System for evaluating lung images

Abstract

The present invention relates to a system for evaluating lung images, which allows for an improved assessment of the detectability of lung nodules. The lung images are smoothed to provide different degrees of smoothing for different lung images 20, 21, 22, wherein signal values ​​and noise values ​​indicative of the detectability of lung vessels and noise in these images are determined and used to determine image quality, which is indicative of the usability of the lung image to be evaluated for detecting lung nodules. Since the lung images display lung vessels with many different vessel sizes and many different image values, generally covering a wide range of potential lung nodules, image quality determination based on different lung images with different degrees of smoothing allows for a reliable assessment of the usability of the lung images for detecting lung nodules.

Description

Technical Field

[0001] The present invention relates to systems, methods and computer programs for evaluating lung images. Background Art

[0002] For detecting lung nodules, lung images, such as lung computed tomography images, are often used, wherein the noise of the lung images can be determined in order to assess the image quality of the lung images with respect to their usability for reliably detecting lung nodules. However, determining the image quality based solely on the noise is often not good enough, since the detectability of lung nodules in the lung images may also depend on other factors. Summary of the invention

[0003] SUMMARY OF THE INVENTION It is an object of the present invention to provide a system, a method and a computer program for evaluating lung images, which allow an improved assessment regarding the detectability of lung nodules in lung images.

[0004] In a first aspect of the present invention, a system for evaluating a lung image is provided, wherein the system comprises:

[0005] a lung image providing unit configured to provide a lung image, the lung image comprising image elements assigned image values ​​and showing lung blood vessels,

[0006] a smoothing unit configured to smooth the provided lung image to provide different lung images with different degrees of smoothing,

[0007] a signal value determination unit configured to determine signal values ​​for different lung images, wherein for a respective lung image one or more signal values ​​are determined based on the image value of the respective lung image, the one or more signal values ​​being indicative of detectability of a lung vessel in the respective lung image,

[0008] a noise value determination unit configured to determine noise values ​​for different lung images, wherein for a corresponding lung image, one or more noise values ​​are determined based on an image value of the corresponding lung image, the one or more noise values ​​indicating noise in the corresponding lung image,

[0009] An image quality determination unit is configured to determine an image quality of the provided unsmoothed image based on the signal value and the noise value determined for different lung images.

[0010] In a lung image, there are lung vessels with many different vessel sizes and many different image values ​​(e.g., Hounsfield density), which generally cover the corresponding range of potential lung nodules well. Since different lung images correspond to different degrees of smoothing, such as including no smoothing and including one or more different degrees of smoothing, the size distribution of the lung vessels is different in different lung images, for example, as the degree of smoothing increases, the "center of mass" of this distribution will move to a larger size. Therefore, signal values ​​and noise values ​​are determined for different distributions of lung vessel sizes in different lung images, wherein it has been found that by using these signal values ​​and noise values ​​to determine the image quality of a provided unsmoothed lung image, it is possible to very accurately determine whether the lung image is suitable for detecting lung nodules.

[0011] The lung image providing unit may be a storage unit in which lung images have been stored and from which lung images may be retrieved. The lung image providing unit may also be a receiving unit for receiving lung images from a lung image generating system such as a computed tomography system, wherein the lung image providing unit may be adapted to provide the received lung images. The lung image providing unit may also be a lung image generating system, which generates a lung image and provides the generated lung image. Preferably, the provided lung image is a computed tomography image generated by applying a radiation dose to the patient. However, the lung image may also be an image generated by another imaging method, which may also show the lung vessels.

[0012] The smoothing unit is configured to perform spatial smoothing on the image values ​​assigned to the image elements. The smoothing may be a Gaussian smoothing or another smoothing. The smoothing unit may be adapted to generate a smoothed lung image or a plurality of smoothed lung images, wherein the plurality of smoothed lung images are smoothed to different degrees of smoothness. For determining the image quality, the image quality determination unit preferably uses one or more differently smoothed lung images and also preferably uses the provided unsmoothed lung image.

[0013] In one embodiment, the signal value determination unit is configured to segment pulmonary vessels in the corresponding lung image based on image values ​​of the corresponding lung image, to determine one or several pulmonary vessel values ​​indicating the number of pulmonary vessels in the corresponding lung based on the segmented lung image of the pulmonary vessels, and to determine one or several signal values ​​of the corresponding lung image based on the determined one or several pulmonary vessel values. In particular, the signal value determination unit is configured to subdivide the segmented pulmonary vessels into cross-sectional sub-elements and to determine one or several pulmonary vessel values ​​based on a plurality of cross-sectional sub-elements. For subdividing the corresponding segmented pulmonary vessels into cross-sectional sub-elements, a predetermined thickness of the respective cross-sectional sub-elements of the corresponding segmented pulmonary vessels is preferably assumed. The number of cross-sectional sub-elements in the corresponding lung image provides a reliable measure for the detectability of the pulmonary vessels in the lung image, which may lead to an improved signal value and thus an improved determination of the image quality.

[0014] The signal value determination unit may be adapted to determine a single signal value for the corresponding lung image or to determine several signal values ​​for the corresponding lung image, wherein, in the latter case, different signal values ​​for the same lung image preferably correspond to different sizes of the lung vessels. Thus, the sizes of the lung vessels may be subdivided into size categories and for each size category and for each lung image a corresponding signal value may be determined. For example, the sizes of the cross-sectional sub-elements may be determined and subdivided into size categories and for the corresponding lung image and for the number of cross-sectional sub-elements of a size in the corresponding size category for the corresponding lung image may be considered as the signal value for the corresponding size category (i.e. for the corresponding structural size and for the corresponding lung image). The sizes of the cross-sectional sub-elements may be determined as regional equivalent diameters. However, other size metrics may also be used.

