JP6061984B2 - Image processing apparatus, image processing method, and program - Google Patents

Description

  The present invention relates to an image processing apparatus, an image processing method, and a program for analyzing a tomographic image of an eye.

  The optic nerve head of the retina is the part where the bundle of optic nerves goes deep into the eye. In the case of diseases such as glaucoma, since changes appear in the optic papilla, it is important for diagnosis to specify the optic papilla.

  Patent Document 1 discloses a technique for specifying an area (disc area) of the optic nerve head by obtaining an end of a retinal pigment epithelium that is one of the layers in the retina. This technique uses the anatomical feature that the retinal pigment epithelium does not exist immediately below the center of the optic nerve head. Patent Document 2 discloses a technique for identifying a recessed region in the optic nerve head using the depth from the anterior segment to the fundus surface. This technique utilizes the fact that the optic papilla is recessed on the surface of the fundus.

Special table 2009-523563 JP 2008-73188 A

  When trying to identify the optic papilla based on the absence of retinal pigment epithelium, a false image occurs under the blood vessel or under the lesion, and the retinal pigment epithelium becomes unclear. There is a risk of becoming a part. In addition, when trying to specify based on the surface shape of the retina, there is a possibility that a recess other than the optic nerve head is mistakenly set as the optic nerve head.

Index image processing apparatus of the present invention, showing an image acquiring means for acquiring a tomographic image of the fundus oculi, a boundary extracting means for extracting a boundary of the retina and vitreous body of the tomographic image, the depressions viewed degree of the extracted boundary A range extracting unit for extracting a range in which at least one of each layer of the nerve fiber layer, GCL, INL, IPL, OPL, or IS / OS or its layer boundary does not exist in the tomographic image; based said index calculating means is calculated in said range the range extraction unit and extracted, characterized Rukoto that having a specifying means for specifying the optic disc.

  By having such a configuration, it is possible to identify a recess at the boundary between the retina and the vitreous body.

  In particular, when the layer structure is changed by a lesion in the retinal layer, the retinal pigment epithelium is unclear, but the boundary between the retina and the vitreous body is often pushed up by the lesion and becomes a convex portion. Since this shape is greatly different from the depressed optic nerve head, by using the shape of the boundary of the retina, it is possible to reduce errors that erroneously specify the lesion as the optic nerve head.

It is a block diagram of an optical coherence tomography system. 1 is a configuration diagram of an optical coherence tomography apparatus 20. FIG. It is a figure which shows an example of the tomographic image of the optic nerve head vicinity. 3 is a flowchart showing a flow of processing executed by the image processing apparatus 10. It is a figure which shows the identification process of the disk area | region which the nipple analysis part 14 performs. It is a flowchart which shows the flow of the process which the retinal analysis part 12 performs. It is a figure which shows the image near an optic nerve head, and an A scan profile. It is a figure which shows ILM and IS / OS specified by the retina analysis part 12. It is a flowchart which shows the flow of the process which the optic nerve head specific | specification part 13 performs. It is a figure which shows the specific process of the area | region where IS / OS does not exist. It is a figure which shows the specific process of the center position of a recessed area. It is a figure which shows the identification process of the disk area | region which the nipple analysis part 14 which concerns on another Example performs. It is a figure which shows the tomographic image of the optic nerve head vicinity when a retina inclines and was image | photographed. It is a figure which shows the identification process of the disk area | region which the nipple analysis part 14 which concerns on another Example performs. It is a block diagram of the image processing apparatus 1500 which concerns on another Example.

Example 1
The configuration of the optical coherence tomography system 1 according to the present embodiment will be described with reference to FIG. In this system, the optical coherence tomography apparatus 20 captures a tomographic image of the eye to be examined. The photographed tomographic image is subjected to optic nerve head analysis processing and displayed on the display unit 40.

  The image processing apparatus 10 has each block shown in FIG. 1 as an ASIC, FPGA, or other hardware, and is configured to execute the processes of FIGS. The retina analyzing unit 12 analyzes the tomographic image acquired by the image acquiring unit 11 and specifies the position of the boundary surface of each layer of the retina. The retinal analysis unit 12 identifies the three-dimensional shapes of ILM and IS / OS. Here, the retinal analysis unit 12 obtains the recessed portion in the ILM and the region where the layer in the IS / OS is interrupted from the three-dimensional shape specified based on the tomographic image. The optic nerve head specifying unit 13 specifies the optic nerve head from the tomographic image based on the specified position of each boundary surface and information on the shape of the ILM. The nipple analyzing unit 14 analyzes the optic nerve head and identifies a disk region (optic nerve disk) surrounded by the outer edge of the optic nerve head. In addition, a cup area (disc depression, nipple depression) that is a depression of the optic nerve head is identified based on the disk area. The teat analyzer 14 further calculates a C / D ratio and an R / D ratio. The image processing apparatus 10 displays the C / D ratio and R / D ratio on the display unit 40 together with the tomographic image.

