Ultrasound diagnostic equipment and medical image processing equipment

Abstract

To effectively reduce noise.SOLUTION: An ultrasonic diagnostic device according to an embodiment comprises: a reception circuit; and a processing circuit. The reception circuit performs the first processing of multiplying a reception signal in an analog format by the gain while changing the gain according to the depth, and performs the second processing of converting the reception signal subjected to the first processing into a reception signal in a digital format. The processing circuit performs the third processing of keeping the level of noise included in the reception signal in the digital format at a constant level with respect to the reception signal in the digital format, and performs the fourth processing of reducing the noise included in the reception signal with respect to the reception signal in the digital format subjected to the third processing.SELECTED DRAWING: Figure 5

Description

[Technical Field]

[0001] The embodiments disclosed in this specification and the drawings relate to an ultrasound diagnostic apparatus and a medical image processing apparatus. [Background technology]

[0002] Ultrasound waves are attenuated as they propagate through a living body. The attenuation coefficient of ultrasound waves in soft tissue is approximately 0.6 dB / cm (round trip distance) / MHz. For example, if the echo signal (received signal) from an object at a depth of 5 cm is a 10 MHz ultrasound wave, the attenuation is -60 dB (-0.6 [dB / cm / MHz] * (5 [cm] * 2) * 10 [MHz]). If the object is at a depth of 10 cm, the attenuation is -120 dB. If the thermal noise generated by the transducer and electronic circuitry, i.e., white noise, contained in the received signal from one element of the ultrasound diagnostic device is at a level of -60 dB, the relationship between the received signal and noise at each depth is as shown in Figure 1. Figure 1 shows an example of the relationship between the received signal and noise at each depth when the white noise level is -60 dB. In Figure 1, "signal" indicates the received signal, and "noise" indicates noise. This is similar in other drawings. For example, if the ADC (Analog to Digital Converter) provided in an ultrasound diagnostic device is a 14-bit ADC, the dynamic range is 84 dB. Therefore, in this state, the ultrasound diagnostic device cannot input received signals from shallow to deep areas to the ADC within the dynamic range of the ADC. Therefore, the ultrasound diagnostic device performs processing to change the gain of the received analog signal according to the depth. This processing is called ATGC (Analog Time Gain Compensation).

[0003] Figure 2 shows an example of the relationship between the received signal and noise when an ultrasound diagnostic device applies a gain of 6 dB / cm (converted to one-sided distance (depth=1 cm on the horizontal axis is a propagation length of 2 cm for a round trip) to the received signal in the ATGC. Here, we consider the conditions under which an object up to a depth of 10 cm can be observed. From Figure 1 above, in the received signal from one element, at a depth of 10 cm, the noise is 60 dB higher than the received signal. When the number of elements in a probe (ultrasound probe) is 256, adding all received signals from all elements at deep locations in receive beamforming results in a signal-to-noise (S / N) ratio of 24 dB. This improves the S / N ratio. This is because when received signals are added N times, the S / N improves by the square root of N. As a result, the noise level after receive beamforming is -36 dB (= (-60 + 24) dB). Therefore, in order to be able to observe signals at a depth of 10 cm, the S / N ratio needs to be improved by another 36 dB. In terms of depth, whereas previously signals could only be observed up to a depth of 7 cm after beamforming, in order to be able to observe signals up to a depth of 10 cm, the S / N needs to be improved by another 36 dB.

[0004] One known method for solving this problem, or improving penetration, is to transmit ultrasound waves multiple times to the same location and add together the multiple received signals. Assuming the signal is constant and the noise varies randomly, adding the signals N times improves the S / N ratio by the square root of N. In the example above, 3981 additions are required to achieve a 36 dB S / N improvement. Adding all the signals requires 3981 frames, which, assuming a frame rate of 60 fps, requires 66 seconds (3981 / 60). However, this calculation requires that the echo signal from the living body be constant. However, it is impossible to keep the living body completely still for 66 seconds, taking into account pulsation and breathing movements. For this reason, the first method described above is practically impossible.

[0005] A second method for improving penetration is to transmit pulses with a long wave train length to input a large amount of energy and then compress the pulses when receiving. However, to achieve a 36 dB S / N improvement, it is necessary to transmit pulses that are 3981 times longer. Therefore, 3981 times the period of one wavelength of 10 MHz ultrasound is 398 μs, which means that the pulse length alone will result in a propagation distance of 30 cm. For this reason, using the second method mentioned above is practically impossible.

[0006] A third method for improving penetration is to apply a spatial low-pass filter to the received signal. Because noise is spatially random, it can be reduced by a spatial low-pass filter. If a simple averaging filter is used as such a low-pass filter, a two-dimensional filter with a width of 63 x 63 (approximately 3981) is required to achieve a 36 dB S / N improvement. However, when such a relatively wide two-dimensional filter is applied to the received signal, the blurring of the two-dimensionally filtered received signal becomes relatively large. Therefore, using the third method described above is practically impossible.

[0007] A fourth method for improving penetration is to reduce noise using a deep neural network (DNN), which has been increasingly applied in recent years. Convolutional neural networks (CNNs) have proven particularly effective in the field of image processing. CNNs are also used for noise reduction. Using a CNN, the ideal received signal for various reflectors can be learned, and any deviations from the ideal received signal characteristics can be removed as noise. However, when using a CNN, the kernel is the same across all regions. This requires that the statistical characteristics of the received signal be consistent across all regions. However, in the case of ultrasound signals (received signals obtained by ultrasound diagnostic equipment), the characteristics of the ultrasound signal change significantly depending on the position in both the depth direction (the direction of ultrasound propagation) and the azimuth direction (the direction perpendicular to the ultrasound propagation).

[0008] Figure 3 is a diagram illustrating an example of resolution. As shown in Figure 3, the azimuth resolution (azimuth resolution) is high near the transmission focus point, but the azimuth resolution is poor in shallow and deep areas due to large lateral flow. Furthermore, the lateral flow is large at the ends of the probe compared to near the center of the probe, resulting in poor azimuth resolution. Under such conditions, even if CNN training is advanced, the noise reduction effect is small and may even degrade the signal, making it unsuitable.

[0009] Some CNNs have been reported to output residuals rather than signals. These output the residual, or noise, from a noisy input image and a noise-free training image. Subtracting the residual image from the original image (input image) results in an image with reduced noise. This type of CNN is called a Denoising Convolutional Neural Network (DnCNN). For example, it has been reported that the PSNR (Peak Signal to Noise Ratio) can be improved by approximately 32 dB while maintaining the edges of the image.

[0010] In the case of DnCNN, since the kernel is the same in all regions, the noise must have the same statistical properties across the entire region. However, as explained with reference to Figure 2, the noise amplitude in the digital input signal (digital received signal) of the ultrasound diagnostic device varies with depth. Therefore, DnCNN, which assumes that the noise is spatially uniform, cannot be applied to such digital input signals.

[0011] One way to deal with the difference between noise during training and noise during use (operation) is to adjust the internal parameters of the neural network during use, but this method assumes that the noise is spatially uniform. [Prior art documents] [Patent documents]

[0012] [Patent Document 1] Japanese Patent Application Publication No. 2018-206382 [Patent Document 2] Japanese Patent Publication No. 2020-114295 [Non-patent literature]

[0013] [Non-Patent Document 1] K. Zhang et al, “Beyond a Gaussian Denoiser: Residual Learning of Deep CNN for Image Denoising”, IEEE Transactions on Image Processing, Vol.26, No.7, 2017 Summary of the Invention [Problem to be solved by the invention]

[0014] One of the problems to be solved by the embodiments disclosed in this specification and the drawings is to effectively reduce noise. However, the problems to be solved by the embodiments disclosed in this specification and the drawings are not limited to the above problem. Problems corresponding to the effects of each configuration shown in the embodiments described below can also be positioned as other problems. [Means for solving the problem]

[0015] An ultrasound diagnostic apparatus according to an embodiment includes a receiving circuit and a processing circuit. The receiving circuit performs a first process of multiplying an analog received signal by a gain while varying the gain according to depth, and a second process of converting the first-processed received signal into a digital received signal. The processing circuit performs a third process of maintaining a constant level of noise contained in the digital received signal, and a fourth process of reducing noise contained in the digital received signal after the third process. [Brief explanation of the drawings]

