Ultrasound diagnostic apparatus, medical information processing apparatus and program

Abstract

To ensure a real-time property of an ultrasound diagnostic apparatus while suppressing an amount of computation.SOLUTION: An ultrasound diagnostic apparatus according to an embodiment comprises an ultrasound probe, a first acquisition unit, and a second acquisition unit. The first acquisition unit acquires first ultrasound data representing a plurality of frames that are consecutive in a time direction, the first ultrasound data being acquired by execution of an ultrasound scan on a subject via the ultrasound probe. The second acquisition unit inputs data based on the first ultrasound data into a trained model, and acquires, from the trained model, second ultrasound data representing a single frame, the second ultrasound data having a spatial resolution of fluid in the subject that is higher than that of the first ultrasound data.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] The embodiments disclosed in the present specification and drawings relate to an ultrasound diagnostic apparatus, a medical information processing apparatus, and a program. [Background technology]

[0002] Ultrasound diagnostic devices are widely used to observe and diagnose blood flow in living bodies. Ultrasound diagnostic devices generate and display blood flow information from reflected ultrasound waves using the Doppler method, which is based on the Doppler effect. Examples of blood flow information generated and displayed by ultrasound diagnostic devices include color Doppler images and Doppler waveforms (Doppler spectra).

[0003] Color Doppler images are captured using the Color Flow Mapping (CFM) method. In the CFM method, ultrasound waves are transmitted and received multiple times on multiple scan lines. Then, by applying an MTI (Moving Target Indicator) filter to the data sequence at the same position, signals originating from stationary or slow-moving tissue (clutter signals) are suppressed and signals originating from blood flow are extracted. The CFM method estimates blood flow information such as blood flow velocity, variance, and power values ​​from this blood flow signal, and displays the distribution of the estimated results as a Doppler image.

[0004] It is known that image quality deteriorates with B-mode and Doppler data due to the point spread function (PSF), which is determined by the wavelength of the transmitted ultrasound and the transmit / receive aperture width, etc. While there are solutions such as increasing the frequency of the transmitted ultrasound, there are also limitations to the frequency band of the probe.

[0005] Non-Patent Document 1 describes a method for generating blood flow data with improved resolution by extracting high signal values ​​(amplitude values) from multiple blood flow data sets that are continuous in the time direction. More specifically, Non-Patent Document 1 describes a method for utilizing the fact that the amplitude distribution characteristics of the speckle pattern of blood flow data can be approximately approximated by a probability distribution called a Rayleigh distribution. As a result, high-intensity (high-resolution) and spatially sparse signals that have a low occurrence probability but a large signal value (amplitude value) are extracted from each blood flow data set and integrated.

[0006] However, in the method described in Non-Patent Document 1, if a large number of frames are used, the load of calculation processing increases, which may result in poor responsiveness and impair the real-time performance of the ultrasound diagnostic apparatus. [Prior art documents] [Non-patent literature]

[0007] [Non-Patent Document 1] Jorgen Arendt Jensen et al., “Fast super resolution ultrasound imaging using the erythrocytes,” Proc. SPIE 12038, Medical Imaging 2022: Ultrasonic Imaging and Tomography,120380E, April 4, 2022 Summary of the Invention [Problem to be solved by the invention]

[0008] One of the problems to be solved by the embodiments disclosed in this specification and the drawings is to reduce the amount of calculations and ensure the real-time performance of an ultrasound diagnostic apparatus. However, the problems to be solved by the embodiments disclosed in this specification and the drawings are not limited to the above problem. Problems corresponding to the effects of each configuration shown in the embodiments described below can also be positioned as other problems. [Means for solving the problem]

[0009] An ultrasound diagnostic apparatus according to an embodiment includes an ultrasound probe, a first acquisition unit, and a second acquisition unit. The first acquisition unit acquires first ultrasound data representing multiple frames consecutive in the time direction, obtained by performing an ultrasound scan on a subject via the ultrasound probe. The second acquisition unit inputs data based on the first ultrasound data into a trained model, and acquires second ultrasound data representing a single frame from the trained model, the second ultrasound data having a higher spatial resolution of fluid in the subject than the first ultrasound data. [Brief explanation of the drawings]

[0010] [Figure 1] FIG. 1 is a block diagram showing the configuration of an ultrasonic diagnostic apparatus according to an embodiment. [Figure 2] FIG. 2 is a diagram for explaining the Rayleigh distribution. [Figure 3] FIG. 3 is a diagram illustrating the processing procedure up to obtaining a high-resolution signal. [Figure 4] FIG. 4 is a diagram for explaining the processing performed by the medical image processing apparatus according to the first embodiment in the learning stage. [Figure 5] FIG. 5 is a diagram for explaining the processing performed by the medical image processing apparatus according to the first embodiment. [Figure 6A] FIG. 6A shows a conventional ultrasound image. [Figure 6B] FIG. 6B is a diagram showing an ultrasound image represented by the second ultrasound data. [Figure 7] FIG. 7 is a diagram for explaining the processing performed by the medical image processing apparatus according to the second embodiment. [Figure 8] FIG. 8 is a diagram for explaining the processing performed by the medical image processing apparatus according to the third embodiment. [Figure 9] FIG. 9 is a diagram for explaining the processing performed by the medical image processing apparatus according to the fourth embodiment. [Figure 10] FIG. 10 is a diagram for explaining the processing performed by the medical image processing apparatus according to the fifth embodiment. [Figure 11] FIG. 11 is a diagram for explaining the processing performed by the medical image processing apparatus according to the sixth embodiment. [Figure 12] FIG. 12 is a diagram for explaining the processing performed by the medical image processing apparatus according to the seventh embodiment. [Figure 13] FIG. 13 is a diagram for explaining the processing performed by the medical image processing apparatus according to the eighth embodiment. [Figure 14] FIG. 14 is a diagram for explaining the processing performed by the medical image processing apparatus according to the eighth embodiment. [Figure 15] FIG. 15 is a diagram for explaining the processing performed by the medical image processing apparatus according to the eighth embodiment. [Figure 16] FIG. 16 is a diagram for explaining the processing performed by the medical image processing apparatus according to the eighth embodiment. [Figure 17] FIG. 17 is a diagram for explaining the processing performed by the medical image processing apparatus according to the eighth embodiment. DETAILED DESCRIPTION OF THE INVENTION

[0011] (First embodiment) Hereinafter, embodiments of a medical information processing apparatus, an ultrasound diagnostic apparatus, and a program will be described in detail with reference to the drawings.

[0012] The ultrasound diagnostic apparatus according to this embodiment causes an ultrasound probe to perform an ultrasound scan and collects multiple frames of data (multiple frames of data within a predetermined time period) that are consecutive in the time direction and obtained by the execution of the ultrasound scan. The frame data is collected at a predetermined frame rate by the execution of the ultrasound scan. The frame data refers to any of received data, measurement data, blood flow data, and tissue data. The received data is, for example, a received signal of ultrasound received by an ultrasound probe (e.g., channel data). The measurement data is data (e.g., IQ data) obtained by performing phased summation and quadrature detection processing on the received ultrasound signal. The blood flow data is data (e.g., power Doppler signal data) in which information derived from blood flow in the measurement data has been extracted or emphasized. The tissue data is data (e.g., B-mode data) in which information derived from tissue has been extracted or emphasized.

[0013] The measurement data includes, for example, information derived from tissue (tissue signal components (clutter)) and information derived from blood flow (blood flow signal components). The information derived from blood flow may include not only information derived from blood but also information derived from contrast agents in blood. Furthermore, the blood flow data is data in which information derived from blood flow is extracted or emphasized, and includes blood flow velocity values, variance values, and power values.

[0014] Extracting information derived from blood flow is, for example, an operation of extracting blood flow signal components from measurement data. Emphasizing information derived from blood flow is, for example, a process of making blood flow signal components more prominent relative to tissue signal components. Note that blood flow data may be obtained by a process of extracting or emphasizing information derived from blood flow, or by a process of removing or reducing information derived from tissue.

[0015] Although the present embodiment will be described taking an ultrasound diagnostic device as an example, the present invention may be applied to modalities (medical information processing devices) other than ultrasound diagnostic devices. For example, the present invention may be applied to medical information processing devices such as workstations and servers that acquire ultrasound data obtained based on the results of ultrasound scans of a subject.

