Vendor agnostic AI image processing

Abstract

The image is processed while keeping agnostic to the image source. An input image to be processed is retrieved. A first value or a first set of values of an image metric associated with the input image is determined, and a first filter is generated based on a relationship between the first value or the first set of values and a target value or a set of target values. The first filter is then applied to the input image to generate a working image having a second value or a second set of values of the image metric, the second value or the second set of values being substantially similar to the target value or the set of target values of the image metric. The work image is processed using a standardized image processing method. An output, which may be an image, is generated based on the processed working image and output.

Description

Technical Field

[0001] The present disclosure generally relates to systems and methods for processing images using trained neural networks. In particular, the present disclosure relates to systems and methods for processing images while remaining agnostic to the image source. Background Art

[0002] Conventionally, images are acquired through standard imaging modalities (e.g., computed tomography (CT) scans) resulting in image artifacts and noise embedded in these images. Additionally, systems processing such images may seek to preemptively obtain some information about the scanned content (e.g., identification of the content). Therefore, the images are typically filtered and reconstructed to initially convert the measurement data into an image, which is then processed using algorithms (e.g., denoising, segmentation, or preemptive identification of the content).

[0003] It is desirable to reduce the radiation levels applied to patients, and AI processing such as denoising is a powerful tool to reduce the radiation dose actually applied to patients and has been implemented regularly. AI-based methods allow for improved image quality in such reduced-dose imaging by removing artifacts associated with low doses. Similar AI-based methods can be used to provide additional imaging benefits (e.g., metal artifact reduction and motion artifact reduction) as well as analytical benefits (e.g., segmentation and classification of image content).

[0004] Therefore, various AI-based image processing methods are often used to process the input images before they are presented to the user. However, AI image processing can usually only be reliably performed when the input images provided to the AI ​​tool (e.g., convolutional neural network (CNN)) are within the image distribution range used to train the corresponding algorithm or network. In addition, given the limited capacity of the neural network, the network will perform better if trained on a relatively narrow image distribution. For example, a network trained on body images will generally perform better on body images than a network trained on body images and head images.

[0005] Another example of such a generalization problem involves using standardized reconstruction filters for AI denoising, for example, to support a wide range of reconstruction filters to be applied later. In some embodiments, a dedicated reconstruction pipeline can be utilized, where the input image to the neural network is generated using standard high-resolution filters, and the desired filter characteristics are applied only after the general AI denoising step.

[0006] While this approach to a generalized reconstruction pipeline may be feasible if the entire reconstruction chain is under control so that the received image has the characteristics expected by the AI-based image processing method to be implemented, this approach cannot be applied to images generated by other vendors that have characteristics different from what is expected.

[0007] There is a need for machine learning algorithms and processes for CT image processing that are independent of the characteristics of the input images to be processed. Summary of the invention

[0008] A method for machine learning based image processing is provided. The provided method allows processing images according to standardized image processing while remaining agnostic to the input image source.

[0009] A method is provided in which an input image to be processed is retrieved, for example from an imaging system. The method then determines a first value or a first set of values ​​of an image metric associated with the input image.

[0010] The method then generates a first filter based on a relationship between the first value or the first set of values ​​of the image metric and a target value or a set of target values ​​for the image metric.

[0011] The method then applies the first filter to the input image to generate a working image having a second value or a second set of values ​​of the image metric that is substantially similar to the target value or the set of target values ​​of the image metric.

[0012] The working image is then processed using a standardized image processing method based on the target value or set of target values ​​for the image metric. The method then generates an output based on the processed working image and outputs the output.

[0013] In some embodiments, the output may be an output image based on the processed working image. In other embodiments, the output may be a segmentation result or a classification result independent of the output image.

[0014] In some embodiments, the output is an output image, and the method further comprises generating the output image by applying a second filter to the working image after the normalized image processing. The second filter may be an inverse filter of the first filter.

[0015] In some embodiments, the determination of the first value or the first set of values ​​is based on acquisition parameters associated with the input image.Such acquisition parameters may be extracted from a DICOM file associated with the retrieved input image.

