Device for testing endothelial function of arterial vessel

Abstract

Provided is a device for testing the endothelial function of an arterial vessel, with which it is possible to easily position a muscle tissue between an ultrasound probe and an arterial vessel even by an operator who is not skilled in the judgement of a muscle tissue in a short-axis image.　A muscle tissue region image generation unit 94 generates, using a trained model LM that has learned a muscle tissue region image GKN appearing in the short-axis image and that has been stored, a muscle tissue region image GKN that indicates the muscle tissue KN in an actual short-axis image from image information that has appeared in the actual short-axis image. A display control unit 92 displays the muscle tissue region image GKN superimposed on short-axis images of a first short-axis image display region G1 and a second short-axis image display region G2. Thus, it is possible to easily set the placement of an ultrasound probe 24 on an upper arm 16 and the position of twisting of the upper arm 16 so that the muscle tissue KN is positioned between the ultrasound probe 24 and the arterial vessel 20.

Description

Arterial endothelial function testing device

[0001] The present invention relates to an endothelial function testing device for arterial blood vessels that displays areas of muscle tissue within an ultrasound image showing a cross section of an arterial blood vessel, so that muscle tissue can be easily positioned between the ultrasound probe and the arterial blood vessel when testing the endothelial function of the arterial blood vessel based on changes in arterial diameter that appear in the ultrasound image showing the cross section of the arterial blood vessel.

[0002] There is known an endothelial function testing device for arterial blood vessels that contacts an ultrasonic probe with the epidermis of a living body, generates a cross-sectional image showing a cross section of an arterial blood vessel located below the epidermis from an ultrasonic reflection signal from below the epidermis, measures a change in the vascular diameter of the arterial blood vessel from the cross-sectional image, and tests the endothelial function of the arterial blood vessel based on the amount of change in the vascular diameter. For example, an endothelial function testing device for arterial blood vessels is described in Patent Document 1.

[0003] In the arterial endothelial function testing device described in Patent Document 1, ultrasonic cross-sectional images are synthesized based on reflected waves detected by an ultrasonic array to generate a transverse cross-sectional image (short-axis image) or a longitudinal cross-sectional image (long-axis image) of the arterial blood vessel, which is an ultrasonic cross-sectional image of the arterial blood vessel under the skin, and the diameter of the arterial blood vessel or the endothelial diameter (intima diameter) is calculated from the image. Furthermore, to evaluate vascular endothelial function, the change rate (%) of the vascular endothelial diameter (intima diameter) [= 100 × (dmax - d) / d] (where d is the vascular endothelial diameter at rest, and dmax is the maximum vascular endothelial diameter after release from occlusion) is calculated, which indicates FMD (flow-mediated vasodilation) after ischemia-reactive hyperemia.

[0004] JP 2011-177571 A

[0005] In the above-described endothelial function testing device, if muscle tissue, which attenuates ultrasound less than other tissues such as fat, is present in the propagation path between the ultrasound probe and the arterial blood vessel, the blood vessel image will be displayed relatively clearly in the cross-sectional image of the arterial blood vessel, which is an ultrasound image, thereby ensuring accurate blood vessel diameter measurement.Therefore, a skilled operator reads the area of ​​muscle tissue from the cross-sectional image of the arterial blood vessel and determines the placement of the ultrasound probe on the arm and the twist position of the arm so that the area of ​​muscle tissue is located in the ultrasound propagation path between the epidermis on which the ultrasound probe is placed and the blood vessel.

[0006] However, cross-sectional images of arterial blood vessels generated based on weak and noisy ultrasonic reflection signals are unclear, and therefore considerable skill is required to read muscle tissue in such ultrasound images. Therefore, it can be difficult for an unskilled operator to accurately grasp the muscle tissue region, obtain a clear blood vessel image, and ensure the accuracy of blood vessel diameter measurements.

[0007] The present invention has been made in light of the above circumstances, and its purpose is to provide an endothelial function testing device for arterial blood vessels that allows even an operator who is not skilled in determining muscle tissue in cross-sectional images of arterial blood vessels to easily locate muscle tissue between the epidermis on which the ultrasound probe is placed and the arterial blood vessel.

[0008] The present inventors have conducted various studies in light of the above circumstances and have found that by inputting image information about a cross-sectional image of an arterial blood vessel, which is an ultrasound image, into a neural network, and having an experienced operator repeatedly teach the neural network areas of muscle tissue in the cross-sectional image of the arterial blood vessel to create a machine learning model that learns, and then using this machine learning model to clearly display the areas of muscle tissue in the cross-sectional image of the arterial blood vessel, even an unskilled operator can accurately grasp the areas of muscle tissue and easily position the areas of muscle tissue between the skin on which the ultrasound probe is placed and the arterial blood vessel. The present invention was made based on this finding.

