Method and system for providing a locally-consistent enhancement of a low-quality image

Abstract

A method of providing a medical image of a ROI of a patient, the method comprising: acquiring a first medical image of a region of interest (ROI) of a patient, the medical image characterized by a first signal to noise ratio (SNR); determining for a given pixel in the first image a plurality of different first image patches in the first image, each having a pixel that is coincident with the given pixel; determining for each first image patch a similar second image patch having a second SNR greater than the first SNR; determining an enhanced pixel value for the given pixel having an enhanced SNR greater than the first SNR responsive to pixel values of pixels in the determined second image patches; and using the determined pixel value to generate a second medical image of the ROI having an enhanced SNR greater than the first SNR.

Claims

1. A method for providing a Locally-Consistent enhancement of a Low-Quality image (L-image) of a ROI to provide an Enhanced Image (E-image) of the ROI, wherein F() returns an enhanced pixel value EPV={circumflex over (p)}.sub.j for the given pixel p.sub.j that is computed as a function of H-patches belonging to a set {NNp.sub.j} of nearest neighbor H-patches, {circumflex over (P)}.sub.i, 1iP.sub.N, where P.sub.N is a number of pixels in the L-patch, the method further comprising: selecting a pixel p.sub.j of the L-image to be enhanced; (ii) defining a patch (L-patch) in the L-image, having P.sub.N pixels and comprising pixel p.sub.j; (iii) defining P.sub.N O-patches that have the same size as the L-patch, contain pixel p.sub.j and overlap at least a portion of the L-patch; (iv) for each O-patch, returning from a relevant high-quality source (H-source), a high quality patch (H-patch) that satisfies a desired similarity criterion wherein the H-patches and L-images are acquired from different imaging systems; (v) computing an enhanced pixel value (EPV) for p.sub.j, based on F(); and (vi) replacing the pixel value of p.sub.j with the computed EPV.

2. A method for providing a Locally-Consistent enhancement of a Low-Quality image (L-image) of a ROI to provide an Enhanced Image (E-image) of the ROI, wherein function () returns an average value of pixels from the H-patches that are homologous with a pixel in an O-patch coincident with the given pixel, in symbols ${\hat{p}}_j=F(.Math.)=\frac{1}{P_S^2}\underset{i=1}{\stackrel{P_N}{.Math.}}{\hat{P}}_{i_j}\mathrm{for}{\hat{P}}_i\left\{{\mathrm{NNp}}_j\right\}.$ the method further comprising: (i) selecting a pixel pj of the L-image to be enhanced; (ii) defining a patch (L-patch) in the L-image, having PN pixels and comprising pixel p.sub.j; (iii) defining P.sub.N O-patches that have the same size as the L-patch, contain pixel p.sub.j and overlap at least a portion of the L-patch; (iv) for each O-patch, returning from a relevant high-quality source (H-source), a high quality patch (H-patch), that satisfies a desired similarity criterion wherein the H-patches and L-images are acquired from different imaging systems; (v) computing an enhanced pixel value (EPV) for p.sub.j, based on function (.Math.); and (vi) replacing the pixel value of p.sub.j with the computed EPV.

3. A method for providing a Locally-Consistent enhancement of a Low-Quality image (L-image) of a ROI to provide an Enhanced Image (E-image) of the ROI, wherein function F(.Math.) returns a sum of {circumflex over (P)}.sub.I.sub.J, i=1 . . . P.sub.N weighted by a distance measure between the overlapping pixels of {circumflex over (P)}.sub.i and the original L-patch defined for the given pixel p.sub.j, formalized as follows: ${\hat{p}}_j=\frac{\underset{i=1}{\stackrel{P_N}{.Math.}}\mathrm{EXP}\left(-\frac{D\left(P_j,{\hat{P}}_i\right)}{h^2}\right){\hat{P}}_{i_j}}{\underset{i=1}{\stackrel{P_S^2}{.Math.}}\mathrm{EXP}\left(-\frac{D\left(P_j,{\hat{P}}_i\right)}{h^2}\right)},{\hat{P}}_i\left\{{\mathrm{NNp}}_j\right\}.$ the method further comprising: (i) selecting a pixel p.sub.j of the L-image to be enhanced; (ii) defining a patch (L-patch) in the L-image, having P.sub.N pixels and comprising pixel P.sub.j; (iii) defining P.sub.N O-patches that have the same size as the L-patch, contain pixel p.sub.j and overlap at least a portion of the L-patch; (iv) for each O-patch, returning from a relevant high-quality source (H-source), a high quality patch (H-patch), that satisfies a desired similarity criterion wherein the H-patches and L-images are acquired from different imaging systems; (v) computing an enhanced pixel value (EPV) for p.sub.j, based on function F(); and (vi) replacing the pixel value of p.sub.j with the computed EPV.

