Ultrasonic diagnostic device, information processing device, doppler data generation method, learning method, and program

Abstract

To generate Doppler data for blood flow calculation without depending on a clutter.SOLUTION: An ultrasonic diagnostic device 100 includes a control unit 18. The control unit 18 acquires first Doppler data. The control unit 18 predicts that normal blood flow calculation cannot be performed later from the Doppler data using a learning model. The learning model performs machine learning using second Doppler data from which normal blood flow calculation on a reception block 31 side is possible, and third Doppler data from which normal blood flow calculation is impossible by a clutter on a reception block 32 side. The learning model is a learning model for predicting that normal blood flow calculation cannot be performed later from the Doppler data.SELECTED DRAWING: Figure 2

Description

[Technical Field]

[0001] The present invention relates to an ultrasound diagnostic apparatus, an information processing apparatus, a Doppler data generating method, a learning method, and a program. [Background technology]

[0002] Conventionally, there has been known an ultrasound diagnostic apparatus that uses an ultrasound probe to irradiate ultrasound into the interior of a subject, receives and analyzes the reflected waves, and displays an ultrasound image of the interior of the subject. The subject is, for example, a living patient.

[0003] Ultrasound diagnostic devices have imaging modes for observing the state of blood flow in a subject. One such imaging mode is a color Doppler mode, which displays changes in blood flow velocity in a specified region of interest (ROI) on a B (brightness) mode image as a colored tomographic image. A B-mode image is a tomographic image made up of pixels with brightness values ​​representing the received energy of ultrasound waves (echoes) reflected from the subject.

[0004] Among the Doppler signals obtained from the subject's received signals in color Doppler mode, unnecessary signals from slow-moving objects such as the heart wall or body movement are called clutter. Clutter appears when organs move during breathing or when the ultrasound probe is moved slightly, and has a large amplitude compared to the weak blood flow signal, interfering with the detection of blood flow. Generally, the frequency (phase change) of clutter signals is lower than the blood flow frequency, so a low-cut filter is used to remove these strong, low-frequency unnecessary signals. Low-cut filters used to remove clutter are called wall filters or MTI (Moving Target Injection) filters.

[0005] FIG. 10 is a diagram showing the frequency characteristics of the Doppler signal and the power of the MTI filter. In generating image data in color Doppler mode, the frequency characteristics of the power of the Doppler signal (packet data) corresponding to the received signal based on the ultrasound reflected by the subject are taken. As shown in FIG. 10, the Doppler signal consists of a clutter component 201, a blood flow component 202, and a noise component (not shown). The clutter component 201 is a signal component of the movement of the subject's tissue and is in a band of several tens of hertz or less. The blood flow component 202 is a signal component of the subject's blood flow. The average velocity 203 is a frequency corresponding to the average velocity of the blood flow component 202.

[0006] In order to visualize blood flow components 202 in a color Doppler mode image, clutter components 201 are removed from the Doppler signal by an MTI filter having characteristics 204. The MTI filter characteristics 204 are a high-pass filter that removes clutter components 201 and extracts only blood flow components. The cutoff frequency of the MTI filter characteristics 204 is adjusted so as to remove clutter components 201 and extract blood flow components 202. The cutoff frequency is a frequency that leaves relatively fast blood flow components and removes slow clutter components from the Doppler signal.

[0007] For example, an ultrasound diagnostic device that removes clutter components using an adaptive MTI filter is known (see Patent Document 1). The adaptive MTI filter is an adaptive filter for removing clutter. The adaptive MTI filter automatically changes its cutoff frequency. Clutter frequency increases or decreases depending on the speed of movement, such as body movement. If the cutoff frequency cannot be changed, clutter components may not be removed or blood flow components may be cut off. Therefore, the adaptive MTI filter automatically adjusts the cutoff frequency so that it falls between the blood flow components and the clutter components depending on the situation. [Prior art documents] [Patent documents]

[0008] [Patent Document 1] Japanese Patent Application Laid-Open No. 2014-158698 Summary of the Invention [Problem to be solved by the invention]

[0009] However, the conventional adaptive MTI filters described above are not always effective enough in removing clutter. For example, if the input received signal is excessive and saturates, the signal originating from tissue spreads into the frequency band of the signal originating from blood flow components. This makes it impossible for filters that control frequency components to remove clutter components. Alternatively, depending on the circuit configuration, a large-amplitude input may temporarily fix the output of an amplifier or other device at the maximum (or minimum) amplitude, resulting in the appearance of no signal being output.

[0010] Even if saturation does not occur, adaptive filters cannot be effective enough if the amplitude of the clutter signal momentarily increases. In this case, the filter cutoff frequency must be adjusted (changed / adapted) to match the clutter situation. However, there is a time limit for adjustment, and the adjustment cannot be made in time.

[0011] Figure 11 shows the time characteristics of clutter amplitude in a Doppler signal and conventional signal processing control. As shown in Figure 11, consider a case where a momentary increase in clutter amplitude occurs, creating a time period requiring action. A conventional ultrasound diagnostic device uses an adaptive MTI filter to control signal processing on the spot (when an increase in clutter amplitude occurs) to remove the momentary clutter component.

[0012] As a result, the adaptive MTI filter cannot completely remove clutter components, and filtering does not work properly. This makes it impossible to calculate normal blood flow, and the quality of the color Doppler image deteriorates. To solve this problem, it is possible to temporarily lower or turn off the gain when large amplitude signals are received. However, in this case, blood flow information is lost.

[0013] An object of the present invention is to generate Doppler data such as Doppler image data for normal blood flow calculations regardless of clutter. [Means for solving the problem]

[0014] In order to solve the above problem, the ultrasonic diagnostic apparatus of the invention described in claim 1 comprises: an acquisition unit that acquires first Doppler data; a prediction unit that predicts that normal blood flow calculation will not be possible later from the first Doppler data, using a trained model that is machine-learned using second Doppler data that allows normal blood flow calculation and third Doppler data that does not allow normal blood flow calculation due to clutter, and that predicts that normal blood flow calculation will not be possible later from the Doppler data; and a setting unit that sets parameters for performing normal blood flow calculation in generating the first Doppler data when it is predicted that normal blood flow calculation will not be possible thereafter.

[0015] The invention described in claim 2 is the ultrasound diagnostic device described in claim 1, The causes of the inability to perform normal blood flow calculation are saturation of Doppler data and a decrease in amplitude of clutter signals, The prediction unit predicts, from the first Doppler data, saturation of subsequent Doppler data and a decrease in amplitude of clutter signals of the subsequent Doppler data.

[0016] The invention described in claim 3 is the ultrasonic diagnostic apparatus described in claim 2, The setting unit decreases the gain of the Doppler data when predicting saturation of the subsequent Doppler data, and increases the gain of the Doppler data when predicting a decrease in amplitude of clutter signals of the subsequent Doppler data.

[0017] The invention described in claim 4 is the ultrasound diagnostic apparatus according to any one of claims 1 to 3, The apparatus further includes an output control unit that displays, on a display unit, Doppler image data corresponding to the Doppler data for which the parameters have been set.

[0018] The invention described in claim 5 is the ultrasonic diagnostic apparatus described in claim 1, The apparatus further includes an output control unit that outputs Doppler sound data corresponding to the Doppler data for which the parameters have been set to a sound output unit.

[0019] The ultrasonic diagnostic apparatus of the invention described in claim 6 comprises: an acquisition unit that acquires first Doppler data; The device is equipped with a control unit that is machine-learned using second Doppler data that allows normal blood flow calculation and third Doppler data that does not allow normal blood flow calculation due to clutter, and that uses a trained model for estimating the second Doppler data from the third Doppler data to estimate first Doppler data that allows normal blood flow calculation from the first Doppler data when normal blood flow calculation cannot be performed using the first Doppler data.

[0020] The invention described in claim 7 is the ultrasonic diagnostic apparatus described in claim 6, The reason why the normal blood flow calculation cannot be performed is that the number of bits of the Doppler data is small. When the number of bits of the blood flow signal of the first Doppler data is small, the control unit uses the trained model to estimate first Doppler data having a large number of bits from the first Doppler data.

[0021] The invention described in claim 8 is the ultrasonic diagnostic apparatus described in claim 6, The reason why the normal blood flow calculation cannot be performed is that the number of samplings of Doppler data is small. When the number of sampling times of the first Doppler data is small, the control unit uses the trained model to estimate first Doppler data having a large number of sampling times from the first Doppler data.

[0022] The invention described in claim 9 is the ultrasound diagnostic apparatus according to any one of claims 6 to 8, The apparatus further includes an output control unit that displays Doppler image data corresponding to the estimated first Doppler data on a display unit.

[0023] The invention described in claim 10 is the ultrasonic diagnostic apparatus described in claim 6, The device includes an output control unit that outputs Doppler sound data corresponding to the estimated first Doppler data to a sound output unit.