[0009] That is, the gist of the present invention is (a) an endothelial function testing device for arterial blood vessels that brings an ultrasonic sensor into contact with the epidermis of a living body, generates a cross-sectional image showing a cross-section of an arterial blood vessel located under the epidermis from an ultrasonic reflection signal from under the epidermis, and measures changes in the vascular diameter of the arterial blood vessels that appear in the cross-sectional image, and that comprises: (b) a short-axis image display device that displays a short-axis image of the arterial blood vessel; (c) a muscle tissue region image generation unit that generates a muscle tissue region image showing the region of muscle tissue located in the actual short-axis image from image information that appears in the actual short-axis image using a learned model that has been previously learned and stored about the region of muscle tissue that appears in the short-axis image; and (d) a display control unit that displays the muscle tissue region image on the image display device, superimposed on the short-axis image.

[0010] In the arterial endothelial function testing device configured in this manner, the muscle region display control unit uses a trained model that has previously learned and stored the muscle tissue region appearing in the short-axis image to generate a muscle tissue region image showing the muscle tissue region in the actual short-axis image from image information appearing in the short-axis image, and the display control unit displays the muscle tissue region image superimposed on the short-axis image. This allows even an operator who is not skilled in determining muscle tissue in short-axis images to accurately and easily grasp the muscle tissue region in the short-axis image, and can easily position the ultrasound probe on the arm or twist the arm so that the muscle tissue region is located in the propagation path between the ultrasound probe and the blood vessel. This allows for the generation of clear blood vessel images and improved accuracy in blood vessel diameter measurement.

[0011] Preferably, the trained model is a machine learning model that has been repeatedly subjected to supervised learning using training data in which a predetermined short-axis image is used as input data and muscle tissue regions identified by a skilled operator in the short-axis image are used as output data, thereby enabling clear identification of muscle tissue regions in the short-axis image.

[0012] Preferably, the machine learning model is a multi-layered neural network that has undergone deep learning, in which weight coefficients between nodes in adjacent layers are sequentially updated when the input data is input to nodes in an input layer and output data indicating the muscle tissue area identified by an experienced operator in the short-axis image is repeatedly taught to nodes in an output layer, thereby enabling the muscle tissue area to be more clearly identified in the short-axis image.

[0013] Preferably, in the multilayer neural network, brightness information for each pixel constituting the predetermined short-axis image is input as the input data to an input layer, and output data indicating pixels that indicate the muscle tissue area is output from an output layer, thereby more clearly identifying the muscle tissue area in the short-axis image.

[0014] Preferably, in the output data output from the output layer of the multilayer neural network, image recognition is performed on a pixel-by-pixel basis constituting the short-axis image, the texture of the region to which the pixel belongs is characterized from information on each pixel and its neighboring pixels, and a meaningful region segmentation is estimated for each pixel based on the texture, indicating whether the pixel belongs to a muscle tissue region or other tissue. This reduces the number of nodes constituting the output layer of the multilayer neural network, making the configuration of the multilayer neural network more compact.

[0015] 1 is a perspective view showing the overall configuration of an endothelial function testing device for arterial blood vessels according to one embodiment of the present invention; FIG. 2 is a diagram explaining XYZ orthogonal coordinate axes used in this embodiment for positioning an ultrasound probe in the endothelial function testing device of FIG. 1; FIG. 3 is an enlarged view schematically showing the layer structure of a blood vessel that is the measurement target of the endothelial function testing device of FIG. 1; FIG. 4 is a diagram illustrating an ultrasound image of an arterial blood vessel displayed on an image display device during the positioning operation of an ultrasound probe by the endothelial function testing device of FIG. 1; FIG. 5 is a time chart explaining the change in vascular lumen diameter after release from blood flow obstruction in FMD evaluation of an arterial blood vessel by the endothelial function testing device of FIG. 1; FIG. 6 is a photograph showing an image in which a muscle tissue region image is superimposed on a short-axis image; FIG. 7 is a functional block diagram explaining the configuration of a muscle tissue region image generating unit provided in the endothelial function testing device of FIG. 1; FIG. 8 is a flowchart explaining the main parts of the control operation of the electronic control device of FIG. 1, showing a muscle tissue region image generation routine; and FIG. 9 is a flowchart explaining the main parts of the control operation of the electronic control device of FIG. 1, showing an FMD measurement routine.

[0016] In the present invention, the endothelial function testing device is preferably adapted to measure the brachial artery, which is an artery beneath the epidermis of the upper arm of a living body, but is also applicable to detecting vascular endothelial function of other arteries, such as the forearm and phacoemulsifier arteries, which can be measured from the epidermal surface.

[0017] The ultrasonic probe provided in the endothelial function testing device in the present invention is preferably an H-shaped probe having, on a single plane, two parallel rows of first and second short-axis ultrasonic array probes and a long-axis ultrasonic array probe connecting the first and second short-axis ultrasonic array probes at their longitudinal centers. However, the present invention can also be similarly applied and effective to endothelial function testing devices equipped with other probes, such as those having a single short-axis ultrasonic array probe or those having two parallel rows of first and second short-axis ultrasonic array probes without a long-axis ultrasonic array probe.

[0018] In the present invention, the trained model is preferably a machine learning model based on a machine learning algorithm. The machine learning algorithm may be a nearest neighbor method, a naive Bayes method, a support vector machine, or the like. Furthermore, deep learning is used, which uses a neural network to generate features and connection weighting coefficients for learning. Furthermore, training data refers to training data, and is composed of pairs of input data and output data.

