JP4951758B2 - Program for removing anisotropic noise and anisotropic noise removing method - Google Patents

Description

  The present invention relates to a program for removing anisotropic noise that functions as an image filter capable of analyzing an elongated linear structure such as a capillary vessel of the fundus with high accuracy, and an anisotropic noise removing method.

  In recent years, the number of lifestyle-related diseases represented by diabetic patients has been increasing year by year, and the population affected by diabetes in Japan was 7.4 million according to the announcement of the Ministry of Health, Labor and Welfare in 2003. It is expected to approach 10 million people. Along with saving diabetics and other lifestyle-related diseases, preventing them and improving QOL (Quality of Life) has become a major issue.

  There are various lifestyle-related diseases. For example, a vascular disorder that develops as a result of hyperglycemia appears prominently in the fundus where there are many capillaries. There are similar parts in other parts. Recently, as an early diagnosis method for fundus abnormality, visualization of oxygen saturation, which is one of characteristic factor information, has been attempted. This utilizes the fact that the spectral characteristics of blood hemoglobin differ depending on the binding state with oxygen when the fundus is illuminated with light sources of different wavelengths. For example, FIG. 12 shows the oxygen saturation dependence of the fundus reflection spectrum. According to this, high and low oxygen saturation levels are the same at 545 nm, 570 nm, and 584 nm, but the reflectivity is different in the vicinity of 545 nm, for example, 539 nm, 548 nm, and 577 nm, and in this vicinity, oxyhemoglobin (oxygen blood) and deoxyhemoglobin (deoxygenated oxygen) It can be seen that the difference in the degree of absorption of blood) is large, and the oxygen absorption state of blood can be selectively detected. Furthermore, it can be seen that the state of oxygen saturation can be visualized by this characteristic.

  Lifestyle-related diseases such as diabetes include blood glucose level, protein amount, lipid amount and the like in addition to the blood oxygen saturation level which is a measure of the vascular disorder described above. Explaining why these are used as indicators of diabetes, diabetes becomes a chronic hyperglycemic state, and the state in which the blood concentration of glucose (glucose) is abnormally high continues. This condition leads to vascular damage, causing complications and loss of visual acuity and reducing kidney function. At this time, metabolic abnormalities occur in the body, and functional and histological abnormalities are caused by biochemical abnormalities such as changes in protein and fat. For example, blood contains various lipid components such as erythrocyte protein, albumin and plasma protein, and various lipid components such as triglycerides, cholesterol, phospholipids and free fatty acids. Therefore, when the blood of a diabetic patient is examined, the amount of protein and the amount of lipid are changed as compared with the state before the disease progresses. In addition, the growth of important growth factors (such as VEGF) as a causative factor of retinopathy and rapid accumulation in the blood due to increased production of late glycation reaction products (AGE) are the direct targets of fundus examination. AGE receptors (RAGE) contained in cells also cause aging of cells by acting with AGE.

  Therefore, if there is a change in the above-mentioned factors such as blood oxygen saturation, blood glucose level, protein level, lipid level, and cellular aging, diabetes can be diagnosed early and preventive measures can be taken. become. The same applies to other lifestyle-related diseases such as hypertension and hyperlipidemia. These pieces of factor information indicate specific light absorption at each individual wavelength when the blood vessel is irradiated with light. For this reason, it is possible to diagnose the progress of this lifestyle-related disease at an early stage based on this result by irradiating light of individual wavelengths for each lifestyle-related disease of interest and detecting the light absorbance of the irradiated part. become. If this can be used in clinical practice, it is possible not only to simply observe the shape of the fundus blood vessel as in a conventional camera, but also to make a high-quality diagnosis using the obtained factor information. This will greatly contribute to the QOL improvement of the aging society expected in the future.

  As a conventional method for obtaining a visible image of the fundus, a white light source is used as a light source, and reflection and scattered light from the fundus are photographed for each wavelength by a filter to obtain a spectral image. Similarly, it is also studied to obtain a spectrum of each wavelength by irradiating a white light source, mechanically sweeping an optical element, and further performing a fast Fourier transform. Optical coherence tomography (OCT) can capture a cross-section in the depth direction of the fundus over 2 mm to 3 mm, but currently it is limited to visualization of only the shape and structure, detecting the factor information of lifestyle-related diseases, A functional image of blood flowing through blood vessels cannot be obtained.

  As an example of a conventional spectroscopic measurement apparatus, a spectroscopic measurement apparatus using a white coherent pulsed electromagnetic wave as a light source has been proposed (Patent Document 1). The spectroscopic measurement device disclosed in Patent Document 1 uses a white coherent pulsed electromagnetic wave as a light source for sample transmission (or reflection) light, and also obtains a spectral spectrum by Fourier-transforming a time-series signal, and using this as a basis, a time spectrum is obtained. Spectral imaging is performed by performing decomposition and / or spatial decomposition. It is for automatically measuring a static dielectric constant in real time in a manufacturing process of an electronic device having a dielectric material as a constituent element. However, the inspection target of this spectroscopic measurement apparatus is completely different from the patient's blood vessel targeted by the blood vessel information analysis apparatus, and the apparatus is large and unsuitable for medical treatment.