[0024] The information processing device of the invention described in claim 11 comprises: The system is equipped with a control unit that performs machine learning using the second Doppler data that allows for normal blood flow calculation and the third Doppler data that prevents normal blood flow calculation due to clutter, and generates a trained model for predicting that normal blood flow calculation will later become impossible from the Doppler data.

[0025] The information processing device of the invention described in claim 12 comprises: The device includes a control unit that performs machine learning using second Doppler data that allows normal blood flow calculation and third Doppler data that does not allow normal blood flow calculation due to clutter, and generates a trained model for estimating the second Doppler data from the third Doppler data.

[0026] The Doppler data generation method of the invention described in claim 13 comprises: an acquisition step of acquiring first Doppler data; a prediction step of predicting, from the first Doppler data, that a normal blood flow calculation will later be impossible, using a trained model that is machine-learned using second Doppler data that allows normal blood flow calculation and third Doppler data that does not allow normal blood flow calculation due to clutter, and that predicts, from the Doppler data, that a normal blood flow calculation will later be impossible; and a setting step of setting parameters for performing normal blood flow calculation in generating the first Doppler data when it is predicted that normal blood flow calculation will not be possible thereafter.

[0027] The invention described in claim 13 is a Doppler data generation method, an acquisition step of acquiring first Doppler data; The method includes a control process for estimating, from the first Doppler data, first Doppler data that allows for normal blood flow calculation when normal blood flow calculation is not possible with the first Doppler data, using a trained model for estimating the second Doppler data from the third Doppler data, the trained model being machine-learned using second Doppler data that allows for normal blood flow calculation and third Doppler data that does not allow for normal blood flow calculation due to clutter.

[0028] The learning method of the invention described in claim 15 comprises: The method includes a control process of performing machine learning using the second Doppler data that allows normal blood flow calculation and the third Doppler data that prevents normal blood flow calculation due to clutter, and generating a trained model for predicting that normal blood flow calculation will later become impossible from the Doppler data.

[0029] The learning method of the invention described in claim 16 comprises: The method includes a control process of performing machine learning using second Doppler data that allows normal blood flow calculation and third Doppler data that does not allow normal blood flow calculation due to clutter, and generating a trained model for estimating the second Doppler data from the third Doppler data.

[0030] The program of the invention described in claim 17 is Computer, an acquisition unit that acquires first Doppler data; a prediction unit that predicts that normal blood flow calculation will not be possible later from the first Doppler data, using a trained model that is machine-learned using second Doppler data that allows normal blood flow calculation and third Doppler data that does not allow normal blood flow calculation due to clutter, and that predicts that normal blood flow calculation will not be possible later from the first Doppler data, using a trained model that is machine-learned using second Doppler data that allows normal blood flow calculation and third Doppler data that does not allow normal blood flow calculation due to clutter, a setting unit that sets parameters for performing normal blood flow calculation in generating the first Doppler data when it is predicted that normal blood flow calculation cannot be performed thereafter; Function as.

[0031] The program of the invention described in claim 18 is Computer, an acquisition unit that acquires first Doppler data; a control unit that is machine-learned using second Doppler data that allows normal blood flow calculation and third Doppler data that does not allow normal blood flow calculation due to clutter, and that estimates first Doppler data that allows normal blood flow calculation from the first Doppler data when normal blood flow calculation cannot be performed using the first Doppler data, using a trained model for estimating the second Doppler data from the third Doppler data; Function as.

[0032] The program of the invention described in claim 19 is Computer, a control unit that performs machine learning using the second Doppler data that allows normal blood flow calculation and the third Doppler data that does not allow normal blood flow calculation due to clutter, and generates a trained model for predicting that normal blood flow calculation will later become impossible from the Doppler data; Function as.

[0033] The program of the invention described in claim 20 is Computer, a control unit that performs machine learning using second Doppler data that allows normal blood flow calculation and third Doppler data that does not allow normal blood flow calculation due to clutter, and generates a trained model for estimating the second Doppler data from the third Doppler data; Function as. [Effects of the Invention]

[0034] According to the present invention, it is possible to generate Doppler data for normal blood flow calculation regardless of clutter. [Brief explanation of the drawings]

[0035] [Figure 1] 1 is a schematic diagram of an ultrasonic diagnostic apparatus according to a first embodiment of the present invention. [Figure 2] FIG. 2 is a block diagram showing the functional configuration of the ultrasound diagnostic apparatus. [Figure 3]4A and 4B are diagrams illustrating the time characteristics of clutter amplitude of a Doppler signal and signal processing control according to the first embodiment. [Figure 4] 10 is a flowchart showing a first learning process. [Figure 5] 10 is a flowchart showing a first blood flow display process. [Figure 6] 10 is a flowchart showing a second learning process. [Figure 7] 10 is a flowchart showing a second blood flow display process. [Figure 8] 10 is a flowchart showing a third learning process. [Figure 9] 10 is a flowchart showing a third blood flow display process. [Figure 10] FIG. 10 is a diagram showing frequency characteristics of the power of a Doppler signal and an MTI filter. [Figure 11] 1 is a diagram showing the time characteristics of clutter amplitude of a Doppler signal and conventional signal processing control. DETAILED DESCRIPTION OF THE INVENTION

[0036] Advantages and features provided by one or more embodiments of the present invention will be more fully understood from the following detailed description and the accompanying drawings. However, these drawings are for illustrative purposes only and are not intended to define the limits of the present invention. Hereinafter, first to third embodiments of the present invention will be described with reference to the drawings. However, the scope of the present invention is not limited to the disclosed embodiments.

[0037] (First embodiment) A first embodiment of the present invention will be described with reference to Figs. 1 to 7. First, the device configuration of this embodiment will be described with reference to Figs. 1 to 3. Fig. 1 is a schematic diagram of an ultrasound diagnostic device 100 of this embodiment. Fig. 2 is a block diagram showing the functional configuration of the ultrasound diagnostic device 100. Fig. 3 is a diagram showing the time characteristics of clutter amplitude of a Doppler signal and signal processing control of this embodiment.

[0038] As shown in FIG. 1, an ultrasound diagnostic device 100 is installed in a medical facility such as a hospital, and generates ultrasound image data by transmitting and receiving ultrasound waves to and from a subject such as a living patient. The ultrasound diagnostic device 100 has two image modes: a color Doppler mode and a pulsed Doppler mode. The color Doppler mode displays color Doppler image data (hereinafter referred to as C-mode image data) showing the state of blood flow in the subject superimposed on B-mode image data. The pulsed Doppler mode displays the blood flow waveform within an arbitrary sample gate over time using pulse waves intermittently received from an ultrasound probe. In the pulsed Doppler mode, Doppler waveform image data of the blood flow waveform is displayed together with B-mode image data for which a sample gate is arbitrarily set. The ultrasound image data includes B-mode image data, C-mode image data, and Doppler waveform image data. The ultrasound diagnostic device 100 also uses a machine learning trained model to predict saturation based on the clutter components of the received signal (Doppler signal) and performs processing to prevent saturation.

[0039] The ultrasound diagnostic device 100 includes an ultrasound diagnostic device main body 1 and an ultrasound probe 2. The ultrasound probe 2 is connected to the ultrasound diagnostic device main body 1. The ultrasound probe 2 transmits ultrasound waves (transmitted ultrasound waves) into a subject and receives reflected waves of the ultrasound waves reflected within the subject (reflected ultrasound waves: echoes). The ultrasound probe 2 has an ultrasound probe main body 21, a cable 22, and a connector 23. The ultrasound probe main body 21 is the head of the ultrasound probe 2 and transmits and receives ultrasound waves. The cable 22 is connected to the ultrasound probe main body 21 and the connector 23. The cable 22 is a cable through which a drive signal for the ultrasound probe main body 21 and a received ultrasound signal flow. The connector 23 is a plug connector for connecting to a receptacle connector (not shown) of the ultrasound diagnostic device main body 1.

[0040] The ultrasound diagnostic device main body 1 is connected to the ultrasound probe main body 21 via a connector 23 and a cable 22. The ultrasound diagnostic device main body 1 transmits an electrical drive signal to the ultrasound probe main body 21, causing the ultrasound probe main body 21 to transmit ultrasound waves to the subject. The ultrasound probe 2 generates a reception signal, which is an electrical signal, in response to the ultrasound reflected from inside the subject and received by the ultrasound probe main body 21. The ultrasound diagnostic device main body 1 creates an image of the internal state of the subject as ultrasound image data based on the reception signal generated by the ultrasound probe 2.