4. A system for providing an Enhanced Image (E-image) of a ROI based on a Locally-Consistent enhancement of a Low-Quality image (L-image) of the ROI, the system comprising: (i) an E-processor having enhancement operation elements, that include set of instructions executable to implement a trained neural network said enhancement comprising: a) selecting a pixel p.sub.j of the L-image to be enhanced; b) defining a patch (L-patch) in the L-image, having P.sub.N pixels and comprising pixel p.sub.j; c) defining P.sub.N O-patches that have the same size as the L-patch, contain pixel p.sub.j, and overlap at least a portion of the L-patch; d) for each O-patch, returning from a relevant high-quality source (H-source), a high quality patch (H-patch) that satisfies a desired similarity criterion; e) computing an enhanced pixel value (EPV) for p.sub.j based on a function F() of the H-patches; and f) replacing the pixel value of p.sub.j with the computed EPV; (ii) a H-database and an L-database; (iii) a PACS (Picture Archiving and Communication System) connected through DICOM protocol to E-processor, H-source and an L-database; (iv) a medical imaging modality connected to the E-processor; (v) a radiologic workstation communicating with the PACS system, for the retrieval of the E-images and their visualization; (vi) a display; and (vii) a human-machine interface (HMI).

5. Claim according to claim 1 and comprising repeating steps (i) to (vi).

6. Claim according to claim 2 and comprising repeating steps (i) to (vi).

7. Claim according to claim 3 and comprising repeating steps (i) to (vi).

8. Claim according to claim 4 and comprising repeating steps (a) to (f).

Description

BRIEF DESCRIPTION OF FIGURES

(1) Non-limiting examples of embodiments of the disclosure are described below with reference to figures attached hereto that are listed following this paragraph. Identical features that appear in more than one figure are generally labeled with a same label in all the figures in which they appear. A label labeling an icon representing a given feature of an embodiment of the disclosure in a figure may be used to reference the given feature. Dimensions of features shown in the figures are chosen for convenience and clarity of presentation and are not necessarily shown to scale.

(2) FIG. 1A schematically shows a CT scanner configured to acquire L-SNR CT-images of a patient and process the images to provide enhanced, E-SNR, CT-images of the patient, in accordance with an embodiment of the disclosure;

(3) FIG. 1B shows schematic distribution of signal variances of H-Patches optionally comprised in a H-SNR patch database, in accordance with an embodiment of the disclosure;

(4) FIG. 2 shows a flow diagram of a procedure that the CT scanner shown in FIG. 1 may use to provide E-SNR images based on L-SNR images, in accordance with an embodiment of the disclosure;

(5) FIG. 3A schematically shows an L-Patch determined for a given pixel in an image of a CT-slice of a patient, in accordance with an embodiment of the disclosure;

(6) FIGS. 3B-3J schematically show overlapping O-Patches determined for the given pixel and L-Patch shown in FIG. 3A, in accordance with an embodiment of the disclosure; and

(7) FIG. 4 schematically shows a system operable to provide enhanced quality images, E-images, based on enhancing low quality, L-images, in accordance with an embodiment of the disclosure.