[0016] In some embodiments, the determination of the first value or the first set of values ​​is based on visual characteristics of the retrieved input image. In some such embodiments, the determination of the first value or the first set of values ​​is based on an assessment of empty space in the input image. The method may then include retrieving a calibration image generated by an imaging system that generated the input image. Such a calibration image may be an air scan from the imaging system.

[0017] In some embodiments, wherein the determination of the first value or the first set of values ​​is based on visual characteristics of the input image, the determination may be based on a trained neural network independent of the standardized image processing method.

[0018] In some embodiments, wherein the determination of the first value or the first set of values ​​is based on visual characteristics of the input image, the method may proceed to identify at least one homogeneous image region and derive a noise power spectrum associated with the identified homogeneous image region. The first value or the first set of values ​​may then be determined based on the derived noise power spectrum.

[0019] In some embodiments, the image metric is defined by one of: sharpness of the image, a noise power spectrum of the image, or a combination thereof.

[0020] In some embodiments, the standardized image processing method is a trained artificial intelligence-based process and the target value or set of target values ​​for the image metric is an average value or a set of average values ​​of the image metric calculated for training material used to train the standardized image processing method. In some such embodiments, the standardized image processing method is a convolutional neural network.

[0021] In some embodiments, the standardized image processing method is used to denoise, segment, or classify a target image. In some such embodiments, the standardized image processing method is tuned based on a hypothetical target image having the target value or the set of target values ​​of the image metric.

[0022] In some embodiments, the image metric is a modulation transfer function of the image, and the relationship between the first value or the first set of values ​​and the target value or the set of target values ​​is defined by a shape of the modulation transfer function relative to a Nyquist frequency of an image grid corresponding to the image. In some such embodiments, the method includes determining that the first value or the first set of values ​​of the modulation transfer function generates a zero value at a frequency at which the target modulation transfer function generates a non-zero value. In such embodiments, the method includes downsampling the input image prior to applying the first filter.

[0023] In some such embodiments, the output is an output image, and the method includes generating the output image by applying a second filter to the working image after processing. The second filter may then be an inverse filter of the first filter. The method may then further include upsampling the output image after applying the second filter and before outputting the output image so as to restore the size of the input image.

[0024] In some embodiments, the image metric is a noise power spectrum (NPS) of the corresponding image and the standardized image processing method is a denoising process.

[0025] In some embodiments, the image metric is based on a resolution or voxel size of the corresponding image and a signal-to-noise ratio (SNR) of the corresponding image, and the standardized image processing method is a segmentation process.

[0026] In some embodiments, the image metric is based on a resolution or voxel size of the corresponding image, a signal-to-noise ratio (SNR) of the corresponding image, and a field of view (FOV) of the corresponding image, and the standardized image processing method is a classification process.

[0027] In some embodiments, the first value or first set of values ​​is an array or matrix of values ​​defining the image metric. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] Figure 1 is a schematic diagram of a system according to one embodiment of the present disclosure.

[0029] Figure 2 An exemplary imaging device according to one embodiment of the present disclosure is illustrated.

[0030] Figure 3 A method for processing an image according to the present disclosure is illustrated. DETAILED DESCRIPTION

[0031] The description of the illustrative embodiments according to the principles of the present invention is intended to be read in conjunction with the accompanying drawings, which are considered part of the entire written description. In the description of the embodiments of the invention disclosed herein, any reference to direction or orientation is only for the convenience of description and is not intended to limit the scope of the invention in any way. Relevant terms such as "below", "above", "horizontally", "vertically", "above", "below", "above", "below", "top" and "bottom" and their derivatives (e.g., "horizontally", "downward", "upward", etc.) should be interpreted as referring to the orientation described later or shown in the drawings discussed. These relevant terms are only for the convenience of description, and unless explicitly stated, it is not required to construct or operate the device in a specific orientation. Terms such as "attach", "attach", "connect", "couple", "interconnect" refer to the relationship in which structures are fastened or attached to each other directly or indirectly through intermediate structures and the attachment or relationship of removable or rigid, unless otherwise explicitly described. In addition, the features and benefits of the present invention are explained by reference to exemplary embodiments. Therefore, the present invention should not be explicitly limited to exemplary embodiments that illustrate certain possible non-limiting feature combinations, which may exist alone or in the form of other feature combinations; the scope of the present invention is defined by the appended claims.

