suitable for this method are MRT images, in particular the like, in which background masking is required, such as, for example, diffusion-weighted images, perfusion-weighted images or MR temperature mapping images. However, since the method is based solely on the image data, it can also be applied to images recorded with other devices, such as, for example, CT images or ultrasound images.
    The method is advantageously applied to an image from a series of diffusion weighted MRT images with different diffusion weighting strengths. The background mask can be produced from the image with the best signal-to-noise ratio. Usually this is the image with the lowest intensity of the diffusion weight. Advantageously, the series of diffusion-weighted MRT images also includes an image with the zero diffusion weight, since the signal-to-noise ratio is the best here.
    The method for automated masking and / or characterizing a noise background area according to claim 16 with an intensity value coded image has the following steps: - producing a background mask according to any of claims 1-15 with the image, and - masking the background mask image or characterizing the noise background area based on the background mask.
    In this context, masking means that the intensity value of all pixels associated with the background mask is set to zero. A characterization of the background areas means that the image information of the background area must be treated differently from the other image data during further processing of the image; for example, a different filter can be applied to background areas than to the other image data.
    Furthermore, a method for automated masking and / or characterizing a noise-backed background area according to claim 17 is claimed with a further image, comprising the following steps: - producing a background mask according to any one of claims 1-15 with an intensity value-coded image wherein the intensity value-coded image and the further image map the same structures in the same intersection, and - masking the further image with the background mask or characterizing the noise-backed background area with the background mask, respectively.
    In this method, the background mask that was made with one image is applied with another image. The advantage is that with certain images to be masked, the background is too strongly noise-proofed, so that the quality of the background mask is too poor and is therefore not achieved directly with this image. However, there are often images which represent the same structure in the same cutting plane, the background being weaker with noise, so that the background mask can be produced on the basis of these images. This is especially the case with a series of diffusion-weighted MRT images, where the background mask is usually created on the image with the slightest diffusion weight. This background mask can then be applied to all images in the series. This background mask can also mask the background for an ADC image calculated from the series of diffusion-weighted MRT images.
    Furthermore, a method for automatically determining a quality measure of an image according to claim 20 is claimed, comprising the following steps: - creating a background mask according to any of claims 1 to 15 with the image, and - determining the quality measure, in that the intensity values of all pixels are associated with the intensity values of those pixels associated with the background mask.
    This quality measure characterizes a signal-to-noise ratio with an image. With a strongly noisy image, for example, the intention values of the background area differ less strongly from the intention values of all pixels than with a weakly noisy image. In a particularly simple embodiment the quotient becomes the quantitative value for the quality measure
    calculated, where Ioem_Totaai is the average intensity value of all image pixels and him. Mask is the average intensity value of the image pixels associated with the background mask. This value can be supplied together with the image, so that the user can immediately recognize the quality of a recording.
    Furthermore, according to the present invention, a core spin tomography apparatus is claimed which is suitable for carrying out a method according to one of claims 1 to 21.
    Similarly claimed is a computer software product that implements a method according to any of claims 1-21 when it runs on a computing device connected to this core spin tomography apparatus.
    Further advantages, features and characteristics of the present invention will now be explained in more detail with reference to exemplary embodiments with reference to the accompanying drawings.
    Figure 1 shows diagrammatically the process progress as it occurs during the masking of a background area with the image,
    Figure 2 shows diagrammatically the curve course of an intensity-value histogram with the fit of a Gaussian curve to a sub-area,
    Figure 3 shows diagrammatically the eurverloop of another intensity value histogram with the fit of a Gaussian curve to a sub-area,
    Figure 4 shows the anatomical structure of the skull with a transverse section, and
    Figure 5 shows the background mask that was made from the anatomical structure of the skull in Figure 4 with a transverse section.
