Information processing method, program, and image diagnostic device
A machine learning-based method for ultrasound CT imaging addresses the challenge of identifying spicules by analyzing line concentration and movement in three-dimensional images, improving diagnostic accuracy by reducing false positives.
Patent Information
- Application Number
- JP2021142604
- Authority / Receiving Office
- JP · JP
- Patent Type
- Applications
- Current Assignee / Owner
- Filing Date
- 2021-09-01
- Publication Date
- 2024-12-04
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
Conventional ultrasound CT imaging for breast cancer detection faces challenges in accurately identifying structural disturbances like spicules due to varying appearances of mammary glands and lesions across different imaging devices, and shadows obstructing tissue observation with handheld ultrasound devices.
An information processing method using a machine learning-based detection model to analyze three-dimensional ultrasound images, identifying spicule regions by calculating line concentration and movement between frames, and correcting for center bias to reduce false positives.
Effectively identifies structural disturbances such as spicules by reducing false positives through frame analysis and center bias correction, enhancing diagnostic accuracy in ultrasound CT imaging.
Smart Images

Figure 2024167449000001_ABST
Abstract
Description
[Technical field]
[0001] The present invention relates to an information processing method, a program, and an image diagnostic apparatus in ultrasound image diagnosis. [Background technology]
[0002] Diagnostic systems using ultrasound are non-invasive and do not require surgical procedures for directly cutting open a living body for observation, and are therefore widely used in the medical field as a technique for diagnosing information about the inside of a subject.
[0003] Ultrasonic CT (Computed Tomography), which is one method of ultrasonic diagnosis, irradiates an object with ultrasonic waves and creates a cross-sectional image of the object using scattered waves and transmitted waves, and recent research has shown that it is useful for detecting breast cancer. For example, in Patent Document 1, a large number of ultrasonic transducers that transmit and receive are arranged in a ring shape to surround the object, and three-dimensional imaging is automatically performed by moving them vertically.
[0004] The acquired tomographic images are interpreted by a diagnostician. The doctor makes a diagnosis by detecting predetermined diagnostic findings from the images. As a device to support image interpretation, a diagnostic support device has been developed that acquires examination images from a medical imaging device, detects abnormal areas such as tumors contained in the images through image processing and machine learning processing, and presents the images to the doctor to support image diagnosis.
[0005] For example, Patent Document 2 discloses a breast image lesion detection system that calculates a normal structure model representing the normal structure (direction) of the mammary gland from the nipple inside the breast toward the pectoralis major muscle from a breast X-ray image, extracts the direction of the mammary gland by applying a Gabor filter to the breast X-ray image, and judges whether or not there is a disturbance in the structure of the mammary gland by comparing the original normal direction of the mammary gland with the extracted direction of the mammary gland. Disturbance in the structure of the mammary gland is one of the image diagnostic findings of breast cancer, and the spicule structure observed when tissue around the tumor is concentrated in the tumor is one indicator for detection. [Prior art documents] [Patent documents]
[0006] [Patent Document 1] International Publication No. 2017 / 051903 [Patent Document 2] Patent No. 6256954 Summary of the Invention [Problem to be solved by the invention]
[0007] In tomographic imaging using ultrasound CT, it is possible to obtain a volume image that captures the entire target area in 3D, like X-ray CT, while retaining the characteristics of the scattered image using reflected waves obtained by conventional handheld ultrasound devices. However, the appearance of diagnostic findings in breast cancer images varies depending on the type of imaging device. For example, in X-ray mammography, the mammary gland and the lesion appear white, whereas in ultrasound images, the mammary gland appears white and the lesion appears dark. In addition, in the reflected wave images obtained by existing handheld ultrasound devices, some of the surrounding tissue cannot be observed due to the posterior shadow of the lesion, whereas ultrasound CT captures the entire surrounding tissue of the lesion. This is because in the former, ultrasound is transmitted from one side of the lesion, so the opposite side of the lesion is in shadow as seen from the transmitting array, whereas in the latter, ultrasound is irradiated from all directions surrounding the lesion, canceling out the shadow. Therefore, in ultrasound CT images, a diagnostic support function is desired to easily identify diagnostic findings that correspond to diagnostic findings in existing diagnostic images.
[0008] In one aspect, an object of the present invention is to provide an information processing method and the like that can suitably identify architectural disturbances such as spicules in three-dimensional ultrasound imaging. [Means for solving the problem]
[0009] An information processing method according to one aspect involves a computer executing a process to acquire multiple frame images of cross sections of a breast captured at multiple positions, calculate a line convergence degree indicating the degree of concentration of line structures for each point in each frame image, extract areas from each frame image where the line convergence degree is equal to or greater than a certain level, and determine whether or not architectural distortion in each frame image is a false positive depending on the amount of movement of the extracted area between the frame images. Effect of the Invention
[0010] In one aspect, architectural disturbances such as spicules can be favorably identified. [Brief description of the drawings]
[0011] [Figure 1] FIG. 1 is an explanatory diagram illustrating an example of the configuration of an image diagnostic system. [Diagram 2] FIG. 2 is a block diagram showing an example of the configuration of a server. [Diagram 3] FIG. 1 is a block diagram showing an example of the configuration of an image processing device. [Figure 4] FIG. 2 is an explanatory diagram showing an overview of a detection model. [Diagram 5] FIG. 11 is an explanatory diagram regarding a false positive determination process. [Figure 6] FIG. 11 is an explanatory diagram showing a calculation process of a line convergence degree. [Figure 7] 11 is an explanatory diagram showing a process of calculating a line convergence degree according to a distance between a point of interest and each point; FIG. [Figure 8] 13 is an explanatory diagram relating to a correction process of the line convergence degree according to the directional distribution. FIG. [Figure 9] FIG. 11 is an explanatory diagram relating to a correction process of a center bias. [Figure 10] FIG. 13 is an explanatory diagram showing a display screen of the spicule detection results. [Figure 11] 13 is a flowchart showing a procedure for generating a detection model. [Figure 12] 13 is a flowchart showing a procedure for spicule detection processing. [Figure 13] FIG. 11 is an explanatory diagram showing another example of the process of calculating the line convergence degree. [Figure 14] 13 is a flowchart showing another example of the procedure of the spicule detection process. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS
[0012] The present invention will now be described in detail with reference to the drawings showing embodiments thereof. (Embodiment) Fig. 1 is an explanatory diagram showing an example of the configuration of an image diagnostic system. In this embodiment, an image diagnostic system that detects spicules (architectural distortions) from an image of a cross section of a subject's breast will be described. The image diagnostic system includes a server 1 and an image diagnostic device 2. The server 1 and the image diagnostic device 2 are communicatively connected via a network N.