[0032] The present disclosure describes one or more currently contemplated best modes for practicing the present invention. The description is not intended to be understood in a limiting sense, but rather provides examples of the present invention presented for illustrative purposes only, with reference to the accompanying drawings, to inform those of ordinary skill in the art of the advantages and configurations of the present invention. In the various views of the accompanying drawings, the same reference numerals represent the same or similar parts.

[0033] It is important to note that the disclosed embodiments are merely examples of many advantageous uses of the innovative teachings herein. In general, the statements in the specification of the present application do not necessarily limit any of the various claimed disclosures. In addition, some statements may apply to some inventive features but not to other inventive features. In general, unless otherwise stated, singular elements may be plural, and vice versa, without loss of generality.

[0034] In general, images are processed using various image processing methods designed to visually improve the image or otherwise add information or functionality to the image. This can take the form of reconstruction filters specialized for specific body parts as well as various denoising, segmentation or classification processes.

[0035] In addition, for example, in the context of medical imaging based on computed tomography (CT), different image processors (e.g., machine learning algorithms that can take the form of convolutional neural networks (CNNs)) can be used to process images. These image processors are then trained on various images of the expected form in the case of machine learning algorithms. In particular, the training images typically come from a known source or a known set of sources and have known characteristics (e.g., sharpness and noise power spectrum (NPS)). These characteristics can be used to define image metrics associated with the corresponding images.

[0036] As discussed in more detail below, this image metric can take various forms. In some embodiments, the image metric can be a modulation transfer function (MTF) associated with the imaging system. Because sharpness and NPS can have a unique relationship with the MTF associated with a particular imaging system and images acquired by the imaging system, the MTF can be a convenient way to reference this image metric used in various embodiments. It should be understood that various other image metrics can also be used.

[0037] Note that sharpness can be measured in different ways. As such, sharpness often refers to perceived image sharpness. However, in some embodiments, sharpness may refer to spatial resolution, such that the measure of sharpness may relate to the extent to which small objects can be detected and distinguished from one another in an image.

[0038] Because CNNs are trained on training images with specific values ​​of the image metric they rely on (e.g., MTF), or possibly with values ​​of such image metric within a relatively narrow range, such training is specific to images with visual properties similar to those used during training. Thus, for a CNN to be generally applicable to processing images, it can be independently trained on images from any source that might have been used to acquire the images. However, due to the training requirements, better performance is often achieved by training the neural network on a relatively narrow distribution of input images when defined in terms of the corresponding image metric (e.g., MTF).

[0039] In situations where a user controls the entire imaging pipeline, including processing instruments and image sources (e.g., a CT scanning unit), the values ​​of defined image metrics for all images may be known. However, assumptions cannot often be made about corresponding image metric values ​​for images acquired from third-party vendors or from unknown image sources. Therefore, a transformation may be applied to input images acquired from third-party vendors or unknown sources in order to force the image metrics to better correspond to expected values ​​of the image metrics, thereby allowing for better performance during image processing.

[0040] In medical imaging other than CT (e.g., magnetic resonance imaging (MRI) or positron emission tomography (PET)), different methods may be used to process images, and the resulting images may take different forms. In the present disclosure, embodiments are discussed in terms of CT imaging. However, it should be understood that the methods and systems described herein may also be used in the context of other imaging modalities.

[0041] Figure 1 1 is a schematic diagram of a system 100 according to an embodiment of the present disclosure. As shown in the figure, the system 100 generally includes a processing device 110 and an imaging device 120.

[0042] The processing device 110 may apply a processing routine to an image or measurement data (e.g., projection data) received from the imaging device 120. The processing device 110 may include a memory 113 and a processor circuit 111. The memory 113 may store a plurality of instructions. The processor circuit 111 may be coupled to the memory 113 and may be configured to execute instructions. The instructions stored in the memory 113 may include a processing routine and data associated with the processing routine (e.g., a machine learning algorithm and various filters for processing images). Although all data is described as being stored in the memory 113, it should be understood that in some embodiments, some data may be stored in a database, which itself may be stored in a memory or in a discrete system, or in the cloud.