[0015] The noise value determination unit is preferably configured to determine one or several noise values ​​of the corresponding lung image based on image values ​​of the corresponding lung image representing the segmented lung vessels. Therefore, preferably, the one or several noise values ​​do not relate to the entire corresponding lung image, but only to the part of the corresponding lung image showing the lung vessels. By using these noise values ​​to determine the image quality, the determination of the image quality can be further improved. In particular, the noise value determination unit is configured to determine, for each cross-sectional sub-element, respective noise sub-values ​​indicating the noise in the respective cross-sectional sub-element, and to determine one or more noise values ​​based on the determined noise sub-values.

[0016] The noise value determination unit may be adapted to determine a single noise value for the respective lung image or to determine several noise values ​​for the respective lung image, wherein, in the latter case, different noise values ​​for the same lung image preferably correspond to different sizes of the lung vessels. Thus, the sizes of the lung vessels may be subdivided into size categories and for each size category and for each lung image a corresponding noise value may be determined. For example, the sizes of the cross-sectional sub-elements may be determined and subdivided into size categories and the average of the noise sub-values ​​of the cross-sectional sub-elements of the size in the respective size category for the respective lung image may be considered as the noise value for the respective size category, i.e. for the respective structure size and for the respective lung image. In order to determine a single noise value for a lung image, all noise sub-values ​​that have been determined for the lung image may be averaged.

[0017] Preferably, the determination of the signal value is based on the number of cross-sectional sub-elements and the determination of the noise value is based on noise sub-values ​​already determined for these cross-sectional sub-elements, wherein this consideration of the cross-sectional sub-elements leads to a reliable determination of signal and noise values ​​and thereby further improves the determination of the image quality based on these signal and noise values.

[0018] In a preferred embodiment, the signal value determination unit is adapted to determine signal values ​​for pulmonary vessels of different sizes in different pulmonary images, wherein the noise value determination unit is adapted to determine noise values ​​for pulmonary vessels of different sizes in different pulmonary images, wherein the image quality determination unit is configured to provide reference signal values ​​and reference noise values ​​for pulmonary vessels of different sizes and to determine the image quality such that it indicates a) a deviation of the determined signal value and the determined noise value from b) the provided reference signal value and reference noise value. Thus, a deviation of the determined signal and noise values ​​from the provided reference signal and noise values ​​is determined, wherein the image quality indicates the deviation.

[0019] In the provided unsmoothed lung image, lung nodules (if present) should be reliably detectable, wherein the detectability of lung nodules in the lung image generally depends on the radiation dose applied to the patient to generate the lung image, i.e., the higher the radiation dose applied to the image, the greater the detectability of the lung nodules in the lung image. On the other hand, the radiation dose applied to the patient should not be higher than the dose actually required for reliable detection of lung nodules, since the radiation dose itself may have an adverse effect on the patient, i.e., the radiation dose should be as low as reasonably achievable, wherein this requirement is also referred to as the "ALARA" principle. Reference signal and noise values ​​are preferably provided so that lung images with these values ​​comply with the ALARA principle. The determined image quality can then be used as a measure of the degree of compliance with the ALARA principle.

[0020] Preferably, the image quality determination unit is configured to: i) determine a first manifold based on signal values ​​and noise values ​​determined for different sizes of pulmonary vessels, wherein the first manifold is determined in a space defined by a signal value dimension, a noise value dimension and a pulmonary vessel size dimension; ii) determine a second manifold in the space based on reference signal values ​​and reference noise values ​​provided for pulmonary vessels of different sizes; iii) determine a distance between the first manifold and the second manifold for a plurality of positions on the first manifold; and iv) determine the image quality based on the distance. Furthermore, the image quality determination unit is preferably configured to determine an area of ​​the first manifold, i.e. an area of ​​the first surface, for which the determined distance is greater than a predetermined distance threshold, and determine the image quality based on the area. The area may be a single area, or it may contain several separate sub-areas. In particular, the image quality determination unit is configured to determine the image quality depending on the size of an area above the second manifold, i.e. above the second surface, and / or depending on the size of an area below the second manifold in the defined space. For example, if the deviation measure indicates that the size of the area below the second manifold is larger than a predetermined size threshold, the image quality determination unit may be configured to determine an insufficient image quality. Thus, as a deviation measure between the determined signal and noise values ​​and the reference signal and noise values, the size of the area above the second manifold defined by the reference values ​​and / or the size of the area below the second manifold may be used to determine the image quality. By using the size of the area (which may include several individual sub-areas), the determination of the image quality may be further improved.

[0021] In one embodiment, the lung image providing unit is configured to provide a slice image with a slice thickness as the lung image, wherein the signal value determination unit is suitable for normalizing the corresponding signal values ​​with respect to the slice thickness. The slice image may have one or several voxels in the slice direction, i.e. in the direction perpendicular to the plane in which the lung image mainly extends in this embodiment. In addition, the signal value determination unit may be configured to determine the lung space as the space occupied by the lungs within at least one of the lung images, and to normalize the individual signal values ​​with respect to the determined lung space. Preferably, the lung space is determined in the provided unsmoothed lung image, and the lung space may be a lung volume. The lung space, i.e. the size of the space within the lung image covered by the lungs, can be determined by using known lung segmentation techniques. By using these normalizations, the quality of the signal values ​​and therefore the determination of the image quality based in particular on the signal values ​​can be further improved.