[0016] FIG. 1 is a block diagram showing the configuration of an ultrasound diagnostic apparatus according to an embodiment. The ultrasound diagnostic apparatus 10 is an apparatus that generates ultrasound data based on received signals (reflected wave signals) received from an ultrasound probe 5. The ultrasound diagnostic apparatus 10 shown in FIG. 1 is an apparatus that can generate two-dimensional ultrasound data based on two-dimensional received signals and three-dimensional ultrasound data based on three-dimensional received signals. However, the embodiment is also applicable to cases where the ultrasound diagnostic apparatus 10 is an apparatus dedicated to two-dimensional data. The ultrasound diagnostic apparatus 10 includes a transmission circuit 9, a reception circuit 11, and a medical information processing apparatus 100.

[0017] The ultrasonic probe 5 is, for example, an electronic scanning probe, and has a plurality of transducers 101 arranged one-dimensionally or two-dimensionally at its tip. The transducers 101 are piezoelectric elements (electromechanical transducers) that convert between electrical signals (voltage pulse signals) and ultrasound waves (acoustic waves). The ultrasonic probe 5 transmits ultrasound waves from the plurality of transducers 101 to a subject and receives reflected ultrasound waves from the subject via the plurality of transducers 101. The reflected acoustic waves reflect differences in acoustic impedance within the subject. When a transmitted ultrasound pulse is reflected by the surface of a moving blood flow, heart wall, or the like, the reflected ultrasound undergoes a frequency shift due to the Doppler effect, depending on the velocity signal component of the moving object relative to the ultrasound transmission direction.

[0018] The probe connection unit 103 connects to the ultrasonic probe 5 and transmits and receives ultrasonic waves to and from the ultrasonic probe 5. The connection means of the ultrasonic probe 5 by the probe connection unit 103 may be either wired or wireless. In the wired case, the probe connection unit 5 has a connector unit (receptacle) for connecting the connector (plug) of the ultrasonic probe 5. In the wireless case, it has a communication unit for wireless communication with the ultrasonic probe 5.

[0019] The transmission circuit 9 is a transmission unit that outputs pulse signals (drive signals) to the multiple transducers 101. By applying pulse signals to the multiple transducers 101 with a time difference, ultrasonic waves with different delay times are transmitted from the multiple transducers 101, thereby forming a transmitted ultrasonic beam. The direction and focus of the transmitted ultrasonic beam can be controlled by selectively changing the transducer 101 to which the pulse signal is applied (i.e., the transducer 101 to be driven) or by changing the delay time (application timing) of the pulse signal. By sequentially changing the direction and focus of this transmitted ultrasonic beam, an observation area inside the subject is scanned. Furthermore, by changing the delay time of the pulse signal, a transmitted ultrasonic beam that is a plane wave (focused at a distance) or a diverging wave (focus point is in the opposite direction of the ultrasonic transmission direction for the multiple transducers 101) may be formed. Alternatively, a transmitted ultrasonic beam may be formed using one transducer or some of the multiple transducers 101. The transmission circuit 9 transmits a pulse signal with a predetermined drive waveform to the transducer 101, causing the transducer 101 to generate a transmitted ultrasonic wave having a predetermined transmission waveform.

[0020] The receiving circuit 11 is a receiving unit that inputs, as a received signal, an electrical signal output from the transducer 101 that has received reflected ultrasound. The received signal is input to the processing circuit 110. In this embodiment, the analog signal output from the transducer 101 and the digital data obtained by sampling (digital conversion) the analog signal are both referred to as the received signal without any particular distinction. However, depending on the context, the received signal may also be referred to as received data or measured data to clearly indicate that it is digital data.

[0021] The medical information processing device 100 is connected to a transmission circuit 9 and a reception circuit 11, and processes signals received from the reception circuit 11 and controls the transmission circuit 9. The medical information processing device 100 includes a processing circuit 110, a memory 132, an input device 134, and a display 135.

[0022] The memory 132 is composed of a semiconductor memory element such as a RAM (Random Access Memory), a flash memory, a hard disk, an optical disk, etc. The memory 132 is a memory for storing data such as image data for display generated by the processing circuit 110. The memory 132 can also store received signals (reflected wave signals) output by the receiving circuit 11. In addition, the memory 132 stores, as necessary, control programs for transmitting and receiving ultrasound, image processing, and display processing, as well as various data such as diagnostic information (e.g., patient ID, doctor's findings, etc.), diagnostic protocols, and various body marks.

[0023] The input device 134 receives various instructions and information input from an operator and is configured from input interface devices such as a mouse, a keyboard, buttons, and a trackball.

[0024] The display 135 displays a GUI (Graphical User Interface) for receiving input of imaging conditions and various images under the control of the processing circuitry 110. The display 135 is configured by a display interface device such as a liquid crystal display, for example.

[0025] The processing circuitry 110 controls each component of the ultrasound diagnostic apparatus 10, thereby controlling the entire ultrasound diagnostic apparatus 10. Although the processing circuitry 110 is described as being implemented as a single circuit in Fig. 1, it may also be implemented as multiple circuits by combining multiple independent processors. Furthermore, it may be configured as an independent circuit dedicated to a specific function, such as an ASIC (Application Specific Integrated Circuit).

[0026] Furthermore, the term "processor" used in the above description refers to circuits such as a CPU (Central Processing Unit), a GPU (Graphical Processing Unit), an Application Specific Integrated Circuit (ASIC), a programmable logic device (e.g., a Simple Programmable Logic Device (SPLD), a Complex Programmable Logic Device (CPLD), and a Field Programmable Gate Array (FPGA)). The processor realizes its functions by reading and executing programs stored in memory 132.

[0027] The processing circuitry 110 causes the ultrasonic probe 5 to perform an ultrasonic scan and collects multiple frames of data (multiple frame data within a predetermined time) that are consecutive in the time direction and obtained by the execution of the ultrasonic scan. The processing circuitry 110 performs delay-and-sum processing and quadrature detection processing on the received signals (channel data) collected via the receiving circuitry 11. The delay-and-sum processing is a process of adding together received signals from multiple transducers 101 by changing the delay time and weight for each transducer 101, and is also called delay-and-sum (DAS) beamforming. The quadrature detection processing is a process of converting received signals into in-phase signals and quadrature signals (IQ data (measurement data)) in the baseband. Note that other processes such as adaptive beamforming, model-based processing, and machine learning may also be performed on the received signals.

[0028] Furthermore, the processing circuitry 110 may estimate the amount of tissue displacement due to the subject's body movement or the like between multiple frame data, and correct each frame data based on the estimated result. Specifically, the processing circuitry 110 calculates the amount of tissue displacement due to the subject's body movement or the like between frames from multiple frame data. The processing circuitry 110 corrects the frame data based on the calculated amount of displacement.

[0029] The processing circuitry 110 also performs envelope detection processing, logarithmic compression processing, etc. to generate B-mode data (data in which tissue-derived information is extracted or emphasized) that represents the signal intensity at each point in the observation region as brightness. The processing circuitry 110 also generates blood flow data (power Doppler signal data) in which blood flow-derived information from the measurement data is extracted or emphasized.

[0030] For example, the processing circuitry 110 applies an MTI (Moving Target Indicator) filter to multiple frame data. This reduces information (tissue signal components (clutter)) derived from tissues that are stationary or move little between frames, and extracts information derived from blood flow (blood flow signal components). The MTI filter may be a filter with fixed filter coefficients, such as a Butterworth-type IIR (Infinite Impulse Response) filter or a Polynomial Regression Filter. The MTI filter may also be an adaptive filter that changes its coefficients according to the input signal using eigenvalue decomposition or singular value decomposition.

[0031] The processing circuitry 110 may also decompose the frame data into multiple bases using eigenvalue decomposition or singular value decomposition, remove tissue-derived information by extracting a specific base, and extract information derived from blood flow. The processing circuitry 110 may also use a method such as vector Doppler, speckle tracking, or vector flow mapping to obtain a velocity vector for each coordinate in the received signal data and obtain a blood flow vector representing the magnitude and direction of blood flow. In addition to the methods exemplified here, any method may be used as long as it can extract or emphasize blood flow-derived information contained in the frame data (received data, measured data) or remove or reduce tissue-derived information.