[0013] In this embodiment, the breast is taken as an example of a body part to be image-diagnosed, but other body parts may be used.
[0014] In addition, in this embodiment, "spicules" are given as an example of a detection target, but the phenomenon to be detected may be any disturbance in the architecture of the mammary gland, and architectural disturbance is not limited to what is called spicules.
[0015] The server 1 is a server computer capable of various information processing and transmitting and receiving information. Note that the device equivalent to the server 1 is not limited to a server computer, and may be, for example, a personal computer. In this embodiment, the server 1 performs machine learning to learn from predetermined training data, and functions as a generating device that generates a detection model 50 (see FIG. 4) for detecting a region where spicules are present (hereinafter referred to as a "spicule region") when multiple frame images obtained by capturing cross sections of a breast at multiple positions are input. Data of the detection model 50 generated by the server 1 is installed in the image diagnostic device 2, and the image diagnostic device 2 detects the spicule region from the captured image using the detection model 50.
[0016] The image diagnostic device 2 is an image diagnostic device for ultrasonic echo examination, and includes an image processing device 20 and an imaging device 30. The image processing device 20 is a computer that functions as a console for the image diagnostic device 2, and generates (reconstructs) and displays an ultrasonic image of a breast cross section. Note that the image processing device 20 is not limited to a computer (console) for ultrasonic image diagnosis, and may be a general-purpose computer such as a personal computer.
[0017] The imaging device 30 is an imaging device that transmits and receives ultrasonic signals. As shown in Fig. 1, the imaging device 30 is configured to be able to image the breasts of a subject lying face down. Specifically, the imaging device 30 has a bed-like shape, and a top plate 31 is provided with a hole 32 for inserting the breast. A water tank 33 is provided below the hole 32, and the subject inserts the breast into the water tank 33 through the hole 32.
[0018] The water tank 33 is provided with a ring array 34. The ring array 34 is a ring-shaped transducer array having a plurality of ultrasonic elements (transducers). A plurality of ultrasonic elements are arranged at equal intervals in the ring array 34, and each ultrasonic element transmits an ultrasonic signal and receives a reflected wave. The image processing device 20 generates an ultrasonic image (frame image) of one cross section by reconstructing the reflected wave data in multiple directions obtained from each ultrasonic element. Since the reflected wave signal has different intensity due to the difference in acoustic impedance occurring at the boundary between tissues, the reconstructed reflected wave image expresses the boundary information of tissues and lesions and the density within the tissue as brightness. The ring array 34 is configured to be movable in the vertical direction, and the image diagnostic device 2 moves the ring array 34 up and down to capture an ultrasonic image at each position (height) of the breast.
[0019] As the image diagnostic apparatus 2 according to this embodiment, the ultrasound diagnostic system described in the cited document 1 can be adopted.
[0020] The image diagnostic device 2 is not limited to the above configuration. For example, instead of the bed-type imaging device 30, an image diagnostic device using a handy scanner may be used.
[0021] In addition, although an ultrasound image is used as the image of the breast in this embodiment, an image of another modality such as an X-ray image (mammography, etc.) may be used. In short, it is sufficient that the device is a device that obtains a plurality of frame images obtained by imaging a cross section of a subject at a plurality of positions, images a continuous biological structure, and produces a three-dimensional medical image that continuously reconstructs slice images of the XY plane on a third axis.
[0022] The image diagnostic apparatus 2 uses the imaging device 30 to capture a cross section of the breast that is approximately perpendicular to the direction from the nipple to the pectoralis major muscle. In this embodiment, the image diagnostic apparatus 2 uses the above-mentioned detection model 50 to detect spicule regions where mammary gland structure is disturbed from the captured image. Specifically, the image diagnostic apparatus 2 detects spicule regions by inputting three-dimensional volume data consisting of frame images captured at multiple positions of a cross section of the breast to the detection model 50. By inputting multiple frame images into the detection model 50 as a single data set, spicule regions can be detected while referring to information from previous and subsequent frames.
[0023] Furthermore, as described below, the image diagnostic device 2 judges whether the detection result of the spicule region by the detection model 50 is valid, i.e., whether it is a false positive. Specifically, the image diagnostic device 2 calculates a line convergence degree, which indicates the degree of convergence of line structures, for each point in each frame image. Then, the image diagnostic device 2 judges whether the spicule region detected by the detection model 50 is a false positive, depending on the amount of movement between frames of a region with a line convergence degree equal to or greater than a certain level.
[0024] In this embodiment, the server 1 generates (learns) the detection model 50, but the local image diagnostic device 2 may generate the detection model 50. In this embodiment, the image diagnostic device 2 detects spicule regions and judges false positives, but the server 1 on the cloud may execute the processes. In other words, the distinction between the two is for convenience, and a single computer may execute a series of processes.
[0025] 2 is a block diagram showing an example of the configuration of the server 1. The server 1 includes a control unit 11, a main storage unit 12, a communication unit 13, and an auxiliary storage unit . The control unit 11 has one or more processors such as a central processing unit (CPU), a micro-processing unit (MPU), a graphics processing unit (GPU), etc., and performs various information processing by reading and executing a program P1 stored in the auxiliary storage unit 14. The main storage unit 12 is a temporary storage area such as a static random access memory (SRAM), a dynamic random access memory (DRAM), a flash memory, etc., and temporarily stores data required for the control unit 11 to execute arithmetic processing. The communication unit 13 is a communication module for performing processing related to communication, and transmits and receives information to and from the outside. The auxiliary storage unit 14 is a non-volatile storage area such as a large-capacity memory, a hard disk, etc., and stores the program P1 and other data required for the control unit 11 to execute processing.
[0026] The auxiliary storage unit 14 may be an external storage device connected to the server 1. The server 1 may be a multi-computer consisting of a plurality of computers, or may be a virtual machine virtually constructed by software.
[0027] In the present embodiment, the server 1 is not limited to the above configuration, and may include, for example, an input unit that accepts operation input, a display unit that displays images, etc. The server 1 may also include a reading unit that reads a non-transitory computer-readable recording medium 1a and reads the program P1 from the recording medium 1a. The program P1 may be executed on a single computer, or on multiple computers interconnected via a network N.