DETAILED DESCRIPTION

(8) In the detailed discussion below features and operation of a medical imaging system configured in accordance with an embodiment of the disclosure to enhance a L-image to obtain a E-image of a patient are discussed with reference to FIG. 1. Details of a procedure by which the imaging system shown in FIG. 1 may process an L-image of a patient that the imaging system acquires to provide an E-image based on the acquired image, are discussed with reference to FIG. 2. By way of example, the medical imaging system shown in FIG. 1 comprises a CT-scanner. It is noted that in addition to certain imager types mentioned in the specific exemplary embodiments, the imager according to the disclosure may comprise at least one or any combination of more than one of a device suitable for: x-ray radiography, mammography, fluoroscopy, angiography, ultra-sound imaging, positron emission tomography (PET), computed tomography (CT), magnetic resonance imaging (MRI), and any combination thereof.

(9) In the discussion, unless otherwise stated, adjectives such as substantially and about modifying a condition or relationship characteristic of a feature or features of an embodiment of the disclosure, are understood to mean that the condition or characteristic is defined to within tolerances that are acceptable for operation of the embodiment for an application for which it is intended. Unless otherwise indicated explicitly or implicitly, the word or in the description and claims is considered to be the inclusive or rather than the exclusive or, and indicates at least one of, or any combination of items it conjoins.

(10) FIG. 1 schematically shows, in accordance with an embodiment of the disclosure, a multislice CT-scanner 20 configured to acquire low-dose CT-images characterized by a relatively low SNR of patients and process the images to provide enhanced images (E-images), which are characterized by enhanced SNR as well, possibly, by other quality enhancements. In FIG. 1, by way of example, CT-scanner 20 is shown imaging a ROI of the chest of a patient 22.

(11) CT-scanner 20 comprises a gantry 34 having a stator 35 to which a rotor 36 is mounted so that the rotor can be controlled to rotate about an axis 37. An X-ray source 24 controllable to provide an X-ray beam schematically indicated by dashed lines 26 and an array 30 of X-ray detectors 32 opposite the X-ray source are mounted to the rotor. Detectors 32 generate signals that provide measures of intensity of X-rays from X-ray source 24 that pass through a patient imaged by CT-scanner 20. CT-scanner 20 is assumed, by way of example to be a multislice CT-scanner that simultaneously images a plurality of CT-slices of the body of a patient and detectors 32 in array 30 are configured in a plurality of contiguous curved rows 33 of the detectors. By way of example, in FIG. 1, for convenience of presentation CT-scanner 20 is shown as controllable to simultaneously image up to four CT-slices of a patient. The patient, such as patient 22, is supported on a couch 38 mounted to a suitable pedestal (not shown) so that the couch is controllable to be translated axially along axis 37 selectively in a direction indicated by a block arrow 39 or in a direction opposite the indicated direction. Detectors 32 in detector array 30 and features of CT-scanner that are shadowed by patient 22 and couch 38, which would not normally be seen in the perspective of FIG. 1, are shown for clarity of presentation by ghost lines.

(12) A controller 50 controls operation of CT-scanner 20 and processing of signals generated by detectors 32 to provide a CT-image of a patient. Controller 50 optionally comprises a scanner controller 51 that controls operation of components of CT-scanner 20, data acquisition circuitry 52, a memory 53, and a processor 55.

(13) In an embodiment of the disclosure controller 50 comprises a H-SNR database 54 comprising a plurality of H-SNR images of different regions of a human body for which CT-scanner 20 may be used to acquire images, and/or H-SNR image patches, H-Patches, extracted from H-SNR images for each of different regions of the body for which the CT-scanner may be used to acquire images. In an embodiment H-SNR database 54 comprises H-Patches that have been extracted from H-SNR images and been selected so that a population of H-Patches comprised in the database exhibits a relatively large number of H-Patches having significant information content. Signal variance of an H-patch may be used as a measure of information content of an H-Patch, with relatively large signal variance indicating relatively large information content. Signal variance of an H-Patch refers to a standard deviation of X-ray intensities represented by values, for example gray level values, of the pixels in the H-Patch. In an embodiment, extracted H-Patches are filtered using a roulette wheel selection filter to provide a population of H-Patches exhibiting greater signal variance than a population of extracted H-Patches filtered randomly. By way of example FIG. 1B shows a distribution 57 of a population of H-Patches filtered randomly and a distribution 58 of a population of H-Patches selected using a roulette wheel filter to increase presence of H-Patches having relatively large signal variance in accordance with an embodiment of the disclosure.

