JP5241937B2 - Image processing apparatus and image processing method - Google Patents

Description

  The present invention relates to an image processing apparatus and an image processing method for processing a tomographic image of a subject, and a program for causing a computer to execute the image processing method.

  For the purpose of early diagnosis of various diseases occupying the top causes of lifestyle-related diseases and blindness, examinations by photographing the eye as a subject have been widely performed. For example, an tomographic image acquisition device for an eye using an optical coherence tomography (hereinafter referred to as “OCT”) or the like can three-dimensionally observe the state inside the retinal layer of the eye. . For this reason, this tomographic image acquisition apparatus is expected to be useful for more accurately diagnosing diseases.

  If the thickness of each layer of the retina such as the nerve fiber layer in the eye or the thickness of the entire retinal layer can be measured in the tomogram (tomographic image) of the eye obtained using OCT, glaucoma, age-related macular degeneration, macular edema Thus, it is possible to quantitatively diagnose the degree of progression of such diseases and the degree of recovery after treatment. Conventionally, in order to measure the thickness of these layers, a doctor or an engineer manually operates a retina on a two-dimensional tomographic image (B-scan image) obtained by cutting a cross section of interest from a tomographic image that is three-dimensionally imaged. It was necessary to specify the boundary of each layer. In addition, in order to obtain a three-dimensional distribution of layer thickness, the three-dimensional tomographic image is regarded as a set of several two-dimensional tomographic images, and the boundary of the layer of interest is specified on each two-dimensional tomographic image. There was a need to do.

  However, the task of manually specifying the layer boundaries is a burden for doctors and engineers. In addition, it is difficult to manually quantify the layer boundaries because it causes variations due to differences in the workers and work dates and times. For the purpose of reducing the burden on workers and variations in work, a technique has been proposed in which a computer detects the boundary of each layer of the retina from a tomographic image and measures the thickness of each layer (for example, Patent Document 1 below). And Patent Document 2).

JP 2008-73099 A JP 2007-325831 A

Here, in diseases such as age-related macular degeneration, the shape of the retinal pigment epithelium layer is deformed into irregularities according to the disease state. Therefore, quantifying the degree of deformation is effective for understanding the medical condition. At this time, it is required to estimate the position of the boundary where the retinal pigment epithelium layer would have existed at the normal time before illness as a normal structure. Conventionally, this operation is often performed manually by a doctor or engineer, and the burden and variations in the results have been problems. Therefore, it is required to automate the detection of the retinal pigment epithelium layer boundary from the tomographic image and the estimation of its normal structure.
However, since the image feature on the tomographic image cannot be observed, the normal structure cannot be estimated by the techniques of Patent Document 1 and Patent Document 2.

  The present invention has been made in view of such problems, and an object of the present invention is to provide a mechanism for accurately estimating the normal structure of layers constituting a subject.

The image processing apparatus according to the present invention includes a layer detection unit that detects a layer from a tomographic image of the fundus of the eye, and a deep portion that is located deeper than a threshold in the depth direction of the fundus in the detected layer. A feature point detecting means for detecting a plurality of feature points including the detected layer shape, and a curve convex in the depth direction of the fundus based on the plurality of feature points including the deep part. And a quantifying means for quantifying the difference between the detected layer and the curve.

The image processing method of the present invention includes a step of detecting a layer from a tomographic image of the fundus of the eye, and the detected layer includes a deep portion located deeper than a threshold in the depth direction of the fundus. Detecting a plurality of feature points based on the shape of the detected layer ; obtaining a convex curve in the depth direction of the fundus based on the plurality of feature points including the deep portion; and Quantifying the difference between the detected layer and the curve.

  The program of the present invention is for causing a computer to execute each step of the image processing method.

  According to the present invention, it is possible to accurately estimate the normal structure of the layers constituting the subject. This makes it possible to quantify the degree of disease progression in the subject, the degree of recovery after treatment, and the like.

1 is a schematic diagram illustrating an example of a schematic configuration (hardware configuration) of an image processing system according to a first embodiment of the present invention. It is a schematic diagram which shows an example of a function structure of the image processing apparatus which concerns on the 1st Embodiment of this invention. It is a schematic diagram which shows an example of the tomographic image input into the image process part shown in FIG. 3 is a flowchart illustrating an example of a processing procedure of a control method of the image processing apparatus according to the first embodiment of the present invention. It is a schematic diagram for explaining the estimation of the normal structure of the layer included in the tomographic image, showing the first embodiment of the present invention. It is a schematic diagram which shows the 1st Embodiment of this invention and shows an example of the weight function utilized when estimating the normal structure of the layer contained in a tomographic image. FIG. 5 is a schematic diagram for explaining planarization processing of a tomographic image according to the first embodiment of this invention. FIG. 3 is a schematic diagram illustrating a display example of an image processing result obtained by the image processing apparatus illustrated in FIG. 1 according to the first embodiment of this invention. It is a schematic diagram which shows an example of schematic structure (hardware structure) of the image processing system which concerns on the 2nd Embodiment of this invention. It is a schematic diagram which shows an example of a function structure of the image processing apparatus which concerns on the 2nd Embodiment of this invention. It is a flowchart which shows an example of the process sequence of the control method of the image processing apparatus which concerns on the 2nd Embodiment of this invention. It is a schematic diagram which shows the 2nd Embodiment of this invention and shows the example of a display of the boundary of the layer contained in a tomographic image. FIG. 10 is a schematic diagram for illustrating estimation of a normal structure of a layer included in a tomographic image according to a third embodiment of the present invention. It is a schematic diagram which shows an example of the layer structure of a general retina.

  Hereinafter, embodiments (embodiments) for carrying out the present invention will be described with reference to the drawings. The embodiment described below is only an example, and the present invention is not limited to the illustrated configuration. In the following description, the boundary between the retinal pigment epithelium layer and the upper or lower layer is referred to as “retinal pigment epithelium layer boundary” or simply “layer boundary”. In addition, the estimated position of the boundary where the same layer boundary would exist if it was normal before illness is referred to as “normal structure of the retinal pigment epithelium layer boundary” or simply “normal structure”.

(First embodiment)
First, a first embodiment of the present invention will be described.
FIG. 1 is a schematic diagram illustrating an example of a schematic configuration (hardware configuration) of an image processing system 100 according to the first embodiment of the present invention.

  As shown in FIG. 1, the image processing system 100 according to the present embodiment includes an image processing device 110, a tomographic image capturing device 120, and a local area network (LAN) 130. Here, in the following description, the image processing apparatus 110 illustrated in FIG. 1 is referred to as “image processing apparatus 110-1”. That is, the image processing system 100 illustrated in FIG. 1 is configured such that the image processing apparatus 110-1 is connected to the tomographic image capturing apparatus 120 via the LAN 130.

