Image registration system and method

Abstract

An image registration method, one of a plurality of scan images of a scanned object is set as a reference image, and the other scan images are set as floating images, and the floating images are aligned with the reference image by transforming the floating images according to registration schemes, includes: based on the floating images and the reference image, a plurality of registration schemes are calculated in such a way that the transformed floating images have high similarity with the reference image; for each registration scheme, the similarity of the floating image and the reference image transformed by the registration scheme is calculated as the registration similarity of the registration scheme, the perturbation registration similarity is calculated based on the similarity of the floating image and the reference image transformed by the registration scheme after adding perturbation, and the reliability index based on the registration similarity and the perturbation registration similarity is calculated; the registration schemes are filtered according to the reliability index; and registration is performed according to the filtered registration schemes.

Description

TECHNICAL FIELD

[0001] The present application relates to an image registration system and method. BACKGROUND

[0002] Currently, there are known various internal structure imaging techniques that image the internal structure of an object (including non-living objects and living objects) by scanning the object. Examples include CT (Computed Tomography) imaging, ultrasonic imaging, MRI (Magnetic Resonance Imaging), and the like. Since various internal structure imaging techniques differ in their characteristics and imaging results, when detailed information of the internal structure of an object is required, it is common to select multiple internal structure imaging techniques to be applied at the same time for imaging. At this time, it is necessary to register the scan images of the object generated by the multiple internal structure imaging techniques so that the spatial coordinates of the object in each scan image match each other.

[0003] There is known an image registration system that sets one of multiple scan images as a reference image and performs transformation on other scan images of the same or different modalities so as to align with the reference image. Patent Literature 1 describes an image registration system that determines a registration scheme of geometric transformation on each scan image by calculating the similarity between the transformed scan image and the reference image, and performs image registration. The image registration system of Patent Literature 1 calculates the similarity between different scan images using a normalized cross-correlation function, and is likely to generate an erroneous registration scheme due to the influence of noise and artifacts included in the scan images. Patent Literature 2 describes an image registration system that calculates the similarity based on the certainty of the combination of corresponding pixels between different scan images, and is capable of improving the reliability of the calculated registration scheme. However, the technical solution of Patent Literature 2 also has a case where an erroneous registration scheme is generated due to the influence of noise and artifacts.

[0004] Prior Art Documents

[0005] Patent Literature

[0006] Patent Literature 1: US 2016 / 0133016 A1

[0007] Patent Literature 2: Japanese Patent Application Publication No. 2013-146540 SUMMARY

[0008] Problems to be Solved by the Invention

[0009] The present application has been made to solve the problem of providing an image registration system and an image registration method capable of generating a registration scheme with high reliability.

[0010] Means for Solving the Problems

[0011] The image registration method of the embodiment sets one of a plurality of scan images of a scanned object as a reference image, sets the other scan images as floating images, and aligns the floating images with the reference image by transforming the floating images by registration schemes. The image registration method includes: a registration scheme calculation step of calculating a plurality of registration schemes in such a manner that the transformed floating images have high similarity to the reference image, based on the floating images and the reference image; a reliability index calculation step of calculating, for each registration scheme, a registration similarity as a similarity of the floating image transformed by the registration scheme to the reference image, calculating a perturbed registration similarity based on a similarity of the floating image transformed by the registration scheme to which a perturbation is added to the reference image, and calculating a reliability index based on the registration similarity and the perturbed registration similarity; a registration scheme filtering step of filtering the registration schemes according to the reliability index; and a registration step of performing registration according to the filtered registration schemes.

[0012] The image registration system of the embodiment sets one of a plurality of scan images of a scanned object as a reference image, sets the other scan images as floating images, and aligns the floating images with the reference image by transforming the floating images by registration schemes. The image registration method includes: a registration scheme calculation section of calculating a plurality of registration schemes in such a manner that the transformed floating images have high similarity to the reference image, based on the floating images and the reference image; a reliability index calculation section of calculating, for each registration scheme, a registration similarity as a similarity of the floating image transformed by the registration scheme to the reference image, calculating a perturbed registration similarity based on a similarity of the floating image transformed by the registration scheme to which a perturbation is added to the reference image, and calculating a reliability index based on the registration similarity and the perturbed registration similarity; a registration scheme filtering section of filtering the registration schemes according to the reliability index; and a registration section of performing registration according to the filtered registration schemes.