  On the other hand, a retinal function camera that determines blood oxygenation in the retina by illuminating blood with light of different wavelengths has been proposed (Patent Document 2). The retinal function camera of Patent Document 2 has a first light source in the first wavelength band and a second light source in the second wavelength band, and the absorbance of the first light source band by oxygenated blood is absorption in the second wavelength band. The absorbance of the first light source band due to deoxygenated blood is less than that of the second wavelength band, and the light from the first light source and the second light source is selectively combined on a part of the retina of the eye. Means for focusing, imaging means for producing respective images of a portion of the retina illuminated in respective wavelength bands, and processing means for examining the images obtained by the imaging means.

  The first light source band and the second light source band are selected from 480 nm to 1000 nm, and a band having a large difference in absorbency between oxygenated blood and deoxygenated blood is selected. For example, the first light source band is substantially 488 nm, the second light source band is one of 600 nm, 630 nm, 635 nm, and 700 nm, the first light source band is substantially 635 nm, the second light source band is 830 nm, and one of 910 nm, Etc. A functional image can be obtained by comparing blue light of 450 nm to 500 nm with near infrared light of 600 nm to 805 nm.

  However, the retinal function camera of Patent Document 2 needs to scan, and the eyeball moves in response to light during measurement, the pupil shrinks, or the pulsation affects, and noise is mixed into the examination result. Therefore, it is inherently difficult to improve detection accuracy with this camera. And it is not a thing of the new idea which is going to acquire information by image processing fundamentally, and cannot respond in order to detect many factors of lifestyle-related diseases other than oxygen absorption.

  By the way, regarding noise removal, one of the present inventors has already proposed a line concentration degree image filter that can extract only line information from an image at high speed (see Patent Document 3). This line concentration degree image filter provides a plurality of measurement points on the image with a predetermined area for measurement on both wings of the search line passing through each measurement point, and a plurality of areas included in the vicinity area. Measure the direction of the brightness gradient vector of the image at the neighboring points, and calculate the degree of concentration that evaluates the difference between the direction of the brightness gradient vector and the direction of the search line at each of the neighboring points. The line concentration degree for the measurement point is calculated, and when the line concentration degree becomes maximum, it is determined that there is line information along the search line. Further, the direction of the brightness gradient vector and the direction of the search line are determined. In addition to discretizing each, a basic addition value that can be shared at each measurement point based on the degree of concentration is calculated in advance, and when the direction of the luminance gradient vector is measured, a candidate for the basic addition value is determined for each direction of the search line. Choose one from the neighborhood Calculate the linear concentrations of each measurement point by adding the inner is to estimate that the line information along the direction of the search line when 該線 concentration ratio becomes maximum exists.

JP 2003-279212 A JP 2005-500870 A Japanese Patent Laid-Open No. 2005-284597

  As described above, the visualization of oxygen saturation, which is a factor of lifestyle-related diseases, has been proposed in the past, and the visualization takes a considerable amount of time. And since it is based on spectroscopic measurement, the spectral luminance of a light source will become weak and noise will mix in a measurement result. In addition, it is necessary to scan the camera, and the apparatus is large. Furthermore, the effectiveness of using green light near 545 nm to effectively detect oxygen absorption has not been recognized. And it cannot be diverted to factors of lifestyle-related diseases other than oxygen absorption.

  That is, the number of factors that cause each lifestyle-related disease is not limited to one, and there are many factors that should be tested at the same time when diabetes, hypertension, hyperlipidemia, and other lifestyle-related diseases occur together. Exists. As an angiography of the fundus, imaging using a fluorescence contrast method in which a fluorescent dye is administered by injection into a blood vessel has been widely used. However, depending on the person, a shock state can be induced by administration of the fluorescent dye. This method is invasive to the patient.

  Therefore, it is possible to examine various factor information of lifestyle-related diseases such as blood oxygen saturation, blood sugar level, protein, lipid, aging of blood vessels, etc. Compact, practical, and inexpensive non-invasive blood vessel that can test up to multiple factors, has low noise in the test results, and can accurately detect the state of the above factors (blood vessel information) from blood vessels through which blood flows It is desirable to develop an information analysis device. For this purpose, an image filter capable of analyzing an elongated linear structure such as a capillary vessel with high accuracy from image information and an anisotropic noise removing method are indispensable.

  Therefore, the present invention provides a program for removing anisotropic noise that functions as an image filter capable of analyzing a narrow linear structure such as a capillary of the fundus with high accuracy, and a method for removing the anisotropic noise. The purpose is to do.

The program according to the present invention is a line concentration degree measuring means for calculating a line concentration degree with respect to a neighborhood area assuming a search line and assuming a neighborhood area in the vicinity of a point of interest on an image. obtaining a search line to be as vectors centralized normal vector concentrate line acquisition means, a line concentration degree segment information acquisition means for acquiring information of all the line segments on an image as a line segment a part of the predetermined value or more, the line segment Functions as a neighborhood region setting means that sets an anisotropic neighborhood for each pixel inside a line segment region that is based on information, and a noise removal means that performs noise processing in an anisotropic neighborhood region The main feature is to remove the anisotropic noise inside the line segment region.

The anisotropic noise removal method of the present invention assumes a neighborhood area in the vicinity of a point of interest on an image, assumes a search line, calculates a line concentration degree for the neighborhood area, and searches for a maximum line concentration degree. A line is determined as a vector concentration line, and information on all line segments is detected based on the vector concentration line with the line concentration degree equal to or greater than a predetermined value as a line segment. The line is configured based on the detected line segment information. An anisotropic neighborhood region is set inside the segment region, and anisotropic noise removal processing is performed on the inside of the anisotropic neighborhood region to remove the anisotropic noise inside the segment region. Main features.