[0022] The system may further include a dose level determination unit configured to determine a radiation dose level to be applied to generate a next lung image based on the determined image quality. In particular, the lung image providing unit may be configured to provide a lung image such that it is an image that has been generated by applying a certain radiation dose level to the lungs, wherein the dose level determination unit may be configured to: if the determined image quality is greater than a predetermined quality threshold, determine the radiation dose level to be applied to generate the next lung image such that the radiation dose level is less than the specific radiation dose level. For example, if the image quality has been determined based on the size of the area above the second manifold of the first manifold as described above, wherein the predetermined quality threshold corresponds to a predetermined size threshold of the area, an image quality greater than the predetermined quality threshold may indicate that the image quality is higher than the image quality required to reliably detect lung nodules in the provided unsmoothed lung image, wherein this allows a reduction in the radiation dose level applied to the lungs. Likewise, if the image quality is determined not by using the size of the region, but by using another method to determine the deviation between the determined signal and noise values ​​and the provided reference signal and noise values, a determined image quality greater than a predefined quality threshold may indicate an unnecessarily high image quality, which allows a reduction in the radiation dose level. This may lead to a better alignment of the generation of the next lung image using the ALARA principle.

[0023] The system may further comprise a user interface allowing a user to modify the provided lung image, wherein the system may be adapted to determine the image quality based on the modified lung image. Thus, all processing steps may be performed based on the modified lung image. This allows the user to interactively modify the lung image to determine how characteristics of the provided unsmoothed lung image affect the image quality. For example, the user may interactively change settings such as dynamic window / level and spatial screen resolution.

[0024] In another aspect of the present invention, a method for evaluating a lung image is provided, wherein the method comprises:

[0025] A lung image is provided by a lung image providing unit, the lung image comprising image elements which are assigned image values ​​and which show lung vessels,

[0026] smoothing the provided lung image by a smoothing unit to provide different lung images with different degrees of smoothing,

[0027] determining, by a signal value determination unit, signal values ​​for different lung images, wherein for a respective lung image one or more signal values ​​are determined based on the image value of the respective lung image, the one or more signal values ​​being indicative of detectability of the lung vessels in the respective lung image,

[0028] determining, by a noise value determination unit, noise values ​​for the different lung images, wherein for a respective lung image, one or more noise values ​​are determined based on the image value of the respective lung image, the one or more noise values ​​being indicative of noise in the respective lung image,

[0029] An image quality of the provided unsmoothed image is determined by an image quality determination unit based on the signal values ​​and noise values ​​determined for the different lung images.

[0030] In another aspect of the present invention, a computer program for evaluating lung images is proposed, wherein the computer program includes a program code unit, and when the computer program is run on a computer controlling the system, the program code unit is used to cause the system for evaluating lung images according to claim 1 to perform the steps of the method for evaluating lung images according to claim 14.

[0031] It shall be understood that the system according to claim 1, the method according to claim 14 and the computer program according to claim 15 have similar and / or identical preferred embodiments, in particular as defined in the dependent claims.

[0032] It shall be understood that a preferred embodiment of the invention can also be any combination of the dependent claims or above embodiments with the respective independent claim.

[0033] These and other aspects of the invention will be apparent from and elucidated with reference to the embodiments described hereinafter. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] In the following figures:

[0035] Figure 1 An embodiment of a system for evaluating lung images is shown schematically and exemplarily,

[0036] Figure 2 The dependence of several detectable pulmonary vessels on image smoothness is shown.

[0037] Figure 3 shows a flow chart exemplarily illustrating an embodiment of a method for evaluating a lung image,

[0038] Figure 4 Graphs are shown by way of example, which illustrate the variation of the number of cross-sectional sub-elements and the noise of a pulmonary vessel with different degrees of smoothing for a single structure size,

[0039] Figure 5A graph is shown by way of example, which shows the relationship between the size of the structure and the Figure 4 The curves shown here are affected by

[0040] Figure 6 and 7 Graphs are exemplarily shown which illustrate how the number of detectable cross-sectional sub-elements and the noise of pulmonary vessels differ for different provided lung images. DETAILED DESCRIPTION

[0041] Figure 1 An embodiment of a system for evaluating a lung image is schematically and exemplarily shown. System 1 comprises: a lung image providing unit, which is configured to provide a lung image, the lung image comprising image elements to which image values ​​are assigned and showing lung vessels. In this embodiment, the lung image providing unit 2 is a storage unit storing a lung image, wherein the storage unit is suitable for providing the stored lung image. Furthermore, in this embodiment, the lung image is a computed tomography image of the lungs of a patient. The patient is preferably a human being. However, the patient may also be an animal.

[0042] The system 1 further comprises a smoothing unit 3 for spatially smoothing the provided lung image with different degrees of smoothing in order to generate different smoothed lung images. In this embodiment, the smoothing unit 3 is adapted to apply Gaussian image smoothing. The obtained lung images correspond to different scale levels, or in other words, to different scale space levels.

[0043] The system 1 further comprises a signal value determination unit 4, which is configured to determine signal values ​​for the differently smoothed lung images and for the provided unsmoothed lung images, wherein for the respective lung images, one or several signal values ​​are determined based on the image values ​​of the respective lung images, the one or several signal values ​​indicating the detectability of the pulmonary vessels in the respective lung images. In this embodiment, the signal value determination unit is configured to segment the pulmonary vessels in the respective lung images based on the image values ​​of the respective lung images, to determine a pulmonary vessel value indicating the number of pulmonary vessels in the respective lungs based on the segmented lung images of the pulmonary vessels, and to determine several signal values ​​of the respective lung images based on the determined pulmonary vessel values. In particular, the signal value determination unit 4 is configured to subdivide the segmented pulmonary vessels into cross-sectional sub-elements, determine the sizes of the cross-sectional sub-elements, subdivide the sizes into size categories, and for each size category determine the respective pulmonary vessel values ​​based on the number of cross-sectional sub-elements having the sizes in the respective size category. For determining the cross-sectional sub-elements, a predetermined thickness of the respective cross-sectional sub-elements may be used. However, another technique may be used to determine the cross-section sub-elements.