[0041] The ultrasound probe main body 21 has transducers (not shown) at the tip side. The number of transducers can be set arbitrarily, and in practice, it is, for example, 192. The transducers are arranged, for example, in a one-dimensional array in the scanning direction (azimuth direction). The transducers may also be arranged in a two-dimensional array. In this embodiment, a linear scanning electronic scanning probe is adopted as the ultrasound probe 2. However, the ultrasound probe 2 may be either an electronic scanning type or a mechanical scanning type. The ultrasound probe 2 may also be any of a linear scanning type, a sector scanning type, or a convex scanning type. The ultrasound diagnostic device main body 1 and the ultrasound probe 2 may be configured to communicate wirelessly instead of by wire via a cable 22. This wireless communication may be UWB (Ultra Wide Band) or the like.

[0042] The operation input unit 11 is a control panel or the like that accepts various operation inputs from users such as doctors, technicians, etc. The operation input unit 11 has operation elements such as push buttons, encoders, lever switches, joysticks, trackballs, keyboards, touchpads, and multifunction switches.

[0043] The display unit 171 has a display panel such as an LCD (Liquid Crystal Display), an organic EL (Electro-Luminescence) display, an inorganic EL display, etc. The display unit 171 displays display information such as ultrasound image data on the display panel.

[0044] 2, the ultrasound diagnostic device main body 1 includes an operation input unit 11, a transmission unit 12, receiving amplifiers 131 and 132, beamformers 141 and 142, a signal processing unit 15, a machine learning calculation unit 16, a storage unit 161, a display unit 171, a sound output unit 172, a control unit 18, and a storage unit 19. The receiving amplifiers 131 and 132 and the beamformers 141 and 142 function as receiving units. The control unit 18 functions as an acquisition unit, a prediction unit, a setting unit, an output control unit, and a control unit.

[0045] The operation input unit 11 accepts various operation inputs from the user and outputs the operation signals to the control unit 18. The operation input unit 11 may be formed integrally with the display screen of the display unit 171 and may include a touch panel that accepts touch inputs from the user.

[0046] The transmitter 12, under the control of the controller 18, supplies a drive signal, which is an electrical signal, to the ultrasonic probe 2, causing the ultrasonic probe 2 to generate a transmission ultrasonic wave. The transmitter 12 includes, for example, a clock generating circuit, a delay circuit, and a pulse generating circuit. The clock generating circuit generates a clock signal that determines the transmission timing and transmission frequency of the drive signal. The delay circuit sets a delay time for each individual path corresponding to each transducer, and delays the transmission of the drive signal by the set delay time. The delay circuit focuses a transmission beam formed by the transmission ultrasonic wave using the delay. The pulse generating circuit generates a pulse signal as a drive signal at a predetermined period. The transmitter 12 generates a transmission ultrasonic wave by, for example, driving a continuous portion (e.g., 64 transducers) of multiple transducers (e.g., 192 transducers) arranged in the ultrasonic probe 2. The transmitter 12 then performs scanning by shifting the driven transducers in the scanning direction each time a transmission ultrasonic wave is generated.

[0047] In this embodiment, the transmitter 12 generates a drive signal corresponding to a specified image mode. In particular, when the image mode is a color Doppler mode, the transmitter 12 generates drive signals for multiple sampling times (repetition times) for the same scanning position for each frame of the color Doppler image. This allows transmission and reception of ultrasound waves for the sampling times to be performed at the same scanning position.

[0048] The receiving amplifiers 131 and 132 each receive a received signal, which is an analog electrical signal received from the ultrasonic probe 2. The receiving amplifiers 131 and 132 each amplify the signal at an arbitrary gain value (amplification factor) under the control of the control unit 18.

[0049] The beam former 141 converts the analog received signals amplified by the receiving amplifier 131 into digital received signals. The beam former 141 adjusts the time phase of the AD (Analog to Digital) converted digital received signals by providing a delay time for each individual path corresponding to each transducer. The beam former 141 adds (phased addition) these processed received signals to generate sound ray data. Similarly, the beam former 142 performs AD conversion and phased addition on the received signals amplified by the receiving amplifier 132 to generate sound ray data.

[0050] In the first learning process described below, the receiving amplifiers 131 and 132 are driven and controlled with different large and small gain values ​​for machine learning. For example, the receiving amplifier 131 is driven with a small gain value. The receiving amplifier 132 is driven with a large gain value. Therefore, the beam former 141 outputs sound ray data corresponding to a received signal with a small gain value. The beam former 142 outputs sound ray data corresponding to a received signal with a large gain value. In this way, there are two systems: a receiving block 31 of the receiving amplifier 131 and the beam former 141, and a receiving block 32 of the receiving amplifier 132 and the beam former 142. The receiving blocks 31 and 32 can simultaneously capture received signals when different receiving gains are set.

[0051] In the first blood flow display process described later, the receiving amplifier 131 is driven and controlled with a gain value for displaying a diagnostic color Doppler image. The receiving amplifier 131 is not driven. Therefore, the beam former 141 outputs sound ray data corresponding to the received signal with the diagnostic gain value.

[0052] When the image mode is the color Doppler mode, the signal processing unit 15 generates B-mode image data from the sound ray data from the beam former 141 or 142 under the control of the control unit 18. First, the signal processing unit 15 performs envelope detection processing, logarithmic compression, and the like on the sound ray data from the beam former 141 or 142. Furthermore, the signal processing unit 15 performs luminance conversion on the sound ray data after these processing by adjusting the dynamic range and gain. The signal processing unit 15 generates B-mode image data through this luminance conversion. In other words, the B-mode image data represents the strength of the received signal by luminance.

[0053] In the color Doppler mode, the signal processing unit 15 generates C-mode image data of the ROI from the sound ray data from the beamformer 141 or 142 under the control of the control unit 18. This ROI is the ROI input via the operation input unit 11. The signal processing unit 15 has, for example, a quadrature detection circuit, a corner turn control unit, an MTI filter, a correlation calculation unit, a data conversion unit, a noise removal spatial filter unit, an inter-frame filter, and a color Doppler image conversion unit for generating C-mode image data.

[0054] The quadrature detection circuit, under the control of the control unit 18, performs quadrature detection on the color Doppler mode reception signal (sound ray data) input from the beamformer 141 or 142. The quadrature detection circuit calculates the phase difference between the acquired color Doppler mode reception signal and a reference signal through quadrature detection, thereby obtaining (complex) Doppler signals I and Q. The corner turn control unit, under the control of the control unit 18, arranges the Doppler signals I and Q input from the quadrature detection circuit. The arrangement is an arrangement of the depth direction from the ultrasound probe to the subject and the ensemble direction of the sampling number n of ultrasound transmission and reception for each acoustic line. The corner turn control unit stores the arranged Doppler signals I and Q in a memory (not shown) and reads out the Doppler signals I and Q in the ensemble direction for each depth. The reception signals (Doppler signals I and Q) contain not only the signal components of blood flow necessary for generating a color flow image, but also unnecessary information (clutter components) such as information on blood vessel walls and tissues. The MTI filter, under the control of the control unit 18, filters the Doppler signals I and Q input from the corner turn control unit to remove clutter components.

[0055] The correlation calculation unit, under the control of the control unit 18, calculates the real part D and imaginary part N of the average value S of the autocorrelation calculation of the Doppler signal from the Doppler signals I and Q from the MTI filter. The Doppler signals I and Q are complex Doppler signals z. The average value S of the autocorrelation calculation of the Doppler signal is the average value of the phase difference vector. The data conversion unit, under the control of the control unit 18, calculates blood flow components from the Doppler signals I and Q from the MTI filter and the real part D and imaginary part N of the average value S of the autocorrelation calculation. The blood flow components are blood velocity, power, and variance.

[0056] The noise reduction spatial filter unit filters the power, blood flow velocity, and variance calculated by the data converter under the control of the control unit 18. The inter-frame filter performs inter-frame filtering of the blood flow components from the noise reduction spatial filter unit under the control of the control unit 18. The inter-frame filter selects blood flow components that constitute the color Doppler image from the noise reduction spatial filter unit according to the color Doppler display mode from the operation input unit 11. The inter-frame filter filters the selected blood flow components to smooth frame-to-frame changes and leave an afterimage. Under the control of the control unit 18, the color Doppler image converter color maps the blood flow components from the inter-frame filter and converts them into C-mode image data of the ROI. For example, in color Doppler image data corresponding to blood flow velocity, blood flow flowing toward the ultrasound probe 2 is displayed in red. In color Doppler image data, blood flow flowing away from the ultrasound probe 2 is displayed in blue.

[0057] In the pulse Doppler mode, the signal processing unit 15 generates B-mode image data from the sound ray data from the beam former 141 or 142 under the control of the control unit 18. In the pulse Doppler mode, the signal processing unit 15 generates Doppler waveform image data and Doppler sound data from the sound ray data from the beam former 141 or 142 under the control of the control unit 18. The signal processing unit 15 has, for example, a bandpass filter, a quadrature detection unit, a lowpass filter, a range gate, an integrating circuit, a wall filter, an FFT (Fast Fourier Transform) analysis unit, and a Doppler sound signal generation unit for generating the Doppler waveform image data and Doppler sound data.