  According to the program and the anisotropic noise removal method of the present invention, a linear structure can be extracted with high accuracy without being affected by noise such as the contrast of an image with respect to an elongated and three-dimensionally complicated linear structure. Further, by taking advantage of the uniformity in the longitudinal direction, noise specific to the linear structure can be removed in the linear structure.

According to a first aspect of the present invention, there is provided a computer, comprising: a line concentration degree measuring unit configured to calculate a line concentration degree with respect to a neighborhood area assuming a search area and assuming a neighborhood area in the vicinity of a point of interest on the image; Vector concentration line acquisition means for obtaining a search line having the maximum degree as a vector concentration line, line segment information acquisition means for acquiring information on all line segments on the image with a line concentration degree equal to or greater than a predetermined value as a line segment, Neighboring area setting means for setting an anisotropic neighborhood area for each pixel inside a line segment area configured based on line segment information, and noise removing means for performing noise processing in the anisotropic neighborhood area , And a program for removing anisotropic noise inside the line segment region. With this configuration, a linear structure can be extracted with high accuracy from an elongated and three-dimensionally complicated linear structure, and noise specific to the linear structure can be removed from the linear structure.

  According to a second aspect of the present invention, there is provided a program characterized in that, in the first aspect, the neighborhood region is a narrow fixed region provided in the vicinity of the point of interest. With this configuration, a linear structure can be extracted with high accuracy without being affected by noise such as image contrast.

  A third aspect of the present invention is a program characterized in that, in the first or second aspect, the image is blood vessel information. With this configuration, it is possible to easily detect many factors of lifestyle-related diseases by performing image processing on an image in which capillaries and the like are involved, and to perform an inexpensive and compact blood analysis.

According to a fourth aspect of the present invention, in any one of the first to third aspects, the anisotropic neighborhood region is a rectangle or an ellipse that is long in the longitudinal direction of the line segment region. . With this configuration, it is possible to remove noise peculiar to the linear structure in the linear structure by utilizing the uniformity in the longitudinal direction.

In the fifth aspect of the present invention , a neighborhood area is assumed in the vicinity of the point of interest on the image, a search line is assumed, a line concentration degree for the neighborhood area is calculated, and a search line that maximizes the line concentration degree is obtained. A line segment area configured based on the detected line segment information, which is obtained as a vector concentrated line , detects the information of all line segments based on the vector concentrated line with the line concentration degree equal to or greater than the specified value as the line segment An anisotropic neighborhood region is set inside the region , and anisotropic noise removal processing is performed on the inside of the anisotropic neighborhood region to remove the anisotropic noise inside the line segment region. This is a method for removing anisotropic noise. With this configuration, it is possible to extract a linear structure with high accuracy without being affected by noise such as the contrast of an image with respect to an elongated and three-dimensionally complicated linear structure. By taking advantage of the above, noise peculiar to the linear structure can be removed in the linear structure.

(Embodiment 1)
A program for removing anisotropic noise that functions as an image filter and an anisotropic noise removing method according to Embodiment 1 of the present invention will be described. Since the program of the first embodiment is particularly effective when installed in a blood vessel information analysis apparatus, a program as an image filter and an anisotropic noise removal method installed in this blood vessel information analysis apparatus will be described. Therefore, the anisotropic noise removing program and the anisotropic noise removing method of the present invention are not limited to the blood vessel information analysis apparatus, but are an apparatus for analyzing an elongated linear structure with high accuracy. If it exists, it can be used for any device. Note that the blood vessel information analysis apparatus according to the first embodiment is intended for testing blood information of capillaries in the fundus. However, blood is emitted by irradiating the blood vessels with light even in the lips and other examination sites. Any site that can measure the factor information of lifestyle-related diseases can be examined.

  FIG. 1 is a block configuration diagram of a blood vessel information analyzer equipped with an image filter according to Embodiment 1 of the present invention, FIG. 2 is a diagram showing an example of an oscillation spectrum of single-wavelength light to be irradiated, and FIG. FIG. 4 is a photograph of a measured blood vessel information analysis image, FIG. 4 is a photograph of an image obtained when a 545 nm green laser beam is irradiated on the lips, and FIG. 5 shows image information of only the blood vessel portion from the image information of the photograph of FIG. A photograph of the extracted image, FIG. 6 (a) is a conceptual diagram of a fixed area near a fixed width assumed in a region where capillaries exist, and FIGS. 6 (a) and 6 (b) are luminance gradient vectors in the fixed area of fixed width. FIG. 7 is a diagram for explaining an ideal distribution in the vicinity of the fixed fixed width of the luminance gradient vector, and FIG. 8 is a program for removing noise in the first embodiment of the present invention. Block diagram, diagram (A) is an explanatory diagram of an isotropic small observation region on the spatial axis, FIG. 9 (b) is an explanatory diagram of an isotropic large observation region on the spatial axis, and FIG. 9 (c) is an explanatory diagram of the spatial axis. FIG. 10 is a first flowchart of the anisotropic noise removing method according to the first embodiment of the present invention, and FIG. 11 is an anisotropic noise removing method according to the first embodiment of the present invention. It is a 2nd flowchart of this.