[0016] [Figure 1] FIG. 1 is a diagram showing an example of the relationship between the received signal and noise at each depth when the level of white noise is −60 dB. [Figure 2] FIG. 2 is a diagram showing an example of the relationship between a received signal and noise when the ultrasonic diagnostic apparatus applies a gain of 6 dB / cm (converted into one-sided distance) to the received signal in the ATGC. [Figure 3] FIG. 3 is a diagram illustrating an example of resolution. [Figure 4] FIG. 4 is a block diagram showing an example of the configuration of the ultrasound diagnostic apparatus according to the first embodiment. [Figure 5] FIG. 5 is a diagram illustrating an example of the configuration of the receiving circuit and the noise reduction processing circuit according to the first embodiment. [Figure 6A] FIG. 6A is a diagram showing an example of the relationship between the level of a signal received by the LNA according to the first embodiment and noise. [Figure 6B] FIG. 6B is a diagram showing an example of the relationship between a received signal and noise when the ATGC processing circuit according to the first embodiment multiplies the received signal in ATGC by a gain of 6 dB / cm (converted into one-sided distance). [Figure 6C] FIG. 6C is a diagram showing an example of the relationship between noise and an IQ signal to which a gain has been applied by the inverse gain execution function according to the first embodiment. [Figure 6D] FIG. 6D is a diagram showing an example of the relationship between the IQ signal output by the subtractor according to the first embodiment and noise. [Figure 6E] FIG. 6E is a diagram illustrating an example of the relationship between noise and an IQ signal transmitted by the gain execution function according to the first embodiment. [Figure 7] FIG. 7 is a flowchart showing an example of the flow of processing executed by the noise reduction processing circuit according to the first embodiment. [Figure 8] FIG. 8 is a diagram showing an example of the configuration of a portion of an ultrasonic diagnostic apparatus according to the second embodiment. [Figure 9] FIG. 9 is a diagram illustrating an example of the configuration of a noise reduction processing circuit according to the second embodiment. [Figure 10] FIG. 10 is a diagram illustrating an example of processing executed by the DnCNN processing function according to the third embodiment. [Figure 11] FIG. 11 is a diagram illustrating an example of processing executed by the DnCNN processing function according to the third embodiment. [Figure 12] FIG. 12 is a diagram illustrating an example of the configuration of a noise reduction processing circuit according to the fourth embodiment. [Figure 13] FIG. 13 is a diagram illustrating an example of processing executed by the noise reduction processing function according to the fourth embodiment. [Figure 14] FIG. 14 is a diagram illustrating an example of the configuration of a medical image processing apparatus according to the fifth embodiment. DETAILED DESCRIPTION OF THE INVENTION

[0017] Hereinafter, an ultrasound diagnostic apparatus and a medical image processing apparatus according to an embodiment will be described with reference to the drawings. The embodiment may be combined with conventional technology, other embodiments, or other modified examples to the extent that no contradiction occurs in the content. Similarly, the modified examples may be combined with conventional technology, other embodiments, or other modified examples to the extent that no contradiction occurs in the content. In the following description, similar components will be assigned common reference numerals, and duplicated descriptions may be omitted.

[0018] (First embodiment) 4 is a block diagram showing an example of the configuration of an ultrasonic diagnostic apparatus 1 according to the first embodiment. As shown in FIG. 4, the ultrasonic diagnostic apparatus 1 according to the first embodiment includes an apparatus main body 100, an ultrasonic probe 101, an input device 102, and a display 103.

[0019] The ultrasonic probe 101 has, for example, a plurality of transducers (piezoelectric elements). These transducers generate ultrasonic waves based on a drive signal supplied from a transmission circuit 111 of a transmission / reception circuit 110 included in the device main body 100. Specifically, when a voltage (transmission drive voltage) is applied to the transducers by the transmission circuit 111, the transducers generate ultrasonic waves having a waveform corresponding to the transmission drive voltage. The ultrasonic probe 101 also receives reflected waves from the subject P, converts the reflected waves into reflected wave signals (received signals), which are electrical signals, and outputs (transmits) the reflected wave signals to the device main body 100. The ultrasonic probe 101 also has, for example, a matching layer provided on the transducers and a backing material that prevents ultrasonic waves from propagating backward from the transducers. The ultrasonic probe 101 is detachably connected to the device main body 100.

[0020] When ultrasonic waves are transmitted from the ultrasonic probe 101 to the subject P, the transmitted ultrasonic waves are reflected successively by discontinuous surfaces of acoustic impedance in the tissues of the subject P and are received as reflected waves by multiple transducers of the ultrasonic probe 101. The amplitude of the received reflected waves depends on the difference in acoustic impedance at the discontinuous surfaces from which the ultrasonic waves are reflected. When the transmitted ultrasonic pulse is reflected by the surface of a moving blood flow, heart wall, or the like, the reflected waves undergo a frequency shift due to the Doppler effect depending on the velocity component of the moving object in the direction of ultrasonic transmission. The ultrasonic probe 101 then transmits the reflected wave signal to the receiving circuit 112 of the transmitting / receiving circuit 110, which will be described later.

[0021] The ultrasonic probe 101 is detachably attached to the device main body 100. When scanning a two-dimensional region inside the subject P (two-dimensional scanning), the operator connects, for example, a 1D array probe in which multiple transducers are arranged in a row to the device main body 100 as the ultrasonic probe 101. Types of 1D array probes include linear ultrasonic probes, convex ultrasonic probes, and sector ultrasonic probes. When scanning a three-dimensional region inside the subject P (three-dimensional scanning), the operator connects, for example, a mechanical 4D probe or a 2D array probe to the device main body 100 as the ultrasonic probe 101. A mechanical 4D probe is capable of two-dimensional scanning using multiple transducers arranged in a row like a 1D array probe, and is also capable of three-dimensional scanning by swinging the multiple transducers at a predetermined angle (swing angle). A 2D array probe is capable of three-dimensional scanning using multiple transducers arranged in a matrix, and is also capable of two-dimensional scanning by focusing and transmitting ultrasonic waves.

[0022] The input device 102 is realized by input means such as a mouse, keyboard, button, panel switch, touch command screen, foot switch, trackball, joystick, etc. The input device 102 receives various setting requests from the operator of the ultrasound diagnostic apparatus 1 and transfers the received various setting requests to the apparatus main body 100.

[0023] The display 103 displays, for example, a GUI (Graphical User Interface) that allows the operator of the ultrasound diagnostic apparatus 1 to input various setting requests using the input device 102, and displays ultrasound images based on ultrasound image data generated in the apparatus main body 100. The display 103 is realized by a liquid crystal monitor, a CRT (Cathode Ray Tube) monitor, or the like.

[0024] The device main body 100 generates ultrasound image data based on reflected wave signals transmitted from the ultrasound probe 101. Note that ultrasound image data is an example of image data. The device main body 100 can generate two-dimensional ultrasound image data based on reflected wave signals transmitted from the ultrasound probe 101 and corresponding to a two-dimensional region of the subject P. The device main body 100 can also generate three-dimensional ultrasound image data based on reflected wave signals transmitted from the ultrasound probe 101 and corresponding to a three-dimensional region of the subject P. As shown in FIG. 1, the device main body 100 includes a transmission / reception circuit 110, a noise reduction processing circuit 120, a beamformer 130, a signal processing circuit 140, an image generation circuit 150, an image memory 160, a storage circuit 170, and a control circuit 180.

[0025] The transmission / reception circuit 110, under the control of the control circuit 180, causes the ultrasonic probe 101 to transmit ultrasonic waves and causes the ultrasonic probe 101 to receive reflected waves of the ultrasonic waves. In other words, the transmission / reception circuit 110 performs scanning via the ultrasonic probe 101. Note that scanning is also referred to as scanning, ultrasonic scanning, or ultrasonic scanning. The transmission / reception circuit 110 is an example of a transmission / reception unit. The transmission / reception circuit 110 has a transmission circuit 111 and a reception circuit 112.