  The configuration of the optical coherence tomography apparatus 20 will be described with reference to FIG. Here, the optical coherence tomography apparatus 20 is an OCT imaging apparatus using the principle of optical coherence tomography (OCT; Optical Coherence Tomography). The instruction acquisition unit 21 acquires instruction information for adjusting a two-dimensional measurement range and measurement depth with respect to the fundus surface of the eye to be examined. Based on the instruction information, the galvanometer mirror driving mechanism 201 is driven by the galvanometer mirror 202. The light from the low coherence light source 203 is split into signal light and reference light by the half mirror 204. The signal light is reflected or scattered by the eye 206 via the galvanometer mirror 202 and the objective lens 205. The reference light is reflected or scattered by a reference mirror 207 that is fixedly arranged. The half mirror 204 superimposes the return light of the signal light and the reference light to generate interference light. The diffraction grating 208 splits the interference light into wavelength components of wavelengths λ1 to λn, and each wavelength component is detected by the one-dimensional photosensor array 209. The image reconstruction unit 210 reconstructs a tomographic image of the retina based on detection signals of each wavelength component of the interference light output from the one-dimensional photosensor array 209.

  Note that obtaining a one-dimensional image by irradiating signal light on an arbitrary position of the fundus is called A-scan. This one-dimensional image in the depth direction is called an A-scan image. Further, scanning the fundus by causing the galvano mirror 202 to make signal light incident on the fundus along an arbitrary line is called B scan. A tomographic image obtained by B-scan is called a B-scan image. Furthermore, an image that intersects with the direction of A scan from images obtained by performing A scan at a plurality of positions within a predetermined region on the fundus surface is referred to as a C scan image.

  FIG. 3 shows an example of a tomographic image near the optic nerve head taken by the optical coherence tomography apparatus 20 described above. These images T1 to Tn are obtained by scanning the surface of the fundus with signal light (B-scan) along n parallel lines. The x-axis direction is a direction parallel to the B-scan, the y-axis direction is the direction in which the scan lines are arranged, and the z-axis direction is the depth direction of the eye.

  The boundary L1 in FIG. 3 is the boundary between the ILM (inner boundary membrane) and the vitreous body (hereinafter referred to as ILM boundary), and is the boundary between the retina and the vitreous body. The boundary L2 is a boundary between the NFL (nerve fiber layer) and its lower layer (hereinafter referred to as NFL boundary). The boundary L3 is a boundary (hereinafter referred to as an IS / OS boundary) between the IS / OS (visual cell inner and outer joint joint) and its upper layer. The boundary L4 is a boundary between the RPE (retinal pigment epithelium) and its lower layer (hereinafter, RPE boundary).

  As shown in FIG. 3, as a characteristic of the optic papilla, first, the ILM boundary is recessed in the optic papilla. Second, there is no IS / OS (IS / OS boundary) or RPE (RPE boundary). Using these two features, the position of the optic nerve head is specified from the tomographic image of the retinal layer.

  The shape of the ILM boundary is known to vary greatly among individuals, and the depression of the optic nerve head is not so large, and there may be a depression of the ILM boundary in a region other than the head. However, RPE does not exist anatomically only in the optic nerve head, with some exceptions such as lesions. Therefore, even if there are a plurality of depressions, the position of the optic nerve head can be accurately identified by determining the presence or absence of RPE.

  Furthermore, the effect of using these two features together is significant when there is a false image area due to blood vessels or lesions. The region v1 is a region called a false image region. The false image region is generated when a blood vessel or a lesion absorbs OCT signal light and signal light reaching a position deeper than the blood vessel or the lesion becomes extremely weak. In the false image area, layers and structures that actually exist disappear from the image.

  In many cases where a false image is generated due to a lesion such as bleeding or vitiligo, the ILM boundary above the lesion is pushed up to the vitreous body. As a result, the ILM boundary forms a convex portion. In addition, even when a blood vessel exists, the upper portion of the blood vessel rarely forms a recess. In this embodiment, by determining whether or not the ILM boundary forms a recess and specifying the optic papilla, it is possible to compensate for the disadvantages of specifying the position of the optic papilla only by the presence or absence of RPE.