  In this blood vessel information analyzing apparatus, one set of a plurality of single wavelengths different from each other on the eyeball or a set of multiple sets of light is used as a video rate, that is, one frame (1/30) of an image that cannot be recognized by a human as one frame. (Second), for example, with a pulse width of several mm seconds to several tens of mm seconds. Measure at several different wavelengths for each factor of lifestyle-related diseases. Repeat multiple sets to test multiple factors. Information common to a plurality of factors can be shared between the factors. The reflected wave reflected from the fundus is imaged by a CCD or the like, and image information is acquired for each wavelength. Based on this image information, a blood function image (blood vessel information image) is generated. Therefore, the conventional method is not to obtain a plurality of pieces of image information by irradiating white light and splitting a reflected wave, but to obtain image information for each light with a single wavelength light. An example of the oscillation spectrum of the irradiated light is shown in FIG. In this specification, single-wavelength light is light having a narrow-band wavelength as shown in FIG.

FIG. 1 shows a block diagram of a blood vessel information analysis apparatus equipped with an image filter in the first embodiment. In FIG. 1, 1 is an eyeball. 2 _ij is a light source for irradiating light of different single wavelengths, and one set is provided when there is one lifestyle-related disease factor to be examined, and a plurality of sets are provided when there are a plurality of factors to be examined. The subscript “i” indicates a factor (i = 1,..., M) when inspecting a plurality of factors, and “j” indicates the wavelength of light (j = 2,..., N) used for each factor to be inspected. ). Accordingly, m × (n−1) light sources 2 _ij are provided as a whole, and light of m × (n−1) wavelengths is irradiated. Light from the light source 2 _ij is emitted toward the eyeball 1.

  Here, the reason why the light sources having two or more wavelengths are provided for one factor is that the oxygen saturation is the same at 545 nm, 570 nm, and 584 nm as shown in FIG. For example, at 539 nm, 548 nm, and 777 nm, the difference in light absorption between oxyhemoglobin and deoxyhemoglobin is large, and the reflectance is different at these locations, so that oxygen saturation can be measured by comparison. That is, at least two points of the other measurement points depending on the oxygen saturation having different reflectances are required, with one point where the high and low oxygen saturations not depending on the oxygen saturation coincided as a reference point for comparison, This is because it is preferable to provide more measurement points in order to further increase the detection accuracy. In this case, the oxygen saturation is calculated by weighting between the measurement points. The same applies to proteins, lipids, and the like, but when the test substance, for example, plasma protein or neutral fat, is specified, the wavelength of light to be irradiated is determined.

Any wavelength range of the light source 2 _ij may be used as long as it can irradiate light capable of measuring a factor of lifestyle-related diseases. However, it is better to avoid the ultraviolet region as much as possible. Lasers such as semiconductor lasers, solid state lasers, optical fiber lasers, and light sources such as LEDs are suitable for easy control. As described above, the wavelength for inspecting the oxygen saturation is green, but it is sufficient to irradiate green light having two wavelengths, three wavelengths, or more. When irradiating with laser light, it is necessary to keep a level of at least ² mW / mm ² or less in light of US standards ANZI, Japanese Industrial Standards JIS, and other standards. Should.

Reference numeral 3 in FIG. 1 denotes an optical coupling unit that sequentially emits pulses emitted from a plurality of light sources 2 _ij as a pulse train. The pulses emitted from the respective light sources 2 _ij are collected here and emitted as a pulse train from the emission end coupled to the optical coupling unit 3. It is important from the viewpoint of eliminating the image shift of the blood vessel information analysis apparatus according to the first embodiment that the emission position of each pulse light and the examination site such as the fundus are fixed and fixedly arranged.

As the optical coupling unit 3, for example, an optical coupler that collects a plurality of optical fibers into one optical fiber may be used. If the output position of each pulsed light can be regarded as a fixed position when the optical fiber is thin and bundled in a cable shape, the light from each light source is simply guided by an individual optical fiber, and the ends are bundled together with the output end. You may do it. Alternatively, the optical coupling unit 3 may be an optical switch that actively switches light emitted from each light source 2 _ij at a predetermined timing and outputs it as a pulse train. In this case, it is controlled by the control unit 5 described later.

  Instead of providing multiple light sources that emit light of different single wavelengths, the light emitted from the light sources is modulated by a light modulator such as an electrical light modulator. You may irradiate from a radiation | emission end as several single wavelength light. Further, a part of the single wavelength light can be emitted from each light source, and the remaining single wavelength light can be emitted from one light source by being modulated by a modulator. With this configuration, the number of light sources can be reduced and a compact configuration can be achieved.

  In FIG. 1, 4 is an imaging unit such as a CCD or CMOS that receives reflected light from the fundus etc., 4a is an imaging control unit, and 4b is an image signal that outputs charges accumulated in the imaging unit 4 by receiving reflected light. A processing unit, 5 is a control unit, and 6 is a storage unit.

When each pixel of the imaging unit 4 receives reflected light by irradiation of the light source 2 _ij , the charge accumulated from the imaging unit 4 is output to the image signal processing unit 4b as an image signal, where imaging noise is removed, After being amplified and A / D converted, it is transferred to the control unit 4 as image information.