[0043] The processing device 110 may also include an input unit 115 and an output unit 117. The input unit 115 may receive information (e.g., images or measurement data) from the imaging device 120. The output unit 117 may output an image (e.g., a filtered image) to a user or a user interface device. Alternatively, the output unit 117 may output information about the image (e.g., a segmentation result or a classification result). The output unit 117 may include a monitor or display.

[0044] In some embodiments, processing device 110 may be directly associated with imaging device 120. In alternative embodiments, processing device 110 may be distinct from imaging device 120 such that it receives image or measurement data at input 115 through a network or other interface for processing.

[0045] In some embodiments, the imaging device 120 may include an image data processing device, and a spectral or conventional CT scanning unit for generating CT projection data when scanning a target (eg, a patient).

[0046] Figure 2An exemplary imaging device 200 is illustrated according to one embodiment of the present disclosure. It should be understood that although a CT imaging device is shown and the following discussion is conducted in the context of CT images, similar methods can also be applied in the context of other imaging devices, and the images to which these methods can be applied can be acquired in a variety of ways.

[0047] In an imaging device according to an embodiment of the present disclosure, the CT scanning unit may be adapted to perform multi-axis scanning and / or spiral scanning of a target to generate CT projection data. In an imaging device according to an embodiment of the present disclosure, the CT scanning unit may include an energy-resolving photon counting image detector. The CT scanning unit may include a radiation source that emits radiation for passing through a target when acquiring projection data.

[0048] exist Figure 2 In the illustrated example, a CT scanning unit 200 (e.g., a computed tomography (CT) scanner) may include a fixed gantry 202 and a rotating gantry 204, which may be rotatably supported by the fixed gantry 202. The rotating gantry 204 may rotate about a longitudinal axis around an examination region 206 of a subject while acquiring projection data. The CT scanning unit 200 may include a support 207 to support a patient in the examination region 206, and is configured to move the patient through the examination region during an imaging procedure.

[0049] The CT scanning unit 200 may include a radiation source 208 (e.g., an X-ray tube), which may be supported by the rotating gantry 204 and configured to rotate with the rotating gantry 204. The radiation source 208 may include an anode and a cathode. A source voltage applied between the anode and the cathode may accelerate electrons from the cathode to the anode. The electron flow may provide an electric current from the cathode to the anode to generate radiation that passes through the examination region 206.

[0050] The CT scanning unit 200 may include a detector 210. The detector 210 may be opposite the radiation source 208 across the examination region 206 at an angular arc. The detector 210 may include a one-dimensional or two-dimensional array of pixels (e.g., direct conversion detector pixels). The detector 210 may be adapted to detect radiation passing through the examination region 206 and generate a signal indicative of the energy of the radiation.

[0051] The CT scanning unit 200 may also include generators 211 and 213. The generator 211 may generate tomographic projection data 209 based on the signal from the detector 210. The generator 213 may receive the tomographic projection data 209, and in some embodiments, the generator 213 may generate a raw image 311 of the target based on the tomographic projection data 209. In some embodiments, the tomographic projection data 209 may be provided to the input 115 of the processing device 110, while in other embodiments, the raw image 311 may be provided to the input of the processing device.

[0052] Various physical characteristics of the CT scanning unit 200 and the processing applied by the system 100 to any output of the CT scanning unit may give rise to values ​​of image metrics that characterize the image being processed. For example, the imaging system may generate images having a particular noise power spectrum (NPS) and a known sharpness level, among other characteristics. Such an image metric may be, for example, a modulation transfer function (MTF) that characterizes the output of the imaging system 100.

[0053] Typically, an automated image processing method (e.g., an AI image processing task) is trained using a set of training images. Such training images are acquired from the imaging system 100, and as described above, such training images have defined values ​​for image metrics. Using MTF as an example (where all training images are extracted from a set of images with exactly the same MTF (because all training images are acquired from a single imaging system), or where the MTF values ​​of all training images are within a narrow band (because all training images are acquired from imaging systems from a single known manufacturer)), the resulting AI image processing method will give good results when used to process images with the same or similar MTF characteristics.