[0026] The transmission circuit 111, under the control of the control circuit 180, supplies a drive signal to the ultrasonic probe 101 and causes the ultrasonic probe 101 to transmit ultrasonic waves. The transmission circuit 111 has a rate pulser generating circuit, a transmission delay circuit, and a transmission pulser. When scanning a two-dimensional region within the subject P, the transmission circuit 111 causes the ultrasonic probe 101 to transmit an ultrasonic beam for scanning the two-dimensional region. When scanning a three-dimensional region within the subject P, the transmission circuit 111 causes the ultrasonic probe 101 to transmit an ultrasonic beam for scanning the three-dimensional region.

[0027] The rate pulser generating circuit, under the control of the control circuit 180, repeatedly generates rate pulses for forming a transmission ultrasound wave (transmission beam) at a predetermined rate frequency (PRF: Pulse Repetition Frequency). The rate pulses pass through a transmission delay circuit, so that a voltage having a different transmission delay time is applied to the transmission pulser. For example, the transmission delay circuit imparts a transmission delay time for each transducer, which is required to focus the ultrasound waves generated from the ultrasound probe 101 into a beam and determine the transmission directivity, to each rate pulse generated by the rate pulser generating circuit. The transmission pulser supplies a drive signal (drive pulse) to the ultrasound probe 101 at a timing based on the rate pulse. The transmission delay circuit arbitrarily adjusts the transmission direction of the ultrasound wave from the transducer surface by changing the transmission delay time imparted to each rate pulse.

[0028] After the drive pulse is transmitted from the transmission pulser via a cable to the transducer in the ultrasonic probe 101, the transducer converts the electrical signal into mechanical vibration. That is, when a voltage is applied to the transducer, the transducer vibrates mechanically. Ultrasound generated by this mechanical vibration is transmitted into the living body. Here, the ultrasound waves, which have different transmission delay times for each transducer, are focused and propagate in a predetermined direction.

[0029] The transmission circuit 111 has a function of being able to instantaneously change the transmission frequency, transmission drive voltage, etc. in order to execute a predetermined scanning sequence under the control of the control circuit 180. In particular, the change in the transmission drive voltage is realized by a linear amplifier type oscillation circuit that can instantaneously switch the value of the transmission drive voltage, or a mechanism that electrically switches between multiple power supply units.

[0030] The reflected waves of the ultrasonic waves transmitted by the ultrasonic probe 101 reach a transducer inside the ultrasonic probe 101, and are converted from mechanical vibrations into an electrical signal (reflected wave signal) in the transducer, and then input to the receiving circuit 112. That is, an analog reflected wave signal is input to the receiving circuit 112. The receiving circuit 112 has an LNA (Low Noise Amplifier), an ATGC (Analog Time Gain Compensation) processing circuit, an ADC (Analog to Digital Converter), a demodulator, etc., and performs various processes on the reflected wave signal transmitted from the ultrasonic probe 101 to generate a baseband in-phase signal (I signal, I: In-phase) and a quadrature signal (Q signal, Q: Quadrature-phase) as a digital reflected wave signal. The I signal and Q signal are called IQ signals. The receiving circuit 112 then transmits the generated IQ signals to the noise reduction processing circuit 120. Details of the receiving circuit 112 will be described later.

[0031] The noise reduction processing circuit 120 receives the IQ signals transmitted by the receiving circuit 112 for each channel. Then, the noise reduction processing circuit 120 reduces noise contained in the received IQ signals and transmits the noise-reduced IQ signals to the beamformer 130. Note that one channel corresponds to one transducer or multiple transducers. The following description will be given taking as an example a case where one channel corresponds to one transducer. The noise reduction processing circuit 120 is realized by, for example, a processor. Details of the noise reduction processing circuit 120 will be described later.

[0032] The beamformer 130 generates reflected wave data by performing phasing addition processing (delay addition processing) on ​​the IQ signals transmitted by the noise reduction processing circuit 120. For example, the beamformer 130 imparts a delay time required to determine the reception directivity to the IQ signals for each channel. The beamformer 130 then generates reflected wave data by adding the IQ signals to which the delay time has been applied. The beamformer 130 transmits the generated reflected wave data to the signal processing circuit 140.

[0033] The signal processing circuit 140 receives the reflected wave data transmitted by the beamformer 130, performs various signal processing on the received reflected wave data, and transmits the processed reflected wave data to the image generation circuit 150 as B-mode data or Doppler data. The signal processing circuit 140 is realized by, for example, a processor. The signal processing circuit 140 is an example of a signal processing unit. An example of the various signal processing operations performed by the signal processing circuit 140 will be described below.

[0034] For example, the signal processing circuit 140 performs various processes, such as envelope detection and logarithmic compression, on the reflected wave data to generate B-mode data in which the signal strength (amplitude strength) at each sample point is expressed as luminance. For example, the signal processing circuit 140 includes an envelope detector and a logarithmic compressor. For example, the envelope detector performs envelope detection on the reflected wave data, and the logarithmic compressor logarithmically compresses data related to the envelope obtained by the envelope detection (for example, data indicating amplitude, etc.). The signal processing circuit 140 transmits the generated B-mode data to the image generation circuit 150.

[0035] The signal processing circuit 140 also performs signal processing for harmonic imaging, which visualizes harmonic components. Examples of harmonic imaging include contrast harmonic imaging (CHI) and tissue harmonic imaging (THI). The following scanning methods are known for contrast harmonic imaging and tissue harmonic imaging. For example, known scanning methods include amplitude modulation (AM), phase modulation (PM), also known as pulse subtraction or pulse inversion, and AMPM, which combines AM and PM to obtain both the effects of AM and PM.

[0036] Furthermore, the signal processing circuit 140 performs frequency analysis on the reflected wave data to extract motion information of the moving object (blood flow, tissue, contrast agent echo components, etc.) based on the Doppler effect from the reflected wave data, and generates Doppler data indicating the extracted motion information. For example, the signal processing circuit 140 extracts average velocity, average variance, average power value, etc. as motion information of the moving object across multiple points, and generates Doppler data indicating the extracted motion information of the moving object. The signal processing circuit 140 transmits the generated Doppler data to the image generation circuit 150.

[0037] Using the functions of the signal processing circuit 140, the ultrasound diagnostic apparatus 1 according to the embodiment can perform a color Doppler method, also known as a color flow mapping (CFM) method. In the color flow mapping method, ultrasonic waves are transmitted and received multiple times along multiple scan lines. The color flow mapping method applies an MTI (Moving Target Indicator) filter to a data sequence at the same position to suppress signals (clutter signals) derived from stationary or slow-moving tissues from the data sequence at the same position, thereby extracting signals derived from blood flow (blood flow signals). The color flow mapping method then estimates blood flow information, such as blood flow velocity, blood flow dispersion, and blood flow power, from the blood flow signals. The signal processing circuit 140 transmits color image data representing the blood flow information estimated by the color flow mapping method to the image generation circuit 150. The color image data is an example of Doppler data.

[0038] The signal processing circuit 140 is capable of processing both two-dimensional reflected wave data and three-dimensional reflected wave data.

[0039] The image generation circuitry 150 generates ultrasound image data from the B-mode data or Doppler data transmitted by the signal processing circuitry 140. The image generation circuitry 150 is realized by a processor.

[0040] For example, the image generation circuit 150 generates two-dimensional B-mode image data that represents the intensity of the reflected wave as brightness from the two-dimensional B-mode data generated by the signal processing circuit 140. The image generation circuit 150 also generates two-dimensional Doppler image data in which motion information or blood flow information is visualized from the two-dimensional Doppler data generated by the signal processing circuit 140. The two-dimensional Doppler image data in which motion information is visualized is velocity image data, variance image data, power image data, or image data that combines these.

[0041] Here, the image generation circuit 150 generally converts (scan converts) a scan line signal sequence of an ultrasound scan into a scan line signal sequence of a video format, such as that of a television, to generate ultrasound image data for display. For example, the image generation circuit 150 generates ultrasound image data for display by performing coordinate conversion on the data transmitted by the signal processing circuit 140 in accordance with the ultrasound scanning form of the ultrasound probe 101. In addition to scan conversion, the image generation circuit 150 also performs various image processing, such as image processing (smoothing processing) that regenerates an average brightness image using multiple image frames after scan conversion, and image processing (edge ​​enhancement processing) that uses a differential filter within the image. The image generation circuit 150 also combines text information of various parameters, scales, body marks, etc. with the ultrasound image data.