  The control unit 5 is one or a plurality of processors that control the system of the entire blood vessel information analysis device in terms of hardware, and executes a function that reads a program stored in the storage unit 6 and executes a predetermined control function Configured as a means.

  The storage unit 6 includes a ROM and a RAM in terms of hardware, and a non-volatile memory that stores image information. The image information obtained by the imaging unit 4 is temporarily stored in the inspection information memory unit 6a of the storage unit 6. For each factor of each living environment disease, reference image information detected by light emission of a predetermined wavelength and measurement image information detected at a measurement wavelength are stored as one set.

  Next, the following function realizing means is mounted on the control unit 5 shown in FIG. 7 is a data calculation means for calculating blood state information of blood vessels based on image information stored in the examination information memory section 6a, and 8 is a blood vessel function image (blood vessel information image) obtained by extracting a blood vessel portion from the image information. This is an image processing means for coloring and performing other image processing. Reference numeral 8a denotes an image filter that removes noise from the image information. In the present invention, image information is acquired at a video rate, and since imaging is completed in a short time, which is tens of times that the pupil contracts upon receiving light irradiation, noise caused by eye movement is generated. The possibility of mixing is low.

  Now, the data calculation means 7 calculates blood state information of blood vessels for each wavelength, that is, factor information of lifestyle-related diseases. In terms of oxidative saturation, there are at least two pieces of image information, that is, 545 nm reference blood vessel image information that does not depend on oxygen saturation, and blood vessel measurement image information detected at 539 nm in the vicinity depending on oxygen saturation. By obtaining the intensity (luminance) and calculating the intensity ratio between the two, the oxygen saturation of blood can be obtained. FIG. 3 is an example of a blood vessel information analysis image obtained by measuring the fundus oxygen distribution. The blood vessel information analysis image of FIG. 3 is not for humans but for monkeys.

  The image processing unit 8 extracts blood vessels and removes noise from the blood vessel image information by using the image filter 8a before the data calculation unit 7 performs the calculation. The image filter 8a is a means capable of performing anisotropic noise removal processing constituted by a plurality of function realizing means, and functions by executing a program for removing the read anisotropic noise on a computer. is there. Details thereof will be described later. Only the blood vessel portion is extracted from the image information acquired at each wavelength, and the noise of the image of the blood vessel portion is corrected so that the background information does not affect it. FIG. 4 is an image obtained when the lips are irradiated with 545 nm green laser light, and FIG. 5 is an image obtained by extracting and correcting image information of only the blood vessel portion from this image. The image as shown in FIG. 5 is created with reference image information and measurement image information, the intensity ratio of the image is obtained at the corresponding position of the blood vessel part, the two-dimensional distribution of the intensity ratio is obtained, and this is color-coded by stage A blood vessel information analysis image as shown in FIG. 3 is obtained.

  In the center of the image shown in FIG. 5, a round and bright gray area is a place where many capillaries of the fundus are present, and white areas in this area are places where oxygen distribution is relatively large. The place where the concentration of oxygen is the highest is shown in black in the white part in the center. On the other hand, the dark part outside the gray round area is a part where the light intensity from each light source 2ij is insufficient and the illuminance is low. However, in the blood vessel information analysis apparatus according to the first embodiment, a sufficient amount of information can be obtained with this central information.

  Next, the data calculation means 7 calculates the intensity ratio of the blood vessel part from the image intensity based on the image information thus obtained, and generates information on the intensity ratio distribution. Based on this, the image processing means 8 generates a blood vessel information analysis image. The blood vessel information analysis image, the original image, and the like can be displayed on the display device 10 described later by the control unit 5. The blood vessel information analysis image data and image information acquired at this time are stored in the examination information memory unit 6a for each factor.

  In addition to measuring the reference image information and the measurement image information one by one, when there is measurement image information taken at a second wavelength, for example, 548 nm, the measurement taken at the second wavelength 548 nm It is preferable to calculate the second intensity ratio obtained from the image information and the reference image information. Further, if there are third and fourth wavelengths, the third and fourth intensity ratios are calculated, and an intensity ratio obtained by weighting them with the first intensity ratio is calculated to obtain an intensity ratio distribution. Based on this, the image processing means 8 generates a blood vessel information analysis image. Accordingly, if the intensity ratio is calculated as a whole by measuring at the second and third wavelengths, the blood vessel information analysis image is more accurate than when measured only at the first wavelength.

  Next, returning to FIG. 1, the display function of the blood vessel information analysis device will be described. Reference numeral 9 is a display control means, and 10 is a display device for displaying numerical data such as a blood vessel information analysis image, an original picture, or oxygen saturation obtained by calculation, as output from the display control means 9. The control unit 5 causes the display control means 9 to output RGB signals such as blood vessel information analysis images and image information in the examination information memory unit 6a to the display device 10. The display device 10 displays a blood vessel information analysis image, an original image, and the like according to the RGB signals.

The light source control function of the blood vessel information analyzer is performed by the control unit 5. In FIG. 1, reference numeral 11 denotes a timing unit such as a timer provided in the control unit 5. At least mm seconds can be counted. The control unit 5 sequentially operates the driving unit (not shown) of the next light source 2 _ij in synchronization with the output of the image signal from the imaging unit 4 to the image signal processing unit 4b. It is also possible to operate the next light source 2 _ij at a timing delayed by a predetermined time. Thereby, the pulse of each light source 2 _ij is made into a pulse train by the optical coupling unit 3, and the reflected light from the fundus is received by the imaging unit 4.