[0044] Figure 2 Schematically and exemplarily, three lung images 20, 21, 22 are shown, which have been smoothed with different degrees of smoothing, wherein the degree of smoothing (i.e., image smoothness) is in the range of Figure 2 It can be seen that the number of detectable pulmonary vessels decreases from left to right (i.e., as the image smoothness increases).

[0045] In this embodiment, the lung image providing unit 2 is configured to provide a slice image with a slice thickness as an unsmoothed lung image, wherein the signal value determination unit 4 is suitable for normalizing the corresponding signal values ​​with respect to the slice thickness. In addition, the signal value determination unit 4 is configured to determine the lung volume within the provided unsmoothed lung image and to normalize the respective signal values ​​with respect to the determined lung volume. Therefore, the size of the volume occupied by the lungs in the provided unsmoothed lung image can be determined by, for example, segmenting the lungs within the lung image, wherein the obtained size can be used to normalize the determined signal values.

[0046] The system 1 further comprises a noise value determination unit 5, which is configured to determine noise values ​​in different smoothed lung images and the provided unsmoothed lung images, wherein for the respective lung images, one or several noise values ​​are determined based on the image values ​​of the respective lung images, the one or several noise values ​​indicating the noise in the respective lung images. In particular, the noise value determination unit 5 is configured to determine one or several noise values ​​such that they indicate the noise in the image values ​​representing the pulmonary vessels in the respective lung images. Therefore, the noise value determination unit 5 is adapted to determine one or more noise values ​​for the respective lung images based on the image values ​​of the respective lung images representing the segmented pulmonary vessels. In this embodiment, the noise value determination unit 5 is configured to determine for each cross-sectional sub-element a corresponding noise sub-value indicating the noise in the corresponding cross-sectional sub-element, and to determine the noise value for the respective size (i.e., for the respective size category) based on the noise sub-value determined for the cross-sectional sub-element in the respective size category. In order to determine the noise value, the noise value determination unit 5 may use the absolute magnitude of the Heinz Laplace operator. However, other known techniques for determining a noise value based on image values ​​of an image may also be used to determine the noise value by the noise value determination unit 5 .

[0047] The system further comprises an image quality determination unit 6, which is configured to determine the image quality of the provided unsmoothed image based on the signal values ​​and noise values ​​determined for the differently smoothed lung images and the provided unsmoothed lung images. In particular, the image quality determination unit 6 is configured to provide reference signal values ​​and reference noise values ​​for different sizes of lung vessels, and to determine the image quality so that it indicates a) the deviation between the signal values ​​and noise determined for the different sizes and b) the reference signal values ​​and reference noise values ​​provided for the different sizes. In this embodiment, the image quality determination unit 6 is configured to: i) determine a first surface based on the signal values ​​and noise values ​​determined for the different sizes, wherein the surface is determined in a three-dimensional space defined by possible values ​​of the signal values, the noise values ​​and the different sizes of the lung vessels, ii) determine a second surface in the three-dimensional space based on the reference signal values ​​and reference noise values ​​provided for the different sizes; iii) determine the distance between the first surface and the second surface for a plurality of positions on the first surface, and iv) determine the image quality based on the distance. For example, the image quality determination unit 6 can be configured to determine an area of ​​the first surface for which the determined distance is greater than a predefined distance threshold, and determine the image quality based on the area, the area can be a single area or it can include several separate sub-areas, wherein all sub-areas can be above the second surface, below the second surface, or wherein some sub-areas can be above the second surface and some areas can be below the second surface. In particular, the image quality determination unit 6 can be configured to determine the image quality based on the size of the area above the second surface in the three-dimensional image and / or based on the size of the area below the second surface. In one embodiment, the image quality determination unit 6 is configured to: if the size of the area above the second surface is greater than a predefined first size threshold, then determine that the image quality is too high and can be reduced. In addition, the image quality determination unit 6 can be configured to: if the size of the area below the second surface for which the determined distance is greater than a predefined distance threshold is greater than a second size threshold, then determine that the image quality is too low and therefore insufficient to reliably detect lung nodules. The first and second size thresholds can be the same or they can be different. All thresholds, such as size thresholds and distance thresholds, can be predetermined by a calibration process and / or can be modified by a user. The image quality determination unit 6 may also be adapted to determine that the image quality is desired, in particular according to the ALARA principle, if a size of an area of ​​the first surface for which the distance determined is greater than a predefined distance threshold is smaller than a first size threshold above the first surface and smaller than a second size threshold below the second surface.

[0048] The system 1 also includes a dose level determination unit 7, which is configured to determine the radiation dose level to be applied to generate the next lung image based on the determined image quality. In this embodiment, the lung image providing unit 2 is configured to provide a lung image so that it is an image that has been generated by applying a certain radiation dose level to the lungs, wherein the dose level determination unit 7 is configured to: if the determined image quality is greater than a predetermined quality threshold, determine the radiation dose level to be applied to generate the next lung image so that the radiation dose level is less than the specific radiation dose level. This means in particular that if the size of the area of ​​the first surface (for which the determined distance is greater than a predefined distance threshold) is greater than a first size threshold above the second surface, the radiation dose level to be applied to generate the next lung image is determined so that it is less than the specific radiation dose level.