  In FIG. 3, an area C5 indicates an optic nerve head area and is called a disk area. A region D7 indicates a region surrounded by the edge of the recessed portion of the optic disc and is called a cup region. In this embodiment, in accordance with the definition of Patent Document 1, the disk area is defined as an area surrounded by the end of the RPE. In addition, as shown in FIG. 3, the cup region is defined as a curved surface connecting the intersection of the ILM and a surface translated by a predetermined distance vertically upward from the surface connecting the RPE ends.

  The C / D ratio (cup-to-disc ratio) is defined by the ratio of the diameter of the outer edge of the cup area to the diameter of the disk area. Glaucoma is suspected when the C / D ratio is large, that is, when the depression is large. The R / D ratio (rim-to-disc ratio) is defined as the ratio of the width of the rim region to the measurement position of the width and the nipple diameter passing through the nipple center. The rim area means an area surrounded by the outer edge of the disk area and the outer edge of the cup area. Glaucoma is suspected when the R / D ratio is small, that is, when the recessed area is larger than the optic disc area.

  FIG. 4 is a flowchart of this embodiment. A specific processing procedure executed by the image processing apparatus 10 will be described with reference to this flowchart.

  <Step S401> In step S401, the image acquisition unit 11 acquires a plurality of two-dimensional tomographic images T1 to Tn (see FIG. 3) captured by the optical coherence tomography apparatus 20.

  <Step S402> In step S302, the retinal analysis unit 12 specifies the positions and shapes of the ILM boundary and the IS / OS boundary for each pixel column arranged in the y-axis direction for each of the two-dimensional tomographic images T1 to Tn. Details of the processing will be described later in steps S501 to S511 of FIG.

  <Step S403> In step S403, the optic disc specifying unit 13 specifies the position where the ILM boundary specified by the retinal analysis unit 12 is depressed and the IS / OS specification has failed as the optic disc. . Further, the center position of the disc area of the optic nerve head is specified based on the shape of the depression. Details of this processing will be described later in steps S501 to S504 in FIG.

  <Step S404> In step S404, the nipple analyzing unit 14 specifies a disk area by the area growth method based on the specified position of the optic nerve head. Also, the cup area is specified based on the disk area.

  The nipple analysis unit 14 specifies a disk area using the center of the disk area specified in step S403 as a base point. In this embodiment, as shown in FIG. 5, A-scan labeled “center of disk area” is used as a seed point S, and A-scan labeled “IS / OS cannot be specified” is targeted for area growth. . Then, the disk area R can be specified by expanding from the inside. Further, it can be said that T1 and T2 corresponding to both ends of the specified disk region R are RPE ends.

  Thus, by specifying the center of the disk area and specifying the disk area from the center (inside the disk), the influence of the false image existing in the retina can be reduced when the disk area and the RPE end are specified. it can.

  In this way, the boundary information between the ILM and IS / OS specified from the image is used to specify the vicinity of the center of the optic disc recess and to search for the RPE end from the inside of the disc. RPE edge specific errors can be reduced without being affected by the image.

  Note that the base point does not necessarily have to be the center of the recessed portion, and may be a predetermined position with a high probability of being a region inside the disc. However, since there is a high probability that the center position of the recess is an area inside the disk, the accuracy of the process of specifying the outer edge of the optic disc area can be improved.

  <Step S405> In step S405, the teat analyzer 14 calculates a C / D ratio and an R / D ratio.

  <Step S406> In step S406, the image processing apparatus 10 superimposes the specified RPE end, the cup area, and the disk area on the tomographic image and causes the display unit 40 to display the image.

  Further, the calculated C / D ratio and R / D ratio are also displayed on the image or otherwise. Thereby, the relationship between the shape of the depression of the optic nerve head and the identified cup area or disk area is clearly shown. Therefore, the user can grasp the ground shape of the depression, the calculated C / D ratio, and the basis of the R / D ratio from the image.

  As another example, a two-dimensional area obtained by projecting a cup area and a disk area onto a C scan image is displayed superimposed on the C scan image. Thereby, the shapes of the cup area and the disk area can be grasped. It should be noted that the present invention is not limited to the C-scan image, and may be a fundus surface image captured by another modality such as a fundus camera, or an accumulated image obtained by integrating the B-scan image group in the depth direction. Moreover, the projection image produced | generated using the B scan image group may be sufficient.