[0013] Effects of the Invention

[0014] According to the image registration system and the image registration method of the present invention, a registration scheme with high reliability can be generated. BRIEF DESCRIPTION OF DRAWINGS

[0015] Figure 1 is a diagram showing an example of the structure of the image registration system of the first embodiment.

[0016] Figure 2 is a flowchart showing an example of the image registration method of the first embodiment.

[0017] Figure 3 is a diagram indicating an example of a reference processing image and a floating processing image.

[0018] Figure 4 is a diagram for explaining a problem that can occur when a registration scheme is generated based on similarity.

[0019] Figure 5 is a diagram for explaining an effect of a perturbation on registration similarity of a registration scheme.

[0020] Figure 6 is a diagram indicating scan images of different modalities of a liver.

[0021] Figure 7 is a diagram indicating scan images of different modalities of a liver that are registered. DETAILED DESCRIPTION

[0022] Hereinafter, an image registration system and method, a registration scheme evaluation system and method according to the present application will be described with reference to the accompanying drawings.

[0023] First Embodiment

[0024] Figure 1 is a diagram indicating an example of a structure of an image registration system 100 according to the first embodiment. The image registration system 100 according to the first embodiment has a display section 110, an input section 120, a storage section 130, a registration scheme generation section 140, a reliability index calculation section 150, a registration scheme filtering section 160, and an image registration section 170. The display section 110, the input section 120, the storage section 130, the registration scheme generation section 140, the reliability index calculation section 150, the registration scheme filtering section 160, and the image registration section 170 are communicably connected to each other.

[0025] The display section 110 displays various information. For example, the display section 110 displays a reference image and a floating image used in image registration, and displays a GUI (Graphical User Interface) for receiving an input operation from a user, and the like. For example, the display section 110 is an LCD (Liquid Crystal Display) or an organic EL (Electroluminescence) display, or the like. Y

[0026] ​The input unit 120 accepts user input operations and outputs a signal based on the accepted input operation to the registration scheme generation unit 140. For example, the input unit 120 can be implemented using a mouse and keyboard, trackball, switch, button, joystick, touchscreen, etc. The input unit 120 can also be implemented using a user interface that accepts voice input, such as a microphone. When the input unit 120 is a touchscreen, the display unit 110 can be integrated with the input unit 120.

[0027] Storage unit 130, for example, uses ROM, flash memory, RAM (Random Access Memory) Y Storage devices such as random access memory (RAM), hard disk drives (HDDs), solid-state drives (SSDs), and registers are used to implement this. Flash memory, HDDs, and SSDs are non-volatile storage media. These non-volatile storage media can be implemented using other storage devices connected via a network, such as network-attached storage (NAS) and external storage servers. These networks include, for example, the Internet, wide area networks (WANs), local area networks (LANs), carrier terminals, wireless communication networks, wireless base stations, and dedicated lines.

[0028] The storage unit 130 stores a plurality of scan image groups, each of which includes a plurality of scan images obtained by scanning the same region of the same scanned object by the same scanning method with different parameters or different scanning methods. Hereinafter, the scan images obtained by different scanning methods are referred to as scan images of different modalities. Furthermore, hereinafter, the scan images are described as three-dimensional gray scale images, but the scan images can be two-dimensional images or color images such as RGB images, as long as the plurality of scan images included in the same scan image group are of the same type in terms of dimensions or colors. In the present embodiment, the scan images are three-dimensional images of width W x height H x depth D, each of which includes voxels that represent structural information of the scanned object at specific positions by gray scales. The positions of the voxels in the scan images can be represented by a rectangular coordinate system determined by mutually orthogonal X, Y, and Z axes, for example, which correspond to the width W, height H, and depth D of the scan images, respectively. Each of the scan images included in the same scan image group represents the same region of the same scanned object, but since the different scan images are obtained by different scanning parameters or scanning methods, the coordinates of the scan images do not match each other, and registration is required. In actual applications, due to limitations of scanning technology, different scanning methods usually have different fields of view and different scanning coverage, for example, MRI technology usually has a large field of view and scanning coverage, while the field of view and scanning coverage of ultrasonic imaging technology are usually small, and thus the proportions of the portions of the scan images of different modalities corresponding to the scanning coverage are different. In three-dimensional images, for the portions corresponding to regions not covered in scanning, for example, zero padding is performed.