[0058] The bandpass filter removes unnecessary frequency components from the sound ray data from the beamformer 141 or 142. The quadrature detection unit generates a quadrature detection signal from the sound ray data (received signal) input from the bandpass filter. The lowpass filter removes high-frequency components from the quadrature detection signal input from the quadrature detection unit to generate a received signal related to the Doppler shift frequency. The range gate acquires only the ultrasonic echo from the Doppler sample gate depth from the received signal output from the lowpass filter. The integration circuit integrates the received signal output from the range gate.

[0059] The wall filter includes an image wall filter and a sound wall filter. The image wall filter and the sound wall filter filter the received signal output from the integrating circuit to cut or suppress clutter components as low-frequency components. The image wall filter and the sound wall filter may be configured as high-pass filters. The image MTI filter and the sound MTI filter may be realized as separate filters, either as hardware or software modules, or as a single filter.

[0060] The FFT analysis unit generates Doppler waveform (spectrum) image data by frequency analyzing the Doppler shift frequency components of the received signal output from the image wall filter, and outputs the Doppler waveform (spectrum) image data to the display unit 171.

[0061] The Doppler sound signal generator separates components approaching or receding from the ultrasound probe 2 using a Hilbert transform or the like from the received IQ signal output from the sound wall filter, and performs frequency conversion as necessary and post-processing such as DA conversion to generate a Doppler sound signal. The Doppler sound signal has, for example, a stereo left signal and a stereo right signal.

[0062] In this embodiment, a case where the image mode is the color Doppler mode will be described in the first learning process and the first blood flow display process described later. Therefore, the signal processing unit 15 will be mainly described as having a configuration for generating B-mode image data and C-mode image data.

[0063] The machine learning calculation unit 16, under the control of the control unit 18, performs machine learning using the high-gain and low-gain C-mode image data input from the signal processing unit 15 as training data, and generates estimated data as a trained model.

[0064] As shown in Figure 3, for example, consider a case where a momentary increase in the amplitude of clutter components occurs, creating a time period requiring action. The ultrasound diagnostic device 100 predicts (reads ahead) the occurrence of a momentary increase in clutter amplitude through machine learning. The ultrasound diagnostic device 100 predicts the increase in the amplitude of the clutter component signal (clutter signal) and performs signal processing control to automatically adjust the gain of the receiving amplifier to remove the momentary clutter components.

[0065] The machine learning performed by the machine learning calculation unit 16 uses, for example, a pair of clutter amplitude increase and saturation. That is, the machine learning calculation unit 16 performs machine learning using high-gain C-mode image data and simultaneously generated low-gain C-mode image data as training data. The subject used is a living organism or a pseudo-living organism in which the amplitude of the clutter signal in the Doppler signal changes over time. Each gain value is set so that a portion of the entire time range of the received signal on the receiving amplifier 132 side is saturated, and so that the received signal on the receiving amplifier 131 side is not saturated over the entire time range.

[0066] The machine learning calculation unit 16 performs machine learning to estimate, from the C-mode image data on the receiving block 32 side that will soon become saturated, that the C-mode image data on the receiving block 31 side will become saturated in the future. This saturation occurs due to an increase in the amplitude of the clutter signal. Furthermore, the machine learning calculation unit 16 performs machine learning to estimate, from the non-saturated C-mode image data on the receiving block 31 side, a state in which the amplitude of the clutter signal will become smaller in the future. The machine learning calculation unit 16 generates estimation data from the results of these machine learning. The estimation data is data for estimating a change from saturated C-mode image data to a later non-saturated state. Furthermore, the estimation data is data for estimating a change from non-saturated C-mode image data to a state in which the amplitude of the clutter signal will later decrease (referred to as a decreased amplitude state). In the case of two-dimensional blood flow measurement (color Doppler mode), a configuration in which machine learning is performed using the situation around the prediction point and the situation of the prediction point is also conceivable. Furthermore, it is also possible to use not only C-mode images but also B-mode images, and use the situation around the predicted point and the situation at the predicted point as input data for machine learning.

[0067] The machine learning calculation unit 16 stores the generated estimation data in the storage unit 161. In the first blood flow display process, the machine learning calculation unit 16 reads out the estimation data from the storage unit 161. The machine learning calculation unit 16 uses the estimation data to estimate a change from a saturated state to a subsequent non-saturated state from the C-mode image data input from the signal processing unit 15. Similarly, the machine learning calculation unit 16 uses the estimation data to estimate a change from a non-saturated state to a subsequent state in which the amplitude of the clutter signal decreases from the C-mode image data.

[0068] The machine learning calculation unit 16, under the control of the control unit 18, performs processing such as coordinate conversion on the ultrasound image data input from the signal processing unit 15 to convert it into an image signal for display. The machine learning calculation unit 16 outputs the image signal to the display unit 171. Furthermore, under the control of the control unit 18, the machine learning calculation unit 16 outputs the Doppler sound signal input from the signal processing unit 15 to the sound output unit 172.

[0069] The storage unit 161 is a semiconductor memory that stores estimation data in a readable and writable manner by the machine learning calculation unit 16.

[0070] The display unit 171 displays an ultrasound image on a display panel in accordance with the image signal output from the machine learning calculation unit 16 under the control of the control unit 18. The display unit 171 also displays various display information input from the control unit 18 on the display panel.

[0071] The sound output unit 172 is composed of a speaker, and outputs the Doppler sound signal input from the machine learning calculation unit 16 as a Doppler sound under the control of the control unit 18.

[0072] The control unit 18 includes, for example, a CPU (Central Processing Unit), a ROM (Read Only Memory), and a RAM (Random Access Memory). The control unit 18 reads various processing programs stored in the ROM, loads them into the RAM, and controls each component of the ultrasound diagnostic apparatus 100 in cooperation with the CPU. The ROM is composed of nonvolatile memory such as a semiconductor. The ROM stores a system program corresponding to the ultrasound diagnostic apparatus 100, various processing programs executable on the system program, and various data such as a gamma table. In particular, the ROM stores a first learning program for executing a first learning process (described below) and a first abnormality discrimination program for executing a first blood flow display process (described below). These programs are stored in the RAM in the form of computer-readable program codes. The CPU sequentially executes operations in accordance with the program codes in the RAM. The RAM forms a work area for temporarily storing various programs executed by the CPU and data related to these programs.

[0073] The storage unit 19 is a storage unit such as an HDD (Hard Disk Drive) or an SSD (Solid State Drive) that stores information such as ultrasound image data in a writable and readable manner.

[0074] Next, the operation of the ultrasound diagnostic apparatus 100 of this embodiment will be described with reference to Figures 4 and 5. Figure 4 is a flowchart showing the first learning process. Figure 5 is a flowchart showing the first blood flow display process.

[0075] 4, a first learning process will be described as a learning process executed by the ultrasound diagnostic apparatus 100. The first learning process is a process in which large / small gain C-mode image data is generated as training data and machine learning is performed.

[0076] The tip of the ultrasound probe main body 21 is placed against the subject in advance. The subject is a living organism or a simulated living organism in which the amplitude of clutter signals in Doppler signals changes over time. In the ultrasound diagnostic device 100, an instruction to execute a first learning process is input by a user, such as a doctor or technician, via the operation input unit 11. In response to the execution instruction, the control unit 18 executes the first learning process in accordance with a first learning program stored in the ROM.

[0077] First, the control unit 18 controls the gain value of the receiving amplifier 132 to a large gain (step S11). In step S11, the control unit 18 controls the transmitting unit 12, the beam former 142, and the signal processing unit 15 to scan the subject and generate and acquire large-gain C-mode image data. The control unit 18 stores the large-gain C-mode image data in the storage unit 19. The large gain is a gain value at which a portion of the entire time range of the received signal on the receiving amplifier 132 side is saturated. The large-gain C-mode image data is Doppler data that makes normal blood flow calculation impossible due to saturation.

[0078] Simultaneously with step S11, the control unit 18 controls the gain value of the receiving amplifier 131 to a small gain (step S12). In step S12, the control unit 18 generates and acquires small-gain C-mode image data from the same received signal as in step S11 by controlling the beamformer 141 and the signal processing unit 15. The control unit 18 stores the small-gain C-mode image data in the storage unit 19. The small gain is a gain value that does not saturate the entire time range of the received signal on the receiving amplifier 131 side. Since the small-gain C-mode image data does not saturate, it is Doppler data that allows normal blood flow calculation.