[0049] The system 1 further comprises a user interface 8, which allows a user to modify the provided unsmoothed lung image, wherein the system 1 is adapted to determine the image quality based on the modified lung image. Hence, all processing steps may be performed based on the modified lung image. Hence, the user may change settings such as dynamic window / level and / or spatial screen resolution, wherein processes like smoothing, signal value determination, noise value determination and image quality determination are then performed based on the modified lung image.

[0050] The system 1 further comprises an input unit 9, like a keyboard, a computer mouse, a touch screen, etc., and an output unit 10, comprising a display for displaying, for example, an image of the lungs and for indicating the determined image quality. The output unit 10 may also comprise an acoustic unit for acoustically indicating the determined image quality.

[0051] In the following, reference will be made to Figure 3 The shown flow chart describes an embodiment of a method for evaluating a lung image.

[0052] In step 101, a lung image is provided by a lung image providing unit 2, the lung image comprising image elements assigned image values ​​and showing lung vessels. In step 102, the provided lung image is smoothed to different degrees of smoothness by a smoothing unit 3, signal values ​​are determined for the differently smoothed lung images and the unsmoothed lung images by a signal value determination unit 4, wherein a signal value for the respective lung images is determined based on the image values ​​of the respective lung images, the signal value indicating the detectability of lung vessels having the respective sizes in the respective lung images, and noise values ​​for the different lung images are determined by a noise value determination unit 5, wherein noise values ​​are determined for the respective lung images based on the image values ​​of the respective lung images for different sizes of the lung vessels, the noise values ​​indicating the noise in the respective lung images. In step 103, an image quality determination unit 6 determines the image quality of the provided unsmoothed image based on the signal values ​​and noise values ​​determined for the different lung images.

[0053] Generally, it is recognized that screening for lung cancer with low-dose computed tomography is effective for detecting lung nodules, wherein image quality is highly desirable, especially to avoid oversight. Screening for lung cancer should, on the one hand, use the lowest radiation dose reasonably achievable, while on the other hand, maintaining an image quality level that is sufficient to ensure that possible lung tumors, i.e., possible lung nodules, are detected. Therefore, screening for lung cancer should comply with the ALARA principle. The above-described system and method for evaluating lung images allow determination of whether the image quality (i.e., image quality level) of the provided lung images is sufficient to achieve the ALARA principle. In addition, the system and method for evaluating lung images are vendor-independent and can be retrospectively applied to a batch of easily reconstructed lung images that can be received from a picture archiving and communication system (PACS). In addition, the determined image quality is specific to the corresponding lung images and therefore to the corresponding patients and the corresponding imaging schemes that have been used to generate the provided lung images. The pulmonary vessels detectable in the provided lung images are of course patient-specific, and they are likely and very reliably present in the lung images, making pulmonary vessels of different sizes very suitable for marking for image quality.

[0054] The system and method for evaluating a lung image automatically quantitatively evaluates the lung image based on the signal values ​​(i.e., detectability) of lung vessels of different sizes in the provided lung image and based on noise values ​​determined for the portion of the lung image represented by the lung vessels, wherein the noise values ​​for the lung vessels of different sizes are also determined. Pulmonary vessels are always present in lung images, wherein the range of sizes of the lung vessels is large, and in the case of computed tomography, the Hounsfield density covers the range of potential lung nodules very well. The idea is that if a significant deficiency is observed in several detectable, especially visible, lung vessels, then this may be a strong indicator that the image quality is not sufficient to reliably detect potential nodules. The density of blood vessels may vary between patients, but the average density in a group of patients is much more uniform than the number of tumors (i.e., the number of lung nodules). Therefore, determining an average image quality by averaging the image qualities determined for different images of different patients can result in a more robust quality metric, i.e., a more robust image quality, which can indicate the quality of the imaging system (or more generally, for example, the screening center at which the image is generated).

[0055] The above-mentioned first surface is a first manifold in a space spanned by three-dimensional signal, noise and structure size (i.e., the size of the pulmonary blood vessels in the lung image), wherein the first manifold is used to evaluate the quality of the provided lung image. If too much of the first manifold is lower than the ideal reference manifold, i.e., lower than the second surface, it is considered that the provided lung image does not meet the necessary image quality requirements. In another embodiment, the second surface, i.e., the reference manifold, may correspond to a manifold from another imaging system, i.e., another imaging system that is not used to generate the provided lung image, such as an imaging system of another screening center, if the provided lung image has been generated by the imaging system of the screening center, thereby comparing the image quality of different screening centers, for example. The system and method for evaluating lung images not only considers a single signal / noise operating point, but establishes and compares a determined first manifold, i.e., a determined first surface, for the possible size range of possible lung nodules in the lung image.

[0056] The lung image providing unit 2 may be adapted to provide a lung image by identifying a lung volume in a provided overall chest computed tomography volume image and by segmenting the identified lung volume from the overall chest computed tomography volume image to provide a lung image which substantially only includes the lung volume and not parts of the lung image outside the lung volume. The segmentation of the pulmonary vessels in the corresponding lung image may be regarded as an automatic detection of image structures in the corresponding lung image corresponding to the pulmonary vessels, wherein the segmentation process may use known segmentation techniques, such as a joint component analysis for multiple Hounsfield thresholds, wherein from the obtained structures, structures having a two-dimensional cross-section and whose area size is less than a predetermined upper area limit may be selected. Figure 2 The corresponding final segmented pulmonary vessels are schematically and exemplarily shown in FIG.

[0057] The system and method for evaluating lung images calculates for each blood-like structure object (i.e., in the above-described embodiment) for each cross-sectional sub-element of the segmented lung vessels, the size and the noise by using the example area equivalent diameter as the size and the absolute amplitude of the Heinz Laplace slice for determining the noise. In a preferred embodiment, the number of detected vessel cross-sectional sub-elements (which can also be regarded as frequencies) for each lung volume is accumulated into a histogram for different lung images according to the structure size and the noise value. The resulting graph is exemplarily shown for a single structure size, which also takes into account the normalization with respect to the slice thickness of the lung image and the normalization for the volume of the lung in the lung image.