<Retina layer boundary identification processing>
Details of the ILM and IS / OS specifying processing in step S402 will be described with reference to FIG. This process is a pre-process for specifying a region where the ILM forms a recess and for specifying a region where the IS / OS layer in the retina is interrupted.

<Step S501>
In step S501, the retinal analysis unit 12 performs image conversion on the OCT tomographic image acquired in step S401. In this embodiment, a median filter and a Sobel filter are respectively applied to a tomographic image to create a median image and a Sobel image. Here, it is assumed that the pixel value is large when the signal strength is strong and small when the signal strength is weak.

  In the present embodiment, the Sobel filter has directionality so as to emphasize a boundary from a low luminance value to a high luminance value when viewed from a shallow direction (above the image) in A-scan. The reason is as follows. In this case, ILM shape characteristics and IS / OS boundary information are used to detect a disk, which is a part necessary for analysis of the optic nerve head. Therefore, it is important to specify the ILM and IS / OS. In the retinal layer structure, ILM is a boundary between a vitreous body having a low luminance value and a retinal tissue having a relatively high luminance value, and IS / OS is also in contact with a relatively dark tissue in a shallow direction. In other words, the ILM and IS / OS are more emphasized by providing the above directionality.

<Step S502>
In step S502, the retinal analysis unit 12 calculates the average luminance value of the background (vitreous body) using the median image created in step S501. In the present embodiment, first, binarization processing is performed on the median image by the P-tile method to specify the background area. Next, the average value of the luminance values of the median image in the background area is calculated.

  The binarization processing by the P-tile method is to create a histogram of an image to be processed, accumulate from the higher or lower luminance value, and determine the luminance value when a predetermined ratio P is reached as a threshold value. As a binary method. In this embodiment, since the ratio of the retinal region in the image is roughly known, the binarization process is empirically performed by setting the P value to 30 percent from the higher luminance value, and the luminance value is equal to or lower than the threshold value. Is a background pixel.

  When the background pixel is specified, the average luminance value of the background is calculated with reference to the luminance value of the median image at the background pixel.

<Step S503>
In step S503, the retinal analysis unit 12 creates a profile from the converted image created in step S501. In this embodiment, a profile is created for each A-scan from both the median image and the Sobel image. By creating a profile from the median image, there is an effect that noise that is particularly problematic in the OCT image is suppressed, and the tendency of the luminance value can be more easily grasped. Also, by creating a profile from the Sobel image, there is an effect that it becomes easier to detect retinal layer boundary candidate points in the specification of the retinal layer boundary performed later. FIG. 7 shows a profile created from a median image and a Sobel image in an A-scan A7 in the tomographic image. As shown in FIG. 7, the tendency of the luminance value can be seen from the profile PM7 of the median image, and the candidate point of the retinal layer boundary can be seen from the profile PS67 of the Sobel image.

  Moreover, it is not always necessary to create a profile from these converted images, as long as an edge having a predetermined strength can be detected from the original image and other converted images.

<Step S504>
In step S504, the retinal analysis unit 12 detects a local maximum point (hereinafter referred to as a peak) from the profile created in step S503. In the present embodiment, a peak in a profile created from a Sobel image is detected. The detection uses a threshold value determined empirically or based on image information. In the retina, ILM and IS / OS reflect or scatter many signals. For this reason, when using a Sobel filter having directionality so as to emphasize the boundary from a low luminance value to a high luminance value as seen from the shallow direction described in step S501, it is easy to detect a strong edge. In addition, since there is no strong edge detected by the Sobel filter having this directionality other than the lesioned part (such as peeling of the vitreous cortex), the ILM and IS / OS are adjusted by this method by adjusting the threshold value. It can be extracted with priority.

<Step S505>
In step S505, the retinal analysis unit 12 counts the peaks detected in step S504, and branches the process based on the number. In this embodiment, when there are two or more peaks that are not specified as the retinal layer boundary or the vitreous cortex at the time of this step input (Yes in step S505), two peaks are selected in order from the shallow direction in A-scan. . And it progresses to Step S506 as the 1st peak and the 2nd peak, respectively. When there is one peak (No in step S505), the largest peak is set as the first peak and the process proceeds to step S508.