[0079] The control unit 18 determines whether the number of accumulated data of C-mode image data stored in the storage unit 19 is equal to or greater than a predetermined number (step S13). The predetermined number in step S13 is a number sufficient for machine learning of large / small gain C-mode image data. The machine learning, for example, estimates the boundaries of its feature amounts using the large / small gain C-mode image data accumulated in the storage unit 19 as training data. The boundaries of the feature amounts are values ​​for estimating, from the C-mode image data, a change from a saturated state to a subsequent non-saturated state and a change from a non-saturated state to a subsequent amplitude reduction state. The machine learning generates, from the boundary values, a trained model for estimating, as estimation data, the change from a saturated state to a subsequent non-saturated state and a change from a non-saturated state to a subsequent amplitude reduction state.

[0080] If the number is less than the predetermined number (step S13; NO), the process proceeds to step S11. If the number is equal to or greater than the predetermined number (step S13; YES), the process proceeds to step S14. In step S14, the control unit 18 reads out a predetermined number or more of C-mode image data with large / small gains from the storage unit 19, and causes the machine learning calculation unit 16 to perform machine learning. The control unit 18 causes the machine learning calculation unit 16 to extract estimation data from the learning results of the machine learning in step S14 (step S15). The estimation data is a trained model for estimating (predicting) a change from a saturated state to a subsequent non-saturated state and a change from a non-saturated state to a subsequent amplitude reduction state from the C-mode image data. The control unit 18 causes the machine learning calculation unit 16 to store the estimation data extracted in step S15 in the storage unit 161 (step S16). The first learning process ends.

[0081] Next, referring to FIG. 5, a first blood flow display process as an operation process executed by the ultrasound diagnostic device 100 will be described. The first blood flow display process estimates a change from a non-saturated state to a saturated state or a decrease in the amplitude of clutter signals from a non-saturated state based on ultrasound image data in color Doppler mode. The first blood flow display process is a process for adjusting gain according to the estimation result. In addition, the reception block 32 is not used in the first blood flow display process.

[0082] Setting information for the color Doppler mode has been input in advance as the image mode via the operation input unit 11. The tip of the ultrasound probe body 21 is placed against the subject. The subject is a living patient. After the first learning process, the ultrasound diagnostic device 100 receives an instruction to execute a first blood flow display process from the user via the operation input unit 11. In response to the execution instruction, the control unit 18 executes the first blood flow display process in accordance with the first blood flow display program stored in the ROM.

[0083] The control unit 18 scans the subject and generates and acquires ultrasound image data under the control of the transmitter 12 to the signal processor 15 in accordance with the input setting information (step S21). The ultrasound image data is B-mode image data and C-mode image data of the subject. The setting information is various setting parameters for the color Doppler mode. In step S21, the control unit 18 causes the machine learning calculation unit 16 to superimpose a C-mode image of the generated C-mode image data on a B-mode image of the B-mode image data and display them on the display unit 171.

[0084] The control unit 18 determines whether to terminate the first blood flow display process based on whether or not a user has input an instruction to terminate the first blood flow display process via the operation input unit 11 (step S22). If the first blood flow display process is to be terminated (step S22; YES), the first blood flow display process is terminated. If the first blood flow display process is not to be terminated (step S22; NO), the control unit 18 causes the machine learning calculation unit 16 to read out the learned estimation data from the storage unit 161 (step S23). In step S23, the control unit 18 causes the machine learning calculation unit 16 to estimate the transition in the amplitude of the clutter signal of the C-mode image data generated in step S21 using the estimation data. The estimation of the transition in the amplitude of the clutter signal is an estimation of a change from a non-saturated state to a later saturated state and a change from a non-saturated state to a later state in which the amplitude of the clutter signal decreases. Note that the estimation of the transition in the amplitude of the clutter signal may be performed by the control unit 18.

[0085] The control unit 18 determines whether the C-mode image data will change from the current non-saturated state to a later saturated state according to the estimation result of step S23 (step S24). If the C-mode image data will later become saturated (step S24; YES), the control unit 18 reduces the gain value of the receiving amplifier 131 by a predetermined amount (step S25). The process returns to step S21.

[0086] If the C-mode image data will not be saturated later (step S24; NO), the process proceeds to step S26. In step S26, the control unit 18 determines whether the C-mode image data will change from the current non-saturated state to the later amplitude-decreased state according to the estimation result of step S23. If the C-mode image data will be in the later amplitude-decreased state (step S26; YES), the control unit 18 increases the gain value of the receiving amplifier 131 by a predetermined amount (step S27). The process proceeds to step S21. If the C-mode image data will not be in the later amplitude-decreased state (step S26; NO), the process proceeds to step S21.

[0087] As described above, according to this embodiment, the ultrasound diagnostic apparatus 100 includes a control unit 18. The control unit 18 generates and acquires C-mode image data as first Doppler data. The control unit 18 uses the estimation data to predict that normal blood flow calculation will not be possible later from the C-mode image data. The estimation data is machine-learned using C-mode image data as second Doppler data that allows normal blood flow calculation, and C-mode image data as third Doppler data that prevents normal blood flow calculation due to clutter. The estimation data is a learning model for predicting that normal blood flow calculation will not be possible later from the C-mode image data.

[0088] The control unit 18 performs machine learning using C-mode image data on the receiving block 31 side that allows normal blood flow calculation, and C-mode image data on the receiving block 32 side that does not allow normal blood flow calculation due to clutter. The control unit 18 generates estimated data as a trained model for predicting that normal blood flow calculation will later become impossible from the C-mode image data.

[0089] Therefore, C-mode image data for normal blood flow calculation can be generated regardless of clutter before normal blood flow calculation cannot be performed later.

[0090] The factors that prevent normal blood flow calculation from being performed are saturation of C-mode image data and a decrease in amplitude of clutter signals in C-mode image data. The control unit 18 predicts, from acquired C-mode image data, saturation of subsequent C-mode image data and a decrease in amplitude of clutter signals in subsequent C-mode image data. This allows appropriate prediction of cases in which normal blood flow calculation will not be performed later.

[0091] When the control unit 18 predicts saturation of the subsequent C-mode image data, it reduces the gain of the receiving amplifier 131 for generating C-mode image data. When the control unit 18 predicts a decrease in the amplitude of the clutter signal of the subsequent C-mode image data, it increases the gain of the receiving amplifier 131. This makes it possible to deal with cases where normal blood flow calculations cannot be performed later.

[0092] The control unit 18 displays the C-mode image data as Doppler image data for which the gain has been set on the display unit 171. This makes it possible to display high-quality C-mode image data that is easy to view and from which clutter components have been appropriately removed, thereby enabling accurate diagnosis of blood flow.

[0093] (Second embodiment) A second embodiment of the present invention will be described with reference to Figures 6 and 7. Figure 6 is a flowchart showing a second learning process. Figure 7 is a flowchart showing a second blood flow display process.

[0094] In the first embodiment, the C-mode image data with high and low gains obtained simultaneously were machine-learned. In the first embodiment, estimation data was used to estimate the change from a non-saturated state to a saturated state and the change from a non-saturated state to a state in which the amplitude of the clutter signal decreases. When the gain of a Doppler signal is reduced by predicting the change from a saturated state to a non-saturated state, the amplitude of the blood flow signal may decrease, resulting in a reduction in the amount of information (number of bits). Specifically, the amplitude of the blood flow signal may decrease when converted to a digital signal by AD conversion in the beamformer due to saturation of the circuit system. As a result, the amplitude of the blood flow signal may have a small number of bits, such as 1 or 2. Alternatively, even if the gain value of the receiving amplifier 131 is normal, the amplitude of the clutter signal mixed with the blood flow signal may be large, resulting in a small number of bits.

[0095] This embodiment uses machine learning to mitigate the influence of a blood flow signal with a small number of bits. Specifically, this embodiment uses machine learning to learn C-mode image data of a blood flow signal with a large number of bits when the Doppler signal is not saturated and C-mode image data of a blood flow signal with a small number of bits when the Doppler signal is saturated. This embodiment deals with the case where C-mode image data of a blood flow signal with a small number of bits in a saturated state is estimated by estimating C-mode image data of a blood flow signal with a large number of bits in a non-saturated state.

[0096] The device configuration of this embodiment is the same as that of the first embodiment, and uses an ultrasound diagnostic device 100. However, in this embodiment, the ultrasound diagnostic device 100 does not use a receiving block 32. Furthermore, a second learning program and a second blood flow display program are stored in the ROM of the control unit 18, instead of the first learning program and the first blood flow display program. The second learning program is a program for executing a second learning process, which will be described later. The second blood flow display program is a program for executing a second blood flow display process, which will be described later.

[0097] Next, the operation of the ultrasound diagnostic device 100 of this embodiment will be described with reference to Figures 6 and 7. First, the second learning process as a learning process executed by the ultrasound diagnostic device 100 will be described with reference to Figure 6. The second learning process is a process of generating C-mode image data of blood flow signals with large / small bit counts as training data and performing machine learning.