[0042] Furthermore, the image generation circuit 150 generates three-dimensional B-mode image data by performing coordinate transformation on the three-dimensional B-mode data generated by the signal processing circuit 140. The image generation circuit 150 also generates three-dimensional Doppler image data by performing coordinate transformation on the three-dimensional Doppler data generated by the signal processing circuit 140. That is, the image generation circuit 150 generates "three-dimensional B-mode image data and three-dimensional Doppler image data" as "three-dimensional ultrasound image data (volume data)." The image generation circuit 150 then performs various rendering processes on the volume data to generate various types of two-dimensional image data for displaying the volume data on the display 103.

[0043] The rendering process performed by the image generation circuit 150 includes, for example, a process of generating MPR image data from volume data using multi-planar reconstruction (MPR). The rendering process performed by the image generation circuit 150 also includes, for example, a volume rendering (VR) process of generating two-dimensional image data reflecting three-dimensional information. The image generation circuit 150 is an example of an image generation unit.

[0044] The B-mode data and Doppler data are ultrasound image data before scan conversion processing, and the data generated by the image generation circuit 150 is ultrasound image data for display after scan conversion processing. Note that the B-mode data and Doppler data are also called raw data.

[0045] The image memory 160 is a memory that stores various types of image data generated by the image generation circuit 150. The image memory 160 also stores data generated by the signal processing circuit 140. The B-mode data and Doppler data stored in the image memory 160 can be called up by an operator after diagnosis, for example, and becomes ultrasound image data for display via the image generation circuit 150. For example, the image memory 160 is realized by a semiconductor memory element such as a RAM or a flash memory, a hard disk, or an optical disk.

[0046] The memory circuitry 170 stores control programs for scanning (transmitting and receiving ultrasound), image processing, and display processing, as well as various data such as diagnostic information (e.g., patient ID, doctor's findings, etc.), diagnostic protocols, and various body marks. The memory circuitry 170 is also used, as necessary, to store data stored in the image memory 160. For example, the memory circuitry 170 is realized by a semiconductor memory element such as a flash memory, a hard disk, or an optical disk.

[0047] The control circuit 180 controls the overall processing of the ultrasound diagnostic apparatus 1. Specifically, the control circuit 180 controls the processing of the transmission / reception circuit 110, the noise reduction processing circuit 120, the beamformer 130, the signal processing circuit 140, and the image generation circuit 150 based on various setting requests input by the operator via the input device 102 and various control programs and various data read from the storage circuitry 170. The control circuit 180 also controls the display 103 to display an ultrasound image based on ultrasound image data for display stored in the image memory 160. For example, the control circuit 180 controls the display 103 to display a B-mode image based on B-mode image data or a color image based on color image data. The control circuit 180 also controls the display 103 to display a color image superimposed on the B-mode image. The control circuit 180 is an example of a display control unit or a control unit. The control circuit 180 is realized by, for example, a processor. An ultrasound image is an example of an image.

[0048] Furthermore, the control circuit 180 controls the ultrasonic probe 101 via the transmission / reception circuit 110, thereby controlling ultrasonic scanning.

[0049] The overall configuration of the ultrasound diagnostic apparatus 1 according to the first embodiment has been described above. Next, an example of the configuration of the receiving circuit 112 and the noise reduction processing circuit 120 will be described. Figure 5 is a diagram showing an example of the configuration of the receiving circuit 112 and the noise reduction processing circuit 120 according to the first embodiment.

[0050] 5, the receiving circuit 112 includes a plurality of LNAs 112a, a plurality of ATGC processing circuits 112b, a plurality of ADCs (Analog to Digital Converters) 112c, and a plurality of demodulators 112d. For example, the LNAs 112a, the ATGC processing circuits 112b, the ADCs 112c, and the demodulators 112d are provided for each channel. For example, if one channel corresponds to one oscillator, one LNA 112a, one ATGC processing circuit 112b, one ADC 112c, and one demodulator 112d are provided for one oscillator (one channel). Below, we will explain an example of processing (operation) performed by one LNA 112a, one ATGC processing circuit 112b, one ADC 112c, and one demodulator 112d corresponding to one channel, but the processing performed by the LNA 112a, ATGC processing circuit 112b, ADC 112c, and demodulator 112d corresponding to other channels is similar.

[0051] The LNA 112a receives a reception signal transmitted by a transducer corresponding to this LNA 112a, amplifies the received reception signal by a preset gain, and transmits the amplified reception signal to the ATGC processing circuit 112b.

[0052] FIG. 6A is a diagram showing an example of the relationship between the received signal and noise level received by the LNA 112a according to the first embodiment. For example, FIG. 6A shows the relationship between the signal and noise at each depth. The attenuation coefficient of ultrasound is approximately 0.6 dB / cm (round trip distance) / MHz in soft tissue. For example, if an echo signal (received signal) from an object at a depth of 5 cm is ultrasound at 10 MHz, the attenuation is −60 dB as shown in FIG. 6A. If the object is at a depth of 10 cm, the attenuation is −120 dB. Here, if the level of thermal noise generated by the transducer and various electronic circuits, i.e., white noise, included in the received signal from one transducer of the ultrasound diagnostic device 1, is −60 dB, the relationship between the received signal and noise at each depth is as shown in FIG. 6A. For example, if the ADC 112c is a “14-bit ADC,” the dynamic range is 84 dB. Therefore, in this state, the ultrasonic diagnostic apparatus 1 cannot input received signals from shallow to deep areas to the ADC 112c within the dynamic range of the ADC 112c.

[0053] Therefore, the ATGC processing circuit 112b performs processing (ATGC) on the received signal, which is an analog signal, to change the gain depending on the depth. Specifically, the ATGC processing circuit 112b attenuates the received signal transmitted by the LNA 112a by different amounts depending on the depth. That is, the ATGC processing circuit 112b multiplies the gain of the received signal transmitted by the LNA 112a using different gain values ​​depending on the depth. In this way, the ATGC processing circuit 112b amplifies the received signal while changing the gain value depending on the time elapsed since the transmission of the ultrasonic wave. Then, the ATGC processing circuit 112b transmits the received signal, to which the gain has been applied, to the ADC 112c.

[0054] 6B is a diagram showing an example of the relationship between a received signal and noise when the ATGC processing circuit 112b according to the first embodiment multiplies the received signal by a gain of 6 dB / cm (converted into one-sided distance) in ATGC. As shown in FIG. 6B, since ATGC has been performed on the received signal, the noise is not spatially uniform.

[0055] The ADC 112c converts the analog received signal sent by the ATGC processing circuit 112b into a digital received signal. For example, the digital received signal after such conversion is a radio frequency (RF) signal. The ADC 112c then sends the RF signal, which is the digital received signal, to the demodulator 112d.

[0056] The demodulator 112d receives the RF signal transmitted by the ADC 112c. The demodulator 112d then demodulates the received RF signal to convert it into I and Q signals in the baseband. The demodulator 112d then transmits the I and Q signals to the noise reduction processing circuit 120. The I and Q signals are digital signals.

[0057] As described above, the receiving circuit 112 according to the first embodiment includes an ATGC processing circuit 112b that performs ATGC, which multiplies an analog received signal by a gain while varying the gain according to the depth, and an ADC 112c that performs processing to convert the received signal that has undergone ATGC into a digital received signal. Here, ATGC is an example of a first processing. Furthermore, processing to convert the received signal that has undergone ATGC into a digital received signal is an example of a second processing.

[0058] 5, the noise reduction processing circuit 120 includes a plurality of inverse gain execution functions 120a, a memory 120b, a DnCNN processing function 120c, a subtractor 120e, and a gain execution function 120f. For example, an inverse gain execution function 120a is provided for each channel. For example, if one channel corresponds to one oscillator, one inverse gain execution function 120a is provided for one oscillator (one channel).

[0059] Here, for example, the processing functions executed by the multiple inverse gain execution functions 120a, DnCNN processing function 120c, and gain execution function 120f, which are components of the noise reduction processing circuit 120 shown in Fig. 5, are recorded in the form of computer-executable programs in the storage circuit 170. The noise reduction processing circuit 120 is, for example, a processor, which reads each program from the storage circuit 170 and executes the read program to realize the function corresponding to each program. In other words, the noise reduction processing circuit 120 in a state in which each program has been read has each function shown in the noise reduction processing circuit 120 in Fig. 5.