  The image processing apparatus 110-1 according to the present embodiment, for example, a layer such as a retinal pigment epithelium layer boundary that constitutes an eye part from a tomographic image of a subject (eye part in this example) imaged by the tomographic image imaging apparatus 120. Detect boundaries. Then, the image processing apparatus 110-1 estimates the normal structure of the retinal pigment epithelium layer and the like based on the feature that the shape of the layer is deformed into irregularities due to a disease such as age-related macular degeneration, and quantifies the disease. It is a thing to do. In this embodiment, an example in which the eye part is applied as an example of the subject is shown. However, the present invention is not limited to this, and can be applied as long as it is an imaging target as a tomographic image. is there.

  As shown in FIG. 1, the image processing apparatus 110-1 includes a central processing unit (CPU) 111, a main memory 112, a magnetic disk 113, a display memory 114, a monitor 115, a mouse 116, a keyboard 117, and a common bus 118. It is comprised. A control program 1131 is stored inside the magnetic disk 113.

  The CPU 111 mainly controls the operation of each component of the image processing apparatus 110-1 and controls the operation of the image processing apparatus 110-1 in an integrated manner.

  The main memory 112 stores, for example, a control program 1131 loaded from the magnetic disk 113 when the CPU 111 executes processing, or serves as a work area when the control program 1131 is executed by the CPU 111.

  The magnetic disk 113 stores (stores) an operating system (OS), device drivers for peripheral devices, a control program 1131, and the like. In addition, the magnetic disk 113 is obtained by performing various types of information and data necessary for the CPU 111 to perform processing using the control program 1131 as well as processing by the CPU 111 using the control program 1131 as necessary. Various information and data are stored. Here, the CPU 111 comprehensively controls various operations performed by the image processing apparatus 110-1 by executing a control program 1131 stored in the magnetic disk 113.

  The display memory 114 temporarily stores display data to be displayed on the monitor 115.

  The monitor 115 is composed of, for example, a CRT monitor or a liquid crystal monitor, and displays an image based on display data in the display memory 114 under the control of the CPU 111.

  A mouse 116 and a keyboard 117 are used for a pointing input and a character input by the user, respectively.

  The common bus 118 connects the internal components of the image processing apparatus 110-1 so that they can communicate with each other, and can communicate the internal components of the image processing apparatus 110-1 with the LAN 130 and the tomographic image capturing apparatus 120. Connect to.

  The tomographic image capturing apparatus 120 is an apparatus that captures a tomographic image of, for example, an eye part, and includes, for example, a time domain OCT or a Fourier domain OCT. The tomographic image capturing apparatus 120 three-dimensionally captures a tomographic image of an eye part or the like of a subject (patient) (not shown) according to an operation by a user (engineer or doctor). Then, the tomographic imaging device 120 outputs the obtained tomographic image to the image processing device 110-1.

  The LAN 130 connects the image processing apparatus 110-1 and the tomographic imaging apparatus 120 so that they can communicate with each other as necessary. The LAN 130 is configured by, for example, Ethernet (registered trademark). Note that the image processing apparatus 110-1 and the tomographic imaging apparatus 120 may be connected directly via an interface such as USB or IEEE 1394, for example, without using the LAN 130.

Next, the functional configuration of the image processing apparatus 110-1 shown in FIG. 1 will be described.
FIG. 2 is a schematic diagram illustrating an example of a functional configuration of the image processing apparatus 110-1 according to the first embodiment of the present invention.

  As shown in FIG. 2, the image processing apparatus 110-1 is configured to have each functional configuration of an image acquisition unit 210, an image processing unit 220, a storage unit 230, and a display unit 240.

  In this embodiment, for example, the CPU 111 shown in FIG. 1 executes the control program 1131 stored in the magnetic disk 113, whereby the image acquisition unit 210 and the image processing unit 220 shown in FIG. 2 are configured. The Further, for example, the CPU 111 and the control program 1131 shown in FIG. 1 and the main memory 112, the magnetic disk 113, or the display memory 114 constitute the storage unit 230 shown in FIG. Further, for example, the display unit 240 includes the CPU 111 and the control program 1131 shown in FIG. 1 and the monitor 115 shown in FIG.

  The image acquisition unit 210 requests the tomographic image capturing apparatus 120 to transmit a tomographic image of the eye that is a subject, and receives and acquires the tomographic image of the eye transmitted from the tomographic image capturing apparatus 120. Do. Then, the image acquisition unit 210 sends the acquired tomographic image to the image processing unit 220 and the storage unit 230.

  The image processing unit 220 performs processing of the tomographic image sent from the image acquisition unit 210, and includes a layer detection unit 221, a normal structure estimation unit 222, and a quantification unit 223 as shown in FIG. Configured. Here, the image processing performed by the image processing unit 220 will be described with reference to FIG.

  FIG. 3 is a schematic diagram illustrating an example of a tomographic image input to the image processing unit 220 illustrated in FIG. Specifically, FIG. 3 shows an example of the layer structure of the retina in the eye. In the following description, the depth direction of the fundus is defined as the positive direction of the z coordinate, and the xy plane is defined in the direction orthogonal to the z coordinate.

  The layer detection unit 221 detects the layer boundary (B1 in FIG. 3) and the retinal pigment epithelial layer boundary (B21 in FIG. 3) constituting the eye part from the input tomographic image. Details of specific processing performed by the layer detection unit 221 will be described later.

  Based on the retinal pigment epithelium layer boundary (B21 in FIG. 3) detected by the layer detection unit 221 and the feature deformed by the lesion of the retinal pigment epithelium layer, the normal structure estimation unit 222 has a normal structure (FIG. 3). B22) is estimated. In FIG. 3, the layer boundaries (B1 and B21) detected by the layer detection unit 221 are indicated by solid lines, and the layer boundaries (B21) estimated by the normal structure estimation unit 222 are indicated by dotted lines. Details of specific processing performed by the normal structure estimation unit 222 will be described later.

  The quantification unit 223 performs a process of calculating the layer thickness T1, the area, and the volume of the layer L1 from the detection result by the layer detection unit 221. Further, the quantification unit 223 is based on the retinal pigment epithelium layer boundary (B21 in FIG. 3) detected by the layer detection unit 221 and the normal structure (B22 in FIG. 3) estimated by the normal structure estimation unit 222. The state of the retinal pigment epithelium layer (specifically, the disturbance of the retinal pigment epithelium layer) is quantified. Details of specific processing performed by the quantification unit 223 will be described later.

Next, the storage unit 230 in FIG. 2 will be described.
As illustrated in FIG. 2, the storage unit 230 includes an image storage unit 231 and a data storage unit 232.