[0032] Furthermore, the processing circuit 110 reads and executes a program stored in the memory 132, thereby causing the first acquisition function 111, the extraction function 112, and the second acquisition function 113 to function and performing processing to output high-resolution ultrasound data.

[0033] Before describing the processing using the ultrasound diagnostic device 10 configured as above, a method for generating high-resolution ultrasound data according to the prior art (for example, Non-Patent Document 1) will be described.

[0034] The amplitude distribution characteristic 50 of the speckle pattern of blood flow data (for example, an IQ signal after clutter removal) is roughly approximated by a probability distribution called a Rayleigh distribution, as shown in Figure 2. A conventional method for generating high-resolution data (high-resolution signal) extracts, from a plurality of blood flow data corresponding to each of a plurality of frame data, high-intensity (high-resolution) and spatially sparse signals (points 51 shown in Figure 2) that have a low occurrence probability but a large signal value, and then synthesizes (integrates, adds, etc.) these to generate a high-resolution signal.

[0035] Specifically, as shown in FIG. 3, during the period from when an ultrasound scan is performed on a subject via an ultrasound probe to when a high-resolution signal 24 is obtained, the following processes are performed to obtain (1) an IQ signal 20 before clutter removal, (2) a blood flow signal 21 (IQ signal), (3) a power Doppler signal 22, and (4) an extracted signal 23.

[0036] The IQ signal 20 is obtained by performing delay-and-sum and quadrature detection processing on received signals (ultrasound reflected wave signals) representing a plurality of frames 20a, 20b, 20c, 20d, and 20e obtained by ultrasound scanning of the subject.

[0037] The blood flow signal 21 is a signal obtained by removing clutter from the IQ signal 20, and is blood flow data obtained by applying a WF (Wall Filter) to the IQ signal 20, for example. The blood flow data is a Doppler signal that represents the velocity, dispersion, and power of a fluid such as blood flow. The blood flow signal 21 corresponds to each of multiple frames 20a, 20b, 20c, 20d, and 20e. The blood flow signal 21 may also be obtained by applying an MTI filter to the IQ signal 20.

[0038] The power Doppler signal 22 is obtained by, for example, calculating the absolute value of an IQ signal (Doppler data) as the blood flow signal 21. The power Doppler signal 22 corresponds to each of the plurality of frames 20a, 20b, 20c, 20d, and 20e.

[0039] The extracted signal 23 is a signal representing an object and is obtained from the power Doppler signal 22. The object may be, for example, a blood flow, which has a low occurrence probability but a large signal value (amplitude value). The extracted signal 23 corresponds to each of multiple frames 20a, 20b, 20c, 20d, and 20e. The extracted signal 23 is a high-intensity (high-resolution) and spatially sparse signal. Specifically, the extracted signal 23 is obtained by peak sharpening, which applies a nonlinear function to each position (pixel value) within the frame represented by the power Doppler signal 22. In such a case, for example, a process may be performed to increase the pixel density and improve the effective resolution by resampling the power Doppler signal 22. The resampling is, for example, a process of replacing each pixel with multiple smaller pixels, and the signal value corresponding to each replaced pixel may be interpolated from the original signal value (the signal value before replacement) by, for example, bicubic interpolation or the like. Furthermore, each position (pixel value) within the frame represented by the resampled power Doppler signal 22 may be raised to a power (e.g., 8th power, 12th power, etc.) to sharpen the signal value corresponding to each pixel representing the object. This makes the signal value corresponding to the pixel representing the object more prominent than the signal values ​​corresponding to the surrounding pixels, and an extracted signal 23 representing a high-intensity, spatially sparse object is obtained from the power Doppler signal 22. The extracted signal 23 may also be referred to as a local peak signal. Furthermore, although the object in this embodiment is described using blood flow as an example, the object may also be internal tissue, contrast agent, or the like.

[0040] The high-resolution signal 24 is obtained by synthesizing (accumulating, adding, etc.) the extracted signals 23 (local peak signals) corresponding to the multiple frames 20a, 20b, 20c, 20d, and 20e. The high-resolution signal 24 represents a single frame. The high-resolution signal 24 has a higher spatial resolution of the fluid within the subject than the IQ signal 20, the blood flow signal 21, the power Doppler signal 22, and the extracted signal 23, and is a signal that provides high definition (high-definition depiction) of the fluid flowing within the subject. In particular, obtaining the extracted signal 23 requires a large load of the signal processing described above, which is time-consuming.

[0041] As described above, the conventional techniques require multiple processes before obtaining the high-resolution signal 24. Therefore, when a large number of frames are used, the real-time performance of high-resolution ultrasound data may be impaired. For example, the number of frames required to obtain the high-resolution signal 24 is several thousand to several tens of thousands, resulting in poor real-time performance.

[0042] The ultrasound diagnostic device 10 according to this embodiment outputs high-resolution ultrasound data using a trained model based on AI (Artificial Intelligence), thereby reducing the amount of calculation and ensuring the real-time performance required for ultrasound diagnosis.

[0043] The processing in the learning stage will be described below. In the learning stage, the neural network is trained using the input data 300 and the high-resolution data 400 as output data as a set of training data. The neural network is trained using a set of data sets, including a signal (corresponding to data based on the first ultrasound data) corresponding to at least one of the above-mentioned (1) IQ signal 20 before clutter removal, (2) blood flow signal 21 (IQ signal after clutter removal), (3) power Doppler signal 22, and (4) extracted signal 23, and a signal corresponding to the high-resolution signal 24 (corresponding to the second ultrasound data). For example, as shown in FIG. 4, the neural network is trained using the power Doppler signal 22 representing multiple frames as input data and the high-resolution data (high-resolution signal 24) representing a single frame as output data. In the neural network training, the weighting coefficients of the trained model 200 are determined. A learning function (not shown) determines the weighting coefficients by calculating the gradient of the weights using back error propagation so as to minimize a given objective function, for example.

[0044] Examples of the neural network of the trained model 200 include a perceptron, a single-layer neural network or a deep neural network using a CNN, a U-net, a transformer, etc. Furthermore, since the power Doppler signals 22 are data consisting of multiple frames, the trained model 200 may be a time-series neural network such as an RNN.

[0045] When the trained model 200 is trained in this manner, the processing circuit 110 stores the determined weight coefficients of the neural network in the memory 132. Note that the learning stage processing is similar in embodiments other than the first embodiment, and in each embodiment, the high-quality image data 400 serving as output data is used as training data in the same manner, but the training data used as input data 300 differs in each embodiment.

[0046] Next, the processing in the operation stage (inference stage) will be described with reference to Fig. 5. The first acquisition function 111 (first acquisition unit) acquires first ultrasound data obtained based on the results of an ultrasound scan on a subject. That is, the first acquisition function 111 acquires first ultrasound data representing multiple frames that are consecutive in the time direction and that are obtained by performing an ultrasound scan on a subject via the ultrasound probe 5. The first ultrasound data corresponds to received signals (ultrasound reflected wave signals) representing the multiple frames.

[0047] The extraction function 112 acquires IQ signals 20 representing multiple frames by performing phasing addition and quadrature detection processing on received signals (ultrasound reflected wave signals) representing multiple frames. The extraction function 112 also acquires blood flow signals 21 representing multiple frames by applying a WF (Wall Filter) to the IQ signals 20. The extraction function 112 also acquires power Doppler signals 22 representing multiple frames by calculating the absolute value of the IQ signals (Doppler data) as the blood flow signals 21.