[0058] exist Figure 4 In FIG. 1 , arrow 30 indicates the direction of increasing number of cross-sectional sub-elements, i.e. increasing signal value, while arrow 31 indicates increasing noise, i.e. increasing noise value. Figure 4 In , point 32 corresponds to a differently smoothed lung image, while point 33 corresponds to the lung image provided initially without smoothing. Line 34 is a point passing through these points. Thus, Figure 4 The number of vessel cross sections, which can also be viewed as the frequency of vessel cross sections due to normalization with respect to the lung volume, and the noise values ​​for different degrees of smoothing and individual structure sizes are shown, where in this example, as described above, the noise is estimated as the absolute magnitude of the Heinz Laplace slice, i.e. the deviation from the linear interpolation of neighboring voxels. As indicated by point 33, the original lung image has the highest noise and the highest number of vessel cross sections before any smoothing. With increasing smoothing, which can be obtained by successively applying a smoothing operation such as a Gaussian smoothing operation to the lung image, the noise level of the cross sections decreases, but the number of detectable cross sections decreases.

[0059] Figure 4 The relationship between the number of vessel cross sections, i.e. the signal value and the noise value, is shown for a specific structure size or for a specific size class (i.e. in this embodiment for an area equivalent diameter or a diameter class of cross-sectional sub-elements detectable in the lung image). Figure 5 The relationship between the number of cross sections and the noise is shown by way of example for different structural sizes, ie in this example for different area equivalent diameters of the segmented pulmonary vessels. Figure 5 In FIG. 3 , arrow 35 indicates the direction of increasing structural size, so that curve 36 corresponds to the smallest structural size and curve 37 corresponds to the largest structural size. Figure 5As shown, for small structure sizes, the number of detectable pulmonary vessels and therefore the number of detectable pulmonary vessel cross-section sub-elements mainly decreases, while for larger structure sizes it remains relatively stable. These curves together span a first surface (i.e., a first manifold) in a three-dimensional space defined by the number of vessel cross-sections (i.e., possible signal values ​​indicated by arrow 30), the possible signal values ​​indicated by arrow 31, and the possible structure sizes indicated by arrow 35. Therefore, the three-dimensional space is spanned by the signal dimension indicated by arrow 30, the noise dimension indicated by arrow 31, and the structure size dimension indicated by arrow 35. Preferably, based on Figure 5 The curve shown determines the first surface, so that the first surface is a directional surface with a direction pointing to a higher detection frequency. The first surface is preferably determined by fitting to a signal value-noise value-structure size point in a three-dimensional space, wherein in this embodiment, the signal value is defined by the number of cross-sectional sub-elements of the segmented pulmonary blood vessels with the corresponding structure size. The fitting can be, for example, piecewise linear fitting, polynomial fitting, spline fitting, support vector fitting, etc.

[0060] Figure 6 Two curves 38, 39 are shown exemplarily, which correspond to different reconstructions of the same computed tomography scan, i.e. the same computed tomography projection data have been used to differently reconstruct two different lung images, wherein the first curve 38 has been determined based on a first reconstructed computed tomography image and the second curve 39 has been determined based on a second reconstructed computed tomography image. In this example, the first reconstructed computed tomography image has been reconstructed such that it corresponds to relatively high spatial frequencies and the second reconstructed image has been reconstructed such that it corresponds to relatively low spatial frequencies. In this sense, the first image can also be regarded as a hard image and the second image as a soft image. The correspondingly provided unsmoothed images are indicated by points 40, 45, respectively. The first high-frequency image has more noise and more structure, i.e. more cross-sectional sub-elements of the segmented lung vessels, than the second low-frequency image. However, with increasing smoothness, i.e. for example after successive Gaussian smoothing, the curves become very close, indicating similar image quality. Figure 6 A curve for a single structural size is shown, i.e., in this embodiment, a curve for a single area equivalent diameter. For other structural sizes, corresponding curves will also be determined so that the curves span corresponding surfaces in the three-dimensional space defined above, wherein the two surfaces are very close as the degree of smoothing increases.

[0061] Figure 7Schematically, four curves 41 ... 44 are illustrated for a single structure size (i.e. for a single size class), wherein the different curves 41 ... 44 correspond to four different provided lung images. It can be seen that the provided lung image corresponding to curve 41 shows significantly more structures than the other provided lung images at almost all noise levels. Here too, for various structure sizes (in Figure 7 , the curves correspond only to a single structural dimension), the four curves 41 . . . 44 are summarized as four two-dimensional surfaces, which can be compared, for example, by height or relative distance.

[0062] In one embodiment, a two-dimensional surface in three-dimensional space, i.e., a first surface, is determined for the provided lung image, and the two-dimensional first surface is compared with a reference surface corresponding to the desired image quality, thereby generating a measure of the image quality of the provided lung image. For example, a marker distance from the first surface to the second reference surface can be determined and used to evaluate the image quality of the provided lung image. The reference surface can also be considered as an image quality surface, which can be taken from a reference base or a specific screening center. The reference can be stratified according to patient groups (e.g., male / female, age, body mass index, etc.). Therefore, for each patient, a reference surface corresponding to the corresponding patient can be provided with respect to features such as gender, age, etc. In one embodiment, if too many areas of the directional surface are below the reference quality surface, i.e., below the second surface, the image quality determination unit 6 is adapted to determine that the image quality of the provided lung image is not sufficient for lung nodule detection. Conversely, if too many areas of the surface are above the second surface, it can be determined that the image acquisition dose can and should be reduced to meet the ALARA principle. Whether there are too many areas below or above the second surface can be determined based on a predefined area size threshold as described above, for example, by calibration. By using a user interface, this quality check can be implemented for interactive viewing software. The user can change settings such as dynamic window / level, spatial screen resolution, etc., wherein a quality check can be performed for each interactively selected setting, i.e., determining the image quality, wherein the user can be warned if the resulting display quality does not appear to be sufficient for thorough detection of nodules.