  The image storage unit 231 stores a tomographic image acquired by the image acquisition unit 210 (and a tomographic image processed by the image processing unit 220 as necessary).

  The data storage unit 232 stores function and parameter data used by the image processing unit 220. Specifically, the data storage unit 232 is used when the normal structure estimation unit 222 estimates the normal structure of the retinal pigment epithelium layer, and when the layer detection unit 221 obtains the layer boundary from the tomographic image. Various parameter data are stored.

  Further, the data storage unit 232 associates various tomographic images stored in the image storage unit 231 with information on various processing results processed by the image processing unit 220, and stores this as patient data. In this example, since the storage unit 230 includes, for example, the magnetic disk 113 and the like, various tomographic images and various processing result information processed by the image processing unit 220 are associated with the magnetic disk 113 and the like. And memorized.

  Specifically, the storage unit 230 includes tomographic image data acquired by the image acquisition unit 210, layer boundary data detected by the layer detection unit 221, normal structure data estimated by the normal structure estimation unit 222, and quantification by the quantification unit 223. Associate numerical data and save it. Various data stored in the storage unit 230 (data storage unit 232) may be stored in an external server (not shown) connected by the LAN 130. In this case, various types of data stored in the storage unit 230 are stored. Is transmitted to the external server.

Next, the display unit 240 will be described.
The display unit 240 displays the tomographic image obtained by the image acquisition unit 210, the layer boundaries obtained by the layer detection unit 221 and the normal structure estimation unit 222, and various quantification results quantified by the quantification unit 223. To do.

Next, a specific processing procedure executed by the image processing apparatus 110-1 according to the present embodiment will be described with reference to FIG.
FIG. 4 is a flowchart illustrating an example of a processing procedure of the control method of the image processing apparatus 110-1 according to the first embodiment of the present invention.

<Step S101>
First, in step S <b> 101, the image acquisition unit 210 requests the tomographic imaging apparatus 120 to transmit a tomographic image of the eye that is the subject, and obtains the tomographic image of the eye that is transmitted from the tomographic imaging apparatus 120. get. Then, the image acquisition unit 210 inputs the tomographic image acquired from the tomographic imaging apparatus 120 to the image processing unit 220 and the storage unit 230 (image storage unit 231).

<Step S102>
Subsequently, in step S102, the layer detection unit 221 performs a process of detecting a boundary of a predetermined layer constituting the eye part that is the subject from the tomographic image acquired in step S101.

Hereinafter, a specific processing method of layer boundary detection by the layer detection unit 221 will be described.
In this example, the input 3D tomographic image is considered as a set of 2D tomographic images (B-scan images), and the following 2D image processing is executed on each 2D tomographic image.

  First, the layer detection unit 221 performs a smoothing filter process on the focused two-dimensional tomographic image to remove noise components. Then, the layer detection unit 221 performs edge detection filter processing, detects an edge component from the tomographic image, and extracts an edge corresponding to a layer boundary in each layer. For example, the layer detection unit 221 searches for an edge in the depth direction of the fundus from the vitreous body side, and sets the first peak position as the boundary between the vitreous body and the retinal layer. Further, for example, the layer detection unit 221 searches for an edge in the depth direction of the fundus and sets the last peak position as the retinal pigment epithelium layer boundary. Through the above processing, the layer boundary can be detected.

  In addition, the layer boundary detection method may be detected by applying a dynamic contour method such as Snakes or level set method. Here, in the case of the level set method, a level set function that is one dimension higher than the dimension of the region to be detected is defined, and the boundary of the layer to be detected is regarded as the zero contour line. Then, the contour is controlled by updating the level set function, and thereby the boundary of the layer can be detected.

  In addition, as a layer boundary detection method, a layer boundary may be detected using graph theory such as GraphCut. In this case, a node corresponding to each pixel of the image and a terminal called sink and source are set, and an edge (n-link) connecting the nodes and an edge (t-link) connecting the terminals are set. Then, the layer boundaries can be detected by giving weights to these edges and obtaining the minimum cut for the created graph.

  The boundary detection method using the dynamic contour method or GraphCut may be performed three-dimensionally for a three-dimensional tomographic image, or the input three-dimensional tomographic image is considered as a set of two-dimensional tomographic images, The two-dimensional tomographic image may be applied two-dimensionally. Further, in order to detect the layer boundary, not only the image feature amount but also a priori knowledge such as a probability atlas or a statistical atlas may be used for detection. The method for detecting the layer boundary is not limited to these methods, and any method may be used as long as the layer boundary can be detected from the tomographic image of the eye.

<Step S103>
Subsequently, in step S103, the normal structure estimation unit 222 is based on the boundary of the predetermined layer (in this example, the retinal pigment epithelium layer boundary) detected in step S102 and the feature deformed by the lesion of the layer. The process of estimating the normal structure is performed.

Hereinafter, a specific processing method for estimating the normal structure of the retinal pigment epithelium layer by the normal structure estimation unit 222 will be described with reference to FIG.
FIG. 5 is a schematic diagram for explaining the estimation of the normal structure of the layer included in the tomographic image according to the first embodiment of the present invention.

The tomographic image shown in FIG. 5 shows the layer boundary B22 estimated as the layer L1, the layer boundary B1, and the layer boundary B21, as in FIG. Further, reference numerals d1 and d2 in FIG. 5 denote coordinates (x ₁ , z ₁ ) and (x ₂ , z ₂ ) of the layer boundary B21 and estimated coordinates (x ₁ , z ₁ ′) of the layer boundary B22, respectively. ), (X ₂ , z ₂ ′). 5, the coordinates x _1, layer boundary B21 was detected, located deep in the retinal pigment epithelium layer than the layer boundary B22 estimated, the coordinates x _2, the layer boundary B21 detected, the layer boundary estimated B22 Rather than the retinal pigment epithelium layer.

  The normal structure estimation by the normal structure estimation unit 222 is performed so that the layer boundary B21, which is the detection result of the layer detection unit 221, does not go below the normal structure B22 except for the detection outlier. This is based on the knowledge that the deformation of the retinal pigment epithelium layer due to age-related macular degeneration is caused by the neovascularization that occurs in the lower part of the layer, and the position of the deformed layer boundary is located above normal .

  Here, as a specific method for estimating the normal structure, the input three-dimensional tomographic image is considered as a set of two-dimensional tomographic images (B-scan images), and the normal structure is estimated for each two-dimensional tomographic image. The case where it performs is demonstrated. The technique used here is to estimate a normal structure by applying a quadratic function to a coordinate point group representing a layer boundary detected in a focused two-dimensional tomographic image. In this case, the normal structure is approximated by a function representing a quadratic curve in each two-dimensional tomographic image.