[0028] 3 is a block diagram showing an example of the configuration of the image processing device 20. The image processing device 20 includes a control unit 21, a main memory unit 22, a communication unit 23, a display unit 24, an input unit 25, a transmission / reception unit 26, an image processing unit 27, and an auxiliary memory unit . The control unit 21 has one or more processors such as CPUs, and performs various information processing by reading and executing the program P2 stored in the auxiliary storage unit 28. The main storage unit 22 is a temporary storage area such as RAM, and temporarily stores data required for the control unit 21 to execute arithmetic processing. The communication unit 23 is a communication module for performing processing related to communication, and transmits and receives information to and from the outside. The display unit 24 is a display screen such as a liquid crystal display, and displays images. The input unit 25 is an operation interface such as a keyboard or mouse, and accepts operation input from the user. The transmission / reception unit 26 controls the ultrasonic elements of the ring array 34 and transmits and receives ultrasonic signals. The image processing unit 27 is a module that performs image processing, and reconstructs reflected wave data received by the ultrasonic elements to generate an ultrasonic image.
[0029] The auxiliary storage unit 28 is a non-volatile storage area such as a hard disk or a large-capacity memory, and stores the program P2 and other data necessary for the control unit 21 to execute processing. The auxiliary storage unit 28 also stores a detection model 50. The detection model 50 is a machine learning model that has learned predetermined training data, and is a model that has learned to detect spicule regions where architectural distortion occurs when three-dimensional volume data consisting of multiple frame images is input. The detection model 50 is expected to be used as a program module that constitutes part of artificial intelligence software.
[0030] The image processing device 20 may include a reading unit that reads the non-transitory computer-readable recording medium 2a and reads the program P2 from the recording medium 2a. The program P2 may be executed on a single computer or on multiple computers interconnected via a network N.
[0031] 4 is an explanatory diagram showing an overview of the detection model 50. The following describes an overview of this embodiment.
[0032] 4 illustrates how spicule regions are detected when multiple frame images of a breast cross section captured at multiple positions are input to detection model 50. Detection model 50 is a model that has been trained to detect spicule regions when three-dimensional volume data consisting of multiple frame images is input. For example, detection model 50 is a neural network generated by deep learning, and is Faster R-CNN, a type of CNN (Convolution Neural Network).
[0033] The detection model 50 may be a neural network other than Faster R-CNN. The detection model 50 may also be a model based on other learning algorithms, such as a decision tree, a random forest, or a support vector machine (SVM).
[0034] The server 1 generates a detection model 50 by using training data to which a label indicating a correct spicule region is added to a group of training three-dimensional volume data. The three-dimensional volume data is image data obtained by imaging a subject's breast with the imaging device 30, and is a group of frame images obtained by imaging cross sections at multiple positions of the breast. The server 1 uses a group of frame images obtained by imaging the breasts of each of multiple subjects in one direction (the direction from the nipple to the pectoral muscle) as a group of training three-dimensional volume data. The label added to the three-dimensional volume data is data indicating the correct coordinate range of the spicule region, and is, for example, a label set by a medical professional (such as a doctor) to surround the spicule region for each frame image. The server 1 generates a detection model 50 by using training data to which a label indicating a spicule region is added to each frame image constituting the three-dimensional volume data.
[0035] The server 1 generates a detection model 50 that detects spicule regions from three-dimensional volume data by providing the detection model 50 with a plurality of frame images as one input data set. That is, the server 1 uses three-dimensional data in which each two-dimensional frame image is arranged along the imaging position (the height of the ring array 34) as input to the detection model 50, as shown on the left side of Fig. 4. This makes it possible to detect spicule regions in each frame image by referring to the features (architectural distortion) of the previous and following frame images.
[0036] When three-dimensional volume data is input, the detection model 50 outputs a detection result of a spicule region. Specifically, the detection model 50 outputs a rectangular bounding box surrounding the spicule region and a mask representing the inside of the boundary of the spicule region for each frame image. Since the detection model 50 according to the present embodiment is Faster R-CNN, the detection model 50 outputs both the bounding box and the mask, but the detection model 50 may output only one of them.
[0037] The server 1 detects spicule regions by inputting training 3D volume data into the detection model 50. The server 1 compares the detected spicule regions with correct labels, and updates parameters such as weights between neurons so that the two are close to each other. The server 1 inputs each training 3D volume data into the detection model 50 in sequence to update the parameters, and finally generates a detection model 50 with optimized parameters.
[0038] Data of the detection model 50 generated by the server 1 is installed in the image diagnostic device 2. When an image of a subject's breast is captured by the imaging device 30, the image diagnostic device 2 detects a spicule region using the detection model 50. That is, the image diagnostic device 2 detects a spicule region by inputting three-dimensional volume data consisting of a plurality of frame images to the detection model 50.
[0039] 5 is an explanatory diagram regarding the false positive determination process. In this embodiment, the image diagnostic device 2 determines whether or not the detection result of the spicule region by the detection model 50 is valid, i.e., whether or not it is a false positive, according to the amount of movement between frames of the region where the line structures are concentrated.
[0040] Specifically, the image diagnostic device 2 calculates the line convergence degree at each point in each frame image, and calculates the amount of movement of an area where the calculated line convergence degree is equal to or greater than a certain level between frames. If the image diagnostic device 2 determines that the amount of movement is equal to or greater than a predetermined threshold, it determines that the spicule area corresponding to that area is a false positive.
[0041] In Fig. 5, a map showing the line convergence degree of each point in a frame image (hereinafter referred to as a "line convergence degree map") is shown on the left, and a frame image showing the movement path between frames of an area with a line convergence degree equal to or higher than a certain level is shown on the right. An outline of the false positive determination process will be described based on Fig. 5.
[0042] The image diagnostic device 2 calculates the line convergence degree of each point in the frame image for each frame image. The line convergence degree is a numerical value indicating the concentration degree of line structures. Line structures are linear structures (edges) that appear in the frame images, and correspond to linear tissues such as mammary glands. Spicules are the concentration of surrounding tissues such as mammary glands in a tumor, so areas highly suspected of being spicules can be extracted depending on the level of line convergence degree.
[0043] The image diagnostic device 2 calculates the line convergence degree by applying a predetermined image filter to the frame image. In this embodiment, the image diagnostic device 2 uses a Gabor filter as the image filter, which extracts a specific directional component from the image. If the horizontal axis of the image is the x-axis and the vertical axis is the y-axis, the filter function g(x, y) of the Gabor filter is defined by the following formula (1).