[0063] The above-described system and method for evaluating lung images do not require scanning of a physical phantom. The quantitative image quality metric provided is preferably not specific to a general scan type or reconstruction type, but rather it is specific to the respective patient, i.e., for example, to the patient's size, weight, bones, etc. In addition, it is specific to the actual imaging protocol, i.e., for example, in the case of computed tomography imaging, to the tube current, tube voltage, etc. It is also specific to the resolution change along the actual process of dynamic dose modulation during the entire scan, i.e., the dose is automatically changed for denser areas and the root and apex of the lung (if present). In addition, it is specific to the loss of resolution that may result from a suboptimal position of the patient relative to the center of the field of view, i.e., specific to the possible gradual reduction of resolution towards off-center positions. It may also be specific to a suboptimal choice of the reconstruction field of view, i.e., an unnecessarily large field of view limits the voxel spacing. Finally, it is specific to the respectively selected reconstruction algorithm, i.e., for example, to the kernel, whether it is an iterative reconstruction, etc. The system and method for evaluating lung images covers a range of sizes and image values ​​of interest, i.e., in the case of computed tomography of Hounsfield density for lung nodule detection. Furthermore, the assessment of the lung image quality can be retroactively applied to all manufacturers of imaging systems. It can also be applied to a batch of images, such as images within a specific time period or in a specific patient population, or to a single image. If a batch of images is considered, a corresponding image quality can be determined for each image, wherein these image qualities can be combined to determine the image quality of the batch of images. For example, the image qualities can be averaged and / or the standard deviation of these image qualities can be determined in order to determine the image quality for the batch of images. In order to determine the image quality for the corresponding image distance between the corresponding first surface and the reference surface, it can be combined, in particular averaged. Thus, the deviation between the two-dimensional noise resolution surface and the reference surface can be condensed into a single signed scalar number by, for example, averaging the distances below and above the reference surface, thereby generating a number for each image, wherein, in order to determine the image quality of the batch of images, the number of different images that may correspond to different patients can be averaged and / or the standard deviation can be determined. The image quality can be determined relatively quickly, so that if necessary, the patient can be rescanned relatively quickly, in particular before the patient leaves the hospital. Systems and methods for evaluating lung images may be applied to, for example, low-dose screening computed tomography images, as well as to other computed tomography images, and may also be applied to other images showing pulmonary vasculature that may not be generated using a computed tomography imaging system.

[0064] Although in the above described embodiments a three-dimensional space is considered, which is defined by a signal dimension, a noise dimension and a structure size dimension, in another embodiment, additional dimensions may be considered, such as structural contrast, wherein, in this case, individual three-dimensional manifolds, i.e., three-dimensional surfaces, are defined in a corresponding four-dimensional space and compared to determine the image quality. In particular, as the smoothness increases, low-contrast structures begin to disappear earlier than high-contrast structures. In one embodiment, the structural contrast is defined as the difference between the average brightness (e.g., average Hounsfield density) of a cross section of a vessel and the average brightness of its neighborhood (i.e., its directly adjacent image background).

[0065] Other variations to the disclosed embodiments can be understood and effected by those skilled in the art in practicing the claimed invention, from a study of the drawings, the disclosure, and the appended claims.

[0066] In the claims, the word "comprising" does not exclude other elements or steps, and the word "a" or "an" does not exclude a plurality.

[0067] A single unit or device may fulfill the functions of several items recited in the claims. The mere fact that certain measures are recited in mutually different dependent claims does not indicate that a combination of these measures cannot be used to advantage.

[0068] Processes such as providing a lung image, smoothing the provided lung image, segmenting lung vessels in the lung image, determining cross-sectional sub-elements of lung values, determining the size of the cross-sectional sub-elements, determining the number of cross-sectional sub-elements, determining the noise value of the cross-sectional sub-elements, determining the image quality performed by one or several units or devices, etc. can be performed by any other number of units or devices. For example, these processes can be performed by a single device. These processes and / or the control of the system for evaluating a lung image according to the method for evaluating a lung image can be implemented as program code modules of a computer program and / or dedicated hardware.

[0069] The computer program can be stored / distributed on suitable media, such as optical storage media or solid-state media provided together with other hardware or as part of other hardware, but the computer program can also be distributed in other forms, such as via the Internet or other wired or wireless telecommunication systems.

[0070] Any reference signs in the claims should not be construed as limiting the scope.

[0071] The present invention relates to a system for evaluating lung images, which allows for an improved assessment of the detectability of nodules. The lung images are smoothed to provide different degrees of smoothing for different lung images, wherein signal values ​​and noise values ​​indicative of the detectability of lung vessels and noise in these images are determined and used to determine image quality, which is indicative of the usability of the lung image to be evaluated for detecting lung nodules. Since lung images display lung vessels with many different vessel sizes and many different image values, generally covering a wide range of potential lung nodules, image quality determination based on different lung images with different degrees of smoothing allows for a reliable assessment of the usability of lung images for detecting lung nodules.