[0060] 5 shows a case where each of the processing functions of the multiple inverse gain execution functions 120a, the DnCNN processing function 120c, and the gain execution function 120f is realized by a single processing circuit (noise reduction processing circuit 120), but the embodiment is not limited to this. For example, the noise reduction processing circuit 120 may be configured by combining multiple independent processors, and each processor may realize each processing function by executing a respective program. Furthermore, each processing function of the noise reduction processing circuit 120 may be realized by being appropriately distributed or integrated into a single or multiple processing circuits.

[0061] Each of the multiple inverse gain executing functions 120a receives the IQ signal transmitted by the corresponding demodulator 112d and multiplies the received IQ signal by a gain that is the inverse of the gain (gain value) multiplied by the ATGC processing circuit 112b. For example, the inverse gain executing function 120a multiplies the IQ signal for each channel by using the inverse of the gain (gain value) multiplied by the ATGC processing circuit 112b as the gain value. In this way, the inverse gain executing function 120a performs processing on the IQ signal, which is an example of a digital received signal, to make the level of noise contained in the IQ signal constant at all depths.

[0062] 6C is a diagram showing an example of the relationship between noise and an IQ signal to which a gain has been applied by the inverse gain applying function 120a according to the first embodiment. As shown in FIG. 6C, the noise does not change depending on the depth and is spatially uniform. Therefore, the IQ signal to which a gain has been applied by the inverse gain applying function 120a can be applied to the DnCNN 120d, which estimates noise.

[0063] The ADC 112c generates noise called quantization noise. However, quantization noise does not change with depth. Therefore, it is more preferable for the inverse gain execution function 120a to apply a gain to the IQ signal such that the sum of the thermal noise, which changes with depth due to ATGC performed by the ATGC processing circuit 112b, and the quantization noise, which does not change with depth, is constant at all depths. However, for example, if the ADC 112c is a "14-bit ADC," the S / N ratio of the ADC 112c is approximately 86 dB (6.02 * 14 + 1.76). Therefore, in the case of Figure 6B, at a depth of 0 cm, the quantization noise is below the thermal noise and can be ignored. This ensures that the noise level is constant at all depths and all channels.

[0064] Then, the inverse gain execution function 120a stores the IQ signal multiplied by the gain in the memory 120b. Specifically, the memory 120b has a two-dimensional storage area configured by a channel axis (channel direction) and a depth axis (depth direction). Then, each of the multiple inverse gain execution functions 120a stores the IQ signal multiplied by the gain in a corresponding position in the storage area of ​​the memory 120b. As a result, two-dimensional IQ signals (two-dimensional data) of channel and depth are stored in the storage area of ​​the memory 120b.

[0065] The DnCNN processing function 120c includes a DnCNN 120d. For example, the DnCNN 120d is stored in the internal memory of the noise reduction processing circuit 120. The DnCNN processing function 120c reads out two-dimensional IQ signals stored in a storage area included in the memory 120b, and inputs the read two-dimensional IQ signals to the DnCNN 120d. In this way, the DnCNN processing function 120c causes the DnCNN 120d to estimate noise (residuals). The DnCNN processing function 120c also transmits the read two-dimensional IQ signals to the subtractor 120e.

[0066] The DnCNN 120d estimates noise (noise components) contained in the input two-dimensional IQ signal and outputs the estimated noise to the subtractor 120e. For example, the DnCNN 120d is a two-dimensional Deep Convolutional Neural Network (DCNN). Various internal parameters such as coefficients of the DnCNN 120d are determined in advance by learning as described below. For example, a learning device that performs learning is "Field II" ([searched April 2, 2021], Internet<http: / / field-ii.dk / > ) to create a noise-free ultrasound signal (for example, the above-mentioned IQ signal). Then, the learning device adds white noise equivalent to the noise level of the ultrasound diagnostic device 1 to this ultrasound signal. The learning device inputs the ultrasound signal with this white noise added to the DCNN, and trains the DCNN so that the output from the DCNN becomes the white noise added to the ultrasound signal. This type of learning is residual learning called DnCNN. The method for training the DnCNN 120d is similar to the method for training a neural network that estimates residuals (noise) in Non-Patent Document 1 and Patent Document 1.

[0067] Here, the input to the DnCNN 120d is a complex IQ signal, but a typical deep learning framework does not support complex numbers. However, in the first embodiment, attention is focused on white noise. The real and imaginary parts of white noise are independent. Therefore, the DnCNN processing function 120c can input the real and imaginary parts of the IQ signal to the DnCNN 120d as independent signals. For example, the DnCNN processing function 120c can input the real and imaginary parts separately to the DnCNN 120d.

[0068] The noise level of the ultrasound diagnostic device 1 varies depending on the ultrasound probe 101 and device conditions (conditions of the ultrasound diagnostic device 1). Therefore, the learning device may train the DnCNN 120d at each of a plurality of noise levels and generate, for each noise level, a DnCNN 120d having internal parameters corresponding to each of the plurality of noise levels. That is, the learning device may generate a plurality of DnCNNs 120d corresponding to the plurality of noise levels. The noise reduction processing circuit 120 may then hold the plurality of DnCNNs 120d generated by the learning device, select from the plurality of DnCNNs 120d a DnCNN 120d that matches the noise level corresponding to the ultrasound probe 101 or device conditions, and estimate the noise contained in the two-dimensional IQ signal using the selected DnCNN 120d.

[0069] In addition, the noise reduction processing circuit 120 may hold SNR-related data such as that disclosed in Patent Document 1 as an index for adjusting the internal parameters of DnCNN 120d, and adjust the internal parameters of DnCNN 120d when estimating noise using a method similar to that disclosed in Patent Document 1.

[0070] The subtractor 120e subtracts the noise estimated and output by the DnCNN 120d from the two-dimensional IQ signal (original signal) transmitted by the DnCNN processing function 120c. As a result, the subtractor 120e obtains a two-dimensional IQ signal with reduced noise. The two-dimensional IQ signal is two-dimensional data configured by a channel axis (channel direction) and a depth axis (depth direction). The subtractor 120e then transmits the two-dimensional IQ signal with reduced noise to the gain execution function 120f.

[0071] Fig. 6D is a diagram showing an example of the relationship between the IQ signal output by the subtractor 120e according to the first embodiment and noise. Fig. 6D shows an example of response in the depth direction. As can be seen from a comparison between Fig. 6C and Fig. 6D, the noise included in the IQ signal is reduced by the subtractor 120e.

[0072] The gain enforcement function 120f multiplies the IQ signal transmitted by the subtractor 120e by a gain that is the inverse of the gain (gain value) that was multiplied by the IQ signal by the inverse gain enforcement function 120a. That is, the gain enforcement function 120f restores the gain. This makes the level of the IQ signal constant or approximately constant at all depths. The gain enforcement function 120f then transmits the IQ signal, which has been multiplied by the gain, to the beamformer 130.

[0073] FIG. 6E is a diagram showing an example of the relationship between the IQ signal transmitted by the gain execution function 120f according to the first embodiment and noise. FIG. 6E shows an example of response in the depth direction. As can be seen from FIGS. 6D and 6E, the noise reduction effect of the DnCNN 120d is high; for example, according to Non-Patent Document 1, the PSNR (Peak Signal to Noise Ratio) improves by approximately 32 dB. Also, as can be seen from FIG. 6E, the level of the IQ signal is constant or approximately constant at all depths. Furthermore, there is almost no blurring of the image edges.

[0074] As described above, the noise reduction processing circuit 120 according to the first embodiment includes an inverse gain execution function 120a that performs processing on a digital IQ signal to make the level of noise contained in the digital IQ signal constant, and DnCNN processing functions 120c and 120d that perform processing on the digital IQ signal that has been processed to make the noise level constant, to reduce the noise contained in the IQ signal. The processing of making the level of noise contained in the IQ signal constant is an example of a third processing. The processing of reducing noise contained in the IQ signal is an example of a fourth processing. The noise reduction processing circuit 120 is an example of a processing circuit.