[0044]
number
[0045] In addition, σ x and σ y is the standard deviation of the Gaussian function, f 0 represents the frequency of the trigonometric function. The Gabor filter (filter function g) itself is not significantly different from known ones, so a detailed description thereof will be omitted in this embodiment.
[0046] The image diagnostic device 2 applies a Gabor filter to the frame image to extract the direction and strength of the line structure at each point in the frame image. Specifically, the image diagnostic device 2 applies a plurality of Gabor filters with different extraction directions θ (Orientation) to the frame image, and calculates the Cartesian product h(x,y) of a filter function g(x,y) corresponding to each direction θ and a pixel value f(x,y) of the frame image. For example, the image diagnostic device 2 extracts the maximum value of h(x,y) as the strength of the line structure, and the direction θ at which h(x,y) is maximum as the direction of the line structure.
[0047] In this embodiment, the image diagnostic device 2 is provided with Gabor filters in eight directions at intervals of 22.5°, 0°, 22.5°, 45°, ..., 157.5°, with the horizontal direction (x-axis direction) of the image being 0°. The image diagnostic device 2 extracts the direction and intensity of the line structure by applying the eight types of Gabor filters to one frame image.
[0048] The image diagnostic device 2 calculates the line convergence degree of each point (each target point) in the frame image based on the direction and intensity of the line structure extracted above. A specific method for calculating the line convergence degree will be described later.
[0049] Through the above-mentioned process, the image diagnostic device 2 generates a line convergence degree map that indicates the line convergence degree of each point, as shown on the left side of Fig. 5. Note that an actual line convergence degree map is expressed as a heat map in which the value of each point (pixel) changes continuously, but for convenience of illustration in Fig. 5, parts (areas) with high line convergence degrees are hatched and other parts are shown in white.
[0050] The image diagnostic device 2 extracts an area where the line convergence degree is equal to or higher than a certain level from the generated line convergence degree map. That is, the image diagnostic device 2 extracts an area where there is a high possibility that spicules (architectural distortion) are occurring.
[0051] The image diagnostic device 2 determines whether or not the spicule region detected by the detection model 50 is a false positive, depending on the amount of movement of the extracted region between frames. Specifically, the image diagnostic device 2 identifies the center (center of gravity) coordinates of a region having a line convergence degree equal to or greater than a certain level from each frame image, and determines whether or not the spicule region is a false positive, depending on the movement distance and / or movement speed of the identified center coordinates between frames.
[0052] On the right side of FIG. 5, the moving path of the center of the area where the line concentration degree is equal to or higher than a certain level is illustrated by a broken line. For example, the image diagnostic device 2 extracts a predetermined number of consecutive frame images, and calculates the total moving distance of the center from the first frame to the last frame (total length of the broken line) and / or the average moving speed between two adjacent frames. The image diagnostic device 2 judges whether the calculated moving distance and / or moving speed is equal to or higher than a predetermined threshold. If the moving distance and / or moving speed is equal to or higher than the threshold, the image diagnostic device 2 judges it to be a false positive. If the moving distance and / or moving speed is less than the threshold, the image diagnostic device 2 judges it to be a positive result.
[0053] For example, the image diagnostic device 2 compares a spicule region detected by the detection model 50 with a region having a certain level of line convergence or more, and identifies the latter region that overlaps with the spicule region. The image diagnostic device 2 determines whether or not the spicule region is a false positive, depending on the amount of movement of the identified region. For example, the image diagnostic device 2 outputs (displays) a spicule region determined not to be a false positive as the final detection result of the spicule region (see FIG. 10). Note that the image diagnostic device 2 may output a spicule region determined to be a false positive in a manner that can be distinguished from a spicule region determined not to be a false positive. Since it is unlikely that the center of a spicule moves significantly between frames, erroneous detections can be reduced by determining a region whose amount of movement is equal to or more than a threshold as a false positive.
[0054] Fig. 6 is an explanatory diagram showing a calculation process of a line convergence degree. Fig. 6 conceptually illustrates a calculation process of a line convergence degree according to this embodiment. In this embodiment, the image diagnostic device 2 performs several processes on the result of extraction of a line structure by a Gabor filter in accordance with the purpose of detecting spicules, and then calculates a line convergence degree. A method of calculating the line convergence degree will be described with reference to Fig. 6.
[0055] As described above, the image diagnostic device 2 extracts the orientation and magnitude of the line structure at each point in the frame image from each frame image. As a result, an intensity map (intensity distribution) and a direction map (direction distribution) of the line structure in the frame image are generated, as shown in the second row from the top in Fig. 6. Note that, although the actual intensity map and direction map are expressed as heat maps, for the sake of convenience, in Fig. 6, the intensity of the line structure is shown as a line, and the direction of the line structure is shown as an arrow.
[0056] Here, the image diagnostic device 2 performs a predetermined pre-processing on the intensity map. Specifically, for each point (pixel) in the frame image, the image diagnostic device 2 calculates the average value of the intensity of each point and the intensity of the surrounding points of each point, and sets the calculated point as the intensity of the target point. For example, the image diagnostic device 2 sets the average value of the intensities of 6 x 6 points (pixels) including the target point as the intensity of each target point.
[0057] Furthermore, the image diagnostic device 2 binarizes the intensity of each point depending on whether the averaged intensity of each point is equal to or greater than a predetermined threshold. Specifically, the image diagnostic device 2 sets a threshold from the intensity distribution of the entire frame image using the Otsu Method, and sets the intensity of points equal to or greater than the threshold to "1" and the intensity of points less than the threshold to "0." Note that the values to be binarized are not limited to "1" and "0," and may be different values depending on whether the intensity is equal to or greater than the threshold.
[0058] The brightness and thickness of the line itself are thought to have little effect on the detection of spicules. Therefore, it is expected that false positives can be reduced by binarizing (and averaging) the intensity.
[0059] The image diagnostic device 2 calculates the product of the binarized intensity of each point and the direction of each point. That is, the image diagnostic device 2 generates a new map by multiplying the binarized intensity map shown in Fig. 6 by the direction map. Since the intensity is binarized, the map represents a direction distribution in which the line structures of points whose intensity is less than a threshold are reduced (ignored) and line structures whose intensity is equal to or greater than a threshold are emphasized.
[0060] For the sake of convenience, in the following description, the product of the binarized intensity and direction will be referred to as a "masked orientation" since it is a direction in which line structures whose intensity is less than a threshold are reduced (ignored).