[0029] The registration scheme generation unit 140 reads the scan image groups stored in the storage unit 130, sets one of the scan images included in the scan image groups as a reference image and the remaining images as floating images, and calculates registration schemes for the floating images based on the reference image. Here, the reference image refers to an image that provides a positional reference. The registration scheme generation unit 140 performs geometric transformation including translation and rotation on each of the floating images so that the positions of the internal structures of the scanned object corresponding to the coordinates of the reference image and the floating image match each other, and sets a transformation scheme representing the above-mentioned geometric transformation as the registration scheme for the floating image.

[0030] The reliability index calculation unit 150 evaluates the reliability of the registration schemes generated by the registration scheme generation unit 140 and calculates a reliability index RI. The reliability index RI is used to confirm whether the registration scheme generation unit 140 has generated a mis-matched registration scheme due to image noise, a small field of view of ultrasonic scanning, or differences in imaging effects between different modalities, and the like.

[0031] The registration scheme filtering section 160 filters out registration schemes that are likely to be mismatched, from among the registration schemes generated for the floating image, based on the similarity between the floating image after transformation by each registration scheme and the reference image, and the reliability index RI of each registration scheme calculated by the reliability index calculation section 150.

[0032] The image registration section 170 displays the floating image after transformation by the registration scheme filtered out by the registration scheme filtering section 160, together with the reference image, to the display section 110.

[0033] The registration scheme generation section 140, the reliability index calculation section 150, the registration scheme filtering section 160, and the image registration section 170 are implemented, for example, by a hardware processor such as a CPU and a GPU executing a program (software) stored in the storage section 130. Some or all of these constituent elements can be implemented by a hardware (circuit section: circuitry) such as an LSI, an ASIC, and an FPGA, and can also be implemented by a cooperative action of software and hardware. The above-mentioned program can be stored in advance in the storage section 130, and can also be stored in a removable storage medium such as a DVD and a CD-ROM, and mounted from the storage medium to the storage section 130 by mounting the storage medium to a drive device of the image registration system 100. Y ) to be implemented. The above-mentioned program can be stored in advance in the storage section 130, and can also be stored in a removable storage medium such as a DVD and a CD-ROM, and mounted from the storage medium to the storage section 130 by mounting the storage medium to a drive device of the image registration system 100.

[0034] Figure 2 is a flowchart showing an example of the image registration method of the first embodiment. Referring to Figure 2 The flow of the image registration method of the present embodiment will be described.

[0035] In step S101, the user selects a group of scan images stored in the storage section 130 that need to be registered, by the input section 120, according to the user interface displayed on the display section 110.

[0036] In step S102, the registration scheme generation section 140 reads the group of scan images selected by the user from the storage section 130, and sets one of the scan images as the reference image R, and the remaining scan images as the floating image F. Preferably, the registration scheme generation section 140 sets the scan image in which the proportion of the portion filled with zeros is the smallest among the scan images in which the scan coverage is the widest, as the reference image R. Hereinafter, for ease of explanation, it is assumed that the group of scan images contains two scan images, and the scan image in which the scan coverage is wider is set as the reference image R, and the other scan image is set as the floating image F.

[0037] In step S103, the registration scheme generation unit 140 preprocesses the reference image R and the floating image F. Preprocessing includes image enhancement, noise reduction, and feature extraction. Feature extraction includes plane extraction, edge extraction, and gradient extraction; when the scanned object is an organ of a living organism, gradient extraction is preferred. Hereinafter, the preprocessed reference image R and the floating image F will be referred to as the preprocessed reference image RP and the preprocessed floating image FP, respectively. For ease of explanation, the three-dimensional versions of the preprocessed reference image RP and the preprocessed floating image FP will sometimes be represented by two-dimensional cross-sections.