[0019] DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS Preferred embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0020] FIG. 1 is a perspective view illustrating the overall configuration of an endothelial function testing device 22 that uses a probe unit 12 supported by a sensor support 10 to non-invasively test the endothelial function of an arterial blood vessel 20 located from above the skin 18 (strictly speaking, the epidermis) to below the skin 18 in the upper arm 16 of a living body 14.

[0021] The probe unit 12 functions as a sensor for examining biological information related to the arterial blood vessel 20, i.e., the endothelial function of the arterial blood vessel 20, and includes an H-shaped ultrasonic probe 24 having a pair of parallel first and second short-axis ultrasonic array probes 24a and 24b, and a long-axis ultrasonic array probe 24c connecting the first and second short-axis ultrasonic array probes 24a and 24b at their longitudinal center, on a single plane, i.e., a flat probe surface 25. The probe unit 12 also includes a multi-axis drive mechanism (positioning mechanism) 26 that positions the ultrasonic probe 24 in the X, Y, and Z directions and positions the rotation angles around the X and Z axes, and a drive device 27 including an XYZ motor that drives the multi-axis drive mechanism in the X, Y, and Z directions, respectively. The first short-axis ultrasonic array probe 24a, the second short-axis ultrasonic array probe 24b, and the long-axis ultrasonic array probe 24c are, for example, as shown in FIG. 2 described later, each of which is made up of a plurality of ultrasonic transducers a made of piezoelectric ceramics. １ ~a ｎ are linearly arranged to form a longitudinal shape.

[0022] 2 is a diagram illustrating the XYZ orthogonal coordinate axes used in this embodiment for positioning the ultrasonic probe 24. With respect to the XYZ orthogonal coordinate axes shown in FIG. 2, the ultrasonic probe 24 is translated in the X-axis direction by the multi-axis drive mechanism 26, for example, and rotated around the X-axis and Z-axis.

[0023] 3 is an enlarged view showing a schematic diagram of the wall structure of an arterial blood vessel 20 in the upper arm, which is the measurement target of the endothelial function testing device 22. The arterial blood vessel 20 is made up of an intima L １ , medial L ２ , and adventitia L ３ Since ultrasonic waves are generally reflected from areas with different acoustic impedances, an image of the arterial blood vessel 20 using ultrasonic waves shows the blood in the lumen of the blood vessel and the intima L. １ and the interface of the tunica media L ２ and adventitia L ３ The interface between the blood vessels and the vascular system is displayed in white, the blood vessel lumen is displayed in black, and the tissue is displayed in black and white mottled areas.

[0024] 1, the endothelial function testing device 22 includes an electronic control device 28 configured from a so-called microcomputer, an image display device 30, an ultrasound drive control circuit 32, and a three-axis drive motor control circuit 34. In measuring the state of blood vessels using the endothelial function testing device 22, when a drive signal is supplied from the ultrasound drive control circuit 32 by the electronic control device 28, beam-shaped ultrasound waves are sequentially emitted from the first short-axis ultrasound array probe 24a, the second short-axis ultrasound array probe 24b, and the long-axis ultrasound array probe 24c of the ultrasound probe 24 in the probe unit 12 by well-known beamforming drive. Then, the first short-axis ultrasonic array probe 24a, the second short-axis ultrasonic array probe 24b, and the long-axis ultrasonic array probe 24c detect reflected ultrasonic signals, and the detected reflected ultrasonic signals are processed in the electronic control device 28 to generate a first short-axis image, a second short-axis image, and a long-axis image, which are ultrasonic images of the arterial blood vessels 20 under the skin 18, and these images are displayed in the first short-axis image display area G1, the second short-axis image display area G2, and the long-axis image display area G3 of the image display device 30.

[0025] In FIG. 1 , the electronic control device 28 functionally comprises an ultrasound drive control unit 80, a detection processing unit 82, an ultrasound image generation unit 84, a three-axis drive motor control unit 86, a cuff pressure control unit 88, a blood vessel condition evaluation unit 90, a display control unit 92, and a muscle tissue region image generation unit 94.

[0026] 4A and 4B show the positional relationship between the ultrasound probe 24 positioned at a predetermined measurement position and the arterial blood vessel 20 when an ultrasound image of the arterial blood vessel 20 is generated in measurement of the blood vessel condition using the endothelial function testing device 22, and also illustrate an example of an ultrasound image of the blood vessel displayed on the image display device 30 in that positional relationship. In the image display device 30, for example, as shown in FIG. 4A, the first short-axis image display area G1, the second short-axis image display area G2, and the long-axis image display area G3 have a common vertical axis indicating the depth dimension from the skin 18. Note that "ImA" and "ImB" in FIG. 4A each indicate a cross section of the arterial blood vessel 20.