[0098] The tip of the ultrasound probe main body 21 is placed against the subject in advance. The subject is a living organism or a simulated living organism in which the amplitude of clutter signals in Doppler signals changes over time. In the ultrasound diagnostic device 100, for example, a user inputs an instruction to execute a second learning process via the operation input unit 11. In response to the execution instruction, the control unit 18 executes the second learning process in accordance with a second learning program stored in the ROM.

[0099] First, the control unit 18 controls the gain value of the receiving amplifier 131 to a gain value that makes the Doppler signal a large-bit blood flow signal that does not saturate (step S31). In step S31, the control unit 18 controls the transmission unit 12, the beam former 141, and the signal processing unit 15 to scan the subject and generate and acquire C-mode image data including a large-bit blood flow signal that does not saturate. The control unit 18 stores the C-mode image data including the large-bit blood flow signal in the storage unit 19. The C-mode image data including the large-bit blood flow signal is Doppler data that allows normal blood flow calculation.

[0100] The control unit 18 controls the gain value of the receiving amplifier 131 to a gain value that saturates the Doppler signal and turns it into a blood flow signal with a small number of bits (step S32). In step S32, the control unit 18 controls the transmitter 12, the beam former 141, and the signal processing unit 15 to scan the subject and generate and acquire C-mode image data including a saturated blood flow signal with a small number of bits. The control unit 18 stores the C-mode image data including the blood flow signal with a small number of bits in the storage unit 19. The control of the gain value in steps S31 and S32 may be configured to be performed by the machine learning calculation unit 16. The C-mode image data including the blood flow signal with a small number of bits is Doppler data that cannot be used for normal blood flow calculation due to saturation.

[0101] The control unit 18 determines whether the number of accumulated data of C-mode image data stored in the storage unit 19 is equal to or greater than a predetermined number (step S33). The predetermined number in step S33 is a number sufficient for machine learning of C-mode image data including blood flow signals with large / small bit counts. The machine learning, for example, estimates a boundary using the C-mode image data including blood flow signals with large / small bit counts stored in the storage unit 19 as training data. The boundary is the boundary between feature quantities of the C-mode image data in a saturated state / non-saturated state. The machine learning generates, from the value of the boundary, a trained model for discriminating between a saturated state / non-saturated state from the C-mode image data as estimation data. Furthermore, the machine learning generates another trained model as estimation data using the C-mode image data including blood flow signals with large / small bit counts as training data. This estimation data is data for estimating C-mode image data including blood flow signals with a large bit count from C-mode image data including blood flow signals with a small bit count. In this embodiment, estimation data having these two functions is generated.

[0102] If the number is less than the predetermined number (step S33; NO), the process proceeds to step S31. If the number is equal to or greater than the predetermined number (step S33; YES), the process proceeds to step S34. In step S34, the control unit 18 reads out C-mode image data including blood flow signals with large / small bit counts from the storage unit 19, and causes the machine learning calculation unit 16 to perform machine learning. The control unit 18 causes the machine learning calculation unit 16 to extract estimation data from the learning results of the machine learning in step S34 (step S35). The estimation data includes a trained model for discriminating between saturated and non-saturated states from the C-mode image data. The estimation data also includes a trained model for estimating C-mode image data including blood flow signals with large bit counts from C-mode image data including blood flow signals with small bit counts.

[0103] The control unit 18 causes the machine learning calculation unit 16 to store the estimation data extracted in step S35 in the storage unit 161 (step S36), and the second learning process ends.

[0104] Next, a second blood flow display process as an operation process executed by the ultrasound diagnostic device 100 will be described with reference to Fig. 5. The second blood flow display process is a process of estimating a saturation state from ultrasound image data in color Doppler mode, and generating and displaying C-mode image data with a large number of bits according to the estimation result.

[0105] Setting information for the color Doppler mode is input in advance as the image mode via the operation input unit 11. The setting information is various setting parameters for the color Doppler mode. The tip of the ultrasound probe body 21 is placed against the subject. The subject is a living patient. After the second learning process in the ultrasound diagnostic device 100, for example, a command to execute a second blood flow display process is input from the user via the operation input unit 11. In response to the execution command, the control unit 18 executes the second blood flow display process in accordance with the second blood flow display program stored in the ROM.

[0106] The control unit 18 scans the subject and generates and acquires ultrasound image data (step S41) by controlling the transmitter 12 to the signal processor 15 in response to input of setting information to the operation input unit 11. The ultrasound image data is B-mode image data and C-mode image data of the subject.

[0107] The control unit 18 causes the machine learning calculation unit 16 to read the learned estimation data from the storage unit 161 (step S42). In step S42, the control unit 18 causes the machine learning calculation unit 16 to determine whether or not the C-mode image data generated in step S41 is saturated, using the estimation data. In step S42, the control unit 18 determines whether or not the C-mode image data is saturated, depending on the determination result. Note that the determination and estimation of the saturated state using the estimation data may be configured to be performed by the control unit 18.

[0108] If the state is saturated (step S42; YES), the process proceeds to step S43. The C-mode image data generated in step S41 includes a saturated blood flow signal with a small number of bits. In step S43, the control unit 18 causes the machine learning calculation unit 16 to estimate C-mode image data with a large number of bits in a non-saturated state from the C-mode image data with a small number of bits using the estimation data. Note that the estimation of C-mode image data with a large number of bits using the estimation data may be configured to be performed by the control unit 18.

[0109] The control unit 18 causes the machine learning calculation unit 16 to superimpose the C-mode image of the C-mode image data generated in step S43 on the B-mode image of the B-mode image data (step S44). In step S44, the control unit 18 causes the machine learning calculation unit 16 to display the C-mode image superimposed on the B-mode image on the display unit 171. If there is no saturation (step S42; NO), the processing proceeds to step S44. However, in this case, the C-mode image of the C-mode image data generated in step S41 is displayed on the B-mode image.

[0110] The control unit 18 determines whether or not to terminate the second blood flow display process based on whether or not an instruction to terminate the second blood flow display process has been input from the user via the operation input unit 11 (step S45). If the second blood flow display process is to be terminated (step S45; YES), the second blood flow display process is terminated. If the second blood flow display process is not to be terminated (step S45; NO), the process proceeds to step S41.

[0111] As described above, according to this embodiment, the ultrasound diagnostic apparatus 100 includes a control unit 18. The control unit 18 generates and acquires C-mode image data as first Doppler data. When normal blood flow calculation cannot be performed on the C-mode image data, the control unit 18 uses the estimation data to estimate C-mode image data on which normal blood flow calculation is possible from the C-mode image data. The estimation data is machine-learned using C-mode image data as second Doppler data on which normal blood flow calculation is possible, and C-mode image data as third Doppler data on which normal blood flow calculation is impossible due to clutter. The estimation data is a trained model for estimating C-mode image data on which normal blood flow calculation is possible from C-mode image data on which normal blood flow calculation is impossible.

[0112] The control unit 18 performs machine learning using C-mode image data on the receiving block 31 side that allows for normal blood flow calculation and C-mode image data on the receiving block 32 side that does not allow for normal blood flow calculation due to clutter. The control unit 18 estimates C-mode image data on which normal blood flow calculation is possible from C-mode image data on which normal blood flow calculation is not possible.

[0113] Therefore, it is possible to generate C-mode image data for normal blood flow calculation regardless of clutter from C-mode image data for which normal blood flow calculation cannot be performed.

[0114] The reason why normal blood flow calculation cannot be performed is that the number of bits of the blood flow signal in the C-mode image data is small. When the number of bits of the C-mode image data is small, the control unit 18 uses estimation data to estimate C-mode image data with a large number of bits from the C-mode image data. Therefore, it is possible to generate C-mode image data with a large number of bits and normal blood flow calculation from C-mode image data with a small number of bits that cannot be used for normal blood flow calculation.

[0115] The control unit 18 displays the C-mode image data with the estimated large number of bits on the display unit 171. This makes it possible to display high-quality C-mode image data that is easy to view and from which clutter components have been appropriately removed, thereby enabling accurate diagnosis of blood flow.

[0116] (Third embodiment) A third embodiment of the present invention will be described with reference to Fig. 8 and Fig. 9. Fig. 8 is a flowchart showing a third learning process. Fig. 9 is a flowchart showing a third blood flow display process.

[0117] In the second embodiment, high-bit / low-bit C-mode image data are machine-learned, and high-bit C-mode image data is estimated from low-bit C-mode image data. Color Doppler mode displays blood flow information within the body in two dimensions. To display C-mode image data in real time, the number of samplings at one location must be limited to a few times. Within this sampled data, blood flow is a signal that changes rapidly (i.e., high frequency), while clutter is a signal that changes slowly (i.e., low frequency). Therefore, to extract only blood flow, it is sufficient to pass the data through an MTI filter, which is a high-pass filter. However, if the number of samplings is small, frequency separation performance tends to deteriorate, making it difficult to separate clutter components from blood flow components. If the number of samplings is large, the frame rate decreases.