[0054] However, trained AI image processing algorithms do not provide good results when applied to images retrieved from imaging systems provided by different manufacturers.

[0055] Figure 3 A method for processing an image according to the present disclosure is shown. As discussed above, a method is provided for applying a standardized image processing method to an input image while remaining agnostic to certain characteristics of the input image. For example, the standardized image processing method can be AI-based (e.g., applying a trained neural network), and the method can be based on an image acquired by a well-known imaging system. The method can then process the input image using the standardized image processing method while remaining agnostic to the source of the input image. The method is able to process even if the standardized image processing method cannot directly process or cannot provide acceptable results with respect to the acquired image.

[0056] As shown, the method includes first retrieving (320) an input image to be processed. The method then determines (at 330) a first value or a first set of values ​​of an image metric associated with the input image. As described above, the image metric can be an MTF associated with an imaging system from which the image was retrieved. Alternatively, the image metric can be a metric defined by or based on various characteristics of the image. In any case, the first value or first set of values ​​determined for the input image can define one or more visual characteristics of the input image, enabling the input image to be compared with other images.

[0057] In some embodiments, the determination of this value or this set of values ​​is based on acquisition parameters associated with the input image. In such embodiments, the acquisition parameters can be extracted from a DICOM file associated with the input image. In such embodiments, as discussed above, this value or this set of values ​​can define the MTF.

[0058] In some embodiments, the determination of this value or set of values ​​may be based on visual characteristics of the retrieved input image. For example, such a determination may be based on an assessment of empty space in the input image. In such embodiments, the method may also include (at 335) retrieving a calibration image generated by the imaging system 100 that generated the input image. For example, the calibration image may be an air scan from the same imaging system 100.

[0059] The input image, or simply the empty space contained therein, may then be compared (at 340) to the air scan retrieved from the imaging system 100 to determine a first value or set of values.

[0060] In some embodiments, the first value or first set of values ​​may be determined by applying a trained neural network or other AI-based algorithm to the input image. Such a trained neural network may be independent of the standardized image processing method applied by the method.

[0061] In some embodiments, the first value or the first set of values ​​may be determined by identifying at least one homogeneous image region and deriving a noise power spectrum (NPS) associated with the identified homogeneous image region.Then, the first value or the first set of values ​​may be determined based on the derived NPS.

[0062] Typically, image metrics can be defined using multiple visual characteristics of the input image. For example, an image metric can be defined by one of the following: the sharpness of the image, the noise power spectrum (NPS) of the image, or a combination of these characteristics. In some embodiments, the image metric can be based on the resolution or voxel size of the input image and the signal-to-noise ratio (SNR) of the corresponding image. Other characteristics (including field of view (FOV)) can also be considered.

[0063] The value or set of values ​​determined (at 330) may take the form of a single variable defining an image metric, or may take the form of a set of values ​​in an array or matrix defining an image metric. As such, when a value or set of values ​​is discussed herein, it refers to any set of values ​​used to characterize an image in the context of such an image metric.

[0064] Once a first value or set of values ​​of the image metric is determined (at 330), the method proceeds to generate a first filter based on a relationship between the first value or set of values ​​of the image metric and a target value or set of target values ​​of the image metric (350). Such a first filter can then be applied (360) to the input image to transform the input image into a working image (370) having a second value or set of values ​​of the image metric, the second value or set of values ​​being different from the first value or set of values.

[0065] The first filter is generated as a custom filter designed for the following function: transforming the first image so that one or more values ​​of the metric correspond to or are similar to a target value or a set of target values. This can be achieved by enhancing or attenuating various frequency components in the image to match corresponding values ​​of the target image. Therefore, after the transformation to the working image is completed (at 370), the second value or second set of values ​​of the image metric is substantially similar to the target value or a set of target values ​​of the image metric. Typically, the target value or a set of target values ​​of the image metric corresponds to a value associated with a known imaging system for which a standardized image processing method is designed. In the case of an AI-based image processing method (e.g., a neural network), the target value or a set of target values ​​corresponds to the value of the image metric of the image on which the neural network was trained.