[0061] In this way, the image diagnostic device 2 generates a mask orientation map (distribution of mask orientations) in which line structures whose intensities are equal to or greater than a threshold value are emphasized. Based on the generated mask orientation map, the image diagnostic device 2 generates a final line concentration degree map shown in the bottom part of FIG.
[0062] In this embodiment, the mask direction is extracted by taking the product of the direction and intensity of the line structure, and the line convergence degree is calculated from the distribution of the mask direction, but the direction (mask direction) according to the intensity is not extracted, and the line convergence degree may be calculated from the direction distribution of the line structure acquired primarily. That is, the image diagnostic device 2 only needs to be able to extract at least the direction of the line structure from the frame image and calculate the line convergence degree based on the direction of the extracted line structure, and the configuration of calculating the line convergence degree using the intensity of the line structure as well is not essential.
[0063] 7 is an explanatory diagram showing a process of calculating the line convergence degree according to the distance between the attention point and each point. The processing contents when calculating the line convergence degree from the mask direction will be described with reference to FIG.
[0064] The image diagnostic device 2 regards each point in the frame image as a point of interest, and obtains a relative vector connecting the point of interest and each point other than the point of interest. The image diagnostic device 2 calculates the degree of line convergence by obtaining the inner product value of the relative vector and the mask direction (direction of the line structure) at the other point that is the starting point of the relative vector.
[0065] A specific calculation process is conceptually illustrated on the left side of Fig. 7. When calculating the line convergence degree for an arbitrary attention point P in a frame image, the image diagnostic device 2 calculates the line convergence degree based on the mask direction at a point Q within a certain distance from the attention point P, specifically, at the point Q within a range from a predetermined inner diameter A to an outer diameter B. The image diagnostic device 2 takes a relative vector between the attention point P and the point Q, and calculates the inner product value of the relative vector and the mask direction at the point Q. The image diagnostic device 2 calculates the inner product values for all points Q within the range from the inner diameter A to the outer diameter B, and calculates the average value of the inner product values for each point Q. The image diagnostic device 2 regards the calculated average value as the line convergence degree of the attention point P.
[0066] In this embodiment, when calculating the line convergence degree as described above, the image diagnostic device 2 weights the inner product value for each point Q so as to place more importance on the line structure located closer to the attention point P. Specifically, when calculating the inner product value between the relative vector and the mask direction for each point Q, the image diagnostic device 2 weights the point Q by multiplying the point Q by a distance function d(r) that increases as the distance r from the point Q decreases. The distance function d(r) is defined, for example, by the following formula (2).
[0067]
number
[0068] In addition, β in formula (2) is a constant. The hatching on the right side of Fig. 7 conceptually illustrates how the distance function d(r) changes depending on the distance r from point Q. The image diagnostic device 2 multiplies the distance function d(r) designed to be higher as the distance r becomes closer, so that the image diagnostic device 2 calculates the line concentration degree by placing more importance on line structures closer to the attention point P.
[0069] 8 is an explanatory diagram regarding the correction process of the line concentration degree according to the directional distribution. The image diagnostic device 2 further performs weighting according to the variation (distribution) of the line structure toward the point of interest in order to extract an area where the line structure is concentrated from all directions. Specifically, the image diagnostic device 2 calculates an index value for evaluating the degree to which the line structure toward the point of interest is evenly distributed in all directions (multiple directions), and corrects the line concentration degree according to the magnitude of the calculated index value.
[0070] For example, as shown in Fig. 8, the image diagnostic device 2 divides a circular region within a predetermined distance centered on the attention point P into multiple fan-shaped regions at a predetermined angle (22.5°) each in accordance with the extraction direction θ of the Gabor filter. Then, the image diagnostic device 2 extracts a directional component toward the attention point P from the line structure at each point in the fan-shaped region, thereby obtaining a probability value P i (i is the number of each sector region, i=1, 2, 3, ..., 8). The image diagnostic device 2 calculates the entropy shown in the following formula (3) as an index value representing the degree to which the line structures pointing to the attention point P are evenly distributed in all sector regions.
[0071]
number
[0072] The image diagnostic apparatus 2 corrects the line convergence degree of the attention point P in accordance with the magnitude of the calculated entropy. Specifically, the image diagnostic apparatus 2 corrects the line convergence degree to be higher as the entropy increases.
[0073] In this embodiment, the index value is entropy, but the index value is not limited to this and may be other values (for example, the probability value P i The average value of the above may be used.
[0074] 9 is an explanatory diagram relating to the correction process of the center bias. The image diagnostic device 2 further performs correction to reduce the influence of the tendency (bias) that linear structures such as mammary glands tend to concentrate toward the center of the breast.
[0075] As described above, the image diagnostic device 2 captures a cross section that is approximately perpendicular to the direction from the nipple to the pectoralis major muscle. However, since the mammary glands radiate from the nipple to the periphery in the cross section, line structures naturally tend to concentrate toward the center of the breast in the image captured by the imaging device 30. Therefore, the image diagnostic device 2 performs the following process to reduce the effect of line structures concentrating at the center of the breast.
[0076] Specifically, the image diagnostic device 2 normalizes the line convergence so that the line convergence is constant (for example, 1) when the line structure faces the point of interest at all points. The image diagnostic device 2 first detects the boundary between the breast and the area other than the breast from the frame image, and generates a map (upper left) in which the breast and the area other than the breast are binarized. The image diagnostic device 2 multiplies (weights) the map by a predetermined distance function so that the value is higher closer to the center of the breast, thereby generating a normalization map (upper right) in which the value is set to increase from the periphery of the breast toward the center.
[0077] The image diagnostic device 2 normalizes the line convergence degree by dividing the line convergence degree map (lower left) by the generated normalization map. That is, the image diagnostic device 2 corrects the line convergence degree by dividing the line convergence degree of each point in the line convergence degree map by the value of each point in the normalization map. This normalizes the value to 1 when the line structure faces the point of interest at all points, reducing the influence of the center bias.
[0078] As described above, the image diagnostic device 2 performs a certain process on the extraction result of the Gabor filter in accordance with spicule detection, and then calculates the line convergence degree. As described in Fig. 5, the image diagnostic device 2 extracts an area having a line convergence degree equal to or higher than a certain degree from the line convergence degree map, and judges whether or not the spicule area detected by the detection model 50 is a false positive according to the amount of movement of the area between frames.