(14) Controller 50 may comprise any electronic and/or optical processing and/or control circuitry, to provide and enable functionalities that the controller may require to support its operation in embodiments that are described below and embodiments similar to the described embodiments. By way of example, processor 55 may comprise any one, or any combination of more than one of, a microprocessor, an application specific circuit (ASIC), field programmable array (FPGA) and/or system on a chip (SOC). Memory 53 may comprise any electronic and/or optical circuitry suitable for storing data and/or computer executable instructions and may, by way of example, comprise any one or any combination of more than one of a flash memory, random access memory (RAM), read only memory (ROM), and/or erasable programmable read-only memory (EPROM).

(15) Whereas memory 53, H-SNR database 54, and/or processor 55, are indicated in FIG. 1A as being centralized local modules comprised in CT-scanner 20, at least one of the modules may be a virtual module and/or may be configured as a distributed module comprising hardware and/or software components at different locations outside of CT-scanner 20, and may be cloud based. For example H-SNR database 54 and/or processor 55 may be a cloud based database and controller 50 may comprise and/or have access to any suitable wire and/or a wireless communication interface (not shown) for communicating with the H-SNR database and/or processor.

(16) To image a ROI of a patient's body, such as the chest region of patient 22, system controller 50 controls couch 38 to translate the patient through gantry 34 so that the ROI passes through the gantry. While translating the patient through the gantry, system controller 50 controls X-ray source 24 to generate fan beam 26 and rotor 36 to rotate about axis 37 so that the fan beam exposes the patent to X-rays from a plurality of different view angles in a range of about 360 about axis 37. For each of the view angles, detectors 32 provide signals that measure intensities of X-rays in fan beam 26 that pass through the ROI of the patient located between detectors 32 and X-ray source 24. For each view angle rows 33 of detectors 32 provide measurements of X-ray intensity useable to generate images of optionally four different CT-slices of the patient. FIG. 1 schematically shows four exemplary CT-slices 40 of patient 22 for which CT-scanner 20 acquires intensity measurements of X-rays that have passed through patient 22 for the position of patient 22 in gantry 34 and a view angle of about 90 at which X-ray source 24 and detector array 30 are shown in the figure. The intensity signals generated by detectors 32 may be received by data acquisition circuitry 52 which may preprocess and store the preprocessed signals in memory 53. Processor 55 may process the stored signals to generate an image of each slice 40 of patient 22 and combine the images of the CT-slices to provide a 3D volume image of the patient's ROI.

(17) In a variation of the embodiment, processor 55 may first enhance the preprocessed signals stored in memory 53, which signals represent a sinogram of the scan, and then, in order to further increase the quality, the processor may enhance again the image reconstructed from the enhanced sinogram. Either or both of the described enhancements procedures may be carried-out by processor 55 according the flow diagram in FIG. 2 discussed below.

(18) In an embodiment, to moderate risks to the health of patient 22 resulting from exposure of the patient to X-ray radiation from X-ray source 24, controller 50 controls CT-scanner 20 to acquire intensity data for a CT-image of the patient with a relatively low dose of radiation. The CT-image acquired using the low radiation dose may be a L-SNR image characterized by relatively low SNR. Processor 55 may operate to improve SNR of the L-SNR image to generate an image, also referred to as an E-Image, of patient 23 exhibiting enhanced SNR by processing images of CT-slices, such as exemplary CT-slices 40, of the patient using H-Patches stored in database 54.

(19) In an embodiment, processor 55 may operate to provide an SNR enhanced CT-image of a ROI of patient 22 in accordance with a procedure illustrated in a flow diagram 100 shown in FIG. 2.