[0075] In addition, the inverse gain execution function 120a performs a process to keep the level of noise contained in the IQ signal constant by multiplying the IQ signal by a gain that is the opposite of the gain that is changed according to depth in ATGC, or by performing a process to keep the sum of the white noise contained in the IQ signal and the quantization noise that occurs when the received signal that has undergone ATGC is converted into a digital received signal constant at all depths.

[0076] Furthermore, as a process for reducing noise contained in the IQ signal, the DnCNN processing function 120c inputs an IQ signal that has been processed to make the level of noise contained in the IQ signal constant to the DnCNN 120d, which is a neural network that outputs noise contained in the input signal, and causes the DnCNN 120d to output the noise contained in the IQ signal. Then, as a process for reducing noise contained in the IQ signal, the subtractor 120e subtracts the noise output from the DnCNN 120d from the IQ signal, thereby reducing the noise contained in the IQ signal.

[0077] As described above, the beamformer 130 according to the first embodiment performs beamforming on the IQ signals whose noise has been reduced by the process of reducing the noise contained in the IQ signals.

[0078] In the above description, two-dimensional data (two-dimensional IQ signals) in the channel direction and depth direction are used as an example of the processing target for each ultrasonic wave transmission. However, the DnCNN processing function 120c may use IQ signals obtained by transmitting ultrasonic waves multiple times to input a three-dimensional IQ signal (three-dimensional data) to the DnCNN 120d, which is configured with an axis of the number of ultrasonic wave transmissions (transmission number direction), a channel axis (channel direction), and a depth axis (depth direction). In other words, the DnCNN 120d may be three-dimensionally compatible (three-dimensionally configured). As a result, the DnCNN 120d estimates noise and outputs the estimated noise to the subtractor 120e.

[0079] Furthermore, if there is sensitivity variation between channels, the noise level will not be spatially constant. In this case, the ultrasound diagnostic device 1 performs sensitivity measurement in advance and performs gain correction to make the noise level constant. For example, the inverse gain execution function 120a first performs sensitivity measurement by observing the received signals of each channel when ultrasound is not being transmitted. Specifically, the inverse gain execution function 120a measures sensitivity variation between channels and calculates the root mean square (rms) of all depths for each channel. The inverse gain execution function 120a then applies gain correction to the IQ signals so that the rms values ​​at the same depth are the same for all channels. After this, the inverse gain execution function 120a applies a gain that is the inverse of the gain (gain value) applied by the ATGC processing circuit 112b to the IQ signals that have been subjected to the gain correction. In this way, the inverse gain execution function 120a performs processing to make the noise level contained in the IQ signals constant across all channels and all depths based on the sensitivity variation between channels.

[0080] FIG. 7 is a flowchart showing an example of the flow of processing executed by the noise reduction processing circuit 120 according to the first embodiment.

[0081] (Step S101) As shown in FIG. 7, in step S101, each of the multiple inverse gain executing functions 120a multiplies the IQ signal transmitted by the corresponding demodulator 112d by a gain that is the inverse of the gain (gain value) multiplied by the ATGC processing circuit 112b.

[0082] (Step S102) In step S102, the plurality of inverse gain executing functions 120a stores the two-dimensional IQ signals multiplied by the gains in the memory 120b.

[0083] (Step S103) In step S103, the DnCNN processing function 120c reads out the two-dimensional IQ signal stored in the memory 120b and inputs the read out two-dimensional IQ signal to the DnCNN 120d, which then estimates noise and outputs the estimated noise to the subtractor 120e.

[0084] (Step S104) In step S104, the subtractor 120e subtracts the noise estimated and output by the DnCNN 120d from the two-dimensional IQ signal transmitted by the DnCNN processing function 120c. As a result, the subtractor 120e obtains a two-dimensional IQ signal with reduced noise. The subtractor 120e then transmits the two-dimensional IQ signal with reduced noise to the gain execution function 120f.

[0085] (Step S105) In step S105, the gain execution function 120f multiplies the IQ signal transmitted by the subtractor 120e by a gain that is the inverse of the gain (gain value) multiplied by the IQ signal by the inverse gain execution function 120a. That is, the gain execution function 120f restores the gain. It multiplies the IQ signal by an appropriate gain such that the level of the IQ signal becomes constant or approximately constant at all depths. Then, the gain execution function 120f transmits the IQ signal with the multiplied gain to the beamformer 130.

[0086] The ultrasound diagnostic device 1 according to the first embodiment has been described above. In conventional ultrasound diagnostic devices, the noise level of the signal input to the DnCNN is not constant depending on the location, making it difficult to effectively reduce noise. On the other hand, in the ultrasound diagnostic device 1 according to the first embodiment, the noise level of the signal input to the DnCNN 120d is constant at all locations, making it possible to reduce noise with the noise level constant at all locations. Therefore, the ultrasound diagnostic device 1 can efficiently reduce noise using a method that uses a neural network. Furthermore, this noise reduction method using DnCNN makes it possible to reduce noise without blurring the signal. As a result, the ultrasound diagnostic device 1 according to the first embodiment can significantly improve the penetration of ultrasound waves.

[0087] (Second embodiment) In the first embodiment, a case has been described in which the ultrasound diagnostic device 1 performs processing to reduce noise before beamforming. However, the ultrasound diagnostic device 1 may also perform processing to reduce noise after beamforming. Therefore, such an embodiment will be described as the second embodiment. Note that the description of the second embodiment will mainly focus on differences from the first embodiment, and descriptions of configurations similar to those of the first embodiment may be omitted.

[0088] FIG. 8 is a diagram showing an example of the configuration of a portion of an ultrasound diagnostic apparatus 1 according to the second embodiment. A noise reduction processing circuit 121 according to the second embodiment differs from the noise reduction processing circuit 120 according to the first embodiment in that it is provided subsequent to the beamformer 130. For example, the signal processing circuit 140 includes an envelope detector 140a and a logarithmic compressor 140b. In the second embodiment, the noise reduction processing circuit 121 is provided between the envelope detector 140a and the logarithmic compressor 140b. That is, the noise reduction processing circuit 121 is provided subsequent to the envelope detector 140a and prior to the logarithmic compressor 140b. The noise reduction processing circuit 121 is an example of a processing circuit.

[0089] For example, in the second embodiment, the envelope detector 140a performs envelope detection on the reflected wave data, the noise reduction processing circuit 121 performs processing to reduce noise contained in data related to the envelope obtained by the envelope detection (e.g., data indicating the amplitude of the reflected wave data), and the logarithmic compressor 140b logarithmically compresses the data related to the noise-reduced envelope, thereby generating B-mode data in the signal processing circuit 140. In this way, the data input to the noise reduction processing circuit 121 and the data output from the noise reduction processing circuit 121 are real number data. Furthermore, the data input to the noise reduction processing circuit 121, the data handled in the noise reduction processing circuit 121, and the data output from the noise reduction processing circuit 121 are examples of received signals.

[0090] In the second embodiment, the beamformer 130 generates reflected wave data by performing phasing addition processing (delay addition processing) on ​​the IQ signals transmitted by the receiving circuit 112. For example, the beamformer 130 imparts a delay time required to determine the reception directivity to the IQ signals for each channel. The beamformer 130 then generates reflected wave data by adding the IQ signals to which the delay time has been imparted. The beamformer 130 transmits the generated reflected wave data to the signal processing circuit 140. Here, the beamformer 130 according to the second embodiment performs variable aperture processing, which changes the number of channels to be added depending on the depth. That is, in the second embodiment, the beamformer 130 performs beamforming on the IQ signals, which changes the number of channels to be added depending on the depth.

[0091] Fig. 9 is a diagram illustrating an example of the configuration of a noise reduction processing circuit 121 according to the second embodiment. As illustrated in Fig. 9, the noise reduction processing circuit 121 includes an inverse gain execution function 121a, a memory 121b, a DnCNN processing function 121c, a subtractor 121e, and a gain execution function 121f.

[0092] Here, for example, the processing functions executed by the inverse gain execution function 121a, the DnCNN processing function 121c, and the gain execution function 121f, which are components of the noise reduction processing circuit 121 shown in Fig. 9, are recorded in the form of computer-executable programs in the storage circuit 170. The noise reduction processing circuit 121 is, for example, a processor, which reads out each program from the storage circuit 170 and executes the read program to realize the function corresponding to each program. In other words, the noise reduction processing circuit 121 in a state in which each program has been read out has each function shown in the noise reduction processing circuit 121 in Fig. 9.