[0079] 10 is an explanatory diagram showing a display screen of the spicule detection result. When a spicule region is detected from a captured breast image, the image diagnostic device 2 displays a screen as shown in FIG.
[0080] For example, the image diagnostic device 2 displays on the screen a breast image on which a label indicating a spicule region is superimposed. Specifically, the image diagnostic device 2 displays a frame image on which a bounding box surrounding the spicule region and a semi-transparent mask that colors the inside of the spicule region are superimposed. For convenience, the colored spicule region is illustrated by hatching in Fig. 10. The image diagnostic device 2 superimposes the bounding box and mask on the spicule region detected by the detection model 50, excluding the spicule region determined to be a false positive.
[0081] Furthermore, the image diagnostic device 2 displays a probability value (indicated as "Pred" in FIG. 10) output from the detection model 50, which indicates the likelihood of the detection result of the spicule region, in association with each spicule region. Furthermore, the image diagnostic device 2 highlights spicule regions whose probability value is equal to or greater than a predetermined value (e.g., 90% or greater). For example, as shown in FIG. 10, the image diagnostic device 2 thickens the bounding box corresponding to the spicule region whose probability value is equal to or greater than a predetermined value, and changes the mask color of the spicule region. For example, the image diagnostic device 2 displays a change button 101 at the bottom of the screen, and switches to another frame image in response to an operation input to the change button 101.
[0082] Furthermore, the image diagnostic apparatus 2 displays on the screen a display button 102 for displaying the positive / false positive determination result of the spicule region. When an operation input to the display button 102 is accepted, the image diagnostic apparatus 2 displays the positive / false positive determination result of the spicule region based on the line convergence degree on the breast image.
[0083] Specifically, the image diagnostic device 2 displays an object (broken line) of the movement path illustrated on the right side of FIG. 5 superimposed on the breast image illustrated in FIG. 10. That is, the image diagnostic device 2 displays the movement path between frames of the area where the line convergence degree is equal to or higher than a certain level. This allows a medical professional viewing the breast image to objectively determine why the area where the label is superimposed was determined to be positive and why the other area was determined to be a false positive. When displaying the object, for example, the image diagnostic device 2 may change the display mode of the object (for example, the color of the broken line) depending on whether the amount of movement related to each movement path is equal to or higher than a threshold, i.e., whether it is a false positive or not.
[0084] In this embodiment, a two-dimensional frame image (an image of a breast cross section) is displayed as the final detection result, but for example, the image diagnostic device 2 may generate a three-dimensional image by reconstructing a plurality of frame images and superimpose a label representing a spicule region on the three-dimensional image. In this case, for example, the image diagnostic device 2 superimposes a lattice-shaped bounding box surrounding the three-dimensional spicule region and superimposes a three-dimensional mask for coloring the spicule region. In this way, the image diagnostic device 2 only needs to be able to display a breast image on which a label representing a spicule region is superimposed based on a captured image (a plurality of frame images), and the breast image is not limited to a two-dimensional frame image.
[0085] As described above, according to this embodiment, spicule regions are detected from multiple frame images, and whether or not the spicule regions are false positives is determined based on the amount of movement between frames of regions with high line concentration. In this way, by using three-dimensional information, architectural distortions such as spicules can be suitably identified.
[0086] In the present embodiment, the image diagnostic device 2 uses a machine learning model (detection model 50) as a means for detecting a spicule region, but the image diagnostic device 2 may detect a spicule region based on rules. For example, the image diagnostic device 2 may detect a region having a line convergence degree equal to or greater than a certain level as a spicule region.
[0087] In addition, in this embodiment, the image diagnostic device 2 determines whether or not a spicule region detected by its own device is a false positive, but it may also determine whether or not a spicule region detected by another device (computer) or a spicule region designated by a user (medical professional) is a false positive. That is, the image diagnostic device 2 only needs to be able to determine whether or not a spicule region (architectural distortion) in at least each frame image is a false positive, and a configuration for detecting the spicule region to be determined is not essential.
[0088] Fig. 11 is a flowchart showing the procedure of the process of generating the detection model 50. The process of generating the detection model 50 by machine learning will be described with reference to Fig. 11. The control unit 11 of the server 1 acquires training data in which a label indicating the correct spicule region is assigned to a group of training 3D volume data (step S11). The training 3D volume data is a group of frame images in which cross sections of a subject's breast are captured at multiple positions. The control unit 11 acquires training data in which a label indicating the correct coordinate range of the spicule region is assigned to each frame image constituting the 3D volume data.
[0089] Based on the training data, the control unit 11 generates a detection model 50 that detects spicule regions where architectural distortion occurs when 3D volume data consisting of multiple frame images is input (step S12). For example, the control unit 11 generates Faster R-CNN as the detection model 50. The control unit 11 detects spicule regions by inputting 3D volume data for training to the detection model 50. The control unit 11 compares the detected spicule regions with the correct label, and generates the detection model 50 by optimizing the weights between neurons so that the two are similar. The control unit 11 ends the series of processes.
[0090] 12 is a flowchart showing the procedure of the spicule detection process, and the process for detecting the spicule region will be described with reference to FIG. The control unit 21 of the image diagnostic device 2 acquires a plurality of frame images obtained by capturing cross sections at a plurality of positions of the breast (step S31). Specifically, as described above, the control unit 21 acquires a group of frame images obtained by capturing a plurality of cross sections perpendicular to the direction from the nipple to the pectoralis major muscle, which are ultrasound images obtained by transmitting and receiving ultrasound signals. The control unit 21 inputs three-dimensional volume data consisting of the acquired plurality of frame images to the detection model 50, thereby detecting spicule regions where architectural distortion occurs (step S32).
[0091] The control unit 21 extracts the direction and intensity of the line structure at each point in the frame image from each frame image (step S33). Specifically, the control unit 21 applies a plurality of Gabor filters having different extraction directions θ to the frame image to extract the direction (Orientation) and intensity (Magnitude) of the line structure at each point in the frame image.