(20) In a block 101 of the procedure, controller 50 controls CT-scanner 20 to acquire a low X-ray dose, L-SNR CT-image of patient 22. The image is assumed to be formed from a number S, of low SNR images, L-PIM(s), 1sS, of CT-slices of the patient, each slice comprising pixels LP(s).sub.p 1pP(s). In a block 103 processor 55 (FIG. 1A) optionally determines for each pixel LP(s).sub.p in an image L-PIM(s) of an s-th CT-slice, an L-Patch(s,p) comprising and optionally centered on pixel LP(s).sub.p, and comprising J pixels in image L-PIM(s). Optionally, L-Patch(s,p) is square. FIG. 3A schematically shows an example square L-Patch(s,p) 60 outlined in a white border and comprising J=9 pixels defined for an LP(s).sub.p pixel 61 marked by an asterisk in an image L-PIM(s) 62 of an s-th CT-slice of patient 22. Optionally, in a block 105 the processor determines J different patches O-Patch(s,p,j), 1jJ, in image L-PIM(s) 62 that overlap L-Patch(s,p) 60, each O-Patch(s,p,j) comprising a different set of J pixels from image L-PIM(s). The pixels in a given overlapping O-Patch(s,p,j) may be referred to as pixels LOP(s,p,j).sub.k, 1kJ. FIGS. 3B-3J schematically show the J=9 overlapping patches O-Patch(s,p,j), in L-PIM(s) 62 defined for L-Patch(s,p) 60 shown in FIG. 3A. Each overlapping patch O-Patch(s,p,j) is bordered by a dashed white border 63, and comprises a different set of J=9 pixels from CT-slice image L-PIM(s) 62, one of which pixels is coincident with pixel 61 indicated by the asterisk.

(21) Optionally, in a block 107, processor 50 identifies for each O-Patch(s,p,j) 63 a high SNR patch H-Patch(s,p,j) in H-SNR database 54 (FIG. 1A) having a same number of pixels and same shape as O-Patch(s,p,j) that exhibits a relatively high degree of similarity with the O-Patch(s,p,j). In an embodiment the identified H-patch(s,p,j) may be an approximate nearest neighbor (ANN) to the O-Patch(s,p,j). An approximate nearest neighbor may be determined using any of various nearest neighbor search algorithms and may for example be determined using a randomized kd-trees algorithm such as implemented in the Fast Library for ANN (FLANN) for a Euclidean norm. Alternatively, a high SNR patch for a given O-Patch(s,p,j) may be generated by a regression neural network trained to generate a high SNR patch when fed at its input with a low SNR patch. For this purpose, it is necessary to have a training set consisting of corresponding low and high SNR patch pairs. Said pairs of patches can be obtained using an ANN algorithm to associate between patches from a low SNR database to patches of a high SNR database. Alternatively, the pairs can be created by adding synthetic noise corresponding to a given probability distribution function depending on the considered imaging modality (for example, Rician noise for MRI, Gaussian and Poisson noise for CT). In a block 109, processor 55 determines an enhanced value ELP(s).sub.p for each pixel LP(s).sub.p as a function of pixel values of pixels in the approximate nearest neighbor high SNR H-Patches(s,p,j), identified for the overlapping patches, O-Patches(s,p,j), 1jJ, associated with LP(s).sub.p. In symbols ELP(s).sub.p=F(H-Patch(s,p,j) j:1jJ)).

(22) Let the pixels in a j-th approximate nearest neighbor H-Patch(s,p,j) identified for an O-Patch(s,p,j) be represented by HNN(s,p,j).sub.m, 1mJ and its value by Val(HNN(s,p,j).sub.m). Let a pixel HNN(s,p,j).sub.m for which m=m*(j) be a pixel, hereinafter also referred to as a H-SNR coincident pixel, in H-Patch(s,p,j) that is homologous with a pixel LOP(s,p,j).sub.k in overlapping patch O-PATCH(s,p,j) that is coincident with pixel LP(s).sub.p in L-PIM(s). ELP(s).sub.p may optionally be determined in accordance with an expression ELP(s).sub.p=F(H-Patch(s,p,j))=(1/J).sub.1.sup.J Val(HNN(s, p, j).sub.m*(j)). A similar way to express the enhanced value for the pixel is

(23) ${\mathrm{ELP}\left(s\right)}_p={\hat{p}}_j=\frac{1}{P_S^2}\underset{i=1}{\stackrel{P_S^2}{.Math.}}{\hat{P}}_{i_j}\mathrm{as}\mathrm{for}{\hat{P}}_i\left({\mathrm{NN}}_{p_j}\right)$
already given in the summary.