[0038] Figure 3 This is a schematic diagram illustrating an example of a reference processed image RP and a floating processed image FP. Reference Figure 3 The legend in the diagram, Figure 3 In the diagram, solid lines represent blood vessels, hollow circles represent tumors, dashed lines represent shadows caused by tumor occlusion, solid circles represent pathological tissue, and areas with dots represent noise / artifact regions. Artifacts are various forms of images that appear on the image but do not actually exist. Noise / artifact regions refer to areas where pixel values ​​have randomly changed due to noise or artifacts, causing blurring or graininess in the region. Figure 3 (a) represents an example of a reference processed image RP. Figure 3 (b) represents an example of floating image processing (FP). For example... Figure 3 As shown, since the reference image RP and the floating image FP are images from different modalities of different scanning techniques, their scanning ranges and the effects presented on each structure are different. The reference image RP has a wide scanning range and is not affected by occlusion effects, but it contains noise / artifact regions. The reference image RP is, for example, a scan image obtained through magnetic resonance imaging (MRI). The floating image FP has a narrow scanning range and is affected by occlusion effects. The floating image FP is, for example, a scan image obtained through ultrasound scanning. Figure 3 In the reference image RP and the floating image FP, the blood vessels and tumors correspond to each other. However, because the scanning range of the reference image R is wider than that of the floating image F, blood vessels and pathological tissues not shown in the floating image FP are present in the reference image RP. Furthermore, although denoising was performed on both the reference image RP and the floating image FP, the influence of noise could not be completely eliminated. Due to noise or artifacts, noise / artifact areas with marked points appear in the reference image RP. Additionally, tumor shadows appear in the floating image FP, but not in the reference image RP.

[0039] In step S104, the registration scheme generation unit 140 generates multiple initial registration schemes within the transformable range of the floating processed image FP based on a pre-set registration algorithm. In this embodiment, n registration schemes are generated for the floating processed image FP, and the registration schemes of the floating processed image FP are respectively designated as registration scheme S1 to registration scheme S2. n (where n is a natural number). Below, without considering registration schemes S1 to S2, the following discussion will focus on registration schemes S1 to S2. n When distinguishing between them, they are collectively referred to as registration scheme S. Registration scheme S is, for example, a transformation matrix that performs geometric transformations on a 3D image, including translation, rotation, and scaling. In this embodiment, registration scheme S performs translation transformations on the X, Y, and Z axes and rotation transformations around the X, Y, and Z axes of the scanned image in a Cartesian coordinate system defined by mutually orthogonal X, Y, and Z axes. The translation amounts on the X, Y, and Z axes of registration scheme S are respectively denoted as t. X t Y t Z Let the rotation of the registration scheme S around the X-axis, Y-axis, and Z-axis be r respectively. X r Y r Z .

[0040] The transformable range of the floating image FP refers to the translation amount t of the registration scheme S calculated by the registration scheme generation unit 140. X t Y t Z and rotation amount r X r Y r Z The maximum allowable range. This variable range can be set according to the object being scanned. For example, the variable range of the floating image FP can be set according to the size and spatial shape of the organ, setting a larger range for large organs and a smaller range for small organs. For example, in the case of the liver being scanned, the translation amount t... X t Y t Z The ranges are (-40cm, 40cm), (-40cm, 40cm), and (-40cm, 40cm), with rotation amount r. X r Y r Z The ranges are (-50°, 40°), (-80°, 30°), and (-30°, 150°). For example, in the case where the object being scanned is a prostate, the translation amount t X t Y t Z The ranges are (-10cm, 10cm), (-10cm, 10cm), and (-10cm, 10cm), with rotation amount r.X Y Z The ranges of the translation amounts t X Y Z The ranges of the translation amounts t X Y Z The ranges of the rotation amounts r

[0041] The registration scheme generation unit 140 can generate a plurality of initial registration schemes randomly or at a certain interval within the transformable range of the floating process image FP, or can subdivide the transformable range into a plurality of sub-ranges and generate initial registration schemes in each of the sub-ranges. Due to image noise, differences in imaging effects between different modalities, and randomness of the registration algorithm, a single registration scheme can not be able to satisfy the accuracy requirement of registration or can cause mis-matching, and therefore, by generating a plurality of different registration schemes and selecting a registration scheme with high registration accuracy from among them, the accuracy of registration can be improved.