[0027] The endothelial function testing device 22 evaluates flow-mediated dilation (FMD), which measures the diameter, intimal thickness, plaque, blood flow velocity, etc. of the arterial blood vessel 20, based on the reflected signals of ultrasound emitted from the ultrasound probe 24 to the arterial blood vessel 20. In this FMD evaluation, the image display device 30 displays the rate of change in the diameter of the intima of the arterial blood vessel 20, i.e., the rate of expansion R of the lumen diameter, in a time series. When evaluating FMD and generating ultrasound images of the arterial blood vessel 20, the ultrasound probe 24 is automatically positioned at a predetermined measurement position P relative to the arterial blood vessel 20, which is the measurement target, by driving the multi-axis drive mechanism 26, which receives a drive signal from the triaxial drive motor control circuit 34, via a triaxial drive motor control unit 86 provided in the electronic control device 28. The predetermined measurement position P is a position where the first short-axis ultrasonic array probe 24a and the second short-axis ultrasonic array probe 24b are perpendicular to the arterial blood vessel 20 and the long-axis ultrasonic array probe 24c is parallel to the arterial blood vessel 20. Explaining with reference to Fig. 4, the predetermined measurement position P is a position where "a = b, c = d, e = f" in Fig. 4. That is, it is a measurement position where the distance from the first short-axis ultrasonic array probe 24a to the center of the arterial blood vessel 20 and the distance from the second short-axis ultrasonic array probe 24b to the center of the arterial blood vessel 20 are equal to each other and where the images of the arterial blood vessel 20 are positioned at the center in the width direction in both the first short-axis image display region G1 and the second short-axis image display region G2.

[0028] When measuring the state of blood vessels using the endothelial function testing device 22, the sensor support 10 holds the probe unit 12 in a fixed position so as to lightly contact the arterial blood vessel 20 located above the skin 18 of the upper arm 16 of the living body 14 and directly below the skin 18 without deforming it.

[0029] As shown in FIG. 1, the sensor support 10 includes, for example, a base 36 fixed to a pedestal (not shown), a unit fixture 38 to which the probe unit 12 is fixed so that it can be pressed toward the upper arm 16 by a pressing actuator 37, a first arm 40 whose base end is connected to the base 36 via a flexible fixture, and a flexible arm 44 having a second arm 42 whose tip end is connected to the unit fixture 38 via a flexible fixture and which is connected to the tip of the first arm 40 so as to be rotatable around one axis.

[0030] The multi-axis drive mechanism 26 is configured to include an X-axis rotation (yawing) mechanism fixed to the unit fixture 38 for positioning the rotation position of the ultrasonic probe 24 about the x-axis by an x-axis rotation actuator, an X-axis translation mechanism for positioning the translation position of the ultrasonic probe 24 in the X-axis direction by an X-axis rotation actuator, and a Z-axis rotation mechanism for positioning the rotation position of the ultrasonic probe 24 about the Z-axis by a Z-axis actuator. With this configuration, the multi-axis drive mechanism 26 positions the ultrasonic probe 24 by a drive device 27 that operates in accordance with commands from an electronic control device 28.

[0031] The ultrasound drive control circuit 32 controls the emission of ultrasound waves from the ultrasound probe 24 to the arterial blood vessel 20 in accordance with commands from an ultrasound drive control unit 80 provided in the electronic control device 28. For example, among the numerous ultrasound transducers a1 to an arranged in a row in the first short-axis ultrasound array probe 24a, a certain number of ultrasound transducers, for example, 15 ultrasound transducers a1 to a15, are simultaneously driven at a frequency of about 10 MHz while applying a predetermined phase difference from the end ultrasound transducer a1, thereby sequentially emitting convergent ultrasound beams toward the arterial blood vessel 20 in the direction of the ultrasound transducer arrangement. Then, the ultrasound beam is scanned while shifting the ultrasound transducers one by one, and reflected waves for each emission are received and input to the electronic control device 28. The reflected wave signal input to the electronic control device 28 is detected by a detection processing unit 82 and processed by an ultrasound image generation unit 84 as information that can be used for image synthesis, as described in detail below.

[0032] Based on the reflected ultrasonic waves received by the ultrasound probe 24, the electronic control device 28 generates cross-sectional images of the arterial blood vessel 20 under the skin 18, i.e., a cross-sectional image (short-axis image) of the arterial blood vessel 20 in a direction intersecting the longitudinal direction thereof and a cross-sectional image (long-axis image) of the arterial blood vessel 20 in the longitudinal direction thereof, and displays these images on the image display device 30. The vascular condition evaluation unit 90 calculates the diameter of the arterial blood vessel 20 or the endothelial diameter (lumen diameter) d1, which is the diameter of the endothelium 70, from the short-axis image or long-axis image of the arterial blood vessel 20 thus generated. Furthermore, to evaluate the endothelial function of the arterial blood vessel 20, the electronic control device 28 calculates the expansion rate (rate of change) R (%) [= 100 × (d1 - da) / da] of the vascular lumen diameter, which represents flow-mediated dilation (FMD) after ischemia-reactive hyperemia. In this formula, "da" represents the vascular lumen diameter (base diameter, resting diameter) at rest.