[0092] The control unit 21 calculates the degree of line convergence at each point in each frame image based on the direction and intensity of the extracted line structure (step S34). Specifically, the control unit 21 calculates the average value of the intensity of each point and the intensity of the surrounding points of each point, and binarizes the intensity of each point in the frame image depending on whether the calculated average value is equal to or greater than a predetermined threshold. The control unit 21 extracts the mask direction by taking the product of the binarized intensity and the direction extracted in step S33. The control unit 21 sets each point in the frame image as a point of interest, takes a relative vector between each point within a predetermined distance from the point of interest and the point of interest, and calculates an inner product value of the relative vector and the mask direction. In this case, the control unit 21 weights the inner product value corresponding to each point using a distance function that becomes higher as the distance between the point of interest and other points becomes closer. The control unit 21 calculates the average value of the weighted inner product values to calculate the degree of line convergence at the point of interest.
[0093] The control unit 21 corrects the line convergence degree so that the line convergence degree becomes higher as the line structure toward the point of interest is distributed evenly in all directions (multiple directions) (step S35). Specifically, the control unit 21 divides a circular region centered on the point of interest into multiple fan-shaped regions corresponding to multiple extraction directions θ, respectively, and calculates a probability value of a line structure toward the point of interest from each direction based on the direction of the line structure at each point in each fan-shaped region. The control unit 21 calculates an index value (entropy) representing the degree to which the line structure toward the point of interest is distributed evenly in each direction based on the calculated probability value, and corrects the line convergence degree depending on the magnitude of the calculated index value.
[0094] The control unit 21 normalizes (corrects) the line convergence degree so that the line convergence degree is constant when the line structure faces the attention point at all points (step S36). Specifically, the control unit 21 generates a normalization map in which the value of each point in the frame image is set so that the value increases from the breast periphery toward the center, and corrects the line convergence degree by dividing the line convergence degree at each point by the value of each point in the normalization map.
[0095] The control unit 21 extracts a region having a line convergence degree equal to or greater than a certain level from each frame image (step S37). Then, the control unit 21 determines whether or not the spicule region detected in step S32 is a false positive, depending on the amount of movement of the extracted region between frames (step S38). Specifically, as described above, when the movement distance and / or movement speed of the spicule region between frames is equal to or greater than a predetermined threshold, the control unit 21 determines that the spicule region is a false positive.
[0096] Control unit 21 displays on display unit 24 a breast image in which a label is superimposed on the spicule region determined not to be a false positive (step S39). Specifically, control unit 21 displays a frame image in which a bounding box surrounding the spicule region and a semi-transparent mask that colors the inside of the spicule region are superimposed. Control unit 21 also displays, superimposed on the frame image, an object (broken line) that indicates the movement path between frames of the region with a line convergence degree equal to or higher than a certain level. Control unit 21 ends the series of processes.
[0097] As described above, according to this embodiment, architectural disturbances such as spicules can be suitably identified.
[0098] (Modification) In the above-mentioned embodiment, the line concentration degree is detected for each frame image, but the line concentration degree may be detected using an integrated image obtained by integrating a plurality of consecutive frame images. FIG. 13 is an explanatory diagram showing another example of the calculation process of the line concentration degree. In this modification, as shown in the first and second rows from the top of FIG. 13, the image diagnostic device 2 generates an integrated image by integrating a plurality of frame images with consecutive imaging positions from the frame images captured by the imaging device 30. Then, the image diagnostic device 2 extracts the direction and intensity of the line structure at each point in the image from each generated integrated image, and generates an intensity map and a direction map of the line structure in the integrated image as shown in the third row from the top of FIG. 13.
[0099] The integrated image is generated by calculating the average brightness (pixel value) of each pixel in, for example, five frame images, and using the calculated average as the pixel value of each pixel. Note that the number of frame images from which an integrated image is generated is not limited to five, and may be any number such as three to seven, or the integrated image may be generated by other methods.
[0100] Fig. 14 is a flowchart showing another example of the procedure of the spicule detection process. The process shown in Fig. 14 is obtained by adding steps S41 to S49 between steps S31 and S32 in the process shown in Fig. 12. Explanations of the same steps as in Fig. 12 will be omitted.
[0101] In this modification, when the control unit 21 of the image diagnostic apparatus 2 acquires a plurality of frame images capturing cross sections of a breast at a plurality of positions (step S31), the control unit 21 accumulates the acquired frame images every predetermined number of times to generate an accumulated image (step S41). For example, when three-dimensional volume data of 60 frame images is acquired in step S31, the control unit 21 performs the above-mentioned accumulation process every 5 images to generate 12 accumulated images. Then, the control unit 21 executes the process of steps S42 to S47 based on the accumulated images generated (for example, 12 accumulated images). The process of steps S42 to S47 is the same as the process of steps S32 to S37.
[0102] In step S42, the detection model used to detect the spicule region in the accumulated image has the same configuration as the detection model 50 shown in FIG. 4, but is a model trained using training data in which 3D volume data consisting of a plurality of accumulated images is used as training data and a label indicating the spicule region in the accumulated image is assigned to each accumulated image constituting the training 3D volume data. The training 3D volume data here is a group of accumulated images obtained by accumulating a predetermined number of frame images captured by the imaging device 30, and the label is a label obtained by accumulating the labels assigned to each frame image for each predetermined number. The label accumulated for each predetermined number is, for example, data including all the correct coordinate ranges of the spicule region indicated by the labels assigned to each frame image. By learning using such training data, a detection model that detects the spicule region when 3D volume data consisting of a plurality of accumulated images is input is generated. Such a detection model is also generated by the server 1 and installed in the image diagnostic device 2. Therefore, the control unit 21 of the image diagnostic apparatus 2 detects spicule regions by inputting the three-dimensional volume data based on the multiple accumulated images generated in step S41 to such a detection model (step S42).
[0103] The control unit 21 also extracts the direction and strength of the line structure at each point in the integrated image generated in step S41 from each integrated image (step S43), and calculates the line convergence degree at each point in each integrated image based on the direction and strength of the extracted line structure (step S44). The control unit 21 also corrects the line convergence degree so that the line convergence degree becomes higher as the line structure toward the point of interest is distributed evenly in all directions (multiple directions) (step S45), and normalizes the line convergence degree so that the line convergence degree becomes constant when the line structure faces the point of interest at all points (step S46). The control unit 21 then extracts areas with a line convergence degree equal to or higher than a certain level from each integrated image (step S47).
[0104] In step S47, the control unit 21 determines whether or not an area with a line concentration degree equal to or greater than a certain level has been extracted from the accumulated image (whether or not there is an extracted area) (step S48), and if it is determined that there is no extracted area (step S48: NO), the control unit 21 terminates the series of processes.