[0118] Therefore, in this embodiment, two types of C-mode image data are acquired: one where the number of samplings at each scanning position is large, and one where the number of samplings is small (normal number). In this embodiment, C-mode image data with a large number of samplings (large number of samplings) and a small number of samplings (small number of samplings (normal number)) are machine-learned. This embodiment is configured to estimate C-mode image data with a large number of samplings from C-mode image data with a small number of samplings.

[0119] The device configuration of this embodiment is the same as that of the first embodiment, and uses an ultrasound diagnostic device 100. However, in this embodiment, the ultrasound diagnostic device 100 does not use a receiving block 32. Furthermore, a third learning program and a third blood flow display program are stored in the ROM of the control unit 18, instead of the first learning program and the first blood flow display program. The third learning program is a program for executing a third learning process, which will be described later. The third blood flow display program is a program for executing a third blood flow display process, which will be described later.

[0120] Next, the operation of the ultrasound diagnostic device 100 of this embodiment will be described with reference to Figures 8 and 9. First, the third learning process as a learning process executed by the ultrasound diagnostic device 100 will be described with reference to Figure 8. The third learning process is a process of generating C-mode image data of blood flow signals with large / small sampling counts as training data and performing machine learning.

[0121] The tip of the ultrasound probe main body 21 is placed against the subject in advance. The subject is a living organism or a simulated living organism in which the amplitude of clutter signals in Doppler signals changes over time. In the ultrasound diagnostic device 100, for example, a user inputs an instruction to execute a third learning process via the operation input unit 11. In response to the execution instruction, the control unit 18 executes the third learning process in accordance with a third learning program stored in the ROM.

[0122] First, the control unit 18 sets the number of samplings in color Doppler mode to a predetermined value that is larger than the normal number of samplings (step S51). In step S51, the control unit 18 controls the transmitter 12, the receiving amplifier 131, the beam former 141, and the signal processor 15 to scan the subject and generate and acquire C-mode image data with a large number of samplings. The control unit 18 stores the C-mode image data with a large number of samplings in the memory unit 19. The C-mode image data with a large number of samplings is Doppler data that allows normal blood flow calculation.

[0123] The control unit 18 controls the number of samplings to a predetermined small value (normal number of samplings) (step S52). In step S52, the control unit 18 controls the transmitter 12 to the signal processor 15 to scan the subject and generate and acquire C-mode image data with a small number of samplings. The control unit 18 stores the C-mode image data with a small number of samplings in the storage unit 19. The C-mode image data with a small number of samplings is Doppler data for which normal blood flow calculation is not possible. The control of the number of samplings in steps S51 and S52 may be configured to be performed by the machine learning calculation unit 16.

[0124] The control unit 18 determines whether the number of accumulated data of C-mode image data stored in the storage unit 19 is equal to or greater than a predetermined number (step S53). The predetermined number in step S33 is a number sufficient for machine learning of C-mode image data with large / small sampling counts. The machine learning generates a trained model as estimation data, for example, using the C-mode image data with large / small sampling counts accumulated in the storage unit 19 as training data. The estimation data is data for estimating C-mode image data with large sampling counts from C-mode image data with small sampling counts.

[0125] If the number is less than the predetermined number (step S53; NO), the process proceeds to step S51. If the number is equal to or greater than the predetermined number (step S53; YES), the process proceeds to step S54. In step S54, the control unit 18 reads out the C-mode image data with large / small sampling counts from the storage unit 19, and causes the machine learning calculation unit 16 to perform machine learning. The control unit 18 causes the machine learning calculation unit 16 to extract estimation data from the learning results of the machine learning in step S54 (step S55). The estimation data is a learned model for estimating C-mode image data with a large sampling count from C-mode image data with a small sampling count.

[0126] The control unit 18 causes the machine learning calculation unit 16 to store the estimation data extracted in step S55 in the storage unit 161 (step S56), and the third learning process ends.

[0127] Next, a third blood flow display process as an operation procedure executed by the ultrasound diagnostic device 100 will be described with reference to Fig. 9. The third blood flow display process is a process in which C-mode image data with a large number of samplings is estimated from ultrasound image data with a normal number of samplings in color Doppler mode, and gain is adjusted according to the estimation result. In the first blood flow display process, the reception block 32 is not used.

[0128] Setting information for the color Doppler mode is input in advance as the image mode via the operation input unit 11. The setting information for the color Doppler mode includes various setting parameters for the color Doppler mode, such as a small sampling count (normal count) and ROI. The tip of the ultrasound probe main body 21 is placed against the subject. The subject is a living patient. After the third learning process in the ultrasound diagnostic device 100, for example, a user inputs an instruction to execute a third blood flow display process via the operation input unit 11. In response to the execution instruction, the control unit 18 executes the third blood flow display process in accordance with a third blood flow display program stored in the ROM.

[0129] The control unit 18 scans the subject and generates and acquires ultrasound image data (step S61) by controlling the transmitter 12 to the signal processor 15 in response to input of setting information to the operation input unit 11. The ultrasound image data includes B-mode image data of the subject and C-mode image data with a small number of sampling times.

[0130] The control unit 18 causes the machine learning calculation unit 16 to read out the estimation data from the storage unit 161 (step S62). In step S62, the control unit 18 causes the estimation unit 16 to estimate C-mode image data with a large number of samplings from the C-mode image data with a small number of samplings generated in step S61, using the estimation data. Note that the estimation of C-mode image data with a large number of samplings using the estimation data may be configured to be performed by the control unit 18.

[0131] The control unit 18 causes the machine learning calculation unit 16 to superimpose the C-mode image of the C-mode image data generated in step S62 on the B-mode image of the B-mode image data (step S63). In step S63, the control unit 18 causes the display unit 171 to display the C-mode image superimposed on the B-mode image.

[0132] The control unit 18 determines whether or not to terminate the third blood flow display process based on whether or not an instruction to terminate the third blood flow display process has been input from the user via the operation input unit 11 (step S64). If the third blood flow display process is to be terminated (step S64; YES), the third blood flow display process is terminated. If the third blood flow display process is not to be terminated (step S64; NO), the process proceeds to step S61.

[0133] As described above, according to this embodiment, a factor that prevents normal blood flow calculation is a small number of samplings of C-mode image data (for example, a normal number of samplings). When the number of samplings of C-mode image data is small, the control unit 18 uses estimation data to estimate C-mode image data with a large number of samplings from the C-mode image data. Therefore, C-mode image data with a large number of samplings and normal blood flow calculation can be generated from C-mode image data with a small number of samplings without reducing the frame rate.

[0134] The control unit 18 displays the C-mode image data with a large estimated sampling count on the display unit 171. Therefore, it is possible to display high-quality C-mode image data that is easy to view and from which clutter components have been appropriately removed without reducing the frame rate, and to accurately diagnose blood flow.

[0135] In the above description, an example has been disclosed in which a ROM is used as a computer-readable medium for the program according to the present invention, but this is not limiting. Other computer-readable media include non-volatile memories such as flash memories and portable recording media such as CD-ROMs. Furthermore, a carrier wave is also applicable to the present invention as a medium for providing data for the program according to the present invention via a communication line.

[0136] The description of the above embodiments is merely an example of the ultrasound diagnostic device, information processing device, Doppler data generation method, learning method, and program according to the present invention, and is not limited to this. For example, at least two of the above embodiments may be appropriately combined.

[0137] In each of the above embodiments, the ultrasound diagnostic device 100 as an information processing device is configured to perform machine learning using ultrasound image data generated by the device itself. However, this configuration is not limited to this. The learning process in each of the embodiments may be performed by an information processing device such as a server connected to the ultrasound diagnostic device for communication. The server may receive C-mode image data from the ultrasound diagnostic device 100, and perform machine learning using the received C-mode image data as training data to generate estimation data.

[0138] Furthermore, in the first embodiment, the estimation data is generated by machine learning using C-mode image data as Doppler data, but this is not limiting. The control unit 18 may also generate estimation data by machine learning using Doppler waveform image data in high / low gain pulse Doppler modes. The estimation data is a trained model for estimating, from the Doppler waveform image data, a change from a non-saturated state to a later saturated state and a change from a non-saturated state to a later amplitude-decreasing state of clutter signals. After machine learning, the control unit 18 uses the estimation data to estimate, from the generated Doppler waveform image data, a change from a non-saturated state to a later saturated state and a change from a non-saturated state to a later amplitude-decreasing state. Depending on the estimation result, the control unit 18 adjusts the gain of the receiving amplifier 131.

[0139] This configuration allows for easy-to-view, high-quality Doppler waveform image data with clutter components appropriately removed to be displayed, enabling accurate diagnosis of blood flow. Furthermore, the control unit 18 outputs Doppler sound corresponding to the gain-adjusted Doppler sound data from the sound output unit 172. This allows for high-quality Doppler sound to be output, enabling accurate diagnosis of blood flow.