  The main point of this method is not only to give a weight according to the distance of the difference between the point group and the function, but also to assign a weight to be used according to the sign of the difference when fitting the function to the detected coordinate point group. It is to change. For example, the normal structure estimation unit 222 selects a weight function to be used based on the following equation (1).

Here, in the equation (1), z _i is the z coordinate of the layer boundary detected at the coordinate x _i , and z _i ′ is the z coordinate of the estimated normal structure at the coordinate x _i . ε _i is the difference between z _i and z _i ′. a, b, and c are parameters of the function to be estimated. ρ1 and ρ2 are weight functions, and ρ1 and ρ2 are positive definite even functions having a unique minimum value at x = 0. As shown in equation (1), a weight function is selected according to the sign of ε _i .

In the case of the codes d1 and d2 in FIG. 5, even if the magnitude of the difference is the same, the signs of the differences are different (ε ₁ is positive and ε ₂ is negative), so the weighting functions used are different. In the case of FIG. 5, the weight function used for the symbol d1 is ρ1, and the weight function used for the symbol d2 is ρ2. In addition, since it is desirable to estimate the normal structure so that the relationship of ε _i ≦ 0 is established, the weight function is set so that the weight when z is positioned deeper than the retina is greater than z ′. . In other words, the weight functions ρ1 and ρ2 are defined so that ρ1 (ε _i ) ≧ ρ2 (−ε _i ) at the difference ε _i . The weight function can be selected as long as it satisfies the above-described conditions (that is, conditions based on the characteristics deformed by the lesion of the layer constituting the eye (in this example, the retinal pigment epithelium layer)). It may be used.

  Next, an example of an evaluation formula for obtaining an approximate function is shown in the following formula (2).

In equation (2), ρ (ε _i ) is a weighting function selected according to the sign of ε _i (ie, ρ1 or ρ2).
FIG. 6 is a schematic diagram illustrating an example of a weight function used when estimating a normal structure of a layer included in a tomographic image according to the first embodiment of this invention.

In FIG. 6A, FIG. 6B, and FIG. 6C, the difference ε _i is taken on the horizontal axis, this is x, and the vertical axis is the weighting function ρ (x). The weighting function used when estimating the normal structure of the layer included in the tomographic image is not all shown in FIG. 6, and any function may be set. In addition, in order to confirm how much weight is actually given by ρ (x), the derivative of ρ (x) may be taken. In this example, the normal structure estimation unit 222 estimates a function so that the evaluation value M is minimized in the above equation (2).

  In this way, by selecting an appropriate weight function according to the sign of the error, as shown in FIG. 3, the layer boundary B21 detected by the layer detection unit 221 is the normal structure B22 estimated by the normal structure estimation unit 222. It can suppress that it protrudes below. As a result, it is possible to accurately estimate the normal structure.

  Here, the input 3D tomographic image is considered as a set of 2D tomographic images (B-scan images), and the method for estimating the normal structure for each 2D tomographic image has been described. However, the normal structure estimation method is not limited to this method, and for example, processing may be performed on a three-dimensional tomographic image. In this case, the ellipsoid may be fitted to the three-dimensional coordinate point group of the layer boundary detected in step S102 using the same weight function selection criteria as described above. The shape approximating the normal structure is not limited to a quadratic function, and may be estimated using a function of another order.

(Another method for estimating normal structure)
Here, a normal structure estimation method different from the above-described method will be described with reference to FIG. FIG. 14A shows an example of the layer structure of the retina in a normal eye. Moreover, FIG.14 (b) is a figure which shows an example of the layer structure of the retina in age-related macular degeneration. The retinal pigment epithelium layer 1400a in the normal eye shown in FIG. 14 (a) has a structure having a smooth curve. On the other hand, the retinal pigment epithelium layer 1400b in age-related macular degeneration shown in FIG. 14 (b) has a shape in which a part of the region undulates vertically. In addition, although the tomographic image of the eye part by OCT is obtained three-dimensionally, in order to simplify description in FIG. 14, the cross section is represented two-dimensionally. Further, since the retinal pigment epithelium layer is a layer having a small thickness, it is represented by a single line in FIG.

  At this time, as a method for diagnosing age-related macular degeneration using OCT, based on the image shown in FIG. 14B, an actual retinal pigment epithelium layer boundary (1401 in FIG. 14C) that can be observed on a tomographic image. ) And its normal structure (1402 in FIG. 14C) on the tomographic image. Then, the area and volume of the difference between the actual retinal pigment epithelium layer boundary (1401) that can be observed on the tomographic image and its normal structure (1402) on the tomographic image (the hatched portion in FIG. 14C). Quantify the state of the disease by

  For example, an elliptic curve fitted to minimize the square error with respect to the retinal pigment epithelium layer boundary detected from a two-dimensional tomographic image (B-scan image) can be used as the normal structure. Note that the normal structure estimated by fitting the elliptic curve by the least square method may be different from the actual normal structure. Therefore, even if the area based on the difference from the detected layer boundary is obtained, the reliability may not be sufficient for use as a scale for quantifying the medical condition. Accordingly, it is preferable to use the above-described normal structure estimation method (a method of changing to an appropriate weighting function according to the sign of the difference between the point group and the function).

<Step S104>
Subsequently, in step S104, the quantification unit 223 performs various quantification processes.
First, the quantification unit 223 performs a process of calculating the layer thickness T1 of the layer L1, the area and the volume of the layer shown in FIG. 3 based on the layer boundary detected by the layer detection unit 221 in step S102. Here, the layer thickness T1 can be calculated by obtaining the absolute value of the difference between the z coordinates of the layer boundary B1 and the layer boundary B21 at each coordinate point on the xy plane. The area of the layer L1 can be calculated by adding the layer thickness at each coordinate point in the x-axis direction for each y coordinate. The volume of the layer L1 can be calculated by adding the obtained area in the y-axis direction.

  Further, the quantification unit 223 is detected in step S102 based on the retinal pigment epithelium layer boundary B21 detected by the layer detection unit 221 in step S102 and the normal structure B22 estimated by the normal structure estimation unit 222 in step S103. Quantify the disturbance of the retinal pigment epithelium layer.

  Specifically, the difference T2 (FIG. 3) between the structure of B21 and the structure of B22 at each coordinate point on the xy plane is calculated as a difference between the z coordinates. Then, by adding the obtained difference T2 in the x-axis direction, an area making a difference is obtained for each y coordinate, and further, by adding the obtained area in the y-axis direction, a volume making a difference is calculated. Note that the feature amount representing the disturbance of the retinal pigment epithelium layer may be any as long as it quantifies the difference between B21 and B22. For example, the maximum value, the median value, the variance, the average, the standard deviation, the number of points equal to or greater than the threshold, the ratio thereof, and the like may be obtained and used. Furthermore, an average density value, a density dispersion value, a contrast, and the like may be obtained as the density characteristics of the region formed by B21 and B22.