[0033] When measuring the state of blood vessels using the endothelial function testing device 22, the measurement site on the living body 14, for example the upper arm 16, is compressed by a pressure device such as a cuff 62 to prevent blood flow, and a part of the living body 14 (the part more peripheral than the blood flow obstruction site) is put into an ischemic state. After that, the blood flow is suddenly released, causing a rapid increase in blood flow in the arterial blood vessel 20 at the measurement site, which causes the production of nitric oxide (NO) from the endothelium due to increased shear stress on the endothelium of the vascular wall, and the endothelial function is assessed by examining the relaxation state of the smooth muscle that depends on this nitric oxide.

[0034] 5 is a time chart illustrating the change in the vascular lumen diameter d1 after the release of blood flow obstruction (avascularization) in the FMD evaluation of the arterial blood vessel 20 by the endothelial function testing device 22. In FIG. 5, time t1 represents the release of blood flow obstruction, and the vascular lumen diameter d1 begins to expand from time t2, and reaches its maximum value d1 at time t3. ＭＡＸ Therefore, the expansion rate R of the blood vessel lumen diameter calculated by the electronic control device 28 reaches a maximum at time t3.

[0035] As shown in FIG. 1 , the endothelial function testing device 22 performs the occlusion of the arterial blood vessel 20 for FMD evaluation by controlling the air pump 58, pressure control valve 60, and the like using a cuff pressure control unit 88 provided in the electronic control device 28. For example, in accordance with a command from the electronic control device 28, the original pressure from the air pump 58 is controlled by the pressure control valve 60 and supplied to a cuff 62 wrapped around the upper arm 16. Specifically, the occlusion for FMD evaluation is performed by increasing the pressure in the cuff 62 (cuff pressure) to a predetermined occlusion cuff pressure that exceeds, for example, the systolic blood pressure of the living body 14. At this time, the cuff pressure control unit 88 detects the pressure in the cuff 62 (cuff pressure) in response to a signal from a pressure sensor 64 that detects the pressure. In FIG. 5 , for example, the cuff pressure control unit 88 maintains the cuff pressure at the occlusion cuff pressure for a predetermined time before the release of the occlusion, i.e., a predetermined time before time t1, and immediately reduces the cuff pressure to atmospheric pressure when the occlusion is released (time t1). As a result, blood flow in the arterial blood vessel 20 at the measurement site P is rapidly initiated, and the endothelial function testing device 22 measures the blood vessel diameter d ＭＡＸ is measured.

[0036] The ultrasound image generation unit 84 generates a first short-axis image representing a cross section of the arterial blood vessel 20 based on the ultrasound reflection signal detected by the first short-axis ultrasound array probe 24a, generates a second short-axis image representing a cross section of the arterial blood vessel 20 based on the ultrasound reflection signal detected by the second short-axis ultrasound array probe 24b, and generates a long-axis image representing a longitudinal section of the arterial blood vessel 20 based on the ultrasound reflection signal detected by the long-axis ultrasound array probe 24c.

[0037] The display control unit 92 displays the first short axis cross-sectional image, the second short axis cross-sectional image, and the long axis cross-sectional image generated by the ultrasound image generation unit 84 in the first short axis image display area G1, the second short axis image display area G2, and the long axis image display area G3, respectively, on the image display device 30.

[0038] Generally, short-axis cross-sectional images obtained using ultrasound have a lot of speckle noise, as shown in Fig. 6 showing the actual first short-axis cross-sectional image or second short-axis cross-sectional image, and skill is required to determine the muscle tissue KN. For this reason, the endothelial function testing device 22 of this embodiment is equipped with a muscle tissue region image generation unit 94 that generates a muscle tissue region image GKN showing the muscle tissue KN located in the first short-axis image display region G1 and the second short-axis image display region G2 from image information of the actual first short-axis image and second short-axis image display region G2, using a learned model LM that has been learned and stored in advance for the muscle tissue KN appearing in the short-axis image.

[0039] As shown in Fig. 7, the muscle tissue region image generation unit 94 includes a learned model unit 96 that stores a learned model LM that has been previously learned and stored for the muscle tissue KN appearing in the first short-axis image and the second short-axis image, and a muscle tissue region image synthesis unit 98 that uses the learned model LM to generate a muscle tissue region image GKN that shows the muscle tissue KN located in the actual first short-axis image and the second short-axis image, respectively, from image information appearing in the actual first short-axis image and the second short-axis image. The display control unit 92 displays the muscle tissue region image GKN showing the muscle tissue KN superimposed on the first short-axis image and the second short-axis image, respectively, in the first short-axis image display area G1 and the second short-axis image display area G2 of the image display device 30, for example, as shown in Fig. 6. In the case of Fig. 6, the muscle tissue region image GKN is displayed surrounded by a curved image LKN that shows the outer edge of the muscle tissue KN. The muscle tissue region image GKN may also be an image that displays the entire region of the muscle tissue KN using diagonal lines or color.

[0040] The trained model LM is a machine learning model MLM obtained by repeatedly performing supervised learning using training data in which the image information of the entire first and second short-axis images displayed in the first and second short-axis image display areas G1 and G2 is used as input data, and a muscle tissue region image GKN specified by an experienced operator using a mouse or pen to draw within the short-axis images displayed on an input display device (not shown) is used as output data. The image display device 30 may be used as the input display device.