<Step S506>
In step S506, the retinal analysis unit 12 compares the profile of the median image between the two peaks selected in step S505 with the average luminance value of the background. In this embodiment, first, for a pixel existing between the first peak and the second peak, a value obtained by multiplying the average luminance value of the background calculated in step S502 by the coefficient 1.2 is set as a threshold value. . Next, pixels having a luminance value larger than the threshold value are counted, and the ratio of the number to the total number of pixels existing between the peaks is calculated from the number.

  This coefficient is determined empirically and is not limited to this. For example, the coefficient may be dynamically determined from the image information by using a ratio between the average luminance value of the background and the average luminance value of an area other than the background (an area greater than or equal to the threshold value in the binarization process).

<Step S507>
In step S507, the retinal analysis unit 12 branches the process based on the ratio calculated in step S506. In the present embodiment, when the calculated ratio is ½ or more (Yes in step S507), it is determined that retinal tissue exists between the peaks, and the process proceeds to step S508. When the calculated ratio is smaller than 1/2 (No in step S507), it is determined that the interval between the peaks is the background, the first peak is not specified as the layer boundary (identified as the vitreous cortex), and the process returns to step S505. Go back and pick the two peaks again.

  In the present embodiment, the retinal tissue or the background is determined from the ratio of pixels equal to or greater than the threshold, but the present invention is not limited to this. For example, a feature amount may be calculated from the profile, and the determination may be performed using the identifier as an input.

<Step S508>
In step S508, the retinal analysis unit 12 identifies one of the peaks as ILM. In this embodiment, since the ILM exists at the upper end of the retinal tissue for the first peak and the second peak determined to have retinal tissue between the peaks in step S507, the first peak is identified as the ILM. In addition, when branching from step S505, the first peak is identified as ILM.

<Step S509>
In step S509, the retinal analysis unit 12 checks whether a feature point equal to or greater than the threshold exists in a deeper direction (below the image) on the same A-scan than the ILM specified in step S508. In this embodiment, a value obtained by multiplying the ILM peak size specified on the same A-scan by a coefficient of 0.8 is set as the threshold value. It is examined whether a peak equal to or greater than this threshold exists in a direction deeper than ILM. If it exists (Yes in step S509), the process proceeds to step S510. If it does not exist (No in step S509), the process proceeds to step S511.

  This threshold value is obtained empirically and is not limited to this. For example, in addition to the peak size, a distance between peaks may be used.

<Step S510>
In step S510, the retinal analysis unit 12 specifies a peak equal to or higher than the threshold set in step S509 as an IS / OS. In the case where there are a plurality of peaks equal to or higher than the threshold, in this embodiment, the peak present at the shallowest position in the peak group equal to or higher than the threshold is defined as IS / OS.

<Step S511>
In step S511, the retinal analysis unit 12 determines that the IS / OS cannot be specified, attaches an “IS / OS cannot be specified” label to A-scan, and ends the process.

  In this way, it is possible to reduce specific errors by determining the structure between the peaks and identifying the type of the layer boundary based on the result. FIG. 8 shows a tomographic image in which ILM and IS / OS are specified using this method. D1 and D3 drawn by thick solid lines in the figure are the specified ILM and IS / OS, respectively. As can be seen from FIG. 8, the ILM is specified in all A-scans. On the other hand, there is an A-scan that cannot identify an IS / OS, and such an A-scan is labeled “IS / OS cannot be identified” as described in step S511.

  In addition, since the position of the ILM in the B-scan image is specified, the shape of the ILM can be specified.

<Optical nerve head depression specific processing>
The details of the optic disc recess identifying process in step S403 will be described with reference to FIG.

<Step S901>
In steps S901 to S903, the optic disc identifying unit 13 identifies the center of the disk area based on the ILM and IS / OS boundary information identified in step S402. In particular, in step S801, an area to be a center candidate of the disk area (hereinafter referred to as a candidate area) is specified. In this embodiment, attention is paid to the fact that there is no IS / OS in the disk area. First, a local region including neighboring A-scans is set for each A-scan. In step S411, the ratio of A-scans labeled “IS / OS cannot be specified” is calculated in the local region. Specifically, as shown in FIG. 10, a target A-scan (A1, A2 in the figure) and a local region (R1, R2 in the figure) including the vicinity of the predetermined range are set. If the “IS / OS cannot be specified” label is attached to ½ or more of the A-scan existing in the local region, the “candidate region” label is attached to the central A-scan. For example, in FIG. 10, it is assumed that an IS / OS is specified for an A-scan whose IS / OS is represented by a solid line BL in the local area, and an “IS / OS cannot be specified” label is assigned to the other A-scans. Suppose that is attached. As can be seen from FIG. 10, in the local region R1, there are 1/2 or more A-scans with the “IS / OS cannot be specified” label, so the “candidate region” label is attached to A1. On the other hand, since the A-scan with the “IS / OS cannot be specified” label does not exist in the local region R2, the “candidate region” label is not attached to A2.