    Figure 1 shows diagrammatically the process progress as applied to the masking of a background area 3 'provided with a mass with an intensity value-coded image 1. The starting point is an intensity value-coded image 1, as is used, for example, in medical imaging for the representation of anatomical structures. The image 1 comprises image areas 3 with relevant image information about the anatomical structures and background areas 3 provided with noise. In a first method step 5, an intensity value histogram 6 is established from the image data. In a second method step 7, a sub-area of the intensity value histogram 6 is fitted with a model curve 8. In a third method step 9, an intensity value is calculated as threshold value 10 from the fit parameters of the model curve 8, which separates the sub-area in the intensity-value histogram 6, which is predominantly correlated with the background noise, from the sub-area predominantly with the anatomical structures is correlated. In the fourth method step 11, those image areas are added to the background mask 12, the pixels of which have an intensity value below the threshold value 10 and which are not in the interior of anatomical structures. The background mask 12 is applied to the image 1 and generates a masked image 13, the background area 3 'being substantially covered by the background mask 12. After applying the background mask 12 to the image 1, certain areas 4 can still occur, which, although associated with the background noise, are not included in the background mask 12. These are mainly those image areas of background noise 3 ", the intensity value of which is above the threshold value of 10. Therefore, in a fifth process step 15, a modified background mask 16 is created from the background mask 12, which is at least exactly as large as the background mask 12 and which covers the background area 3 'better than the original background mask 12. In this process step 15, all pixels that are not masked by the background mask 12 and that are topographically related, are merged into a cluster. Such a cluster is added to the background mask 12 depending on its size and / or its average intensity value. A modified background mask 16 is thereby established. In the sixth process step 17, the modified background mask 16 is applied to the image 1.
    However, the background mask or the modified background mask 12, 16 need not necessarily be applied to the original image 1. Similarly, the background mask or the modified background mask 12, 16, for example, may mark the background noise area 3 'with a different image representing the same structure in the same intersection, as, for example, images of a series of diffusion weighted MRT images with different diffusion intensities weighting. Also, the background mask or the modified background mask 12, 16 can only be used to describe the quality of an image in a quantitative value, because, for example, the average intensity of the pixels associated with the background mask 12, 16 is due to the average intensity of all pixels is shared.
    Figure 2 shows an example of an intensity value histogram of an MRT image with the fit of a Gaussian curve 44 to a sub-area. The frequency n at which a certain intensity occurs in the image is indicated here depending on the intensity I. In this example rather carried out according to the booklet, two local maxima 42 and 42 'are clearly recognizable in the intensity value histogram, the subarea around the left maximum 42 characterizing the intensity distribution of the background noise, while the subarea around the right maximum 42' intensity distribution of the actual image information. The area around the left maximum 42 is fitted with a Gaussian curve 44, which is a suitable model curve for this subarea. From the parameters μ and σ (average value and standard deviation respectively) of the Gaussian curve, a threshold value of S 46 is obtained according to the relationship
    calculated, which separates the background noise from the actual image information. The threshold value 10.46 is characterized in that the function value of the model curve 8 - in this case the Gaussian curve 44 - at the threshold value 10.46 has clearly disappeared.
    In practice, intensity value histograms do not always have this clear distinction between background noise and image information. Figure 3 shows the intensity value histogram of an MRT image that originates from a series of diffusion-weighted images with differently strong diffraction weights, the strength of the diffusion weight being zero in the image shown here. Here the areas around the left and right maximum 52 and 52 ’respectively characterize the intensity distributions of the background noise and the actual image information respectively. However, the right-hand sub-area is clearly flatter and wider than the left. The fit of the Gaussian curve 44 on the left sub-area is also possible without any problems, the threshold value 46 can be calculated according to the same formula.
    While in Fig. 2 the right-hand sub-region or even the transition region with the minium can be fitted with suitable model curves 8, it is only possible to a lesser extent in Fig. 3, since then the stability of the method is no longer guaranteed.
    In the case of particularly strong noise images with a very weak signal from anatomical structures, the intensity distributions of the background noise and the actual image information are strongly superimposed. In the intensity value histogram, the area assigned to the background noise can no longer be clearly separated from the area that contains the actual image information. By applying the method to such an image, areas would be assigned to the background noise, which may include clinically relevant image information. Such a picture is due to a high quotient
    characterized in that S is the threshold value and Imox is the maximum intensity value of the image. This quotient simultaneously represents a measure of the quality of the background mask 12, 16. A separate characterization of the image when the quotient exceeds a certain value - for example 0.3 - makes it clear to the user that the method reaches its limits. In a possible embodiment, the user can then decide for himself whether he wants to keep the (modified) background mask 12, 16 or not. In another possible embodiment, the (modified) background mask 12, 16 can be automatically removed again if the quotient is above the determined value.