[0041] Preferably, the machine learning model MLM is a multi-layered neural network that has undergone deep learning, in which the weight coefficients between nodes belonging to adjacent layers are sequentially updated when the input data is input to the nodes of the input layer and the output data is taught to the nodes of the output layer as pixels indicating the muscle tissue KN identified by an experienced operator in the short-axis image.

[0042] In the above multilayer neural network, preferably, brightness information Pin(x, y) for each pixel constituting the predetermined short-axis image is input as input data to the input layer, and output data Pout(x, y, t) indicating a pixel that indicates muscle tissue KN is output from the output layer. The t of the output data is a texture indicating whether or not the pixel indicated by the output data Pout(x, y) is within the area of ​​muscle tissue KN (if t = 1, it is within the area of ​​muscle tissue KN, and if t = 0, it is a feature indicating other tissue).

[0043] That is, in the output data output from the output layer of the multilayer neural network, image recognition is performed on a pixel-by-pixel basis that constitutes the short-axis image, and the texture t of the region (tissue) to which the pixel belongs is characterized from information about each pixel and its adjacent pixels, and a meaningful region division, whether it belongs to muscle tissue KN or other tissue, is estimated for each pixel P(x, y) based on the texture t.

[0044] 8 and 9 are flowcharts illustrating the main control operations of the electronic control device 28, which are repeatedly executed at a predetermined cycle. Fig. 8 shows a muscle tissue region image display control routine that, prior to FMD measurement, displays a muscle tissue region image GKN superimposed on the short-axis images of the first short-axis image display area G1 and the second short-axis image display area G2, thereby enabling the region of muscle tissue KN to be easily positioned between the ultrasound probe 24 and the arterial blood vessel 20. Fig. 9 shows an FMD measurement routine that performs FMD evaluation to measure the diameter, intimal thickness, plaque, blood flow velocity, etc. of the arterial blood vessel 20 based on the reflected signals of ultrasound output from the ultrasound probe 24 to the arterial blood vessel 20.

[0045] 8 (hereinafter, step will be omitted), prior to FMD evaluation, it is determined whether an ultrasound image display start operation has been performed, for example, based on whether any screen of the image display device 30 has been touched. If the determination in S1 is negative, the system is put on standby. If the determination in S1 is positive, however, in S2 corresponding to the ultrasound image generation unit 84, a first short-axis image representing a cross section of the arterial blood vessel 20 is generated based on the ultrasound reflection signals detected by the first short-axis ultrasound array probe 24a, a second short-axis image representing a cross section of the arterial blood vessel 20 is generated based on the ultrasound reflection signals detected by the second short-axis ultrasound array probe 24b, and a long-axis image representing a longitudinal section of the arterial blood vessel 20 is generated based on the ultrasound reflection signals detected by the long-axis ultrasound array probe 24c, and these are displayed in the first short-axis image display area G1, the second short-axis image display area G2, and the long-axis image display area G3, respectively.

[0046] Next, in S3, it is determined whether an operation to activate control for generating and displaying a muscle tissue region image GKN has been performed, for example, based on whether the screen of either the first short-axis image display area G1 or the second short-axis image display area G2 displaying the short-axis image of the image display device 30 has been touched. If the determination in S3 is negative, S1 and subsequent steps are repeated. If the determination is positive, in S4 corresponding to the muscle tissue region image generation unit 94, the trained model LM is used to generate muscle tissue region images GKN showing the muscle tissue KN located in the first short-axis image display area G1 and the second short-axis image display area G2, respectively, from the image information of the first short-axis image and the second short-axis image displayed in the actual first short-axis image display area G1 and the second short-axis image display area G2. Then, in S5 corresponding to the display control unit 92, the muscle tissue region image GKN is clearly displayed superimposed on the first short-axis tomographic image and the second short-axis tomographic image in the first short-axis image display area G1 and the second short-axis image display area G2 where the first short-axis tomographic image and the second short-axis tomographic image are displayed, respectively. This state is shown in FIG. 6.

[0047] While checking the first and second short axis cross-sectional images and the muscle tissue region image GKN superimposed on the first and second short axis cross-sectional images as shown in Figure 6, the operator can easily determine the placement of the ultrasonic probe 24 on the upper arm 16 and the twist position of the upper arm 16 so that the muscle tissue KN shown in the muscle tissue region image GKN is located as close as possible between the skin 18 on which the ultrasonic probe 24 is placed and the arterial blood vessel 20.