[0048] Then, the second acquisition function 113 inputs data based on the first ultrasound data to the trained model 200. Here, the second acquisition function 113 inputs the power Doppler signal 22 (power Doppler data) corresponding to multiple frames acquired by the extraction function 112 to the trained model 200 as data based on the first ultrasound data. In other words, the power Doppler signal 22 representing multiple frames becomes the input 70 of the trained model 200. The second acquisition function 113 also acquires second ultrasound data from the trained model 200, which has a higher spatial resolution of the fluid in the subject than the first ultrasound data and represents a single frame. The second acquisition function 113 acquires a high-resolution signal 24 (second ultrasound data) as the output of the trained model 200. The high-resolution signal 24 corresponds to a signal obtained by combining (accumulating, adding, etc.) the local peak signals of the power Doppler signal 22 representing the multiple frames input to the trained model 200, and represents a single frame. The high-resolution signal 24 has a higher spatial resolution of the fluid within the subject than the received signal, the IQ signal 20, the blood flow signal 21, and the power Doppler signal 22. For example, FIG. 6A shows an ultrasound image based on the power Doppler signal 22, and FIG. 6B shows an ultrasound image based on the high-resolution signal 24. As shown in FIGS. 6A and 6B, the spatial resolution of the fluid (blood flow) within the subject is higher in the ultrasound image based on the high-resolution signal 24 than in the ultrasound image based on the power Doppler signal 22.

[0049] As described above, in the first embodiment, the power Doppler signal 22 representing multiple frames is input to the trained model 200 as data based on the first ultrasound data, and the high-resolution signal 24 representing a single frame, which has a higher spatial resolution of the fluid in the subject than the first ultrasound data, is acquired from the trained model 200. This eliminates the need for the process of extracting the extracted signal 23 for each frame from the power Doppler signal 22 shown in FIG. 3 , thereby reducing the amount of calculation required to obtain the high-resolution signal 24 and ensuring real-time performance of the high-resolution signal 24. In other words, the ultrasound diagnostic apparatus 10 according to the embodiment can obtain a high-resolution signal with a smaller calculation load than when performing normal processing on each frame.

[0050] (Second embodiment) In the first embodiment, a case where a power Doppler signal 22 (power Doppler data) corresponding to a plurality of frames is used as input data for the trained model 200 has been described. However, the input data for the trained model 200 is not limited to this. For example, the second acquisition function 113 may input at least one of the following, corresponding to a plurality of frames: (1) IQ data before application of a wall filter (data corresponding to the IQ signal 20 shown in FIG. 3 ); (2) Doppler data which is the IQ data after application of a wall filter (data corresponding to the blood flow signal 21 shown in FIG. 3 ); (3) power Doppler data representing the absolute value of the Doppler data (data corresponding to the power Doppler signal 22 shown in FIG. 3 ); and (4) data obtained by performing local peak extraction processing on the power Doppler data (data corresponding to the extraction signal 23 shown in FIG. 3 ), as first ultrasound data to the trained model 200. In the second embodiment, a case where the trained model 200 is a trained model trained using the blood flow signal 21 as input data will be described. Using FIG. 7, a process for generating a high-resolution signal 24 performed by the ultrasound diagnostic apparatus 10 according to the second embodiment will be described. Note that repeated explanations of processes similar to those in the previous embodiments will be omitted.

[0051] In the second embodiment, the first acquisition function 111 acquires first ultrasound data obtained based on the results of an ultrasound scan of a subject, and the second acquisition function 113 inputs data obtained based on the first ultrasound data into the trained model to acquire second ultrasound data with higher resolution than the first ultrasound data from the trained model. Here, the first ultrasound data obtained based on the results of an ultrasound scan of a subject is, for example, an IQ signal 20 before clutter removal, and the second ultrasound data is, for example, a high-resolution signal 24. Also, in the second embodiment, as shown in FIG. 7, the data obtained based on the first ultrasound data and input to the trained model 200 is a blood flow signal 21. That is, in the second embodiment, the trained model 200 is a trained model trained using the blood flow signal 21 as input data.

[0052] The following describes the processing in the learning stage when the trained model 200 is a neural network. In this case, the trained model 200 uses a blood flow signal 21 and a high-resolution signal 24 corresponding to the blood flow signal 21 as one set of data, and performs neural network training using data consisting of multiple sets as a data set of training data to be used for training. Here, training the neural network means determining the weight coefficients of the trained model 200, and the processing circuitry 110 determines the weight coefficients by using a learning function (not shown) to calculate the gradient of the weights using back error propagation so as to minimize a given objective function, for example.

[0053] Once the trained model 200 has been trained in this manner, the processing circuit 110 stores the determined weighting coefficients of the neural network in the memory 132.

[0054] Next, the processing during execution of the trained model 200 will be described. The processing circuit 110 calls the weighting coefficients of the neural network determined by the first acquisition function 111 from the memory 132. The second acquisition function 113 inputs the blood flow signal 21 to the trained model 200, and acquires, as the output result of the trained model 200, a high-resolution signal 24, which is ultrasound data equivalent to data after a synthesis process of data that has been subjected to a local peak extraction process corresponding to multiple frames, from the trained model as second ultrasound data.

[0055] As described above, in the second embodiment, the process of obtaining the high-resolution signal 24 from the blood flow signal 21 via the power Doppler signal 22 and the extraction signal 23 is performed using the trained model 200, which inputs the blood flow signal 21 and outputs the high-resolution signal 24. In this case, the second acquisition function 113 inputs Doppler data, which is IQ data after application of a wall filter, to the trained model 200 as data based on the first ultrasound data. This allows the process of generating the power Doppler signal 22 and the extraction signal 23 to be omitted, thereby reducing the amount of calculation and ensuring the real-time performance of the ultrasound diagnostic device 10. In other words, a high-resolution signal can be obtained with a smaller calculation load than when performing normal processing on each frame. In addition, in the second embodiment, the trained model 200 inputs the blood flow signal 21, which includes phase information, and generates the high-resolution signal 24. This allows the processing circuitry 110 to obtain the high-resolution signal 24 by utilizing the phase information.

[0056] (Third embodiment) In the second embodiment, a case where a blood flow signal 21 is used as input data for the trained model 200 has been described. However, the embodiment is not limited to this. In the third embodiment, a case where the trained model 200 is a trained model trained using an IQ signal 20 before clutter removal as input data will be described. Using FIG. 8, a process for generating a high-resolution signal 24 performed by an ultrasound diagnostic apparatus 10 according to the third embodiment will be described. Note that repeated description of processes similar to those in the previous embodiments will be omitted.

[0057] In the third embodiment, the first acquisition function 111 acquires first ultrasound data obtained based on the results of an ultrasound scan of the subject, and the second acquisition function 113 inputs data obtained based on the first ultrasound data into the trained model to acquire second ultrasound data, which is more precise than the first ultrasound data, from the trained model. Here, in the third embodiment, the first ultrasound data obtained based on the results of an ultrasound scan of the subject is, for example, the IQ signal 20 before clutter removal (data corresponding to the IQ signal 20 shown in FIG. 3), and the second ultrasound data is, for example, the high-resolution signal 24. Also, in the third embodiment, as shown in FIG. 8, the data obtained based on the first ultrasound data and input to the trained model 200 is, for example, the IQ signal 20 before clutter removal, i.e., a blood flow IQ signal. That is, in the third embodiment, the trained model 200 is a trained model trained using the IQ signal 20 before clutter removal as input data.

[0058] The following describes the processing in the learning stage when the trained model 200 is a neural network. In this case, the trained model 200 performs neural network training using a set of data consisting of an IQ signal 20 before clutter removal and a high-resolution signal 24 corresponding to the IQ signal 20 before clutter removal as one set of data, and multiple sets of data as a data set of training data to be used for training. The processing circuitry 110 determines weight coefficients by using a learning function (not shown) to calculate the gradient of the weights using back error propagation so as to minimize a given objective function, for example.

[0059] Once the trained model 200 has been trained in this manner, the processing circuit 110 stores the determined weighting coefficients of the neural network in the memory 132.

[0060] Next, the processing during execution of the trained model 200 will be described. The processing circuit 110 calls the weighting coefficients of the neural network determined by the first acquisition function 111 from the memory 132. The second acquisition function 113 inputs the IQ signal 20 before clutter removal into the trained model 200, and acquires, as the output result of the trained model 200, a high-resolution signal 24, which is ultrasound data equivalent to data after a synthesis process of data that has been subjected to a local peak extraction process corresponding to multiple frames, from the trained model as second ultrasound data.