[0105] When it is determined that there is an extraction region (step S48: YES), the control unit 21 specifies the frame image used for generating the integrated image in which it is determined that there is an extraction region as the image to be processed in the false positive determination process of step S38 (step S49). Note that the control unit 21 may specify, in addition to the integrated image in which it is determined that there is an extraction region, the frame image used for generating three integrated images including the integrated image and two integrated images whose imaging positions are adjacent to each other, as the image to be processed. Thereafter, the control unit 21 executes the processes of steps S32 to S39 for the frame image specified as the processing target. The above-mentioned process makes it possible to narrow down the frame images to be subjected to the false positive determination process for the detected spicule region based on the results of the detection of the spicule region in the integrated image and the calculation of the line convergence degree. In addition, by performing the process on a frame image basis for the false positive determination process target, it is possible to suitably specify the structural disturbance of spicules, etc., as in the above-mentioned embodiment.
[0106] As described above, according to this modified example, it is possible to narrow down the frame images to be processed for determining whether or not a detected spicula region is a false positive based on the processing results of the accumulated image, thereby making it possible to efficiently determine whether or not a false positive is occurring and reducing the time required for calculation processing.
[0107] The embodiments disclosed herein are illustrative in all respects and should not be considered as limiting. The scope of the present invention is defined by the claims, not by the above meaning, and is intended to include all modifications within the scope and meaning equivalent to the claims. [Explanation of symbols]
[0108] 1 Server 11 Control section 12 Main memory 13. Communications Department 14 Auxiliary storage P1 Program 2. Diagnostic imaging equipment 20 Image processing device 21 Control section 22 Main memory 23 Communications Department 24 Display section 25 Input section 26 Transmitter / receiver 27 Image Processing Section 28 Auxiliary storage P2 Program 50 Detection Models 30 Imaging device 31 Tabletop 32 holes 33 Aquarium 34 Ring Array
Claims
1. A plurality of frame images are obtained by capturing cross sections of the breast at a plurality of positions; A line concentration degree indicating the concentration degree of the line structure is calculated for each point in each frame image; extracting an area where the line convergence degree is equal to or greater than a certain level from each of the frame images; Whether or not architectural distortion in each of the frame images is a false positive is determined according to the amount of movement of the extracted region between the frame images. An information processing method in which processing is performed by a computer.
2. Detecting an area where architectural distortion occurs by inputting the acquired multiple frame images into a model that has been trained to detect an area where architectural distortion occurs when three-dimensional volume data consisting of the multiple frame images is input; Whether or not the region where architectural distortion occurs is a false positive is determined according to the amount of movement between the frame images of the region where the line convergence degree is equal to or greater than a certain level. The information processing method according to claim 1 .
3. If the region in which architectural distortion occurs is determined to be positive, a breast image is output in which a label indicating architectural distortion is superimposed on the region. The information processing method according to claim 2 .
4. outputting the breast image on which an object representing a moving path between the frame images in an area where the line convergence degree is equal to or greater than a certain level is superimposed; The information processing method according to claim 3.
5. extracting a direction of the line structure at each point in each of the frame images by applying a predetermined image filter to each of the frame images; Calculating the line convergence degree at each point in each of the frame images based on the direction of the extracted line structure. The information processing method according to any one of claims 1 to 4.
6. Extracting the direction and strength of the line structure by applying a Gabor filter to each of the frame images; The line concentration is calculated based on the direction and strength of the extracted line structure. The information processing method according to claim 5.
7. binarizing the intensity of the line structure depending on whether the intensity of the line structure is equal to or greater than a predetermined threshold; The line convergence is calculated based on the product of the binarized intensity and the direction.
7. The information processing method according to claim 5 or 6.
8. Calculating an average value of the intensity of each point in the frame image and the intensity of each point's surrounding points; The intensity of each of the points is binarized according to whether the average value calculated for each of the points is equal to or greater than the threshold value. The information processing method according to claim 7.
9. Each point in the frame image is set as a point of interest, and an inner product value of a vector connecting the point of interest and another point within a certain distance from the point of interest and a direction of the line structure at the other point is calculated for each point within the certain distance; weighting the dot product values corresponding to each point within the certain distance using a distance function that increases as the distance between the point of interest and the other points decreases; The line convergence degree of the point of interest is calculated by calculating an average value of the weighted inner product values. The information processing method according to any one of claims 5 to 8.
10. determining a probability value of the line structure directed toward the point of interest from each of a plurality of directions based on a direction of the line structure at each point within a circular region centered on the point of interest, and calculating an index value representing a degree to which the line structure toward the point of interest is evenly distributed in each of the plurality of directions based on the calculated probability value; The line concentration is corrected based on the calculated index value. The information processing method according to any one of claims 5 to 9.
11. the frame image is an image obtained by capturing a cross section perpendicular to a direction from the nipple to the pectoralis major muscle, A normalization map is generated in which the values of each point in the frame image are set so that the values increase from the periphery toward the center of the breast; The line convergence degree is corrected by dividing the line convergence degree at each point by the value of each point in the normalization map. The information processing method according to any one of claims 1 to 10.
12. generating an integrated image by integrating the acquired plurality of frame images every predetermined number of times; A line concentration degree indicating the concentration degree of the line structure is calculated for each point in each generated integrated image; When an area where the line concentration degree is equal to or greater than a certain level is extracted from the integrated image, a line concentration degree indicating a concentration degree of line structures is calculated for each point in each of the frame images used to generate the integrated image. The information processing method according to any one of claims 1 to 11.
13. A plurality of frame images are obtained by capturing cross sections of the breast at a plurality of positions; A line concentration degree indicating the concentration degree of the line structure is calculated for each point in each frame image; extracting an area where the line convergence degree is equal to or greater than a certain level from each of the frame images; Whether or not architectural distortion in each of the frame images is a false positive is determined according to the amount of movement of the extracted region between the frame images. A program that causes a computer to carry out processing.
14. an acquisition unit that acquires a plurality of frame images obtained by capturing cross sections of the breast at a plurality of positions; A calculation unit that calculates a line concentration degree indicating a concentration degree of a line structure for each point in each frame image; an extracting unit that extracts an area in which the line convergence degree is equal to or greater than a certain level from each of the frame images; a determination unit that determines whether architectural distortion in each of the frame images is a false positive or not according to the amount of movement of the extracted region between the frame images; An imaging diagnostic device comprising:
Citation Information
Patent Citations
Silver halide photographic sensitive material
JP1987056954A
Ultrasonic diagnostic system and ultrasonic diagnostic method
WO2017051903A1