Further, as a method of quantification by the quantification unit 223, the boundary of the layer may be converted into a plane at the time of quantification, and the quantification may be performed using this as a reference position. An example is shown in FIG.
FIG. 7 is a schematic diagram for explaining planarization processing of a tomographic image according to the first embodiment of this invention. Specifically, FIG. 7 converts the layer boundary in FIG. 3 into the deepest boundary B22 of the retinal layer (for example, the retinal pigment epithelium layer boundary) into a plane, and converts other layers in accordance with the deformation. This is an example.

  Further, when there is a layer thickness, area, volume, shape feature, concentration feature, or the like quantified in the past, the quantification unit 223 performs comparison with past quantification data. Below, the example of the comparison method with the past quantification data is demonstrated.

  In this case, first, the quantification unit 223 aligns the past and current tomographic images. At this time, a known method such as rigid affine deformation or non-linear deformation FFD (Free form deformation) can be used for alignment of tomographic images. Then, if the correspondence between the tomographic images taken in the past and the current tomographic image can be obtained, the comparison of the layer thickness, area, etc. in any cross-section, and the volume of any region, etc. Can do.

  Here, an example is shown in which various data are compared after the past and current tomographic images are aligned. For example, the tomographic image capturing apparatus 120 has a tracking function and the like. If the same position can be photographed, it is not necessary to perform alignment. Furthermore, the comparison between the slice numbers of the B-scan images or the comparison in 3D may be simply performed without performing the alignment.

  The quantification unit 223 only needs to obtain at least one of the numerical data described above, and the data to be quantified can be arbitrarily set. If none can be obtained, in the next step S105, it may be displayed on the display unit 240 that it cannot be quantified.

<Step S105>
Subsequently, in step S105, the display unit 240 displays the tomographic image, the result of detecting the layer-to-layer boundary of the tomographic image, the result of estimating the normal structure, and the result of quantifying the layer of the tomographic image. Process. In this example, the display unit 240 includes, for example, the monitor 115, and thus various data are displayed on the monitor 115 in the process of step S105. Here, when there is past data, it may be displayed in comparison with past data.

  FIG. 8 is a schematic diagram illustrating a display example of an image processing result obtained by the image processing apparatus 110-1 illustrated in FIG. 1 according to the first embodiment of this invention. In the example shown in FIG. 8, a tomographic image 801 relating to a layer-to-layer boundary detected by the image processing unit 220 or an estimated normal structure is displayed on the upper part of the monitor 115, and an analysis of the tomographic image 801 is performed on the lower part of the monitor 115. As a result, a graph 802 of retinal layer thickness is displayed.

  As the tomographic image display, the layer detected by the image processing unit 220 and the estimated normal structure may be superimposed on the tomographic image acquired by the image acquisition unit 210. Further, as a tomographic image display, a boundary of a specific layer may be converted into a plane, and a tomographic image (FIG. 7) based on the boundary may be generated and displayed. The range of the layer to be deformed may be one B-scan image or a three-dimensional tomographic image.

<Step S106>
Subsequently, in step S106, the storage unit 230 (such as the data storage unit 232) associates the various types of data acquired in the above steps S101 to S104 and saves them as patient data of a patient on the magnetic disk 113. Specifically, the tomographic image data obtained in step S101, the layer and layer boundary data obtained in step S102, the normal structure data of the layer estimated in step S103, and the layer quantification obtained in step S104. The result of the digitizing process is associated and saved. The data stored on the magnetic disk 113 may be at least one of these data.

  Here, when the estimation result of the normal structure is stored, the weight function ρ (Equation (1)) used at the time of estimation and the function parameters (for example, a, b, and c in Equation (1)) are also associated at the same time. It may be stored and estimated using the same function or parameter during follow-up observation.

  Further, these data may be stored in an external server (not shown). In this case, the storage unit 230 transmits these data to the external server.

  In the present embodiment, the layers detected by the layer detection unit 221 are not limited to the inner boundary membrane and the retinal pigment epithelium layer, and each layer or membrane such as a nerve fiber layer, a photoreceptor layer, or an outer boundary membrane May be detected. Furthermore, the boundary of the layer for estimating the normal structure is not limited to the retinal pigment epithelium layer, and any layer and layer boundary may be estimated as long as the shape changes into irregularities due to a disease. .

  According to the first embodiment described above, it is possible to accurately estimate the normal structure of a layer (in this embodiment, the retinal pigment epithelium layer) constituting the eye part in the tomographic image of the eye part that is the subject. . This makes it possible to accurately quantify the degree of disease progression in the eye and the degree of recovery after treatment.

(Modification 1 of the first embodiment)
The image processing apparatus 110-1 in the first embodiment is connected directly or via a LAN 130 to an image storage server (not shown) that holds tomographic images of the eye that have been captured and stored. The configuration may be such that a tomographic image is input from the image storage server. In this case, in step S101 of FIG. 4, the image acquisition unit 210 requests the image storage server (not shown) to transmit the tomographic image, acquires the tomographic image transmitted from the image storage server, and stores the storage unit 230. To enter. Thereafter, as described above, layer boundary detection (S102) and quantification processing (S104) are performed.

(Modification 2 of the first embodiment)
The image processing apparatus 110-1 in the first embodiment may have a configuration in which one or a plurality of two-dimensional tomographic images are input and the above-described analysis processing (image processing) is performed thereon. Moreover, the structure which selects the cross section which should be noticed from the input three-dimensional tomographic image, and processes with respect to the selected two-dimensional tomographic image may be sufficient. In this case, the selection of the two-dimensional tomographic image may be performed by the user using the mouse 116 or the keyboard 117 or using an interface (not shown), or a predetermined specific part of the fundus (for example, the center of the macular region). The cross-section including) may be processed.

(Second Embodiment)
Next, a second embodiment of the present invention will be described.
FIG. 9 is a schematic diagram illustrating an example of a schematic configuration (hardware configuration) of an image processing system 200 according to the second embodiment of the present invention.

  As shown in FIG. 9, the image processing system 200 according to the present embodiment includes an image processing device 110, a tomographic image capturing device 120, a local area network (LAN) 130, and a wide area image capturing device 140. Has been. Here, in the following description, the image processing apparatus 110 illustrated in FIG. 9 is referred to as “image processing apparatus 110-2”. That is, the image processing system 200 shown in FIG. 9 is configured such that the image processing apparatus 110-2 is connected not only to the tomographic image capturing apparatus 120 but also to the wide area image capturing apparatus 140 via the LAN. Note that the image processing apparatus 110-2 and the wide area image capturing apparatus 140 may be directly connected via an interface such as USB or IEEE1394.