[0061] As described above, in the third embodiment, the process of obtaining the high-resolution signal 24 from the IQ signal 20 before clutter removal via the blood flow signal 21, power Doppler signal 22, and extraction signal 23 is performed using the trained model 200, which inputs the IQ signal 20 before clutter removal and outputs the high-resolution signal 24. In this case, the second acquisition function 113 inputs the IQ data before application of the wall filter to the trained model as the first ultrasound data. This allows the process of generating the blood flow signal 21, power Doppler signal 22, and extraction signal 23 to be omitted, thereby reducing the amount of calculation and ensuring the real-time performance of the ultrasound diagnostic device 10. In other words, a high-resolution signal can be obtained with a smaller calculation load than when performing normal processing on each frame. In addition, in the third embodiment, the trained model 200 inputs the IQ signal 20 before clutter removal and generates the high-resolution signal 24. This allows the processing circuitry 110 to incorporate clutter removal processing into the processing of the trained model 200.

[0062] (Fourth embodiment) In the first to third embodiments, a case where one type of data is used as input data to the trained model 200 has been described. However, the embodiments are not limited to this. In the fourth embodiment, a case where multiple types of data are used as the input 70 to the trained model 200 will be described. That is, the second acquisition function 113 may input, as first ultrasound data to the trained model 200, data that combines two or more of the following, which correspond to multiple frames: (1) IQ data before application of a wall filter (data corresponding to the IQ signal 20 shown in FIG. 3), (2) Doppler data that is IQ data after application of a wall filter (data corresponding to the blood flow signal 21 shown in FIG. 3), (3) power Doppler data that represents the absolute value of the Doppler data (data corresponding to the power Doppler signal 22 shown in FIG. 3), and (4) data obtained by performing local peak extraction processing on the power Doppler data (data corresponding to the extraction signal 23 shown in FIG. 3). In the fourth embodiment, a case where the trained model 200 is a trained model trained using the power Doppler signal 22 and the blood flow signal 21 as input data will be described. The process of generating a high-resolution signal 24 performed by the ultrasound diagnostic apparatus 10 according to the fourth embodiment will be described with reference to Fig. 9. Note that repeated description of the same processes as those in the previous embodiments will be omitted.

[0063] In the fourth embodiment, the first acquisition function 111 acquires first ultrasound data obtained based on the results of an ultrasound scan of a subject, and the second acquisition function 113 inputs data obtained based on the first ultrasound data into the trained model to acquire second ultrasound data, which is more precise than the first ultrasound data, from the trained model. In the fourth embodiment, the first ultrasound data obtained based on the results of an ultrasound scan of a subject is, for example, an IQ signal 20 before clutter removal, and the second ultrasound data is, for example, a high-resolution signal 24. Also, in the fourth embodiment, as shown in FIG. 9 , the data obtained based on the first ultrasound data and input to the trained model 200 is, for example, a blood flow signal 21 and a power Doppler signal 22. That is, in the fourth embodiment, the trained model 200 is a trained model trained using the blood flow signal 21 and the power Doppler signal 22 as input data.

[0064] The following describes the processing in the learning stage when the trained model 200 is a neural network. In this case, the trained model 200 uses a set of data consisting of a blood flow signal 21, a power Doppler signal 22, and a high-resolution signal 24 corresponding to the blood flow signal 21 and the power Doppler signal 22 as one set of data, and performs neural network training using a set of training data to be used for training. The processing circuitry 110 determines weight coefficients by using a learning function (not shown) to calculate the gradient of the weights using back error propagation so as to minimize a given objective function, for example.

[0065] Here, an example of the neural network configuration of the trained model 200 may be, for example, a first layer connected to an input layer related to the blood flow signal 21, a second layer connected to an input layer related to the power Doppler signal 22, a third layer connected to the first and second layers, and so on, where n is a natural number greater than or equal to 3, and the n+1th layer may be connected to the nth layer.

[0066] As another example of the configuration of the neural network of the trained model 200, for example, the first layer may be connected to one of the input layer for the blood flow signal 21 and the input layer for the power Doppler signal 22, and the n+1th layer may be connected to the nth layer, where n is a natural number greater than or equal to 2, and at the same time, the other of the input layer for the blood flow signal 21 and the input layer for the power Doppler signal 22 may be connected to all layers.

[0067] As another example of the neural network configuration of the trained model 200, both the input layer for the blood flow signal 21 and the input layer for the power Doppler signal 22 may be connected to the first layer of the hidden layer.

[0068] Once the trained model 200 has been trained, the processing circuit 110 stores the determined neural network weight coefficients in the memory 132.

[0069] Next, the processing during execution of the trained model 200 will be described. The processing circuit 110 calls up the weighting coefficients of the neural network determined by the first acquisition function 111 from the memory 132. The second acquisition function 113 inputs the blood flow signal 21 and the power Doppler signal 22 as input 70 of the trained model 200, and acquires, as the output result of the trained model 200, a high-resolution signal 24, which is ultrasound data equivalent to data after a synthesis process of data that has been subjected to a local peak extraction process corresponding to multiple frames, as second ultrasound data from the trained model.

[0070] As described above, in the fourth embodiment, a high-resolution signal is generated using a trained model 200 that receives as input 70 a blood flow signal 21 and a power Doppler signal 22 and outputs a high-resolution signal 24. In this case, the second acquisition function 113 inputs data that combines Doppler data and power Doppler data corresponding to multiple frames into the trained model as first ultrasound data. The fourth embodiment combines the advantages of the first and second embodiments, reduces the amount of calculation, and can utilize phase information, resulting in an output signal that is a high-resolution image.

[0071] (Fifth embodiment) In the fourth embodiment, the trained model 200 is described as a trained model trained using the power Doppler signal 22 and the blood flow signal 21 as the input 70. In the fifth embodiment, the trained model 200 may be a trained model trained using the IQ signal 20 before clutter removal, the power Doppler signal 22, and the blood flow signal 21 as input data. The generation process of the high-resolution signal 24 performed by the ultrasound diagnostic apparatus 10 according to the fifth embodiment will be described with reference to FIG. 10 .

[0072] In the fifth embodiment, the first acquisition function 111 acquires first ultrasound data obtained based on the results of an ultrasound scan of a subject, and the second acquisition function 113 inputs data obtained based on the first ultrasound data into the trained model to acquire second ultrasound data, which is more precise than the first ultrasound data, from the trained model. In the fifth embodiment, the first ultrasound data obtained based on the results of an ultrasound scan of a subject is, for example, an IQ signal 20 before clutter removal, and the second ultrasound data is, for example, a high-resolution signal 24. Also, in the fifth embodiment, as shown in FIG. 10 , the data obtained based on the first ultrasound data and input to the trained model 200 is, for example, the IQ signal 20 before clutter removal, a blood flow signal 21, and a power Doppler signal 22. In other words, in the fifth embodiment, the trained model 200 is a trained model trained using the IQ signal 20, blood flow signal 21, and power Doppler signal 22 before clutter removal as input 70.

[0073] The following describes the processing in the learning stage when the trained model 200 is a neural network. In this case, the trained model 200 uses a set of data consisting of an IQ signal 20 before clutter removal, a blood flow signal 21, a power Doppler signal 22, and a corresponding high-resolution signal 24 as a set of data, and performs neural network training using a set of training data to be used for training. The processing circuitry 110 determines weight coefficients by using a learning function (not shown) to calculate the gradient of the weights using back error propagation so as to minimize a given objective function, for example.

[0074] Here, an example of the neural network configuration of the trained model 200 may be, for example, a first layer connected to an input layer related to the blood flow signal 21, a second layer connected to an input layer related to the power Doppler signal 22, a third layer connected to an input layer related to the IQ signal 20 before clutter removal, a fourth layer connected to the first, second and third layers, and below, where n is a natural number greater than or equal to 4, the n+1th layer may be connected to the nth layer.

[0075] As another example of the configuration of the neural network of the trained model 200, for example, the first layer may be connected to one or more of the input layer for the IQ signal 20 before clutter removal, the input layer for the blood flow signal 21, and the input layer for the power Doppler signal 22, and the n+1th layer may be connected to the nth layer, where n is a natural number greater than or equal to 2, and at the same time, the other input layers among the input layer for the IQ signal 20 before clutter removal, the input layer for the blood flow signal 21, and the input layer for the power Doppler signal 22 may be connected to all layers.

[0076] As another example of the neural network configuration of the trained model 200, the input layer before clutter removal, the input layer relating to the blood flow signal 21, and the input layer relating to the power Doppler signal 22 may all be connected to the first layer of the hidden layer.