[0042] In step S105, the registration scheme generation unit 140 performs a certain number of iterations of optimization of the registration scheme S1 to the registration scheme S n in the transformable range of the floating process image FP in such a manner that the similarity of the transformed reference process image RP and the floating process image FP becomes higher, based on a pre-set iterative optimization algorithm. In each iteration of optimization, the optimization algorithm slightly corrects the translation amounts t X Y Z and the rotation amounts r X Y Z For example, the optimization algorithm can be a stochastic gradient descent method, a particle swarm optimization algorithm, a simulated annealing algorithm, or the like. The similarity can be, for example, Cross Correlation, Normalized Cross Correlation, Sum of Square Differences, or the like.

[0043] Next, a problem in evaluating registration schemes based on similarity will be described with reference to Figure 4 Figure 4 ​​​​​​​​​​​(A) indicates the reference processed image RP and the floating processed image FP that have been registered by a correct registration scheme. Figure 4 (B) indicates the reference processed image RP and the floating processed image FP that have been registered by an incorrect registration scheme. In the reference processed image RP and the floating processed image FP that have been registered by a correct registration scheme Figure 4 In (A), the tumor and the blood vessel in the reference processed image RP and the floating processed image FP match each other, but the similarity of the reference processed image RP and the floating processed image FP that have been registered is not high because the noise / artifact region and the pathological tissue in the reference processed image RP have no matching parts in the floating processed image FP, and the shadow of the tumor in the floating processed image FP has no matching part in the reference processed image RP. In contrast, in (B), the similarity of the reference processed image RP and the floating processed image FP that have been registered is high because the blood vessel parts in the reference processed image RP and the floating processed image FP match each other, and the noise / artifact region and the pathological tissue in the reference processed image RP and the blood vessel in the floating processed image FP are incorrectly matched because of the similar pixel values, and the shadow of the tumor in the floating processed image FP and another blood vessel part in the reference processed image RP are incorrectly matched because of the similar pixel values. Figure 4

[0044] Therefore, if the registration scheme is evaluated based only on the similarity of the reference processed image RP and the floating processed image FP that have been registered, the correct registration scheme can be evaluated as low, and the incorrect registration scheme can be evaluated as high, and the evaluation result can not truly reflect the reliability of the registration scheme.

[0045] In contrast, in steps S106 to S108 of the image registration method of the present application, the reliability index calculating section 150 calculates the reliability index RI that indicates the reliability of the registration scheme S for each of the plurality of registration schemes S that have been optimized by the registration scheme generating section 140 in step S105. The reliability of the registration scheme is evaluated by the reliability index RI, and the incorrect registration scheme with low reliability can be avoided.

[0046] The calculation process of the reliability index RI in the image registration method of the present application will be described in detail below.

[0047] First, in step S106, the reliability index calculating section 150 calculates the reliability index RI for each of the registration scheme S1 to the registration scheme Sn by the following equation 1. n ​The transformed floating process image FP is transformed, and the similarity of the transformed floating process image FP to the reference process image RP is calculated as the registration similarity of each registration scheme S. Specifically, the similarity calculation section 151 of the reliability index calculation section 150 applies, to the floating process image FP, the transformation matrix corresponding to each of the registration scheme S1 to the registration scheme S n , translates the floating process image FP in the X-axis, the Y-axis, and the Z-axis, and rotates the floating process image FP around the X-axis, the Y-axis, and the Z-axis. Then, the similarity calculation section 151 calculates, for each of the registration scheme S1 to the registration scheme S n , the similarity of the floating process image FP transformed by the registration scheme to the reference process image RP. The similarity can be, for example, Cross Correlation, Normalized Cross Correlation, or Sum of Square Differences. The similarity used in step S105 and step S106 can be the same or different.

[0048] In step S107, the reliability index calculation section 150 respectively adds a perturbation to each of the registration scheme S1 to the registration scheme S n , and calculates, based on the registration similarity of each of the registration scheme S1 to the registration scheme S n to which the perturbation is added, the perturbation registration similarity of each of the registration scheme S1 to the registration scheme S n .