Claims

1. A system for evaluating a lung image, the system (1) comprising: a lung image providing unit (2) configured to provide a lung image, the lung image comprising image elements to which image values ​​are assigned and which show lung vessels, a smoothing unit (3) configured to smooth the provided lung image to provide different lung images with different degrees of smoothing, a signal value determination unit (4) configured to determine signal values ​​for the different lung images (20, 21, 22), wherein for a respective lung image one or several signal values ​​are determined based on the image value of the respective lung image, the one or several signal values ​​indicating the detectability of the lung vessels in the respective lung image, a noise value determination unit (5) configured to determine noise values ​​for the different lung images (20, 21, 22), wherein for a respective lung image one or several noise values ​​are determined based on the image value of the respective lung image, the one or several noise values ​​being indicative of noise in the respective lung image, An image quality determination unit (6) configured to determine the image quality for the provided unsmoothed image based on the signal value and the noise value determined for the different lung images (20, 21, 22), wherein the system (1) also includes a dose level determination unit (7) configured to determine the radiation dose level to be applied to generate a next lung image based on the determined image quality.

2. The system according to claim 1, wherein: The signal value determination unit (4) is configured to segment the pulmonary vessels in the corresponding lung image based on the image values ​​of the corresponding lung image, determine one or several pulmonary vessel values ​​indicating the amount of pulmonary vessels in the corresponding lung image based on the segmented pulmonary vessels, and determine the one or several signal values ​​for the corresponding lung image based on the determined one or several pulmonary vessel values.

3. The system according to claim 2, wherein: The signal value determination unit (4) is configured to subdivide the segmented pulmonary vessels into cross-sectional sub-elements and determine the one or several pulmonary vessel values ​​based on the number of cross-sectional sub-elements.

4. The system according to claim 2, wherein: The noise value determination unit (5) is configured to determine the one or several noise values ​​for the corresponding lung image based on image values ​​of the corresponding lung image representing the segmented lung vessels.

5. The system according to claim 3, wherein: The noise value determination unit (5) is configured to determine, for each cross-section sub-element, a corresponding noise sub-value indicating noise in the corresponding cross-section sub-element, and determine the one or several noise values ​​based on the determined noise sub-values.

6. The system according to claim 1, wherein: The signal value determination unit (4) is adapted to determine the signal values ​​for the pulmonary vessels of different sizes in the different pulmonary images, wherein the noise value determination unit (5) is adapted to determine the noise values ​​for the pulmonary vessels of different sizes in the different pulmonary images, wherein the image quality determination unit (6) is configured to provide reference signal values ​​and reference noise values ​​for pulmonary vessels of different sizes, and to determine the image quality such that the image quality indication is a) a deviation of the determined signal value and the determined noise value from b) the provided reference signal value and reference noise value.

7. The system according to claim 6, wherein: The image quality determination unit (6) is configured to: i) determine a first manifold based on the signal values ​​and the noise values ​​determined for the different sizes of the pulmonary blood vessels, wherein the first manifold is determined in a space defined by a signal value dimension, a noise value dimension, and a pulmonary blood vessel size dimension; ii) determine a second manifold in the space based on the reference signal values ​​and the reference noise values ​​provided for the pulmonary blood vessels of the different sizes, iii) determine a distance between the first manifold and the second manifold for a plurality of positions on the first manifold, and iv) determine the image quality based on the distance.

8. The system according to claim 7, wherein: The image quality determination unit (6) is configured to: determine an area of ​​the first manifold, for which the determined distance is greater than a predefined distance threshold; and determine the image quality based on the area.

9. The system according to claim 8, wherein: The image quality determination unit (6) is configured to determine the image quality in three-dimensional space according to the size of the area above the second manifold and / or according to the size of the area below the second manifold.

10. The system according to claim 1, wherein: The signal value determination unit (4) is configured to determine the lung space as the space occupied by the lungs within at least one of the lung images (20, 21, 22), and to normalize the corresponding signal values ​​with respect to the determined lung space.

11. The system according to claim 1, wherein: The lung image providing unit (2) is configured to provide a slice image with a slice thickness as the lung image, wherein the signal value determination unit (4) is adapted to normalize the corresponding signal value with respect to the slice thickness.

12. The system of claim 1, wherein: The lung image providing unit (2) is configured to provide the lung image so that the lung image is an image that has been generated by applying a specific radiation dose level to the lungs, wherein the dose level determining unit (7) is configured to determine the radiation dose level to be applied to generate the next lung image if the determined image quality is greater than a predefined quality threshold, so that the radiation dose level is less than the specific radiation dose level.

13. A method for evaluating a lung image, the method comprising: A lung image is provided by a lung image providing unit (2), the lung image comprising image elements which are assigned image values ​​and which show lung vessels, The provided lung image is smoothed by a smoothing unit (3) to provide different lung images with different degrees of smoothing, signal values ​​for the different lung images (20, 21, 22) are determined by a signal value determination unit (4), wherein for a respective lung image one or more signal values ​​are determined based on the image values ​​of the respective lung image, the one or more signal values ​​being indicative of the detectability of the lung vessels in the respective lung image, Noise values ​​for the different lung images (20, 21, 22) are determined by a noise value determination unit (5), wherein for a respective lung image one or more noise values ​​are determined based on the image value of the respective lung image, the one or more noise values ​​being indicative of noise in the respective lung image, An image quality determination unit (6) determines an image quality for the provided unsmoothed image based on the signal value and the noise value determined for the different lung images (20, 21, 22), A radiation dose level to be applied to generate a next lung image is determined by a dose level determination unit (7) based on the determined image quality.

14. A computer program product for evaluating lung images, the computer program product comprising a program code unit, which, when the program code unit is run on a computer controlling the system, causes the system for evaluating lung images according to claim 1 to perform the steps of the method for evaluating lung images according to claim 13.

Images