[0066] For example, when an AI-based image processing algorithm or a neural network such as a convolutional neural network (CNN) is trained on a set of images acquired from a known imaging system to produce a known MTF, and when input images are acquired from imaging systems from different manufacturers to produce different MTFs, the first filter can transform the input image into a working image having an MTF similar to or the same as the MTF of the image from the known imaging system. This transformation allows the target image to be effectively processed using the AI-based image processing algorithm, even if the input image cannot be so processed.

[0067] In some cases, not all training images share a value or set of values ​​for the image metric. This may occur when the image metric is MTF, where the training images were acquired from different imaging systems. Alternatively, this may occur where the image metric is based on the images themselves rather than on the source system. In such embodiments, the target value or set of target values ​​for the image metric may be an average value or set of average values ​​of the image metric calculated for the training material used to train the standardized image processing method.

[0068] In some embodiments, the standardized image processing method may be tuned based on a hypothetical target image having a target value or set of target values ​​for an image metric. This may occur where the training images do not directly correspond to an expected set of images, or where the standardized image processing method may be used in a different scenario than that for which it was initially trained.

[0069] When the second value or the second set of values ​​is described as being substantially similar to a target value or a set of target values, it should be understood that this similarity is sufficient to allow an AI-based image processing algorithm trained on an image with a target value or a set of target values ​​for image metrics to process a working image. In some embodiments, a first filter can be generated (at 350) so that the second value or the second set of values ​​corresponds precisely to the target value or a set of target values. In other embodiments, such a second value or the second set of values ​​can be within the range of the target value or a set of target values. The limits of this basic similarity can vary according to the specific standardized image processing method applied in a specific embodiment.

[0070] Thus, after the input image is transformed (at 370) into a working image, the working image may be processed (380) using a standardized image processing method based on a target value or set of target values ​​for an image metric.

[0071] Once the working image has been processed, the method may generate an output (410) in the form of information or an output image based on the processed working image, and output (420) the resulting information or output image. In some embodiments where the output is an output image, the method may optionally include generating a second filter (390) to restore the second value or set of values ​​of the image metric to the first value. This may be achieved by inverting the first filter (generated at 350). The second filter may then be applied (400) to the working image after processing (at 380) the working image to generate the output image (at 410).

[0072] Where the output is an output image, then such an output image may be output (generated at 410) having image metrics of a known imaging system for which the image processing method is designed, or such an output image may be transformed (at 400) by applying a second filter and thus may be output having image metrics associated with the imaging system from which the image was originally retrieved. As such, when the imaging system from which the image was originally retrieved had a unique aesthetic associated with its output, the methods described herein may be used to apply a standardized image processing method while still presenting an output image (at 420) that is familiar to users of the source system.

[0073] The methods described herein can be used to apply a variety of standardized image processing methods. For example, in some embodiments, the standard image processing method can be a denoising process, a segmentation process, or a classification process applied to the image content, and the image metric of the retrieved value is defined (at 330) based on the processing method to be applied.

[0074] For example, in some embodiments, as discussed above, the image metric is the NPS of the corresponding image and the standardized image processing method is a denoising process.

[0075] Alternatively, in some embodiments, the image metric is based on a resolution or voxel size of the corresponding image and a SNR of the corresponding image, and the standardized image processing method is a segmentation process.

[0076] Alternatively, in some embodiments, the image metric is based on the resolution or voxel size of the corresponding image, the SNR and FOV of the corresponding image, and the standardized image processing method is a classification process.

[0077] Additionally, in some embodiments, the method may be used to apply various standardized image processing methods. Thus, prior to determining (at 330) the value or set of values ​​for the image metric, a user may select a standardized image processing method to be applied. The method may then use the user selection to define the image metric and then determine (at 330) the value or set of values ​​only before proceeding (at 350) with generating the first filter.

[0078] As described above, a standardized image processing method may include multiple filters applied in succession. This approach can be used to create a more generalized image processing method that is capable of processing, for example, different body parts or tissue types. In this embodiment for AI denoising, a first standard high-resolution filter can be used to generate an input image for a neural network. Only after the first standard filter is applied can the desired filter characteristics be applied. For example, where different reconstruction filters can typically be used to denoise and process images of bones and soft tissues, a first generalized filter can be used to denoise the two images, and only after such a generalized filter is applied is a second filter used to emphasize the desired characteristics.