[0049] Specifically, for each of the registration scheme S1 to the registration scheme S n , first, the perturbation section 152 of the reliability index calculation section 150 reads, from the storage section 130, the search space Ω for the current scan image group, and selects one or more perturbations D from the search space Ω. In the present embodiment, the perturbation section 152 selects k perturbations D from the search space Ω, and denotes the k perturbations as the perturbation D1 to the perturbation D k . Then, the perturbation section 152 applies, to the registration scheme S, each of the perturbation D1 to the perturbation D k , and calculates the registration similarity of the registration scheme S to which the perturbation D is added. Here, the registration similarity calculation method of the registration scheme S to which the perturbation D is added is the same as the registration similarity calculation method of the registration scheme S in step S106, and thus detailed description is omitted. Then, the similarity calculation section 151 calculates, for the registration scheme S, the registration similarity of the registration scheme S to which each of the perturbation D1 to the perturbation D kThe average registration similarity of the registration scheme S is then used as the perturbation registration similarity of the registration scheme S. Alternatively, the perturbation registration similarity can be calculated by adding each perturbation D1 to perturbation D2. k The median and mode of the registration similarity of the subsequent registration scheme S are used as the perturbation registration similarity.

[0050] The search space Ω is a set of perturbations D, containing multiple perturbations D with different perturbation amounts. Perturbations D are used to fine-tune the transformation amounts of the registration scheme S. In this embodiment, the perturbations D adjust the translation and rotation amounts. Let the perturbation amount of perturbation D on the translation amounts in the X, Y, and Z axes be Δt. X , Δt Y Δt Z Let the disturbance D affect the rotational magnitudes along the X, Y, and Z axes by Δr. X , Δr Y , Δr Z In this case, the translations of the registration scheme S with perturbation D along the X, Y, and Z axes are respectively t. X +Δt X t Y +Δt Y t Z +Δt Z Let the rotations around the X, Y, and Z axes be r respectively. X +Δr X r Y +Δr Y r Z +Δr Z .

[0051] Below, refer to Figure 5 The impact of perturbations on the registration similarity of the registration scheme is explained. Figure 5 The (A) overlap represents the reference image RP and the floating image FP after registration using the correct registration scheme with added perturbation. Figure 5 The (B) overlap represents the reference image RP and the floating image FP after registration using an incorrect registration scheme with added perturbations. Figure 5 The registration scheme used and Figure 4 Same. (Refer to...) Figure 5 In (A), after adding perturbation, the tumors and blood vessels that originally overlapped in the reference image RP and the floating image FP are slightly misaligned due to the perturbation, reducing the matching degree. On the other hand, the noise / artifact regions and pathological tissues in the reference image RP still cannot match the floating image FP, and the shadows of the tumors in the floating image FP still cannot match the reference image RP. Therefore, the registration similarity of this registration scheme changes little compared to before adding perturbation. Figure 5of the reference processed image RP and the floating processed image FP are slightly misaligned due to the influence of the added disturbance, the matching degree is reduced, and the blood vessels in the floating processed image RP that were originally matched with the noise / artifact region and the pathological tissue are displaced due to the influence of the added disturbance, and the pixel values thereof no longer match the pixel values of the corresponding positions in the reference processed image RP, so the registration similarity of this registration scheme changes greatly compared to before the disturbance is added.

[0052] In step S108, the difference value calculation section of the reliability index calculation section 150 calculates, for each of the registration scheme S1 to the registration scheme S n the registration similarity of the registration scheme S and the disturbance registration similarity calculated in step S107 as the reliability index RI. For example, the difference value such as the difference, quotient, or standard deviation of the registration similarity of the registration scheme S and the disturbance registration similarity can be calculated as the reliability index RI. In the present embodiment, the quotient value obtained by dividing the registration similarity of the registration scheme S by the disturbance registration similarity is used as the reliability index RI. In this case, the closer the reliability index RI is to 1, the smaller the difference between the registration similarity of the registration scheme S and the disturbance registration similarity, the higher the reliability of the registration scheme S, and the farther the reliability index RI is from 1, the larger the difference between the registration similarity of the registration scheme S and the disturbance registration similarity, the lower the reliability of the registration scheme S.