  In this way, the blood vessel information analysis device equipped with the image filter in Embodiment 1 sequentially irradiates one or more sets of light with different wavelengths to the eyeball at a video rate, and makes this light a pulse in a very short time. Thus, the blood state information of the blood vessel is measured based on the image information of the reflected wave by removing noise such as eye movement. Therefore, it does not irradiate white light and divide the reflected wave to acquire and analyze a plurality of pieces of image information as in the past, but obtain image information as it is for each light of different single wavelengths, and perform image processing. It measures blood status information. For this reason, it is possible to provide a practical non-invasive blood vessel information analyzing apparatus with an inexpensive, compact and extremely simple configuration. In addition, the blood vessel information analysis apparatus according to the first embodiment uses the fact that the examination target is a blood vessel, and uses this characteristic to obtain highly accurate information by image processing.

  Therefore, the image filter 8a in the first embodiment will be described in more detail. In the blood vessel information analysis apparatus equipped with the image filter 8a in the first embodiment, a spotlight-like image of the fundus is acquired by irradiating light instantaneously, thereby extracting blood vessel information. At this time, since the image is a spotlight image, it is inevitable that a large contrast difference occurs in the image. In normal image processing, pre-processing is necessary to remove the influence of contrast, the calculation is complicated and time consuming, the processing results are noisy, and it is difficult to extract blood vessel information.

Therefore, in the image filter 8a of the first embodiment, the obtained image information (input image information) is regarded as a line-convergence vector field, and the contrast and line width are determined using a model of a linear region. Blood vessel information (elongated linear structure) is detected without being affected by the above. At this time, the intensity of the first derivative vector of luminance (hereinafter referred to as luminance gradient vector) is ignored, and only the direction distribution of the vector is focused. Focusing only on the direction of the vector, when the blood vessel is an ideal linear convex region, the ridge line has the maximum value of luminance. In this state, the luminance gradient vector has a direction orthogonal to the ridge line, is distributed opposite to both sides of the ridge line, and is line-concentrated together toward the ridge line. By utilizing such properties, blood vessel information (direction and width) is extracted, and unnecessary noise is removed. A ridge line on which such luminance gradient vectors are concentrated is called a vector concentration line. FIG. 6A is a conceptual diagram of a narrow fixed region assumed in a region where a capillary is present, and FIG. 6B is a diagram for explaining an actual distribution of luminance gradient vectors in the narrow fixed region. is there. FIG. 7 is a diagram for explaining an ideal distribution of the luminance gradient vector in the vicinity of the narrow fixed width. 6B and FIG. 7, V is a vector concentration line, r ₁ and r ₂ are regions near a fixed rectangle, φ is the direction of the vector concentration line V (ridge line), and W is a line described later. This is the width determined as a line segment (blood vessel) in the calculation of the degree of concentration.

Therefore, the configuration of the image filter 8a according to the first embodiment, which is a feature of the present invention, will be described with reference to FIG. In the image filter 8a, a program for executing each function described below is stored in the storage unit 6 (storage medium), and a processor (in the first embodiment, a blood vessel information analysis apparatus) of the processor (the present invention) And the respective means cooperate to execute anisotropic noise removal processing. In FIG. 8, reference numeral 81 denotes attention point setting means for assuming the attention point P on the image. Reference numeral 82 denotes a line concentration degree measuring means for obtaining a line concentration degree in the vicinity of the attention point P set by the attention point setting means 81. The line concentration degree measuring means 82 calculates a line concentration degree C _N assuming a search line (assumed vector concentration line) assuming near fixed areas r ₁ and r ₂ in the vicinity of the point of interest P. To do.

Next, 83 is a vector concentration ray acquisition means for obtaining a maximum and becomes the search line (vector concentrate line) in the linear degree of concentration C _N of the line concentration degree measuring unit 82 was calculated. The ridge of the line is obtained by the action of the vector concentrated line acquisition means 83. Reference numeral 84 denotes line segment information acquisition means. Segment information acquisition unit 84 is a line concentration degree C _N determines part of the more noise immunity value as a line segment. Thereby, when all vector concentrated lines V are acquired, the existence, direction, and width of all line segments (elongated linear structures such as blood vessels) on the image can be known.

  Subsequently, reference numeral 85 designates an anisotropic neighborhood region (observation region) that is long in the longitudinal direction as shown in FIGS. 9D and 9E, which will be described later, for performing anisotropic noise removal processing for noise removal. This is a neighborhood region setting means. The neighborhood region setting unit 85 assumes a point of interest P in a line segment whose direction and width are known, and determines an observation region for noise removal as an anisotropic filter for the point of interest P. Reference numeral 86 denotes noise removal means, and reference numeral 87 denotes processing determination means, which performs noise removal processing such as a median filter or a morphological filter that minimizes the noise evaluation value. The process determination unit 87 repeats the noise removal process one after another at all the points of interest P, checks whether or not the noise removal is completed, and ends the noise removal process if completed.