    Figure 4 shows schematically an anatomical structure 21 of the skull, as it occurs with a transverse section through the skull with diffusion-weighted MRT images. The two brain halves 23 and the ventricular system 25 are usually found to be signal strong. However, signal-weak structures 27 also exist in the interior of the skull, such as, for example, a blood-flowed sinus. The intensity values thereof can be below the threshold value 10.46. In this image, an image angle 29 always represents an image area which only comprises background noise. Starting from this area, all pixels are added to the background mask 12, the intensity of which is below the threshold value 10.46 and which are associated with the image area via pixels, the intensity of which is also below the threshold value 10.46. In this way it is achieved that signal weak areas 27 in the interior of the anatomical structure 21 are not added to the background mask 12.
    Figure 5 outlines a background mask associated with the anatomical structure 21 shown in Figure 4. The background mask saves the total anatomical structure 21 of the skull. In the present case, in addition to the anatomical structure 21, even smaller areas 33 are saved, since in these areas 33 the intensity of the background noise was above the threshold value 10.46. These regions, which actually belong to the background noise, are added by a modification of the background mask 12 to a modified background mask 16, as described in the method step 15. For this purpose, a cluster is formed from topographically coherent pixels that are not detected by the background mask 12.
    In Figure 5, a cluster is formed from the anatomical structure of the skull 21 and the smaller areas 33. Each cluster is evaluated according to its size and / or its average intensity. Anatomical structures 21, such as those of the skull in Figure 4, form comparatively large clusters and have a comparatively high average intensity value. Similarly, clusters formed by pixels belonging to background noise 3 "form small clusters and have a comparatively low average intensity value.
    To decide whether a cluster is added to the background mask 12 or not, the size of a cluster and / or the average intensity value can be evaluated with a number value. If the number value is above or below a set limit value, the cluster is added to the background mask 12 or not.
    In the special case of diffusion-weighted MRT images, the following formulas are suitable for evaluating the size and intensity of a cluster: the size of the cluster Sc is determined according to the relationship
    evaluated with the numerical value Ps. The average intensity Iaem_c is according to the relationship
    evaluated with the numerical value Pi, where luax is the maximum intensity value of the image and S is the threshold value. The cluster is now added to the background mask 12 when the product of the two number values Ps Pi is above the value 10.
    权利要求:
    Claims (23)
    [1]
    A method for automated production of a background mask (12) from an intensity value coded image with a noise background background (3 '), comprising the steps of: - producing an intensity value histogram (6) of the image, - fitting a model curve (8) to a sub-area of the intensity value histogram (6), - calculating a threshold value (10) in the intensity value histogram (6) from the fit parameters of the model curve (8), and - producing the background mask (12), by starting from at least one defined image area associated with the background noise, all pixels whose intensities are below the threshold value (10) are added to the background mask (12) when they are image area are connected via pixels, the intensities of which are also below the threshold value (10).
    [2]
    Method according to claim 1, characterized in that the sub-area of the intensity value histogram (6) to be fitted is the area around a local maximum of the intensity value histogram (6) associated with the background noise.
    [3]
    Method according to claim 1 or 2, characterized in that the model curve (8) to be used is a Gaussian curve (44).
    [4]
    Method according to claim 3, characterized in that the fit parameters from which the threshold value (10) is calculated are the standard deviation σ and the average value μ of the Gaussian curve (44).
    [5]
    Method according to one of Claims 3-4, characterized in that the threshold value S (10) according to the formula

    [6]
    Method according to one of claims 1 to 5, characterized in that the defined image area associated with the background noise is a part of the image edge.
    [7]
    Method according to one of claims 1 to 6, characterized in that the defined image area associated with the background noise is a user-defined area noise background.
    [8]
    Method according to one of Claims 1 to 7, characterized in that a modified background mask (16) is produced from the background mask (12) in that topographically coherent pixels which are not included in the background mask (12) , to at least one cluster, and that such a cluster is added to the modified background mask (16) depending on its size and / or its average intensity value.
    [9]
    Method according to one of claims 1 to 8, characterized in that the cluster is added to the modified background mask (16) when a product from Pj Ps is above a limit value, Ps being a number value that is the size of the cluster evaluates, and Pi is a number value that evaluates the average intensity of the cluster.
    [10]
    Method according to claim 9, characterized in that the number value Ps from the size of the cluster Sc by the formula

    [11]
    Method according to claim 9, characterized in that the number value Pj from the mean intensity of the cluster Iem_c by the formula

    [12]
    Method according to one of claims 1 to 11, characterized in that a measure of the quality of the background mask is calculated by calculating the quotient of threshold value (10) and a maximum intensity value of the image.