[0048] 9 , the arterial blood vessels 20 are identified from the lumens of multiple tubular organs in short-axis cross-sectional images obtained based on the reflected signals received by the ultrasound probe 24. For example, a video of short-axis cross-sectional images consisting of multiple frames synchronized with the diastole of the pulse is continuously calculated within a period in which the ultrasound probe 24 applies pressure to the biological tissue directly below the ultrasound probe 24 until the vein in the short-axis cross-sectional image is sufficiently compressed by the pressing actuator 37, and then the pressure is released. Next, from the video of multiple consecutive frames of short-axis cross-sectional images, the compressed and released states of the biological tissue directly below the ultrasound probe 24 are determined using a motion vector indicating the movement of a pixel or a local region consisting of a group of pixels between frames. Next, a smoothing process is performed on the compressed short-axis cross-sectional image, and the smoothed short-axis cross-sectional image is binarized into a binarized short-axis cross-sectional image, which is then subjected to a predetermined number of expansion processes and then a predetermined number of contraction processes. Next, multiple lumens are generated in the closed short-axis cross-sectional image, and a labeling process is performed to label the lumens in the closed short-axis cross-sectional image, resulting in a compressed short-axis cross-sectional image with the lumens superimposed. Next, the lumen of the arterial blood vessel 20 is determined from the multiple lumens generated in the compressed short-axis cross-sectional image based on whether a predetermined arterial blood vessel lumen determination condition is satisfied. When any of the lumens displayed in the compressed short-axis cross-sectional image is determined to be the lumen of the arterial blood vessel 20, a template of a local region including the entire arterial blood vessel 20 is created, and the template is used to track the movement of the arterial blood vessel 20 as the short-axis cross-sectional image transitions from the compressed state to the released state, thereby identifying the arterial blood vessel in the released short-axis cross-sectional image. The arterial blood vessel 20 in FIG. 6 shows a schematic image of a perfect circle identified in this way.

[0049] Using the arterial blood vessel 20 determined in S11, the FMD of the arterial blood vessel 20 is evaluated in S12-S18 corresponding to the vascular condition evaluation unit 90. First, in S12, the sensor support 10 lightly contacts the arterial blood vessel 20 located just below the skin 18 of the upper arm 16 of the living body 14 with the probe unit 12, without deforming the skin 18. The probe unit 12 is then positioned by the three-axis drive motor control unit 86 so that the position of the ultrasound probe 24 in three-dimensional space is the measurement position shown in FIG. 4B relative to the arterial blood vessel 20. Next, in S13, the lumen diameter da, which is the diameter of the endothelium 70 of the artery 20, is measured during the rest period. The period before t0 in FIG. 5 shows this state. Then, in S14, the upper arm 16 is compressed by the cuff 62, causing hemostasis of the arterial blood vessel 20 in the upper arm 16. The period from t0 to t1 in FIG. 5 shows this state. Next, in S15, after a predetermined time has elapsed since hemostasis, for example, 5 minutes, the pressure applied by the cuff 62 is released, causing the blood flow in the arterial blood vessel 20 to suddenly release and resume at the measurement site, as shown in the period from time t1 in Fig. 5. This causes the blood flow to suddenly increase, increasing shear stress on the endothelium of the vascular wall of the arterial blood vessel 20, which in turn causes the production of nitric oxide (NO) from the endothelium, and the relaxation of smooth muscle dependent on this nitric oxide causes a temporary increase in the endothelial diameter of the arterial blood vessel 20.

[0050] In S16, the lumen diameter d of the arterial blood vessel 20 after the release of hemostasis reaches the maximum value d as shown at time t4 in FIG. ＭＡＸ When it is determined in S17 that the maximum value dMAX has been reached, the maximum value d determined in S17 is ＭＡＸ and the lumen diameter da, which is the diameter of the endothelium 70 of the artery 20 at rest obtained in S3, the expansion rate (change rate) R (%) of the vascular lumen diameter, which represents the FMD (flow-mediated vasodilation response) after ischemia reactive hyperemia for evaluating the endothelial function of the arterial blood vessel 20 [= 100 × (d ＭＡＸ −da) / da] is calculated and displayed on the image display device 30 by the display control unit 92.

[0051] As described above, according to the endothelial function testing device 22 of this embodiment, the muscle tissue region image generation unit 94 generates a muscle tissue region image GKN representing the muscle tissue KN in the actual short-axis image from image information appearing in the actual short-axis image using the learned model LM that has previously learned and stored the muscle tissue region image GKN appearing in the short-axis image, and the display control unit 92 displays the muscle tissue region image GKN superimposed on the short-axis images in the first short-axis image display area G1 and the second short-axis image display area G2. This allows even an operator who is not skilled in determining the muscle tissue KN in short-axis images to accurately and easily grasp the muscle tissue KN in the short-axis image, and can easily determine the position of the ultrasound probe 24 on the upper arm 16 or the twist position of the upper arm 16 so that the muscle tissue KN is located between the ultrasound probe 24 and the arterial blood vessel 20. This allows for the generation of clear blood vessel images, improving the accuracy of blood vessel diameter measurement.

[0052] Furthermore, according to the endothelial function testing device 22 of this embodiment, the learned model LM is a machine learning model MLM that has repeatedly undergone supervised learning using training data in which a predetermined short-axis image is used as input data and the area of ​​muscle tissue KN identified by a skilled operator in the short-axis image is used as output data. As a result, the area of ​​muscle tissue KN can be clearly identified in the short-axis image.