[0077] Once the trained model 200 has been trained, the processing circuit 110 stores the determined neural network weight coefficients in the memory 132.

[0078] Next, the processing during execution of the trained model 200 will be described. The processing circuit 110 calls up the weighting coefficients of the neural network determined by the first acquisition function 111 from the memory 132. The second acquisition function 113 inputs the IQ signal 20, blood flow signal 21, and power Doppler signal 22 before clutter removal as input 70 of the trained model 200, and acquires, as the output result of the trained model 200, a high-resolution signal 24, which is ultrasound data equivalent to data after a synthesis process of data that has been subjected to local peak extraction processing corresponding to multiple frames, as second ultrasound data from the trained model.

[0079] As described above, in the fifth embodiment, a high-resolution signal is generated using a trained model 200 that receives as input 70 the IQ signal 20, blood flow signal 21, and power Doppler signal 22 before clutter removal and outputs a high-resolution signal 24. In this case, the second acquisition function 113 inputs data that combines IQ data, Doppler data, and power Doppler data corresponding to multiple frames as first ultrasound data to the trained model. The fifth embodiment has the advantages of the first, second, and third embodiments, reduces the amount of calculation, can utilize phase information, and can include processing such as motion correction using tissue signals in the processing of the trained model 200.

[0080] (Sixth embodiment) In the fourth embodiment, a case has been described in which the trained model 200 is a trained model trained using a power Doppler signal 22 and a blood flow signal 21 as input 70. In the sixth embodiment, a case has been described in which the trained model 200 is a trained model trained using an IQ signal 20 and a blood flow signal 21 before clutter removal as input data. The generation process of a high-resolution signal 24 performed by an ultrasound diagnostic apparatus 10 according to the sixth embodiment will be described with reference to FIG. 11 .

[0081] In the sixth embodiment, the first acquisition function 111 acquires first ultrasound data obtained based on the results of an ultrasound scan of a subject, and the second acquisition function 113 inputs data obtained based on the first ultrasound data into the trained model to acquire second ultrasound data, which is more precise than the first ultrasound data, from the trained model. In the sixth embodiment, the first ultrasound data obtained based on the results of an ultrasound scan of a subject is, for example, an IQ signal 20 before clutter removal, and the second ultrasound data is, for example, a high-resolution signal 24. Also, in the sixth embodiment, as shown in FIG. 11 , the data obtained based on the first ultrasound data and input to the trained model 200 is, for example, the IQ signal 20 before clutter removal and the blood flow signal 21. That is, in the sixth embodiment, the trained model 200 is a trained model trained using the IQ signal 20 before clutter removal and the blood flow signal 21 as input 70.

[0082] The following describes the processing in the learning stage when the trained model 200 is a neural network. In this case, the trained model 200 uses a set of data consisting of an IQ signal 20 before clutter removal, a blood flow signal 21, and a corresponding high-resolution signal 24 as a set of data, and performs neural network training using a set of data consisting of multiple sets of training data used for training. The processing circuitry 110 determines weight coefficients by using a learning function (not shown) to calculate the gradient of the weights using back error propagation so as to minimize a given objective function, for example.

[0083] Here, an example of the neural network configuration of the trained model 200 may be, for example, a first layer connected to an input layer relating to the IQ signal 20 before clutter removal, a second layer connected to an input layer relating to the blood flow signal 21, a third layer connected to the first and second layers, and so on, where n is a natural number greater than or equal to 3, and the n+1th layer may be connected to the nth layer.

[0084] As another example of the configuration of the neural network of the trained model 200, for example, the first layer may be connected to one of the input layer relating to the IQ signal 20 before clutter removal and the input layer relating to the blood flow signal 21, and the n+1th layer may be connected to the nth layer, where n is a natural number greater than or equal to 2, and at the same time, the other of the input layer relating to the IQ signal 20 before clutter removal and the input layer relating to the blood flow signal 21 may be connected to all layers.

[0085] As another example of the neural network configuration of the trained model 200, both the input layer relating to the IQ signal 20 before clutter removal and the input layer relating to the blood flow signal 21 may be connected to the first layer of the hidden layer.

[0086] Once the trained model 200 has been trained, the processing circuit 110 stores the determined neural network weight coefficients in the memory 132.

[0087] Next, the processing during execution of the trained model 200 will be described. The processing circuit 110 calls up the weighting coefficients of the neural network determined by the first acquisition function 111 from the memory 132. The second acquisition function 113 inputs the IQ signal 20 and the blood flow signal 21 before clutter removal into the trained model 200, and acquires, as the output result of the trained model 200, a high-resolution signal 24, which is ultrasound data equivalent to data after a synthesis process of data that has been subjected to a local peak extraction process corresponding to multiple frames, as second ultrasound data from the trained model.

[0088] As described above, in the sixth embodiment, a high-resolution signal is generated using a trained model 200 that receives an IQ signal 20 and a blood flow signal 21 before clutter removal and outputs a high-resolution signal 24. In this case, the second acquisition function 113 inputs data that combines IQ data and Doppler data corresponding to multiple frames as first ultrasound data to the trained model. The sixth embodiment has the advantages of the second and third embodiments, and can utilize phase information and can include motion correction using tissue signals in the processing of the trained model 200.

[0089] (Seventh embodiment) In the seventh embodiment, a case will be described in which the trained model 200 is a trained model trained using an extracted signal obtained by performing peak sharpening processing on a power Doppler signal as input data. The generation processing of a high-resolution signal 24 performed by an ultrasound diagnostic apparatus 10 according to the seventh embodiment will be described with reference to FIG. 12 .

[0090] In the seventh embodiment, the first acquisition function 111 acquires first ultrasound data obtained based on the results of an ultrasound scan of a subject, and the second acquisition function 113 inputs data obtained based on the first ultrasound data into a trained model to acquire second ultrasound data with higher resolution than the first ultrasound data from the trained model. Here, the first ultrasound data obtained based on the results of an ultrasound scan of a subject is, for example, an IQ signal 20 before clutter removal, and the second ultrasound data is, for example, a high-resolution signal 24. Also, in the seventh embodiment, as shown in FIG. 12 , the data obtained based on the first ultrasound data and input to the trained model 200 is an extracted signal 23 obtained by performing peak sharpening processing on a power Doppler signal, such as multiplication by a power function. That is, in the seventh embodiment, the trained model 200 is a trained model trained using the extracted signal 23 as input data.

[0091] That is, in the seventh embodiment, the trained model 200 receives the extracted signal 23 for each frame as input and outputs the high-resolution signal 24. In this case, the second acquisition function 113 inputs data obtained by performing local peak extraction processing on the power Doppler data to the trained model as first ultrasound data. That is, the extracted signal 23 is input to the trained model 200 for each multiple frames. The high-resolution signal 24 is considered to be a linear sum of the extracted signals 23, and by optimizing the weighting coefficients using an AI-based model, it is possible to adaptively adjust the weight for each frame and obtain the high-resolution signal 24.

[0092] The following describes the processing in the learning stage when the trained model 200 is a neural network. In this case, the trained model 200 uses a set of data, consisting of an extracted signal 23 and a high-resolution signal 24 corresponding to the extracted signal 23, as a set of data, and performs neural network training using a plurality of sets of data as a data set of training data to be used for training. Here, training the neural network means determining the weight coefficients of the trained model 200, and the processing circuitry 110 determines the weight coefficients by using a learning function (not shown) to calculate the gradient of the weights using back error propagation so as to minimize a given objective function, for example.

[0093] Once the trained model 200 has been trained in this manner, the processing circuit 110 stores the determined weighting coefficients of the neural network in the memory 132.

[0094] Next, the processing during execution of the trained model 200 will be described. The processing circuit 110 calls the weighting coefficients of the neural network determined by the first acquisition function 111 from the memory 132. The second acquisition function 113 inputs the extraction signal 23 to the trained model 200, and acquires, as the output result of the trained model 200, a high-resolution signal 24, which is ultrasound data equivalent to data after a synthesis process of data that has been subjected to a local peak extraction process corresponding to multiple frames, as second ultrasound data from the trained model.