<Step S902>
In step S902, the optic disc identifying unit 13 calculates an ILM gradient in the candidate region identified in step S901. In this embodiment, as in FIG. 11, a local area including the target A-scan and its neighboring A-scan is set, and the A-scan labeled “candidate area” is set as the center of the local area. Process. The gradient calculation method obtains the difference from the ILM coordinate value in the nearby A-scan with reference to the ILM coordinate value in the central A-scan. The gradient here focuses only on the vertical component of the image (z-coordinate value in FIG. 11) and takes the sum of the differences between the central A-scan and all neighboring A-scans, with the downward direction being positive. If the ILM z-coordinate value at the center A-scan is Ic and the ILM z-coordinate value at the nearby A-scan is Ii, the gradient calculation method is as follows.

  The gradient value calculated by the equation 1 takes a large value when the center of the local region comes near the center of the recessed structure of the ILM as shown in FIG.

<Step S803>
In step S803, the optic disc specifying unit 13 checks the ILM gradient calculated in step S802, and sets A-scan having the maximum gradient as the center of the optic disc recess. A-scan, which is the center of the disk area, is labeled “center of disk area”.

  With this process, the position of the optic papilla can be specified, and at the same time the center position can be specified.

  In another method, the region of the recess in the ILM can be specified by specifying the region where the gradient value is positive in both the left and right regions of the central A-Scan position. The region of the depression and the region where the IS / OS cannot be specified are set as candidate regions of the optic nerve head. When there is only one candidate area, the candidate area is the optic disc area. When there are a plurality of candidate regions, the region having the largest degree of depression, that is, the region having the largest gradient value of the above equation 1 is specified as the optic nerve head. Also by this method, the position of the optic nerve head can be specified. It is effective for retinas with few confusing blood vessels and normal eye retinas with flat ILM and few lesions.

  In this embodiment, the gradient value near the specific position is used as an index for evaluating the degree of depression. However, the present invention is not limited to this, and the boundary surface in a predetermined range including the position is not limited to this. It is also possible to use the degree of bending or the amount of change in the depth direction. As another example, a reference plane approximated to the curved surface of the ILM using the least square method or the like is set, and the difference in position in the A-scan direction (z-axis direction) between the reference plane and the curved surface of the ILM is used as an index. It is also possible to calculate the degree of depression.

(Example 2)
In the second embodiment, an example of using the region growing method in consideration of the shape of the disk region in step S404 of the first embodiment will be described. Thick blood vessels are gathered in the vicinity of the optic nerve head, and therefore there are many false image regions due to blood vessels. Considering that the disk area is specified by a simple area growth method as in the first embodiment, in the case of a case where a thick blood vessel extends from the optic nerve head, the disk area includes a false image area due to the blood vessel. Therefore, in the present embodiment, the disk area is specified with higher accuracy by adding “area shape” as a constraint condition of the area growth method in step S404. Since the processing is common except for the optic disc depression specifying process, the description thereof is omitted. Since the configuration of the apparatus is the same as that of the first embodiment, description thereof is omitted.

  The details of the disk area and cup area processing corresponding to step S404 of the first embodiment will be described with reference to FIG.

<Step S1201>
In step S1201, the optic disc specifying unit 13 specifies a disc area search range of a predetermined size based on the disc area specified up to the previous step. In the present embodiment, an ellipse with a predetermined magnification circumscribing an identified area is considered, and the inside of the ellipse is set as a search range. If there is no disk area up to the previous step, a circle with a predetermined radius centered on the “center of disk area” label obtained in step S503 is taken as the search range.

<Step S1202>
In step S1202, the optic nerve head specifying unit 13 checks whether there is an area that has not yet been subjected to the disk area specifying process (hereinafter referred to as an undetermined area) in the disk area search range specified in step S1201. In the present embodiment, a disk area specifying process based on the area growing method is considered using a pixel on the contour line of the specified disk area as a seed point. Therefore, it is examined whether or not an undetermined region exists in the vicinity of each seed point on the contour line. If an undetermined area exists (Yes in step S1202), the process proceeds to step S1203. If an undetermined area does not exist (No in step S1202), the process ends.