The image filter 8a constituted by each means described above, when extracting an elongated linear structure from the image, the vector concentrated line V (ridge) that is a linear convex region that appears in the image by the vector concentrated line acquisition means 83. Line). Specifically, the hypothetical attention point P is successively moved with respect to the blood vessel image to be processed, and the line concentration degree C is assumed by assuming the narrow-fixed neighboring regions r ₁ and r ₂ at each attention point P. _A search line that maximizes _N is obtained. The line concentration degree _CN is an evaluation value that is evaluated using a function such as a cosine function that simply decreases within a predetermined range in which the vector concentration can be evaluated. In the case of the cosine function, the line concentration C _N has a value of −1 to +1. If the ridge line can be extracted, the line information acquisition unit 84 defines that the line concentration degree _CN is a noise resistance value, for example, a part of 0.5 or more is a line segment. It is determined as a linear structure, and the presence, width, and direction of the line segment can be acquired.

In this way, only a value where the line concentration degree is larger than the noise tolerance value is determined as line information, and if it is lower than that, it is removed without considering it as a line, so that it is possible to leave a line segment and clear the image. . Because hardly affected by the contrast because it calculates linear concentrations of C _N in the direction facing the vector, a line concentration degree C _N by the region near r _1, r ₂ of the narrow fixed is only very narrow small area, noise It is possible to extract only the ridge line sharply without being affected by. As in the first embodiment, the line concentration degree C _N such that extremely short time irradiation with light of a single wavelength, even when having a contrast difference of 10 times or more, detects individually a line crossing in different directions It is possible to detect the direction and spread of a stable line even in a complicated structure where a real image is complicated.

  Subsequently, the neighborhood region setting means 85 assumes a point of interest P in a line segment whose direction and width are known, and performs anisotropic noise removal specific to the elongated linear structure (here, specific to blood vessels) to noise. This is done by removing means 86.

  The anisotropic noise removal characteristic of this linear structure will be described. When measuring a fine linear structure at high speed, the temporal and spatial occurrence probability of noise (disturbance) mixed from equipment generally follows a normal distribution. Therefore, this removal is possible by the smoothing process in the time axis and the space axis.

  However, when the time changes in a very short time, it is not possible to secure a sufficient measurement time necessary for estimating and removing noise on the time axis. In addition, when the measurement target has a very elongated shape such as a capillary, as shown in FIGS. 9A and 9B, isotropic observation with a general spatial axis An appropriate observation area cannot be secured. That is, in the measurement in such a case, if the observation region A as shown in FIG. 9A is used, the region is too small and the noise removal capability is lowered, as shown in FIG. 9B. Even if a large observation area is used, the observation area A is too large, and the deterioration of the image becomes large due to the influence of the background image.

  Therefore, as shown in FIG. 9C, anisotropic noise removal processing for determining the target area according to the shape of the target area is effective. However, in an actual image, there are a variety of contrasts, and the structure becomes complicated due to the influence of crossing of lines and the like, and it is difficult to accurately detect the shape of the target region with a conventional image filter.

  However, the image filter unit 8a according to the first embodiment has information on the direction and width of the line segment by the line segment information acquisition unit 84 as described above. Therefore, based on the direction and width information of the linear structure obtained by calculating the degree of line concentration, the neighborhood region setting unit 85 sets an observation region A having a sufficiently large size necessary for the anisotropic noise removal processing. An anisotropic filter with sufficient noise removal capability is constructed while avoiding deterioration due to background effects.

  In particular, in the case of a blood vessel information analyzer or the like, since the blood flow along the blood vessel is kept uniform, this anisotropic filter can be an image filter that most reflects the characteristics of the blood vessel. . The observation region A is preferably a rectangular region that is long along the blood vessel as shown in FIG. 6 (d) that falls within the width W of the line segment (blood vessel) or an elliptical region as shown in FIG. 9 (e). In the observation region, it is better to remove noise as a Gauratian filter or a median filter. Note that it is sufficient that the width w of the observation region has a relationship of W = (4-6) × w with the width W of the blood vessel. By using this anisotropic filter, blood vessel noise can be sufficiently removed.

The anisotropic noise removal method (noise processing procedure) in the first embodiment described above will be described with reference to the flowcharts of FIGS. The difference between FIG. 10 and FIG. 11 is only the difference in the mounting method of the anisotropic noise removal filter, and both are not substantially changed. First, referring to FIG. 10, a point of interest P is assumed in the input image (step 1), and the search points are assumed assuming near-fixed neighboring regions r ₁ and r ₂ at the point of interest P. linear concentrations of _{C N} is the maximum search line Request (vector intensive line) (step2). The point of interest P is moved one after another, and it is determined whether or not all vector concentrated lines V have been acquired (step 3). If all vector concentrated lines V have not been acquired, the process returns to step 1, and all vector concentrated lines V are obtained. Is acquired, the direction and width of all the line segments (blood vessels) on the image are acquired (step 4).

  Next, an attention point P is assumed in the blood vessel whose direction and width are known, and an observation region A for noise removal with respect to the attention point P is assumed (step 5). Noise removal processing is performed using values in the observation area A (step 6), and data is stored at the position of the target point P in the output image. Thereafter, the noise removal process is repeated assuming that the observation region A is successively applied to all the pixels (attention point P), and it is checked whether or not the noise removal is completed. If not, the process returns to step 5 and repeated. If completed, the noise removal process ends. It should be noted that general noise processing is preferably performed on regions other than blood vessels. In addition, it is difficult to set the observation region A on both sides of the blood vessel width, and there is a possibility that a narrow region where noise cannot be removed may remain, but this is a future issue for further noise removal, such as reducing the width w of the observation region. is there. As for the noise removal process in step 6, there is also a recursive process that performs a search that minimizes the noise evaluation value, such as a median filter or a morphological filter.