[0095] As described above, in the seventh embodiment, the trained model 200 is trained using the extracted signal 23 obtained by performing peak sharpening processing on a power Doppler signal as input data. This allows adaptive adjustment of the weight for each frame. Note that the peak sharpening processing is not limited to using multiplication by a power function, and may also be processing that extracts and emphasizes local peaks. In either case, the high-resolution signal 24 has a higher spatial resolution of the fluid than the extracted signal 23.

[0096] (Eighth embodiment) In the first embodiment, an example was described in which a signal (extracted signal 23) obtained by performing peak sharpening on a power Doppler signal 22 is used as input data (data based on the first ultrasound data) of the trained model 200, and a high-resolution signal 24 corresponding to a signal representing a single frame obtained by combining extracted signals 23 corresponding to multiple frames is used as output data (second ultrasound data) of the trained model 200. However, the data based on the first ultrasound data and the second ultrasound data may be other data. For example, as shown in FIG. 13, a power Doppler signal 22 representing multiple frames may be used as input data (data based on the first ultrasound data) of the trained model 200, and a high-resolution signal 25 corresponding to a high-resolution signal 25' representing a single frame obtained by performing the processing shown in FIG. 13 on the power Doppler signal 22 may be used as output data (second ultrasound data) of the trained model 200.

[0097] The processing shown in Fig. 13 will be described below. The power Doppler signal 22 is the same as in the first embodiment, and is a signal representing multiple frames obtained by calculating the absolute value of an IQ signal (Doppler data) as a blood flow signal. The power Doppler signals 22 are converted into a sum signal 221 by being added up for each multiple frames (every five frames in Fig. 13). The sum signal 221 represents a single frame with an improved S / N ratio compared to the power Doppler signal 22. The sum signal 221 is generated for multiple frames by power Doppler signals 22 that are continuous in the time direction (Fig. 13 shows a sum signal 221 representing three frames).

[0098] Then, identification information 222 corresponding to each frame is generated from the sum signal 221 representing the multiple frames. The identification information 222 is information that identifies, for each position in the frame represented by the sum signal 221, the magnitude relationship between the signal value corresponding to that position and the signal values ​​corresponding to the surrounding area of ​​that position. For example, as shown in FIG. 14 , a kernel 230 is placed at a position of interest in the frame (image) represented by the sum signal 221, and the magnitude relationship between the signal value corresponding to that position of interest and the signal values ​​corresponding to each position in the range in which the kernel 230 is placed (the surrounding area of ​​the position of interest) is compared. Then, for each position in the frame, information that identifies whether the signal value is greater than the signal values ​​corresponding to the surrounding area of ​​that position is acquired as the identification information 222.

[0099] Specifically, when a kernel 230 based on a 3×1 ratio is used, the center (second square) of the kernel 230 is positioned as a position of interest, and it is determined whether the signal value corresponding to this position is greater than the signal values ​​corresponding to the positions at both ends of the kernel 230 (the first and third squares). If the signal value corresponding to the position of interest is greater than the signal values ​​corresponding to the surrounding areas of the position, the extraction function 114 represents the position of interest as, for example, "1." On the other hand, if the signal value corresponding to the position of interest is smaller than the signal values ​​corresponding to the surrounding areas of the position, the extraction function 114 represents the position of interest as, for example, "0." By performing this process at each position in the frame, identification information 222 for the sum signal 221 is extracted. The identification information 222 is a binary image in which each position (each pixel) of the frame represented by the sum signal 221 is represented by "0" or "1," and is information (local peak position information) for identifying the position where fluid (blood flow) exists within a local range for each frame.

[0100] As shown in Fig. 15, a position information map 223 is generated by analyzing a plurality of pieces of identification information 222. The position information map 223 is information that statistically represents positions within frames where fluid is present within a time period equivalent to a plurality of frames (Fig. 13 shows three frames, and Fig. 15 shows nine frames). For example, the more times the plurality of pieces of identification information 222 show "1" at corresponding positions, the larger the numerical value at that position, and the more times the plurality of pieces of identification information 222 show "0", the smaller the numerical value at that position. In this way, a weighting coefficient is set at each position in the position information map 223.

[0101] Then, a high-definition signal 25' representing a single frame is generated based on an addition signal 221 representing the latest frame among the multiple frames and the position information map 223. The high-definition signal 25' has a higher spatial resolution of the fluid within the subject than the power Doppler signal 22, and is a signal that provides high definition (high-definition depiction) of the fluid flowing within the subject.

[0102] 13 requires multiple processes before obtaining the high-resolution signal 25'. Therefore, if a large number of frames are used, the real-time nature of high-resolution ultrasound data may be impaired. In particular, depending on the number of frames, it may take a long time to obtain the identification information 222 and the position information map 223, which may reduce the real-time nature of the high-resolution signal 25'.

[0103] The ultrasound diagnostic device 10 according to this embodiment outputs high-resolution ultrasound data using a trained model based on AI (Artificial Intelligence), thereby reducing the amount of calculation and ensuring the real-time performance required for ultrasound diagnosis.

[0104] The processing in the learning stage will be described below. Learning of the neural network is performed by using, as a data set, power Doppler signals 22 representing multiple frames as input data and high-definition data (high-resolution signal 25') representing a single frame as output data, as shown in FIG. 16. In learning of the neural network, weighting coefficients of the trained model 200 are determined. A learning function (not shown) determines the weighting coefficients by calculating the gradient of the weights using back error propagation so as to minimize a given objective function, for example.

[0105] Next, the processing in the operation stage (inference stage) will be described with reference to Fig. 17. The first acquisition function 111 (first acquisition unit) acquires first ultrasound data obtained based on the results of an ultrasound scan on a subject. The first ultrasound data is obtained by performing an ultrasound scan on the subject via the ultrasound probe 5, and represents multiple frames that are consecutive in the time direction. The first ultrasound data corresponds to received signals (ultrasound reflected wave signals) that represent the multiple frames.

[0106] The extraction function 112 acquires IQ signals representing multiple frames by performing phasing addition and quadrature detection processing on received signals (ultrasound reflected wave signals) representing multiple frames. The extraction function 112 also acquires blood flow signals representing multiple frames by applying a WF (Wall Filter) to the IQ signals. The extraction function 112 also acquires power Doppler signals 22 representing multiple frames by calculating the absolute value of the IQ signals (Doppler data) as blood flow signals.

[0107] Then, the second acquisition function 113 inputs the power Doppler signal 22 (data based on the first ultrasound data) into the trained model 200'. The second acquisition function 113 also acquires a high-definition signal 25 (second ultrasound data) as an output of the trained model 200. The high-definition signal 25 corresponds to the high-definition signal 25' obtained by the processing shown in FIG. 13. The ultrasound image generated based on the high-definition signal 25 has image quality equivalent to that of the ultrasound image shown in FIG. 6B, and has a higher spatial resolution of the fluid (blood flow) within the subject than the ultrasound image shown in FIG. 6A based on the power Doppler signal 22.

[0108] As described above, in the eighth embodiment, the power Doppler signal 22 representing multiple frames is input to the trained model 200 as data based on the first ultrasound data, and the high-resolution signal 25 representing a single frame, which has a higher spatial resolution of the fluid in the subject than the first ultrasound data, is acquired from the trained model 200. This reduces the amount of calculation required to obtain the high-resolution signal 25, ensuring real-time performance of the high-resolution signal 25. In other words, the ultrasound diagnostic apparatus 10 according to the embodiment can obtain a high-resolution signal with a smaller calculation load than when performing normal processing on each frame.

[0109] (Other embodiments) In the fourth to sixth embodiments, the second acquisition function 113 has been described as inputting data that combines two or more of (1) IQ data, (2) Doppler data, (3) power Doppler data, and (4) data that has been subjected to local peak extraction processing, which correspond to a plurality of frames, into the trained model 200 as first ultrasound data, but the embodiments are not limited to the above combinations. As an example, the second acquisition function 113 may input data that combines (1) the IQ data and (3) the power Doppler data, which correspond to a plurality of frames, into the trained model 200 as first ultrasound data.

[0110] The embodiment is not limited to the above example, and the trained model 200 may be a trained model trained using a signal prior to the IQ signal before clutter removal, such as an RF signal or channel data, as input data.