[0079] This type of standardized image processing method is only feasible if the entire reconstruction chain is under the control of the user implementing the system. Therefore, the method described herein can be used to initially modify an input image acquired from a third-party imaging system so that the input image corresponds to the expected image parameters of this type of standardized image processing method.

[0080] In some embodiments, as discussed above, the image metric is the modulation transfer function (MTF) of the image. The first filter is then generated using the relationship between the MTF of the input image and the target MTF. This relationship can be defined by the shape of the MTF relative to the Nyquist frequency of the image grid corresponding to the image. For example, a 512 mm filter with a 250 mm field of view (FOV) and a Gaussian shaped MTF of 50% @ 8 line pairs per centimeter (lp / cm) may be used to generate the first filter. 2 The image appears to the neural network as a 512mm image with a 500mm FOV and a Gaussian-shaped MTF of 50% @ 4lp / cm. 2 image.

[0081] Initially, it is noted that a neural network implementing a standardized image processing method will view an image differently due to different MTFs and different FOVs. Therefore, the first filter may include a cropping operation so that the working image has the FOV at which the standardized image processing method was trained.

[0082] In some embodiments, the relationship between the MTF of the input image and the target MTF is evaluated to determine that the MTF of the input image generates zero values ​​at frequencies at which the target MTF generates non-zero values. In such embodiments, the input image may be downsampled (at 345) prior to applying the first filter (at 360). Such downsampling may be before or after generating the first filter (at 350).

[0083] This downsampling takes advantage of the fact that if the MTF drops to zero below the Nyquist frequency defined by the image grid, the image can be downsampled without loss of information. Due to scaling invariance, the downsampled image can then be preprocessed using the first filter (at 360) as discussed above to match the desired frequency response.

[0084] In such an embodiment, after applying the normalized image processing method (at 380) and after applying the second filter (at 400), the working image may be upsampled (at 405) to restore the original image size. The output image may then be output (at 420) at its original size.

[0085] In some embodiments where the shape of the MTF is evaluated with respect to the Nyquist frequency, in addition to or instead of triggering a downsampling operation (at 345), the method may also implement a deconvolution or image sharpening or deblurring process (as part of the first filter), which may aid segmentation or classification. This may be in addition to cropping of the image to ensure that the working image has the FOV when training the standardized image processing method.

[0086] The method according to the present disclosure can be implemented on a computer as a computer-implemented method, or implemented in dedicated hardware, or implemented in a combination of the two. The executable code of the method according to the present disclosure can be stored on a computer program product. Examples of computer program products include storage devices, optical storage devices, integrated circuits, servers, online software, etc. Preferably, the computer program product may include a non-transient program code stored on a computer-readable medium for executing the method according to the present disclosure when the program product is executed on a computer. In one embodiment, the computer program may include a computer program code, which is suitable for executing all steps of the method according to the present disclosure when the computer program is running on a computer. The computer program may be contained on a computer-readable medium.

[0087] While the present disclosure has been described at some length and with a certain particularity with respect to several described embodiments, it is not intended that the present disclosure should be limited to any such particular content or embodiment or any particular embodiment, but rather the present disclosure should be interpreted with reference to the appended claims so as to provide the broadest possible interpretation of such claims in light of the prior art and thereby effectively cover the intended scope of the present disclosure.

[0088] All examples and conditional language recorded herein are intended to be used for teaching purposes, to help readers understand the principles of the present disclosure and the conception contributed by the inventor to promote the development of this area, and should be interpreted as not being limited to the examples and conditions of these specific records. In addition, all statements and specific examples thereof of the principle, various aspects and embodiments of the present disclosure are recorded herein and are intended to cover its structure and function equivalents. In addition, this article is intended to make such equivalents include currently known equivalents and equivalents developed in the future (that is, the elements of the same function of any development, regardless of the structure).