[0093] 9 shows a case where the inverse gain execution function 121a, the DnCNN processing function 121c, and the gain execution function 121f are each realized by a single processing circuit (noise reduction processing circuit 121), but the embodiment is not limited to this. For example, the noise reduction processing circuit 121 may be configured by combining multiple independent processors, and each processor may realize each processing function by executing each program. Furthermore, each processing function of the noise reduction processing circuit 121 may be realized by being appropriately distributed or integrated into a single or multiple processing circuits.

[0094] As described above, the beamformer 130 according to the second embodiment performs variable aperture, changing the number of channels to be added according to the depth. Therefore, the inverse gain execution function 121a according to the second embodiment receives data related to the envelope transmitted by the envelope detector 140a. The inverse gain execution function 121a then multiplies the received data related to the envelope by the gain (gain value) multiplied by the IQ signal by the ATGC processing circuit 112b and a gain based on the number of channels to be added, which is changed according to the depth in the variable aperture.

[0095] For example, the inverse gain execution function 121a multiplies the data relating to the envelope by a gain GainComp(x,z) obtained by the following equation (1).

[0096]

number

[0097] In equation (1), ATGC(x,z) represents the gain (linear scale) applied to the IQ signal by the ATGC processing circuit 112b at depth z with received raster number x, and M(x,z) represents the number of channels added at depth z.

[0098] The inverse gain execution function 121a multiplies the data relating to the envelope by the product of the reciprocal of the gain (gain value) multiplied by the ATGC processing circuit 112b and the reciprocal of the positive square of the number of channels added at depth z, according to equation (1). In this way, the inverse gain execution function 121a performs processing on the data relating to the envelope to make the level of noise contained in the data relating to the envelope constant at all depths. This makes the noise uniform spatially without varying with depth. Therefore, the data multiplied by the gain by the inverse gain execution function 121a can be applied to the DnCNN 121d, which estimates noise. As described above, in the second embodiment, the inverse gain execution function 121a performs processing on the data obtained by beamforming by the beamformer 130 to make the level of noise contained in the data relating to the envelope constant, based on the number of channels added, which changes depending on the depth.

[0099] The inverse gain execution function 121a then stores the data multiplied by the gain in the memory 121b. Specifically, the memory 121b has a two-dimensional storage area configured by the axis of the received raster (received raster direction) and the axis of the depth (depth direction). The inverse gain execution function 121a then stores the data multiplied by the gain in a corresponding position in the storage area of ​​the memory 121b. As a result, two-dimensional data of the received raster and depth is stored in the storage area of ​​the memory 121b.

[0100] The DnCNN processing function 121c includes a DnCNN 121d. For example, the DnCNN 121d is stored in an internal memory of the noise reduction processing circuit 121. The DnCNN processing function 121c reads out two-dimensional data stored in a storage area included in the memory 121b, and inputs the read-out two-dimensional data to the DnCNN 121d. In this way, the DnCNN processing function 121c causes the DnCNN 121d to estimate noise (residual). The DnCNN processing function 121c also transmits the read-out two-dimensional data to the subtractor 121e.

[0101] The DnCNN 121d estimates noise (noise components) contained in the input two-dimensional data and outputs the estimated noise to the subtractor 121e. For example, the DnCNN 121d is a two-dimensional DCNN. Various internal parameters such as coefficients of the DnCNN 121d are determined by, for example, learning similar to the learning performed on the above-mentioned DnCNN 120d.

[0102] The noise level of the ultrasound diagnostic device 1 varies depending on the ultrasound probe 101 and device conditions (conditions of the ultrasound diagnostic device 1). Therefore, the learning device may train the DnCNN 121d at each of a plurality of noise levels and generate, for each noise level, a DnCNN 121d having internal parameters corresponding to each of the plurality of noise levels. That is, the learning device may generate a plurality of DnCNNs 121d corresponding to the plurality of noise levels. The noise reduction processing circuit 121 may then hold the plurality of DnCNNs 121d generated by the learning device, select from the plurality of DnCNNs 121d a DnCNN 121d that matches the noise level corresponding to the ultrasound probe 101 or device conditions, and estimate noise contained in the two-dimensional data using the selected DnCNN 121d.

[0103] In addition, the noise reduction processing circuit 121 may hold SNR-related data such as that disclosed in Patent Document 1 as an index for adjusting the internal parameters of DnCNN 121d, and adjust the internal parameters of DnCNN 121d when estimating noise using a method similar to that disclosed in Patent Document 1.

[0104] The subtractor 121e subtracts the noise estimated and output by the DnCNN 121d from the two-dimensional data (original signal) transmitted by the DnCNN processing function 121c. As a result, the subtractor 121e obtains two-dimensional data with reduced noise. The two-dimensional data is two-dimensional data configured with a received raster axis (received raster direction) and a depth axis (depth direction). The subtractor 121e then transmits the two-dimensional data with reduced noise to the gain execution function 121f.

[0105] The gain execution function 121f multiplies the two-dimensional data transmitted by the subtractor 121e by a gain that is the inverse of the gain (gain value) that the inverse gain execution function 121a multiplied the IQ signal with. That is, the gain execution function 121f restores the gain. Then, the gain execution function 121f transmits the two-dimensional data multiplied by the gain to the beamformer 130.

[0106] In the first embodiment, the data input to the DnCNN 120d is two-dimensional data (two-dimensional IQ signals) of channel and depth. For this reason, in the first embodiment, the DnCNN 120d is operated for each ultrasonic wave transmission to reduce noise, so it is not affected by movement. In contrast, in the second embodiment, the DnCNN 121d is operated after one frame of ultrasonic waves is transmitted to obtain a total reflected wave signal, so if there is movement in the living body, it is affected by the movement of the living body. However, unlike CNNs for normal signals, the DnCNN 121d, which outputs noise, is less affected by the movement of the living body.

[0107] In the above description, two-dimensional data in the received raster direction and depth direction is used as an example of the processing performed for each frame. However, the DnCNN processing function 121c may use data from multiple frames to input three-dimensional data composed of the frame direction, received raster direction, and depth direction to the DnCNN 121d. In other words, the DnCNN 121d may be configured to support three dimensions (three-dimensional configuration). This allows the DnCNN 121d to estimate noise and output the estimated noise to the subtractor 121e. The DnCNN processing function 121c may also perform processing to improve the S / N ratio by adding the same location in multiple frames.

[0108] The noise reduction processing circuit 121 also includes a beam former 130, an envelope detector 140a, and In this case, the data input to the noise reduction processing circuit 121 and the data output from the noise reduction processing circuit 121 with noise reduced are complex number data.

[0109] Furthermore, the noise reduction processing circuit 121 may be provided at a stage subsequent to the logarithmic compressor 140b. In this case, the inverse gain execution function 121a or the DnCNN processing function 121c of the noise reduction processing circuit 121 performs an inverse transformation of the logarithmic compression performed by the logarithmic compressor 140b on the input data. The DnCNN processing function 121c then inputs the inversely transformed data to the DnCNN 121d, causing the DnCNN 121d to estimate noise and output the data. The gain execution function 121f then performs logarithmic compression again on the noise-reduced data. Note that the DnCNN processing function 121c may directly input the logarithmic compressed data to the DnCNN 121d.

[0110] The above has described the ultrasonic diagnostic device 1 according to the second embodiment. The ultrasonic diagnostic device 1 according to the second embodiment provides the same effects as the ultrasonic diagnostic device 1 according to the first embodiment.

[0111] (Third embodiment) In the first embodiment, a case has been described in which the DnCNN processing function 120c reads out two-dimensional IQ signals stored in the memory 120b and inputs the read out two-dimensional IQ signals to the DnCNN 120d without processing them. However, the DnCNN processing function 120c may process the two-dimensional IQ signals and input them to the DnCNN 120d. Therefore, such an embodiment will be described as the third embodiment. Note that the description of the third embodiment will mainly focus on differences from the first embodiment, and a description of the same configuration as the first embodiment may be omitted.