    [13]
    Method according to one of claims 1 to 12, characterized in that the image (1) is an MRT image.
    [14]
    Method according to one of claims 1 to 13, characterized in that the image (1) is an image from a series of diffusion-weighted MRT images with different intensities of the diffusion weight.
    [15]
    Method according to claim 14, characterized in that the series of diffusion weighted MRT images also comprises an image with the diffusion weight of the strength zero.
    [16]
    A method for automated masking and / or characterizing a noise background background (3 ') at an intensity value coded image (1), comprising the following steps: - producing a background mask (12, 16) according to any of the claims 1 -15 at the image (1), and - masking the image (1) with the background mask (12, 16) or characteristics of the background background noise (3 ') based on the background mask (12, 16) .
    [17]
    A method for automated masking and / or characterizing a noise-backed background area (3 ') in a further image, comprising the following steps: - producing a background mask (12, 16) according to any of claims 1 to 15 in a intensity value coded image (1), wherein the intensity value coded image (1) and the further image map the same structures in the same cutting plane, and - mask the further image with the background mask (12, 16), respectively Noisy background area (3 ') based on the background mask (12, 16).
    [18]
    A method of automatically masking and / or characterizing a noise background background (3 ') according to claim 17, characterized in that the further image and the intensity value coded image are two images from an image series that have the same structure in the same intersect.
    [19]
    A method for automatically masking and / or characterizing a noise-backed background area (3 ') according to claim 18, characterized in that the further image is an image produced by further processing of the image series.
    [20]
    A method for automatically determining a quality measure of an image, comprising the following steps: - producing a background mask (12, 16) according to one of claims 1 to 15 with the image, and - determining the quality measure, in that the intensity values of all pixels are associated with the intensity values of those pixels associated with the background mask (12,16).
    [21]
    A method for automatically determining a quality measure of an image according to claim 20, characterized in that the quality measure is the quotient

    [22]
    A nuclear spin tomography apparatus suitable for carrying out a method according to any of claims 1-21.
    [23]
    A computer software product, characterized in that it implements a method according to any of claims 1-21 when it runs on a computer unit connected to a core spin tomography device.
    类似技术:
    公开号 | 公开日 | 专利标题
    NL1032161A1|2007-01-23|Method for automatically producing a background mask for images with noisy background areas, their applications, and a core spin tomography device for performing the method.
    El-Melegy et al.2014|Tumor segmentation in brain MRI using a fuzzy approach with class center priors
    Joshi et al.2010|Classification of brain cancer using artificial neural network
    Sheng et al.2018|Retinal vessel segmentation using minimum spanning superpixel tree detector
    Kollem et al.2019|A review of image denoising and segmentation methods based on medical images
    Kashyap et al.2017|Energy-based active contour method for image segmentation
    Hussain et al.2012|Segmentation of tissues in brain MRI images using dynamic neuro-fuzzy technique
    Balan et al.2012|Smart histogram analysis applied to the skull-stripping problem in T1-weighted MRI
    Milenković et al.2015|Automated breast-region segmentation in the axial breast MR images
    Forbes et al.2010|Adaptive weighted fusion of multiple MR sequences for brain lesion segmentation
    Kazerooni et al.2011|Segmentation of brain tumors in MRI images using multi-scale gradient vector flow
    Xia et al.2004|A knowledge-driven algorithm for a rapid and automatic extraction of the human cerebral ventricular system from MR neuroimages
    Kaur et al.2017|Brain tumor detection using self-adaptive K-means clustering
    Soltanian‐Zadeh et al.2003|MRI tissue characterization of experimental cerebral ischemia in rat
    Huang et al.2006|MRI brain extraction with combined expectation maximization and geodesic active contours