  Next, noise processing by the second mounting method according to the raster scan shown in FIG. 11 will be described. In FIG. 11, first, the position of the attention point P is initialized (step 11). Next, the degree of line concentration on the attention point P of the input image is measured, and information on the vector concentration line and the concentration vector field is acquired (step 12).

  Next, the existence, direction, and width of the line segment (blood vessel) are acquired (step 13), and the observation area A that is the vicinity area of the point of interest P is determined (step 14). And the noise of the attention point P is removed using the information of the observation area A of the input image, and stored in the position of the attention point P of the output image (step 15). Further, the attention point P is moved to the next pixel (step 16). Here, it is determined whether or not calculation has been performed for all pixels (step 17). If calculation has not been completed for all pixels, the process returns to step 12, and if completed, the process is terminated.

  Note that the image filter 8a described above detects the direction and the width of the line based on the line concentration degree of the luminance vector of the image information, even if the image filter 8a is an elongated linear structure, and the detected linear region. An anisotropic noise removal process can be performed on the above. Therefore, not only the blood vessel information analysis apparatus but also an elongated linear structure generally enables high-precision analysis. When blood vessels are analyzed, noise inherent to blood vessels can be removed very effectively.

  As described above, according to the program and the anisotropic noise removal method in the first embodiment, a long and three-dimensionally complicated linear structure such as a fundus blood vessel is not affected by noise such as image contrast. Further, the linear structure can be extracted with high accuracy, and further, noise unique to the linear structure can be removed from the linear structure by utilizing the uniformity in the longitudinal direction.

  The present invention can be applied to an image filter that can analyze an elongated linear structure with high accuracy, a blood vessel information analysis device such as a fundus, and the like.

1 is a block configuration diagram of a blood vessel information analysis device equipped with an image filter according to Embodiment 1 of the present invention. The figure which shows an example of the oscillation spectrum of the light of the single wavelength to irradiate Photograph of blood vessel information analysis image measuring fundus oxygen distribution Photo of image obtained when lips are irradiated with 545nm green laser light Photograph of image obtained by extracting image information of only blood vessel part from image information of photograph of FIG. (A) is a conceptual diagram of a narrow fixed region assumed in a region where capillaries exist, and (b) a diagram for explaining the actual distribution of luminance gradient vectors in the narrow fixed region. The figure explaining the ideal distribution in the near fixed area of the brightness gradient vector The block block diagram of the program for noise removal in Embodiment 1 of this invention (A) An explanatory diagram of an isotropic small observation region on the spatial axis, (b) An explanatory diagram of an isotropic large observation region on the spatial axis, (c) An anisotropy observation region on the spatial axis Illustration First flowchart of anisotropic noise elimination method according to Embodiment 1 of the present invention Second flowchart of the anisotropic noise removing method according to Embodiment 1 of the present invention. Explanatory drawing showing oxygen saturation dependence of fundus reflection spectrum

Explanation of symbols

1 eyeball 2 _ij (i = 1,..., M; j = 2,..., N) light source 3 optical coupling unit 4 imaging unit 4a imaging control unit 4b image signal processing unit 5 control unit 6 storage unit 6a inspection information memory Unit 7 Data operation means 8 Image processing means 8a Image filter means 9 Display control means 10 Display device 11 Timing means 81 Attention point setting means 82 Line concentration degree measuring means 83 Vector concentrated line acquisition means 84 Line segment information acquisition means 85 Neighborhood region setting Means 86 Noise removal means 87 Processing determination means

Claims (5)

Computer
A line concentration degree measuring means that assumes a neighborhood area in the vicinity of a point of interest on the image, calculates a line concentration degree for the neighborhood area assuming a search line,
A vector concentrated line acquisition means for obtaining a search line having the maximum degree of line concentration as a vector concentrated line;
Line segment information acquisition means for acquiring information on all line segments on the image, with the line concentration degree being a part equal to or greater than a predetermined value,
Neighborhood region setting means for setting an anisotropic neighborhood region for each pixel inside the line segment region configured based on the line segment information ;
A noise removing means for performing noise processing in a region near the anisotropy ;
And a program for removing anisotropic noise inside the line segment region. The program according to claim 1, wherein the neighboring area is a narrow fixed area provided in the vicinity of the point of interest. The program according to claim 1, wherein the image is blood vessel information. The program according to claim 1, wherein the anisotropic vicinity region is a rectangle or an ellipse that is long in a longitudinal direction of the line segment region. Assuming a neighborhood area in the vicinity of the point of interest on the image, assuming a search line, calculating a line concentration degree for the neighborhood area, obtaining a search line maximizing the line concentration degree as a vector concentration line, Detects information on all line segments based on the vector concentrated line, with the portion where the degree of concentration is a predetermined value or more as a line segment, and anisotropy inside the line segment area configured based on the detected line segment information An anisotropic noise is characterized in that an anisotropic noise is removed by performing an anisotropic noise removal process on the inside of the anisotropic neighborhood area and setting the neighborhood area of the line segment area. Removal method.