  Similar to the image processing apparatus 110-1 according to the first embodiment, the image processing apparatus 110-2 according to the present embodiment detects a layer boundary such as a retinal pigment epithelium layer boundary from a tomographic image of the eye. And the image processing apparatus 110-2 is based on the characteristic that the shape of a layer deform | transforms into an unevenness | corrugation by diseases, such as age-related macular degeneration, for example like the image processing apparatus 110-1 which concerns on 1st Embodiment. The normal structure of the retinal pigment epithelium layer is estimated and the disease is quantified. However, the image processing apparatus 110-2 according to the present embodiment determines whether the shape of the retinal pigment epithelium layer detected from the tomographic image of the eye is normal or abnormal for each local region, and the range determined as normal (normal range) ) To estimate its normal structure. Accordingly, in the image processing apparatus 110-2 shown in FIG. 9, a control program 1132 different from the first embodiment is stored inside the magnetic disk 113.

  The wide area image capturing apparatus 140 is an apparatus that captures a wide area image such as an eye part, and includes, for example, a fundus camera. The wide-area image capturing device 30 captures a wide-area image such as an eye part of a subject (patient) (not shown) according to an operation by a user (engineer or doctor). Then, the wide area imaging device 140 outputs the obtained wide area image to the image processing apparatus 110-2.

Next, a functional configuration of the image processing apparatus 110-2 illustrated in FIG. 9 will be described.
FIG. 10 is a schematic diagram illustrating an example of a functional configuration of an image processing apparatus 110-2 according to the second embodiment of the present invention.

  As illustrated in FIG. 10, the image processing device 110-2 is configured to have each functional configuration of an image acquisition unit 210, an image processing unit 220, a storage unit 230, and a display unit 240. The functional configuration of the image processing apparatus 110-2 according to the second embodiment is basically the same as the functional configuration of the image processing apparatus 110-1 according to the first embodiment shown in FIG. However, in the image processing apparatus 110-2 according to the second embodiment, the image acquisition unit 210 includes a wide area image acquisition unit 211 and a tomographic image acquisition unit 212, and the normal structure estimation unit 222 includes layers. A determination unit 2221 is configured to determine whether the shape is normal or abnormal.

Next, a specific processing procedure executed by the image processing apparatus 110-2 of the present embodiment will be described with reference to FIG.
FIG. 11 is a flowchart illustrating an example of a processing procedure of a control method of the image processing apparatus 110-2 according to the second embodiment of the present invention.

<Step S201>
First, in step S <b> 201, the tomographic image acquisition unit 212 requests the tomographic imaging apparatus 120 to transmit a tomographic image of the eye that is the subject, and transmits the tomographic image of the eye that is transmitted from the tomographic imaging apparatus 120. To get. The tomographic image acquisition unit 212 inputs the tomographic image acquired from the tomographic image capturing apparatus 120 to the image processing unit 220 and the storage unit 230 (image storage unit 231). Similarly, the wide area image acquisition unit 211 requests the wide area image capturing apparatus 140 to transmit a wide area image of the eye that is the subject, and acquires the wide area image of the eye that is transmitted from the wide area image capturing apparatus 140. . Then, the wide area image acquisition unit 211 inputs the acquired wide area image to the image processing unit 220 and the storage unit 230 (image storage unit 231).

<Step S202>
Subsequently, in step S202, the layer detection unit 221 performs a process of detecting a boundary of a predetermined layer constituting the eye part that is a subject from the tomographic image acquired in step S201. Here, the layer detection method is the same as step S102 of FIG. 4 in the first embodiment. In addition, in order to detect a layer boundary, a lesion or blood vessel may be extracted not only from a tomographic image but also from a wide-area image and detected using the information.

<Step S203>
Subsequently, in step S203, the determination unit 2221 of the normal structure estimation unit 222 determines the normal range and abnormalities for the layer boundary (in this example, the retinal pigment epithelial layer boundary (B21)) detected by the layer detection unit 221 in step S202. Determine the range and detect. Then, the determination unit 2221 classifies the layer boundary detected by the layer detection unit 221 in step S202 into a normal range and an abnormal range.

  Specifically, first, as a condition that the layer has a normal structure, for example, the curvature of the layer boundary is defined to be smooth. In this case, in the detection of the abnormal range by the determination unit 2221, the local curvature of the detected layer boundary is calculated and the boundary shape of the layer is evaluated, so that the portion where the curvature does not change smoothly is the abnormal range. And Alternatively, in the detection of the abnormal range by the determination unit 2221, vitiligo, bleeding, and drusen are detected from the wide area image of the eye, and the retina in the z direction of the tomographic image corresponding to the (x, y) region of the wide area image where the lesion exists. The layer structure is assumed to be abnormal. Then, the determination unit 2221 classifies the positions of the normal range and the abnormal range in the tomographic image.

<Step S204>
Subsequently, in step S204, the normal structure estimation unit 222 determines the normal structure of the layer based on the layer boundary included in the range determined to be normal in step S203 and the feature deformed by the lesion of the layer. The process which estimates is performed.

  Here, as a method for estimating the normal structure in step S204, the normal structure is approximated using a polynomial as in step S103 of FIG. 4 in the first embodiment. At this time, the normal structure estimation unit 222 calculates polynomial parameters using only the coordinate group determined to be normal by the determination unit 2221 and approximates the normal structure. Alternatively, the normal structure estimation unit 222 may estimate the normal structure by performing interpolation processing such as spline interpolation on the coordinate group determined to be normal by the determination unit 2221 without being approximated by a polynomial.

<Step S205>
Subsequently, in step S205, the quantification unit 223 performs various types of quantification processing as in step S104 of FIG. 4 in the first embodiment.

<Step S206>
Subsequently, in step S206, the display unit 240 performs basically the same display process as step S105 of FIG. 4 in the first embodiment. At this time, in the present embodiment, any information may be selected and displayed in displaying the estimated normal structure and the detected layer boundary. This will be described with reference to FIG.

FIG. 12 is a schematic diagram illustrating a display example of a layer boundary included in a tomographic image according to the second embodiment of this invention.
In FIG. 12, the layer boundary detected by the layer detection unit 221 is indicated by a solid line, and the normal structure estimated by the normal structure estimation unit 222 is indicated by a dotted line.