[0053] Next, an example of calculating the reliability index RI for the registration schemes of the liver nuclear magnetic resonance scan image and the ultrasound scan image will be described. Figure 6 is a graph showing different modality scan images of the liver, Figure 6 (a) of (a) shows the nuclear magnetic resonance scan image of the liver, Figure 6 (b) of (a) shows the ultrasound scan image of the liver. Figure 7 is a graph showing different modality scan images of the liver after registration, Figure 7 (a) of (a) shows the nuclear magnetic resonance scan image and the ultrasound scan image after registration by the correct registration scheme, Figure 7 (b) of (a) shows the nuclear magnetic resonance scan image and the ultrasound scan image after registration by the incorrect registration scheme. As Figure 7 the registration similarity of the correct registration scheme shown in (a) is -43.484, and the reliability index RI is 1, as Figure 7 the registration similarity of the incorrect registration scheme shown in (b) is -44.0004, and the reliability index RI is 0.455196. Although the registration similarity of the incorrect registration scheme is high, the reliability index RI is low, so the reliability of this registration scheme is evaluated as low.

[0054] To increase the difference between the reliability index RI of the correct registration scheme and the reliability index RI of the false registration scheme. Different search spaces Ω can be set for different scanned objects. In the case of the scanned object being a pancreas, since the pancreas and the aorta below it grow in a strip shape, increasing the rotational disturbance can increase the difference between the reliability index RI of the correct registration scheme and the reliability index RI of the false registration scheme, and thus the rotational amount of the disturbance D included in the search space Ω of the pancreas can be increased. In the case of the scanned object being a prostate, since the prostate is roughly elliptical, increasing the translational disturbance can increase the difference between the reliability index RI of the correct registration scheme and the reliability index RI of the false registration scheme, and thus the translational amount of the disturbance D included in the search space Ω of the prostate can be increased. In the case of the scanned object being a liver, since the structure of the liver is complex, the translational disturbance and the rotational disturbance need to be increased at the same time, and thus the translational amount and the rotational amount of the disturbance D included in the search space Ω of the prostate can be increased. Therefore, the present application determines the variation of the search space according to the size and the spatial shape and the like of the scanned object.

[0055] In step S109, the registration scheme filtering section 160 filters the registration schemes according to the reliability index RI, and updates the transformable range of the floating process image FP. Specifically, the registration scheme filtering section 160 sorts the plurality of registration schemes S1 to S n in order from high to low in terms of the reliability index RI, removes a predetermined proportion of the registration schemes ranked low, and removes a certain range around the amount of transformation corresponding to the removed registration schemes from the transformable range of the floating process image FP. For example, the registration schemes S of the first half (the first 50%) of the registration schemes S1 to S n with high reliability index RI can be retained, and the registration schemes S of the second half (the second 50%) can be removed. For example, if the registration schemes with a translational amount of t X 0, a rotational amount of r Y 0, a rotational amount of r Z 0, a rotational amount of r X 0, a rotational amount of r Y 0, a rotational amount of r Z 0) are removed, a certain range around (t X 0, t Y 0, t Z 0, r X 0, r Y 0, r Z 0) is removed from the transformable range of the floating process image FP. The size of the removed range can be appropriately set according to the size of the transformable range of the floating process image FP, and for example, in the case of the scanned object being a human organ, a range of t X 0, t Y 0, t Z0, r X 0, r Y 0, r Z 0, r After the transformable range of the floating process image FP is updated by the registration scheme filter 160, in the subsequent step S105, each registration scheme is optimized within the updated transformable range, and the registration scheme is no longer corrected to the removed range. Thus, it is possible to avoid generating again the false registration scheme with low reliability index RI in step S105.

[0056] In step S110, the registration scheme generation unit 140 determines whether the number of remaining registration schemes is one, and in the case where the number of remaining registration schemes is one (step S110: YES), the registration scheme generation unit 140 transmits the registration scheme to the image registration unit 170, proceeds to step S111, and in the case where the number of remaining registration schemes is not one (step S110: NO), returns to step S105 and continues the iterative optimization of the remaining plurality of registration schemes.

[0057] In step S111, the image registration unit 170 registers the scan images according to the registration scheme output from the registration scheme generation unit 140, and displays the registered scan images on the display unit 110.