Claims

1. A computer-implemented method for processing an image, comprising: Retrieve the input image to be processed; determining a first value or a first set of values ​​of an image metric associated with the input image; generating a first filter based on a relationship between the first value or the first set of values ​​of the image metric and a target value or a set of target values ​​of the image metric; applying the first filter to the input image to generate a working image having a second value or a second set of values ​​for the image metric that is substantially similar to the target value or the set of target values ​​for the image metric; processing the working image using a standardized image processing method based on the target value or set of target values ​​for the image metric; as well as An output is generated based on the processed working image.

2. The method according to claim 1, wherein: The output is an output image, and the method further comprises generating the output image by applying a second filter to the working image after processing, the second filter being an inverse filter of the first filter.

3. The method according to claim 1, wherein: The determination of the first value or the first set of values ​​is based on acquisition parameters associated with the input image.

4. The method according to claim 3, wherein: The acquisition parameters are extracted from the DICOM file associated with the retrieved input image.

5. The method according to claim 1, wherein: The determination of the first value or the first set of values ​​is based on visual characteristics of the retrieved input image.

6. The method according to claim 5, wherein: Determination of the first value or the first set of values ​​is based on an assessment of empty space in the input image, and wherein the method further comprises retrieving a calibration image generated by an imaging system that generated the input image, the calibration image being an air scan from the imaging system.

7. The method according to claim 5, wherein: The determination of the first value or the first set of values ​​is based on a trained neural network independent of the standardized image processing method.

8. The method according to claim 5, further comprising: At least one homogeneous image region is identified, and a noise power spectrum associated with the identified at least one homogeneous image region is derived, and wherein the derived noise power spectrum is used to determine the first value or at least one of the first set of values.

9. The method according to claim 1, wherein: The image metric is defined by one of: sharpness of the image, a noise power spectrum of the image, or a combination thereof.

10. The method according to claim 1, wherein: The standardized image processing method is a trained artificial intelligence based process, and wherein the target value or the set of target values ​​of the image metric is an average value or a set of average values ​​of the image metric calculated for training material used to train the standardized image processing method.

11. The method according to claim 10, wherein: The standardized image processing method is a convolutional neural network, and wherein the standardized image processing method is used to denoise, segment or classify a target image.

12. The method according to claim 11, wherein: The standardized image processing method is tuned based on a hypothetical target image having the target value or the set of target values ​​for the image metric.

13. The method according to claim 1, wherein: The image metric is a modulation transfer function of the image, and the relationship between the first value or the first set of values ​​and the target value or the set of target values ​​is defined by a shape of the modulation transfer function relative to a Nyquist frequency of an image grid corresponding to the image.

14. The method according to claim 13, further comprising: The first value or the first set of values ​​of the modulation transfer function is determined to generate zero values ​​at frequencies at which the target modulation transfer function generates non-zero values, and the input image is downsampled prior to applying the first filter.

15. The method according to claim 14, wherein: The output is an output image, and the method also includes: generating the output image by applying a second filter to the working image after processing, wherein the second filter is an inverse filter of the first filter; and upsampling the output image after applying the second filter and before outputting the output image so as to restore the size of the input image.

16. The method according to claim 1, wherein: The image metric is a noise power spectrum (NPS) of the corresponding image, and wherein the standardized image processing method is a denoising process.

17. The method according to claim 1, wherein: The image metric is based on a resolution or voxel size of the corresponding image and a signal-to-noise ratio (SNR) of the corresponding image, and wherein the standardized image processing method is a segmentation process.

18. The method according to claim 1, wherein: The image metrics are based on a resolution or voxel size of the corresponding image, a signal-to-noise ratio (SNR) of the corresponding image, and a field of view (FOV) of the corresponding image, and wherein the standardized image processing method is a classification process.

19. A computer-implemented system for processing an image, comprising: a memory storing a plurality of instructions; as well as a processor circuit coupled to the memory and configured to execute the plurality of instructions to: Retrieve the input image to be processed; determining a first value or a first set of values ​​of an image metric associated with the input image; generating a first filter based on a relationship between the first value or the first set of values ​​of the image metric and a target value or a set of target values ​​of the image metric; applying the first filter to the input image to generate a working image having a second value or a second set of values ​​for the image metric that is substantially similar to the target value or the set of target values ​​for the image metric; processing the working image using a standardized image processing method based on the target value or set of target values ​​for the image metric; as well as An output is generated based on the processed working image.

Images