    Pelapur et al.2014|Multi-focus image fusion using epifluorescence microscopy for robust vascular segmentation
    Kirsch et al.2019|A probabilistic atlas of the human inner ear’s bony labyrinth enables reliable atlas-based segmentation of the total fluid space
    Yang et al.2011|Automated detection of Focal Cortical Dysplasia lesions on T1-weighted MRI using volume-based distributional features
    Hussain et al.2011|Segmentation of brain MRI with statistical and 2D wavelet features by using neural networks
    Chen et al.2015|Fiber segmentation using a density-peaks clustering algorithm
    Öziç et al.2017|A new model to determine asymmetry coefficients on MR images using PSNR and SSIM
    Chiverton et al.2008|Adaptive partial volume classification of MRI data
    Luo2006|Automated medical image segmentation using a new deformable surface model
    Angulakshmi et al.2018|Walsh Hadamard kernel‐based texture feature for multimodal MRI brain tumour segmentation
    Gtifa et al.2019|3D brain tumor segmentation in MRI images based on a modified PSO technique
    同族专利:
    公开号 | 公开日
    DE102005034374B3|2007-03-01|
    US7760939B2|2010-07-20|
    US20070019861A1|2007-01-25|
    引用文献:
    公开号 | 申请日 | 公开日 | 申请人 | 专利标题
    US5900732A|1996-11-04|1999-05-04|Mayo Foundation For Medical Education And Research|Automatic windowing method for MR images|
    US6246783B1|1997-09-17|2001-06-12|General Electric Company|Iterative filter framework for medical images|
    DE10122874B4|2001-05-11|2004-09-23|Siemens Ag|Process for extracting spin collectives with different chemical shifts from phase-coded individual images, taking into account field inhomogeneities, and device therefor|
    KR100339691B1|2001-11-03|2002-06-07|한탁돈|Apparatus for recognizing code and method therefor|
    US7343193B2|2003-06-16|2008-03-11|Wisconsin Alumni Research Foundation|Background suppression method for time-resolved magnetic resonance angiography|
    US8380280B2|2003-07-08|2013-02-19|The United States Of America, As Represented By The Secretary Of The Department Of Health And Human Services|Non-invasive in vivo MRI axon diameter measurement methods|
    EP1528407A1|2003-10-31|2005-05-04|Mitsubishi Electric Information Technology Centre Europe B.V.|Decomposition of a wideband random signal|
    DE10356275B4|2003-11-28|2008-04-17|Siemens Ag|Method for automatic segmentation of phase-coded flow images in magnetic resonance tomography|
    US20050283070A1|2004-06-21|2005-12-22|Celina Imielinska|Systems and methods for qualifying symmetry to evaluate medical images|JP5105786B2|2006-07-07|2012-12-26|株式会社東芝|Magnetic resonance imaging apparatus and image processing apparatus|
    DE102008023915A1|2008-05-16|2009-12-10|Siemens Aktiengesellschaft|Filtering control variable setting method for noise reduction in medical image in X-ray diagnostic facility, involves comparing standard noise deviation and standard structure deviation and setting control variable as function of comparison|
    US8265390B2|2008-11-11|2012-09-11|Siemens Medical Solutions Usa, Inc.|Probabilistic segmentation in computer-aided detection|
    US9014449B2|2011-10-04|2015-04-21|Siemens Aktiengesellschaft|Method and system for segmentation and removal of pulmonary arteries, veins, left atrial appendage|
    US8918159B2|2012-04-18|2014-12-23|Mayo Foundation For Medical Education And Research|System and method for improved accelerated magnetic resonance imaging using ROI masking|
    KR101382625B1|2012-10-17|2014-04-10|고려대학교 산학협력단|Apparatus and method for computing dynamic variation of target object|
    US10290092B2|2014-05-15|2019-05-14|Applied Materials Israel, Ltd|System, a method and a computer program product for fitting based defect detection|
    FR3021441B1|2014-05-26|2017-09-29|Snecma|METHOD AND DEVICE FOR ESTIMATING A QUALITY INDEX OF A 3D IMAGE OF A PIECE OF COMPOSITE MATERIAL|
    JP6215894B2|2015-10-22|2017-10-18|株式会社Ｓｃｒｅｅｎホールディングス|Image processing method and shading reference data creation method|
    法律状态:
    2007-04-02| AD1A| A request for search or an international type search has been filed|
    2008-11-03| RD2N| Patents in respect of which a decision has been taken or a report has been made (novelty report)|Effective date: 20080827 |
    优先权:
    申请号 | 申请日 | 专利标题
    DE102005034374A|DE102005034374B3|2005-07-22|2005-07-22|A method for automatically creating a background mask in images with noisy background areas, applications therefor and an MRI apparatus for performing the method and computer software product|
    DE102005034374|2005-07-22|
    [返回顶部]