  Specifically, FIG. 12A is an example in which the detected layer boundary and the estimated normal structure are displayed simultaneously. FIG. 12B is an example in which only the estimated normal structure is displayed. FIG. 12C is an example in which the detected layer boundary is displayed for the normal range and the detected normal layer and the estimated normal structure are simultaneously displayed for the abnormal range within the same layer boundary. FIG. 12D shows an example in which the detected layer boundary is displayed for the normal range and only the estimated normal structure is displayed for the abnormal range within the same layer boundary.

  As a display method in step S206, the detection results and the estimation results shown in FIGS. 12A to 12D may be superimposed on the tomographic image, or the detection results and the estimation results may be displayed as a tomographic image. You may make it display separately. In this way, it is possible to select a display method of information related to the layer boundary so that arbitrary information can be displayed.

<Step S207>
Subsequently, in step S207, the storage unit 230 performs basically the same storage process as step S106 of FIG. 4 in the first embodiment. At this time, the acquisition range of the tomographic image on the wide area image and the wide area image may be stored and presented together for reasons such as confirmation of the imaging region.

  According to the second embodiment described above, in the tomographic image of the eye that is the subject, whether the structure of the retinal pigment epithelial layer is normal or abnormal is determined for each local region. Therefore, the normal structure of the layer that has changed due to a disease can be accurately estimated. This makes it possible to accurately quantify the degree of disease progression and the degree of recovery after treatment using the results.

(Third embodiment)
Next, a third embodiment of the present invention will be described.
The schematic configuration of the image processing system according to the third embodiment is the same as that of the image processing system 100 according to the first embodiment shown in FIG. The functional configuration of the image processing apparatus 110 according to the third embodiment is basically the same as the functional configuration of the image processing apparatus 110-1 according to the first embodiment shown in FIG. However, since the processing of the normal structure estimation unit 222 is different, the image processing apparatus 110 according to the third embodiment is referred to as “image processing apparatus 110-3”.

  In addition, the image processing apparatus 110-3 according to the present embodiment detects layer boundaries such as the retinal pigment epithelium layer boundary from the tomographic image of the eye, as with the image processing apparatus 110-1 according to the first embodiment. To do. And the image processing apparatus 110-3 is based on the characteristic that the shape of a layer deform | transforms into an unevenness | corrugation by diseases, such as age-related macular degeneration, similarly to the image processing apparatus 110-1 which concerns on 1st Embodiment. The normal structure such as the retinal pigment epithelium is estimated to quantify the disease. However, in the image processing apparatus 110-3 according to the present embodiment, the shape of the boundary of the retinal pigment epithelium layer detected from the tomographic image of the eye part is a function related to one polynomial representing a normal structure, and the upper number n of lesions. The normal structure is estimated on the assumption that the distribution is a mixture of Gaussian functions. This is because the unevenness of the layer generated by the bulge from the bottom is approximated by a Gaussian function, and other global shapes are estimated by a polynomial.

  In the present embodiment, when the normal structure estimation unit 222 in FIG. 2 applies a function to the detected coordinate point group, the shape of the layer boundary is a mixture of a quadratic function representing a normal structure and n Gaussian functions. Assume a distribution. And the normal structure estimation part 222 in this embodiment estimates the normal structure of the layer in which the shape deform | transformed into the unevenness | corrugation by estimating these mixed distributions. An example of this is shown in FIG.

  FIG. 13 is a schematic diagram for explaining the estimation of the normal structure of a layer included in a tomographic image according to the third embodiment of the present invention. Specifically, 1300 in FIG. 13A shows an example of the shape of the boundary of the retinal pigment epithelium layer in age-related macular degeneration. Moreover, 1301 and 1302 of FIG.13 (b) have each shown the example of the polynomial calculated | required from the retinal pigment epithelium layer boundary 1300 of Fig.13 (a), and a Gaussian function.

  A mixed distribution of a quadratic function and a Gaussian function can be obtained using an EM algorithm. At this time, the mixed distribution is a probability model in which a polynomial and a plurality of Gauss functions are combined, and the probability of the polynomial and the Gauss function is added with a weight.

  In this embodiment, the normal structure estimation unit 222 calculates the quadratic function parameters (a, b, c), Gaussian function parameters (mean, variance-covariance matrix), and the mixture ratio of the quadratic function and the Gaussian function. Estimate using EM algorithm. At this time, n Gaussian functions may be set by setting n according to the number of negative peaks in the sign of the difference between the polynomial and the detected layer boundary, or between 1 and N. n may be increased in order and a number n that maximizes the evaluation value may be set.

  In the EM algorithm, parameters are temporarily set as initial values, the validity of parameters is determined in Step, and parameters are selected so that the validity is improved in Mstep. In this example, these Steps and Steps are repeatedly calculated, and the parameters at the time of convergence are selected. Here, the validity is determined by the current likelihood, and the parameter is selected by selecting a parameter that is larger than the current likelihood.

  In the present embodiment, n upward convex Gaussian functions are mixed into one smooth quadratic function. This is because the unevenness of the layer generated by the bulge from below is approximated by the former Gaussian function, and the other global shape is estimated by a quadratic function. Here, a method of estimating a normal structure for each two-dimensional tomographic image is shown, assuming that the input three-dimensional tomographic image is a set of two-dimensional tomographic images (B-scan images). However, the normal structure estimation method is not limited to this, and processing may be performed on a three-dimensional tomographic image. The shape approximating the normal structure is not limited to a quadratic function, and may be estimated using a function of another order. Furthermore, the boundary of the layer for estimating the normal structure is not limited to the retinal pigment epithelium layer, and any layer and layer boundary may be estimated as long as the shape changes into irregularities due to a disease.

(Other embodiments)
2 and 10 constituting the image processing apparatus 110 according to each embodiment of the present invention described above, and the steps shown in FIGS. 4 and 11 are controlled by the CPU 111 stored in the magnetic disk 113. This can be realized by executing a program. This control program and a computer-readable recording medium (magnetic disk 113 in the embodiment) storing the control program are included in the present invention.

  In addition, the present invention can be implemented as, for example, a system, apparatus, method, program, storage medium, or the like. Specifically, the present invention may be applied to a system including a plurality of devices. You may apply to the apparatus which consists of one apparatus.

  In the present invention, a software program (in the embodiment, a program corresponding to the flowcharts shown in FIGS. 4 and 11) for realizing the functions of the above-described embodiments is directly or remotely supplied to the system or apparatus. including. The present invention also includes a case where the system or the computer of the apparatus is achieved by reading and executing the supplied program code.

  Accordingly, since the functions of the present invention are implemented by computer, the program code installed in the computer also implements the present invention. In other words, the present invention includes a computer program itself for realizing the functional processing of the present invention.