[0111] Furthermore, the first to eighth embodiments may be implemented after applying a storage process in the frame direction or the space direction to the IQ signal 20, the blood flow signal 21, the power Doppler signal 22, the extracted signal 23, and the high-resolution signal 24 before clutter removal. In this case, the trained model 200 is a trained model trained using data obtained by applying an interpolation process in the frame direction or the space direction as input data.

[0112] Although the case where the output data of the trained model 200 is the high-resolution signal 24 has been described, the embodiment is not limited to this. As an example, the output data of the trained model 200 may be differential data for generating the high-resolution signal 24, mask data indicating information about data points on which correction processing is performed to generate the high-resolution signal 24, or the like.

[0113] In the embodiment, the case where the trained model 200 is a neural network has been described, but the embodiment is not limited to this, and the processing circuit 110 may generate the high-resolution signal 24 based on the input data using machine learning such as linear regression, logistic regression, decision tree, gradient boosting, PCA, etc. instead of the trained model 200.

[0114] Furthermore, the first ultrasound data acquired by the first acquisition function 111 is not limited to data obtained based on a scan by the ultrasound diagnostic apparatus 10, but may be data received from an external workstation, for example.

[0115] Although the explanation so far has mainly focused on an example of generating high-resolution blood flow data, the method according to the embodiment can also generate high-resolution internal tissue data and contrast data.

[0116] According to at least one of the embodiments described above, it is possible to reduce the amount of calculation required for ultrasound data processing, or to obtain high-resolution ultrasound data.

[0117] With respect to the above embodiment, the following supplementary notes are disclosed as one aspect and optional features of the invention.

[0118] (Appendix 1) One aspect of the present invention provides an ultrasound diagnostic device comprising an ultrasound probe, a first acquisition unit, and a second acquisition unit. The first acquisition unit acquires first ultrasound data representing multiple frames consecutive in the time direction, obtained by performing an ultrasound scan on a subject via the ultrasound probe. The second acquisition unit inputs data based on the first ultrasound data into a trained model, and acquires second ultrasound data representing a single frame from the trained model, the second ultrasound data having a higher spatial resolution of fluid in the subject than the first ultrasound data.

[0119] (Appendix 2) The second acquisition unit may input at least one of (1) IQ data before application of a wall filter, (2) Doppler data which is IQ data after application of a wall filter, (3) power Doppler data which represents the absolute value of the Doppler data, and (4) data obtained by performing local peak extraction processing on the power Doppler data, which correspond to the plurality of frames, into the trained model as the first ultrasound data, and acquire from the trained model as the second ultrasound data ultrasound data which corresponds to data obtained by performing synthesis processing on the data obtained by performing local peak extraction processing on the plurality of frames.

[0120] (Appendix 3) The second acquisition unit may input data that combines two or more of (1) the IQ data, (2) the Doppler data, (3) the power Doppler data, and (4) data that has been subjected to the local peak extraction process, which correspond to the multiple frames, into the trained model as the first ultrasound data.

[0121] (Appendix 4) The second acquisition unit may input data that combines (1) the IQ data and (2) the Doppler data, which corresponds to the plurality of frames, into the trained model as the first ultrasound data.

[0122] (Appendix 5) The second acquisition unit may input data that combines (1) the IQ data and (3) the power Doppler data, which corresponds to the plurality of frames, into the trained model as the first ultrasound data.

[0123] (Appendix 6) The second acquisition unit may input data that combines (2) the Doppler data and (3) the power Doppler data, which corresponds to the plurality of frames, into the trained model as the first ultrasound data.

[0124] (Appendix 7) The second acquisition unit may input data that combines (1) the IQ data, (2) the Doppler data, and (3) the power Doppler data, corresponding to the multiple frames, into the trained model as the first ultrasound data.

[0125] (Appendix 8) One aspect of the present invention provides a medical image processing device including an ultrasound probe, a first acquisition unit, and a second acquisition unit. The first acquisition unit acquires first ultrasound data representing multiple frames consecutive in the time direction, obtained by performing an ultrasound scan on a subject via the ultrasound probe. The second acquisition unit inputs data based on the first ultrasound data into a trained model, and acquires second ultrasound data representing a single frame from the trained model, the second ultrasound data having a higher spatial resolution of fluid in the subject than the first ultrasound data.

[0126] (Appendix 9) A program provided in one aspect of the present invention causes a computer to perform the following process: acquire first ultrasound data representing multiple frames consecutive in time obtained by performing an ultrasound scan on a subject via an ultrasound probe; input data based on the first ultrasound data into a trained model; and acquire second ultrasound data representing a single frame from the trained model, the second ultrasound data having a higher spatial resolution of fluid within the subject than the first ultrasound data.

[0127] Although several embodiments have been described, these embodiments are presented as examples and are not intended to limit the scope of the invention. These embodiments can be implemented in various other forms, and various omissions, substitutions, modifications, and combinations of embodiments can be made without departing from the spirit of the invention. These embodiments and their modifications are included within the scope and spirit of the invention, as well as within the scope of the invention and its equivalents as defined in the claims. [Explanation of symbols]

[0128] 110 Processing circuit 111 First Acquisition Function 112 Extraction function 113 Second Acquisition Function

Claims

1. an ultrasound probe; a first acquisition unit that acquires first ultrasound data representing a plurality of frames that are consecutive in a time direction and are obtained by performing an ultrasound scan on a subject via the ultrasound probe; a second acquisition unit that inputs data based on the first ultrasound data into a trained model and acquires second ultrasound data representing a single frame from the trained model, the second ultrasound data having a higher spatial resolution of the fluid in the subject than the first ultrasound data; Ultrasound diagnostic equipment.

2. The second acquisition unit inputs at least one of (1) IQ data before application of a wall filter, (2) Doppler data which is IQ data after application of a wall filter, (3) power Doppler data which represents an absolute value of the Doppler data, and (4) data obtained by performing a local peak extraction process on the power Doppler data, which correspond to the plurality of frames, into the trained model as data based on the first ultrasound data; Acquire, as the second ultrasound data from the trained model, ultrasound data corresponding to data obtained after a synthesis process has been performed on data that has been subjected to local peak extraction processes corresponding to the plurality of frames. The ultrasonic diagnostic apparatus according to claim 1 .

3. The second acquisition unit inputs data, which is a combination of two or more of: (1) the IQ data, (2) the Doppler data, (3) the power Doppler data, and (4) the data on which the local peak extraction processing has been performed, corresponding to the plurality of frames, into the trained model as the first ultrasound data. The ultrasonic diagnostic apparatus according to claim 2 .

4. The second acquisition unit inputs data that is a combination of (1) the IQ data and (2) the Doppler data, which corresponds to the plurality of frames, into the trained model as the first ultrasound data. The ultrasonic diagnostic apparatus according to claim 2 .

5. The second acquisition unit inputs data that is a combination of (1) the IQ data and (3) the power Doppler data, which correspond to the plurality of frames, into the trained model as the first ultrasound data. The ultrasonic diagnostic apparatus according to claim 2 .

6. The second acquisition unit inputs data that is a combination of (2) the Doppler data and (3) the power Doppler data, which correspond to the plurality of frames, into the trained model as the first ultrasound data. The ultrasonic diagnostic apparatus according to claim 2 .

7. 3. The ultrasound diagnostic device of claim 2, wherein the second acquisition unit inputs data that is a combination of (1) the IQ data, (2) the Doppler data, and (3) the power Doppler data, which correspond to the plurality of frames, into the trained model as the first ultrasound data.

8. a first acquisition unit that acquires first ultrasound data representing a plurality of frames that are successive in a time direction and are obtained by performing an ultrasound scan on a subject via an ultrasound probe; a second acquisition unit that inputs data based on the first ultrasound data into a trained model and acquires second ultrasound data representing a single frame from the trained model, the second ultrasound data having a higher spatial resolution of the fluid in the subject than the first ultrasound data; Medical information processing equipment.

9. On the computer, acquiring first ultrasound data representing a plurality of frames successive in a time direction, the first ultrasound data being obtained by performing an ultrasound scan on a subject via an ultrasound probe; A program that inputs data based on the first ultrasound data into a trained model and executes a process of obtaining second ultrasound data representing a single frame from the trained model, the second ultrasound data having a higher spatial resolution of the fluid within the subject than the first ultrasound data.

Images