<Step S1203>
In step S1203, the optic disc identifying unit 13 performs a disc area identifying process on the undetermined area. In the present embodiment, the disk area is expanded by the area growth method using a pixel on the contour line of the specified disk area as a seed point. The process is repeated until there is no undetermined area in the search range specified in step S1201.

<Step S1204>
In step S1204, the optic disc identifying unit 13 evaluates the identified disc area and calculates an index. In the present embodiment, the evaluation is performed based on the knowledge that the disk area is elliptical. Specifically, an ellipse that circumscribes the specified disk area is obtained, and the ratio of the area of the ellipse to the area of the specified disk area (hereinafter referred to as a filling rate) is calculated as an evaluation index.

  Note that the evaluation index of the specified disk area is not limited to this. For example, in the disk area specifying process such as the area growth method in step S1203, the area or shape change of the area before and after the process may be used as an evaluation index.

<Step S1205>
In step S1205, the optic disc identifying unit 13 branches the process based on the evaluation index calculated in step S1204. In this embodiment, the filling rate is obtained as an evaluation index, and when the filling rate is equal to or lower than a predetermined value (Yes in step S1205), it is considered that the specified disc area is not elliptical, and the disc area specifying process is performed. End the iteration. If the filling rate is larger than the predetermined value (No in step S1205), the process returns to step S1201, and the disk area specifying process is repeated.

  As described above, by performing the disk area specifying process considering the shape, the disk area can be specified with high accuracy even if there is a structure that changes the retinal layer structure such as a blood vessel around the optic papilla. .

Example 3
In the third embodiment, an example in which the disk area and the RPE end are specified more precisely using the trace of the edge component in step S404 of the first embodiment will be described. A tomographic image may be taken with the retina tilted as shown in FIG. When an image is taken with the retina tilted, the signal of a part having a characteristic structure such as an RPE end may be lowered. For example, at the RPE end T62 in FIG. 13, the brightness value decreases toward the tip, so the edge component becomes weak. That is, in the specifying method using a fixed threshold, the RPE end may be specified by being shifted to the outside of the disk area from the actual position. Therefore, in the present embodiment, the RPE end is specified with higher accuracy by adding a precise extraction process of the RPE end by edge tracing after step S504.

  The optic nerve head specifying unit 13 traces the edge component from the RPE end specified in advance to the inside of the disk, thereby specifying the exact position of the RPE end. In the present embodiment, first, the coordinate value and the edge component are examined for each identified RPE end. Next, the edge is traced toward the inside of the disk, starting from the position of each RPE end. The trace refers to the edge component at the position of each RPE end, updates the search point to the position where the edge component existing inside is closest, and also updates the edge component to be referenced. By repeating this, an accurate RPE edge and disk area are specified. Details of the processing will be described later in steps S1401 to S1403. As described above, the RPE end can be specified with higher accuracy by re-searching from the RPE end specified once in consideration of the decrease in luminance due to the photographing condition or the like.

<Precise extraction of optic nerve head depression>
The details of the precise extraction process of the optic disc depression will be described with reference to FIG.

<Step S1401>
In step S1401, the optic nerve head specifying unit 13 refers to the edge component at the RPE end position specified in step S1304 from the Sobel image, and determines a threshold to be used in edge tracing. In the present embodiment, the threshold value is a value obtained by multiplying the edge component at the RPE end position by a coefficient of 0.3.

  This coefficient is determined empirically and is not limited to this.

<Step S1402>
In step S1402, the optic disc identifying unit 13 uses the threshold value obtained in step S1401 and the edge component at the RPE end to obtain an edge component closest to the edge component at the RPE end that is greater than or equal to the threshold value for the nearby A-scan. Search for pixels with In the present embodiment, a nearby Asscan is an A-scan that is adjacent to the inner side of the disc from the RPE end in the B-scan image. As for the search, a search is performed for a nearby A-scan within a predetermined range above and below the RPE at the search start point, and the pixel that best meets the above conditions is searched.

  If there is a pixel equal to or greater than the threshold within the search range (Yes in step S1402), the pixel whose edge component is closest to the edge component at the position of the RPE end among the pixels equal to or greater than the threshold is updated as a new RPE end. The process returns to step S1401. If there is no pixel equal to or greater than the threshold (No in step S1402), the RPE end is not updated and the process proceeds to step S1403.

<Step S1403>
In step S1403, the optic disc identifying unit 13 identifies the RPE end identified in step S1402 as the final RPE end.