[0058] According to the image registration system and the image registration method of the present embodiment, it is possible to filter out the false registration scheme with low reliability affected by noise or artifacts in the process of generating the registration scheme, and thus it is possible to avoid useless calculation, improve the efficiency of registration, and at the same time, generate a registration scheme with high reliability.

[0059] Further, in the above embodiment, the registration of scan images of different modalities is exemplified, but the present application can also be applied to the registration of scan images of the same modality.

Claims

1. An image registration method of setting one of a plurality of scan images of a scanned object as a reference image, setting the other scan images as floating images, and aligning the floating images with the reference image by a registration scheme, the image registration method comprising the steps of: a registration scheme calculation step of generating a plurality of initial registration schemes within a transformable range of a floating process image based on a pre-set registration algorithm, and calculating the registration scheme by performing a certain number of iterative optimizations on the plurality of initial registration schemes in such a manner that the similarity between the floating image transformed by the registration scheme and the reference image becomes higher based on a pre-set iterative optimization algorithm; a reliability index calculation step of calculating, for each of the registration schemes, the similarity between the floating image transformed by the registration scheme and the reference image as a registration similarity of the registration scheme, calculating a perturbed registration similarity based on the similarity between the floating image and the reference image transformed by the registration scheme with a perturbation added thereto, and calculating a reliability index based on the registration similarity and the perturbed registration similarity; a registration scheme filtering step of filtering the registration schemes based on the reliability index; and a registration step of performing registration based on the filtered registration schemes.

2. The image registration method according to claim 1, wherein the reliability index is a difference value between the registration similarity and the perturbed registration similarity.

3. The image registration method according to claim 1, wherein in the registration scheme calculation step, the registration scheme is calculated within a transformable range of the floating image, in the registration scheme filtering step, the transformable range is updated to remove a range corresponding to the filtered registration scheme from the transformable range.

4. The image registration method according to claim 1, wherein in the reliability index calculation step, a plurality of different perturbations are generated for each of the registration schemes, and an average value of a plurality of similarities between the floating image and the reference image transformed by the registration scheme with each of the different perturbations added thereto is calculated as the perturbed registration similarity.

5. The image registration method according to claim 2, wherein the difference value is any one of a difference, a quotient, and a standard deviation.

6. The image registration method according to claim 1, wherein the similarity is any one of cross-correlation, normalized cross-correlation, and total sum of squared differences.

7. The image registration method according to claim 1, wherein the perturbation includes at least one of a translation transformation, a rotation transformation, and a scaling transformation.

8. The image registration method according to claim 7, wherein the scanned object is a pancreas, and the perturbation includes a rotation transformation.

9. The image registration method according to claim 7, wherein the scanned object is a prostate, and the perturbation includes a translation transformation.

10. The image registration method according to claim 7, wherein the scanned object is a liver, and the perturbation includes a translation transformation and a rotation transformation. ​ 11. The image registration method according to any one of claims 1 to 10, wherein the floating image is generated by ultrasonic imaging.

12. The image registration method according to any one of claims 1 to 10, wherein the reference image is generated by any one of ultrasonic imaging, nuclear magnetic resonance imaging, and computed tomography.

13. The image registration method according to any one of claims 1 to 10, wherein the reference image and the floating image are scanning images of different modalities.

14. An image registration system that sets one of a plurality of scanning images of a scanned object as a reference image, sets the other scanning images as floating images, and aligns the floating images with the reference image by a registration scheme, the image registration system comprising: a registration scheme calculation section that generates a plurality of initial registration schemes within a transformable range of a floating processing image based on a registration algorithm set in advance, performs a certain number of iterations of optimization based on an iterative optimization algorithm set in advance in such a manner that the similarity of the floating image after transformation and the reference image is high, and calculates the registration scheme; a reliability index calculation section that, for each registration scheme, calculates the similarity of the floating image after transformation by the registration scheme and the reference image as a registration similarity of the registration scheme, calculates a perturbed registration similarity based on the similarity of the floating image and the reference image after transformation by the registration scheme with a perturbation added, and calculates a reliability index based on the registration similarity and the perturbed registration similarity; a registration scheme filtering section that filters the registration schemes according to the reliability index; and a registration section that performs registration according to the filtered registration schemes.

Images