[0140] In the second embodiment, the estimation data is generated by machine learning C-mode image data as Doppler data, but this is not limiting. The control unit 18 may also generate estimation data by machine learning pulse Doppler mode Doppler waveform image data including large / small bit blood flow signals. The estimation data is a trained model for estimating Doppler waveform image data including large bit blood flow signals from Doppler waveform image data including small bit blood flow signals. After machine learning, if the generated Doppler waveform image data is saturated, the control unit 18 uses the estimation data to estimate Doppler waveform image data including large bit blood flow signals from Doppler waveform image data including small bit blood flow signals. Depending on the estimation result, the control unit 18 displays the Doppler waveform image data including large bit blood flow signals on the display unit 171.

[0141] According to this configuration, it is possible to display easy-to-view, high-quality Doppler waveform image data from which clutter components have been appropriately removed, and to accurately diagnose blood flow.

[0142] In addition, in each of the above embodiments, trained estimation data is generated by machine learning C-mode image data as Doppler data. However, this configuration is not limited to this. Sound ray data or intermediate data generated between sound ray data generation and image data generation may be used as the Doppler data.

[0143] In particular, in the second embodiment, in the case of pulse Doppler mode, the control unit 18 may be configured to perform machine learning on the sound ray data or intermediate data to generate estimation data. In this configuration, the control unit 18 uses the estimation data to estimate sound ray data or intermediate data, and outputs Doppler sound data corresponding to the estimated sound ray data or intermediate data to the sound output unit 172. This makes it possible to output high-quality Doppler sound, enabling accurate diagnosis of blood flow.

[0144] Furthermore, the first and second embodiments may be configured to be applied to other Doppler imaging modes, such as a continuous wave Doppler mode.

[0145] While embodiments of the present invention have been described and illustrated in detail, the disclosed embodiments are made for purposes of illustration and example only, and are not intended to be limiting. The scope of the present invention should be interpreted by the terms of the appended claims. [Explanation of symbols]

[0146] 100 Ultrasound diagnostic equipment 1. Ultrasound diagnostic device 11 Operation input section 12 Transmitter 131,132 Receiving amplifier 141,142 Beamformer 15 Signal processing section 16 Machine Learning Calculation Unit 171 Display section 172 Sound output section 18 Control Unit 19,161 storage section 2 Ultrasonic probe 21 Ultrasonic probe body 22 Cable 23 Connector 31,32 Receive Block

Claims

1. an acquisition unit that acquires first Doppler data; a prediction unit that predicts that normal blood flow calculation will not be possible later from the first Doppler data, using a trained model that is machine-learned using second Doppler data that allows normal blood flow calculation and third Doppler data that does not allow normal blood flow calculation due to clutter, and that predicts that normal blood flow calculation will not be possible later from the Doppler data; and a setting unit that sets parameters for performing normal blood flow calculation in generating the first Doppler data when it is predicted that normal blood flow calculation will not be possible thereafter.

2. The causes of the inability to perform normal blood flow calculation are saturation of Doppler data and a decrease in amplitude of clutter signals, The ultrasonic diagnostic apparatus according to claim 1 , wherein the prediction unit predicts, from the first Doppler data, saturation of subsequent Doppler data and a decrease in amplitude of clutter signals of the subsequent Doppler data.

3. 3. The ultrasonic diagnostic apparatus according to claim 2, wherein the setting unit decreases the gain of the Doppler data when saturation of the subsequent Doppler data is predicted, and increases the gain of the Doppler data when a decrease in amplitude of clutter signals of the subsequent Doppler data is predicted.

4. The ultrasound diagnostic apparatus according to claim 1 , further comprising an output control unit that displays Doppler image data corresponding to the Doppler data for which the parameters have been set on a display unit.

5. The ultrasonic diagnostic apparatus according to claim 1 , further comprising an output control unit that outputs Doppler sound data corresponding to the Doppler data for which the parameters have been set to a sound output unit.

6. an acquisition unit that acquires first Doppler data; and a control unit that is machine-learned using second Doppler data that allows normal blood flow calculation and third Doppler data that does not allow normal blood flow calculation due to clutter, and that estimates first Doppler data that allows normal blood flow calculation from the first Doppler data when normal blood flow calculation is not possible using the first Doppler data, using a trained model for estimating the second Doppler data from the third Doppler data.

7. The reason why the normal blood flow calculation cannot be performed is that the number of bits of the Doppler data is small.

7. The ultrasound diagnostic device of claim 6, wherein the control unit, when the number of bits of the blood flow signal of the first Doppler data is small, uses the trained model to estimate first Doppler data having a larger number of bits from the first Doppler data.

8. The reason why the normal blood flow calculation cannot be performed is that the number of samplings of Doppler data is small. The ultrasound diagnostic device according to claim 6, wherein when the number of samplings of the first Doppler data is small, the control unit uses the trained model to estimate first Doppler data with a large number of samplings from the first Doppler data.

9. The ultrasound diagnostic apparatus according to claim 6 , further comprising an output control unit that displays Doppler image data corresponding to the estimated first Doppler data on a display unit.

10. The ultrasonic diagnostic apparatus according to claim 6 , further comprising an output control unit that outputs Doppler sound data corresponding to the estimated first Doppler data to a sound output unit.

11. An information processing device having a control unit that performs machine learning using second Doppler data that allows normal blood flow calculation and third Doppler data that prevents normal blood flow calculation due to clutter, and generates a trained model for predicting that normal blood flow calculation will later become impossible from the Doppler data.

12. An information processing device including a control unit that performs machine learning using second Doppler data that allows normal blood flow calculation and third Doppler data that does not allow normal blood flow calculation due to clutter, and generates a trained model for estimating the second Doppler data from the third Doppler data.

13. an acquiring step of acquiring first Doppler data; a prediction step of predicting, from the first Doppler data, that a normal blood flow calculation will later become impossible, using a trained model that is machine-learned using second Doppler data that allows normal blood flow calculation and third Doppler data that does not allow normal blood flow calculation due to clutter, and that predicts, from the Doppler data, that a normal blood flow calculation will later become impossible; a setting step of setting parameters for performing normal blood flow calculation in generating the first Doppler data when it is predicted that normal blood flow calculation will not be possible thereafter.

14. an acquiring step of acquiring first Doppler data; a control step of estimating first Doppler data for which normal blood flow calculation is possible from the first Doppler data when normal blood flow calculation is not possible from the first Doppler data, using a trained model for estimating the second Doppler data from the third Doppler data, the trained model being machine-learned using second Doppler data for which normal blood flow calculation is possible and third Doppler data for which normal blood flow calculation is not possible due to clutter.

15. A learning method including a control step of performing machine learning using second Doppler data that allows normal blood flow calculation and third Doppler data that does not allow normal blood flow calculation due to clutter, and generating a trained model for predicting that normal blood flow calculation will later become impossible from the Doppler data.

16. A learning method including a control step of performing machine learning using second Doppler data that allows normal blood flow calculation and third Doppler data that does not allow normal blood flow calculation due to clutter, and generating a trained model for estimating the second Doppler data from the third Doppler data.

17. Computer, an acquisition unit that acquires first Doppler data; a prediction unit that predicts that normal blood flow calculation will later become impossible from the first Doppler data using a trained model that is machine-learned using second Doppler data that allows normal blood flow calculation and third Doppler data that does not allow normal blood flow calculation due to clutter, and that predicts that normal blood flow calculation will later become impossible from the first Doppler data using a trained model that is machine-learned using second Doppler data that allows normal blood flow calculation and third Doppler data that does not allow normal blood flow calculation due to clutter, a setting unit that sets parameters for performing normal blood flow calculation in generating the first Doppler data when it is predicted that normal blood flow calculation will not be possible thereafter; A program to function as a

18. Computer, an acquisition unit that acquires first Doppler data; a control unit that is machine-learned using second Doppler data that allows normal blood flow calculation and third Doppler data that does not allow normal blood flow calculation due to clutter, and that estimates first Doppler data that allows normal blood flow calculation from the first Doppler data when normal blood flow calculation cannot be performed using the first Doppler data, using a trained model for estimating the second Doppler data from the third Doppler data; A program to function as a

19. Computer, a control unit that performs machine learning using the second Doppler data that allows normal blood flow calculation and the third Doppler data that does not allow normal blood flow calculation due to clutter, and generates a trained model for predicting that normal blood flow calculation will later become impossible from the Doppler data; A program to function as a

20. Computer, a control unit that performs machine learning using second Doppler data that allows normal blood flow calculation and third Doppler data that does not allow normal blood flow calculation due to clutter, and generates a trained model for estimating the second Doppler data from the third Doppler data; A program to function as a

Images