  In that case, as long as it has the function of a program, it may be in the form of object code, a program executed by an interpreter, script data supplied to the OS, and the like.

  Examples of the recording medium for supplying the program include a flexible disk, hard disk, optical disk, magneto-optical disk, MO, CD-ROM, CD-R, and CD-RW. In addition, there are magnetic tape, nonvolatile memory card, ROM, DVD (DVD-ROM, DVD-R), and the like.

  As another program supply method, a browser on a client computer is used to connect to an Internet home page. The computer program itself of the present invention or a compressed file including an automatic installation function can be downloaded from the homepage by downloading it to a recording medium such as a hard disk.

  It can also be realized by dividing the program code constituting the program of the present invention into a plurality of files and downloading each file from a different homepage. That is, a WWW server that allows a plurality of users to download a program file for realizing the functional processing of the present invention on a computer is also included in the present invention.

  In addition, the program of the present invention is encrypted, stored in a storage medium such as a CD-ROM, distributed to users, and key information for decryption is downloaded from a homepage via the Internet to users who have cleared predetermined conditions. Let me. It is also possible to execute the encrypted program by using the downloaded key information and install the program on a computer.

  Further, the functions of the above-described embodiments are realized by the computer executing the read program. In addition, based on the instructions of the program, an OS or the like running on the computer performs part or all of the actual processing, and the functions of the above-described embodiments can also be realized by the processing.

  Further, the program read from the recording medium is written in a memory provided in a function expansion board inserted into the computer or a function expansion unit connected to the computer. Thereafter, the CPU of the function expansion board or function expansion unit performs part or all of the actual processing based on the instructions of the program, and the functions of the above-described embodiments are realized by the processing.

  Note that each of the above-described embodiments is merely a specific example for carrying out the present invention, and the technical scope of the present invention should not be construed as being limited thereto. . That is, the present invention can be implemented in various forms without departing from the technical idea or the main features thereof.

110-1 Image Processing Device 210 Image Acquisition Unit 220 Image Processing Unit 221 Layer Detection Unit 222 Normal Structure Estimation Unit 223 Quantification Unit 230 Storage Unit 231 Image Storage Unit 232 Data Storage Unit 240 Display Unit

Claims (25)

Layer detecting means for detecting a layer from a tomographic image of the fundus of the eye,
In the detected layer, a plurality of feature points including a deep portion located deeper than the threshold with respect to the depth direction of the fundus, a featured portion detection means for detecting on the basis of the shape of the detected layer ,
Based on the plurality of feature locations including the deep location, computing means for obtaining a convex curve in the depth direction of the fundus;
Quantification means for quantifying the difference between the detected layer and the curve;
An image processing apparatus comprising:   The image processing apparatus according to claim 1, further comprising means for obtaining a straight line passing through the deep portion. The image processing apparatus according to claim 1, wherein the calculation unit obtains the curve so that the deep portion of the detected layer is not deeper than the curve.   The image processing apparatus according to claim 1, wherein the calculation unit obtains a quadratic curve based on the plurality of characteristic locations as the curve. The image processing apparatus according to claim 4, wherein the calculation unit obtains the quadratic curve that minimizes a difference from the detected layer by a least square method. The calculating means assigns a function to each of the plurality of feature locations, changes the weight of the function according to a sign of a difference between the plurality of feature locations and the function, and detects the curve . The image processing apparatus according to claim 1, wherein the image processing apparatus estimates the normal structure of the layer. The feature location detecting means detects a coordinate point group representing a boundary of the detected layer as the plurality of feature locations,
The image according to any one of claims 1 to 6, wherein the calculation means applies a quadratic function to the coordinate point group to estimate the curve as a normal structure of the detected layer. Processing equipment.   8. The image processing apparatus according to claim 1, further comprising a display control unit that causes a display unit to display a display form indicating a result of quantification by the quantification unit.   The image processing apparatus according to claim 8, wherein the display control unit displays the result quantified by the quantification unit on the display unit in comparison with past quantification data. 10. The image according to claim 8, wherein the display control unit causes the display unit to display a difference area indicating a difference between the detected layer and the curve on the tomographic image. Processing equipment.   The image processing apparatus according to claim 8, wherein the display control unit causes the display unit to display a display form indicating the curve so as to be superimposed on the tomographic image. The result of quantification by the quantification means is at least one of a distance, an area, and a volume between the detected layer and the curve. The image processing apparatus described.   The image processing apparatus according to claim 1, wherein the calculation unit obtains the curve for each of a plurality of two-dimensional tomographic images constituting the three-dimensional tomographic image of the fundus. .   The image processing apparatus according to claim 1, wherein the layer detection unit detects a layer in the tomographic image corresponding to the retinal pigment epithelium layer of the fundus.   The image processing apparatus according to claim 1, wherein the layer detection unit detects an edge component from the tomographic image as a boundary of the layer. Detecting a layer from a tomographic image of the fundus of the eye,
In the detected layer, detecting a plurality of characteristic points including a deep part located deeper than a threshold with respect to the depth direction of the fundus based on the shape of the detected layer ;
Obtaining a convex curve in the depth direction of the fundus based on the plurality of feature points including the deep part; and
Quantifying the difference between the detected layer and the curve;
An image processing method comprising:   The image processing method according to claim 16, further comprising a step of obtaining a straight line passing through the deep portion. The image processing method according to claim 16 or 17, wherein, in the step of obtaining the curve, the curve is obtained so that the deep portion of the detected layer does not become deeper than the curve.   19. The image processing method according to claim 16, wherein in the step of obtaining the curve, a quadratic curve based on the plurality of characteristic locations is obtained as the curve. 20. The image processing method according to claim 19, wherein, in the step of obtaining the curve, the quadratic curve that minimizes the difference from the detected layer is obtained by a least square method. In the step of obtaining the curve, a function is assigned to each of the plurality of feature locations, the weight of the function is changed according to a sign of a difference between the plurality of feature locations and the function, and the curve is detected. 21. The image processing method according to claim 16, wherein the normal structure of the formed layer is estimated. In the step of detecting the plurality of feature locations, a coordinate point group representing a boundary of the detected layer is detected as the plurality of feature locations,
The step of obtaining the curve includes applying a quadratic function to the coordinate point group to estimate the curve as a normal structure of the detected layer. Image processing method.   23. The image processing method according to claim 16, further comprising a step of causing a display unit to display a display form indicating a result of quantification in the quantification step. 24. The step of detecting a layer from the tomographic image of the fundus oculi comprises detecting a layer in the tomographic image corresponding to the retinal pigment epithelium layer of the fundus oculi. Image processing method. 25. A program for causing a computer to execute each step of the image processing method according to any one of claims 16 to 24 .

Images