  In the case of processing as a three-dimensional image, an ellipse that is internal to the RPE end specified on the C-scan image plane is obtained in consideration of the disk shape, and a point on the ellipse may be used as the final RPE end. . By doing so, there is an effect of correcting the RPE end shifted due to forgery by the blood vessels extending from the disk area.

  In this way, by tracing using the edge component and position information, the RPE end can be specified with high accuracy and without being mistaken for other retinal layer boundaries even when the luminance is reduced due to imaging conditions or the like. it can. With the above configuration, even if there is an area where IS / OS or RPE cannot be detected due to a false image, the disk area can be specified with higher accuracy.

(Other embodiments)
In the above-described embodiment, the optic nerve head is specified by using the depression in the shape of the boundary surface of the ILM. However, the ILM is an example, and the vitreous cortex may be used as the boundary surface. In short, an area where the interface between the vitreous body and the retina is recessed may be used.

  Further, although the optic nerve head is identified using a region where the IS / OS layer is interrupted, the IS / OS is not limited thereto. A region in which at least one of the nerve fiber layer, GCL, INL, IPL, OPL, or IS / OS layer or at least one of the layer boundaries does not exist may be acquired as a region where the layer is interrupted. Further, if the optic papilla is specified on the condition that a plurality of layers or layer boundaries do not exist among them, the specifying accuracy can be further improved.

  The processing performed in the image processing apparatus 10 may be distributed by a plurality of apparatuses and realized as an image processing system. Further, the processing performed in the circuit corresponding to one block shown in FIG. 1 may be realized by being distributed by a plurality of circuits or functional blocks.

  When the present invention is realized by the cooperation of computer hardware and software, a hardware configuration can be considered as shown in FIG. The image processing apparatus 1500 includes a CPU 1501, a RAM 1502, a ROM 1503, an HDD 1504, and an I / F 1505. In addition, a keyboard 1506 and a mouse 1507 for receiving input from the user to the image processing apparatus are provided. Here, the ROM 1503 or the HDD 1504 stores a program for executing the processing shown in the flowcharts of FIGS. By developing this program in the RAM 1502 and being executed by the CPU 1501, the processing shown in the flowchart is realized.

  Note that the present invention may be realized by distributing the image processing apparatus with a plurality of CPUs. Further, a case where an operating system (OS) running on a computer performs part or all of actual processing and the above-described functions are realized by the processing is also included in the scope of the present invention. Further, a recording medium recording a program or program code which is software also constitutes the present invention.

  The description in this embodiment described above is an example of a suitable image processing apparatus according to the present invention, and the present invention is not limited to this.

DESCRIPTION OF SYMBOLS 1 Optical coherence tomography system 10 Image processing apparatus 20 Optical coherence tomography apparatus 30 Storage part 40 Display part 11 Image acquisition part 12 Retina analysis part 13 Optic nerve head specific part 14 Nipple analysis part

Claims (5)

Image acquisition means for acquiring a tomographic image of the fundus;
Boundary extraction means for extracting the boundary between the retina and the vitreous body of the tomographic image;
Calculating means for calculating an index indicating the depressions viewed degree of the extracted boundaries,
A range extracting means for extracting a range in which at least one of each layer of the nerve fiber layer, GCL, INL, IPL, OPL or IS / OS or its layer boundary does not exist in the tomographic image;
Based on the aforementioned range in which the calculating means is the range extraction means and the index calculated is extracted, the image processing apparatus according to claim Rukoto that having a specifying means for specifying the optic disc. The index is the amount of change in curvature, the gradient or the depth direction of the boundary in the range, the image processing apparatus according to claim 1, at least any one Tsudea Rukoto features. The image processing apparatus according to claim 1, wherein the calculation unit calculates the index based on a coordinate value indicating an optical axis direction of signal light when the tomographic image is captured. Acquiring a tomographic image of the fundus;
Extracting a boundary between the retina and the vitreous body of the tomographic image;
A range extraction step of extracting a range in which at least one of each layer of the nerve fiber layer, GCL, INL, IPL, OPL or IS / OS or its layer boundary does not exist in the tomographic image;
A calculation step of calculating an index indicating the depressions viewed degree of the extracted boundaries,
Image processing method according to claim Rukoto which have a a step of the basis and the index calculated by the calculating step in said range extracted by the range extraction step, to identify the optic disc. The program for functioning a computer as each means of the image processing apparatus of any one of Claims 1 thru | or 3 .

Images