[0053] Furthermore, according to the endothelial function testing device 22 of this embodiment, the machine learning model MLM is a multi-layered neural network that has undergone deep learning, in which weight coefficients between nodes belonging to adjacent layers are sequentially updated when the input data is input to the nodes of the input layer and output data indicating pixels indicating muscle tissue KN identified by an experienced operator in the short-axis image is repeatedly taught to the nodes of the output layer, thereby enabling the area of ​​muscle tissue KN to be more clearly identified in the short-axis image.

[0054] Furthermore, in the endothelial function testing device 22 of this embodiment, the brightness information for each pixel constituting a predetermined short-axis image is input to the input layer as the input data in the multilayer neural network, and output data indicating pixels that represent muscle tissue KN in the predetermined short-axis image is output from the output layer. This allows for more clear identification of the muscle tissue region image GKN that represents muscle tissue KN in the short-axis image.

[0055] Furthermore, according to the endothelial function testing device 22 of this embodiment, image recognition is performed on a pixel-by-pixel basis constituting the predetermined short-axis image in the output data sequentially output from the output layer of the multilayer neural network, and the texture of the brachial tract to which the pixel belongs is characterized from information on each pixel and its neighboring pixels, and a meaningful region segmentation is estimated for each pixel based on the texture, indicating whether the pixel belongs to muscle tissue KN or other tissue. This reduces the number of nodes constituting the output layer of the multilayer neural network, making the configuration of the multilayer neural network more compact.

[0056] Although the preferred embodiment of the present invention has been described in detail above with reference to the drawings, the present invention is not limited to this and may be implemented with various modifications within the scope of the spirit of the present invention.

[0057] 10: Sensor support, 12: Probe unit, 14: Living body, 16: Upper arm (living body), 18: Skin, 20: Arterial blood vessel, 22: Endothelial function testing device, 24: Ultrasonic probe, 24a: First short-axis ultrasonic array probe, 24b: Second short-axis ultrasonic array probe, 24c: Long-axis ultrasonic array probe, 25: Probe surface, 26: Multi-axis driving mechanism, 27: Driving device, 28: Electronic control device, 30: Image display device, 32: Ultrasonic driving control circuit, 34: Three-axis driving motor control circuit, 36: Base, 37: Pressing actuator (living body compression device), 38: Unit Knit fixture, 40: first arm, 42: second arm, 44: flexible arm, 58: air pump, 60: pressure control valve, 62: cuff, 64: pressure sensor, 70: endothelium, 80: ultrasound drive control unit, 82: detection processing unit, 84: ultrasound signal processing unit, 86: three-axis drive motor control unit, 88: cuff pressure control unit, 90: blood vessel state evaluation unit, 92: display control unit, 94: muscle tissue region image generation unit, 96: learned model unit, 98: muscle tissue region image synthesis unit, G1: first short-axis image display area, G2: second short-axis image display area, G3: long-axis image display area, L １ : Endometrium, L ２ : Media, L ３ : Adventitia, KN: Muscle tissue, GKN: Muscle tissue region image, LKN: Curve image, LM: Learning model, MLM: Machine learning model

Claims

1. An endothelial function examination device for arterial blood vessels that brings an ultrasonic sensor into contact with the epidermis of a living body, generates a cross-sectional image showing a cross-section of an arterial blood vessel located under the epidermis from an ultrasonic reflection signal from under the epidermis, and measures changes in the vascular diameter of the arterial blood vessel that appears in the cross-sectional image, comprising: a short-axis image display device that displays a short-axis image of the arterial blood vessel; a muscle tissue region image generation unit that generates a muscle tissue region image showing the muscle tissue located in the actual short-axis image from image information displayed in the actual short-axis image using a learned model that has been previously learned and stored about the area of ​​muscle tissue that appears in the short-axis image; and a display control unit that displays the muscle tissue region image on the image display device by superimposing it on the short-axis image.

2. The arterial vascular endothelial function testing device of claim 1, characterized in that the trained model is a machine learning model that has been repeatedly subjected to supervised learning using training data in which a specified short-axis image is used as input data and an area of ​​muscle tissue identified by an experienced operator within the short-axis image is used as output data.

3. The device for examining endothelial function of arterial blood vessels according to claim 2, characterized in that the machine learning model is a multi-layered neural network that has undergone deep learning in which weight coefficients between nodes belonging to adjacent layers are sequentially updated when the input data is input to nodes in an input layer and output data is repeatedly taught to nodes in an output layer as pixels that indicate an area of ​​muscle tissue identified by an experienced operator in the short-axis image.

4. The arterial vascular endothelial function inspection device of claim 3, characterized in that in the multi-layer neural network, brightness information for each pixel constituting the specified short-axis image is input to an input layer as the input data, and output data indicating pixels that indicate the area of ​​muscle tissue is output from an output layer.

5. The device for examining endothelial function of arterial blood vessels as described in claim 2, characterized in that, in the output data output from the output layer of the multi-layer neural network, image recognition is performed on a pixel-by-pixel basis constituting the short-axis image, the texture of the brachial region to which the pixel belongs is characterized from information on each pixel and its adjacent pixels, and a meaningful region division is estimated for each pixel based on the texture, indicating whether the pixel belongs to a muscle tissue region or other tissue.

Images