(24) In an embodiment, the enhanced pixel ELP(s).sub.p is determined as a function of the HNN(j).sub.m*(j) weighted by a measure of similarity between the L-PATCH(s,p) determined for LP(s)p and H-PATCH(j) to which pixel HNN(j).sub.m*(j) belongs. If D(s,p,j) represents a measure of similarity between L-PATCH(s,p) and H-PATCH(j) then ELP(s).sub.p may be determined in accordance with an expression,
ELP(s).sub.p=F(Val(HNN(s,p,j).sub.m*(j)))=[.sub.1.sup.JVal(HNN(j).sub.m*(j))exp(D(s,p,j)/h.sup.2))]/.sub.1.sup.J exp(D(s,p,j)/h.sup.2,
where h=J and is a constant. The Similarity function D can alternatively be replaced by a Distance function D for which D=QD, where Q is a positive constant number. In that case, a similar formulation for ELP(s).sub.p can be given by

(25) ${\mathrm{ELP}\left(s\right)}_p={\hat{p}}_j=\frac{\underset{i=1}{\stackrel{P_S^2}{.Math.}}\exp \left(-\frac{D\left(P_j,{\hat{P}}_i\right)}{h^2}\right){\hat{P}}_{i_j}}{\underset{i=1}{\stackrel{P_S^2}{.Math.}}\exp \left(-\frac{D\left(P_j,{\hat{P}}_i\right)}{h^2}\right)},{\hat{P}}_i\left({\mathrm{NN}}_{p_j}\right),$
as already given in the summary.

(26) In a block 111 processor 55 uses the enhanced pixel values ELP(s).sub.p determined for each CT-slice image L-PIM(s) to provide a CT-slice image EL-PIM(s) having enhanced SNR. The processor uses the CT-slice images EL-PIM(s) for all the S CT-slices, optionally in a block 113, to provide an E-SNR CT-image of patient 22 having an SNR that is enhanced relative to the SNR of the originally acquired L-SNR CT-image of patient 22.

(27) It is noted that the above description of FIGS. 1A-3 describes embodiments of the disclosure with reference to processing CT-slices acquired by CT-scanners, however, as noted above practice of an embodiment of the disclosure is not limited to CT-scanners and CT-slices. For example, an embodiment of the disclosure may be configured to process sinogram images acquired by a CT-scanner prior to the sinograms being processed to provide images of CT-slices of a patient. For example, H-SNR database 54 may comprise high SNR sinograms and/or sinogram H-Patches extracted from high SNR sinograms. Processor 55 may process low SNR sinograms of CT-slices of patient 22 acquired by CT-scanner 20 using sinogram H-Patches in a procedure similar to procedure 100 to provide high SNR sinograms for patient 22. The high SNR sinograms may then be processed to provide a picture CT-image of patient 22 having an SNR that is enhanced relative to an SNR of a CT-image of the patient that might have been provided using the L-SNR sinograms. And systems and methods in accordance with embodiments of the disclosure may be configured to enhance for example, low-SNR satellite images, subsurface Earth images acquired for example using seismic tomography, or ultrasound images of the human body or of inanimate bodies.

(28) In the description and claims of the present application, each of the verbs, comprise include and have, and conjugates thereof, are used to indicate that the object or objects of the verb are not necessarily a complete listing of components, elements or parts of the subject or subjects of the verb.

(29) Descriptions of embodiments of the disclosure in the present application are provided by way of example and are not intended to limit the scope of the disclosure. The described embodiments comprise different features, not all of which are required in all embodiments. Some embodiments utilize only some of the features or possible combinations of the features. Variations of embodiments of the disclosure that are described, and embodiments comprising different combinations of features noted in the described embodiments, will occur to persons of the art.