[0112] 10 and 11 are diagrams for explaining an example of processing executed by the DnCNN processing function 120c according to the third embodiment. In the third embodiment, the DnCNN processing function 120c reads out a two-dimensional IQ signal (two-dimensional data) 20 shown in Fig. 10 from the memory 120b. Hereinafter, the two-dimensional IQ signal 20 is data in which "CH0" IQ signals are arranged in the channel direction and "D0" IQ signals are arranged in the depth direction, and an example will be described in which the data is composed of "CH0" x "D0" IQ signals.

[0113] Then, the DnCNN processing function 120c transmits the read two-dimensional IQ signal 20 to the subtractor 120e.

[0114] Furthermore, the DnCNN processing function 120c divides the read two-dimensional IQ signal 20 into a plurality of regions 21. Below, the region 21 is data in which "CH1" IQ signals are arranged in the channel direction and "D1" IQ signals are arranged in the depth direction, and an example will be described in which the data is composed of "CH1" x "D1" IQ signals. In this case, the DnCNN processing function 120c divides the two-dimensional IQ signal 20 into a plurality of "CHn" x "Dn" regions 21. Note that "CHn" = "CH0" / "CH1" and "Dn" = "D0" / "D1".

[0115] Then, the DnCNN processing function 120c performs a two-dimensional discrete cosine transform (DCT) on each of the multiple regions 21, converting each of the multiple regions 21 into frequencies 22 (see FIG. 11). Note that one region 21 is converted into "CH1" x "D1" frequencies 22 by the two-dimensional discrete cosine transform.

[0116] The DnCNN processing function 120c sets "CH1" x "D1" frequencies 22 corresponding to one region 21 as depth data of data (input data) input to the DnCNN 120d. For example, the DnCNN processing function 120c arranges "CH1" x "D1" frequencies 22 corresponding to each region 21 in the depth direction, as shown in FIG. 11. That is, in the example of FIG. 11, a set of "CH1" x "D1" frequencies 22 arranged in the depth direction is arranged as "CHn" in the channel direction and as "Dn" in the depth direction.

[0117] In the field of CNN, the term "channel" is generally used instead of "depth." However, to distinguish it from the above-mentioned channels in the ultrasound diagnostic device 1, it is referred to here as the depth of the input data or feature map.

[0118] The DnCNN processing function 120c inputs a set of "CH1" x "D1" frequencies 22 arranged in the depth direction corresponding to each region 21 to the DnCNN 120d. In this way, the DnCNN processing function 120c causes the DnCNN 120d to estimate noise (residuals). The DnCNN processing function 120c also transmits the set of "CH1" x "D1" frequencies 22 arranged in the depth direction corresponding to each region 21 to the subtractor 120e.

[0119] The DnCNN 120d estimates noise (noise components) contained in the input set of "CH1" x "D1" frequencies 22 and outputs the estimated noise. That is, the DnCNN 120d outputs noise corresponding to each region 21. Various internal parameters such as coefficients of the DnCNN 120d according to the third embodiment are determined by, for example, learning similar to the learning performed on the DnCNN 120d according to the first embodiment described above.

[0120] The DnCNN processing function 120c acquires the noise corresponding to each region 21 output by the DnCNN 120d. Then, the DnCNN processing function 120c performs a two-dimensional discrete inverse cosine transform on the acquired noise corresponding to each region 21, converting the noise corresponding to each region 21 into "CH1" x "D1" signals. For example, such converted signals are noise included in the IQ signals corresponding to each region 21.

[0121] The DnCNN processing function 120c performs this processing for all "CHn" x "Dn" regions 21. As a result, the DnCNN processing function 120c acquires "CH0" x "D0" two-dimensional signals corresponding to all regions 21. Here, these "CH0" x "D0" two-dimensional signals are noise contained in the two-dimensional IQ signal 20. Then, the DnCNN processing function 120c transmits the acquired "CH0" x "D0" two-dimensional signals to the subtractor 120e.

[0122] The subtractor 120e subtracts the "CH0" x "D0" two-dimensional signals (noise) transmitted by the DnCNN processing function 120c from the two-dimensional IQ signal (original signal) 20 transmitted by the DnCNN processing function 120c. As a result, the subtractor 120e obtains a two-dimensional IQ signal with reduced noise. The subtractor 120e then transmits the two-dimensional IQ signal with reduced noise to the gain execution function 120f. The gain execution function 120f according to the third embodiment performs processing similar to the processing performed by the gain execution function 120f according to the first embodiment.

[0123] As described above, in the third embodiment, the DnCNN processing function 120c divides the IQ signal 20, which has been subjected to processing for making the noise level constant as processing for reducing noise contained in the IQ signal 20, into a plurality of regions 21. The DnCNN processing function 120c then performs a predetermined orthogonal transform on each of the plurality of regions 21 to acquire a plurality of frequencies 22 corresponding to each of the plurality of regions 21. The DnCNN processing function 120c then uses the plurality of frequencies 22 corresponding to each of the plurality of regions 21 as depth data of the data to be input to the neural network, and performs an inverse transform of the predetermined orthogonal transform on the output of the neural network.

[0124] The above has described the ultrasonic diagnostic device 1 according to the third embodiment. The ultrasonic diagnostic device 1 according to the third embodiment provides the same effects as the ultrasonic diagnostic device 1 according to the first embodiment.

[0125] The noise reduction processing circuit 120 according to the third embodiment divides a two-dimensional IQ signal 20 into multiple regions 21 and performs a discrete cosine transform on the signal, thereby more strongly conveying the statistical properties of the signal to the DnCNN 120d and making it easier for the DnCNN 120d to separate signal components and noise components. The noise reduction processing circuit 120 may also perform a Karhunen-Loeve transform instead of the discrete cosine transform. The Karhunen-Loeve transform is an orthogonal transform that concentrates energy as much as possible in low-order regions, and the discrete cosine transform has similar properties. By performing such an orthogonal transform, the noise reduction processing circuit 120 makes it easier for the DnCNN 120d to efficiently extract residuals (noise). The noise reduction processing circuit 120 can also use other orthogonal transforms, such as a discrete Fourier transform or a discrete wavelet transform.

[0126] In addition, the noise reduction processing circuit 120 according to the third embodiment can also learn to output a noise-reduced signal after inverse transformation instead of a residual, and output the output as is without subtracting it from the original signal.

[0127] (Fourth embodiment) In the first to third embodiments, the noise reduction processing circuits 120, 121 have been described as reducing noise using a neural network. However, the noise reduction processing circuits 120, 121 may reduce noise without using a neural network. Therefore, such an embodiment will be described as the fourth embodiment. Note that the description of the fourth embodiment will mainly focus on differences from the first embodiment, and a description of the same configuration as the first embodiment may be omitted.

[0128] FIG. 12 is a diagram showing an example of the configuration of a noise reduction processing circuit 122 according to the fourth embodiment. The ultrasound diagnostic apparatus 1 according to the fourth embodiment differs from the ultrasound diagnostic apparatus 1 according to the first embodiment in that it includes a noise reduction processing circuit 122 shown in FIG. 12 instead of the noise reduction processing circuit 120 shown in FIG. 5. The noise reduction processing circuit 122 according to the fourth embodiment also differs from the noise reduction processing circuit 120 according to the first embodiment in that it includes a noise reduction processing function 122a instead of the DnCNN processing functions 120c, 120d, and the subtractor 120e. The noise reduction processing circuit 122 according to the fourth embodiment also differs from the noise reduction processing circuit 120 according to the first embodiment in that it includes a gain execution function 122b instead of the gain execution function 120f. The noise reduction processing circuit 122 is an example of a processing circuit.

[0129] Here, for example, the processing functions executed by the inverse gain execution function 120a, the noise reduction processing function 122a, and the gain execution function 122b, which are components of the noise reduction processing circuit 122 shown in Fig. 12, are recorded in the form of computer-executable programs in the storage circuit 170. The noise reduction processing circuit 122 is, for example, a processor, which reads out each program from the storage circuit 170 and executes the read program to realize the function corresponding to each program. In other words, the noise reduction processing circuit 122 in a state in which each program has been read out has each function shown in the noise reduction processing circuit 122 in Fig. 12.

[0130] 12 shows a case where the inverse gain execution function 120a, the noise reduction processing function 122a, and the gain execution function 122b are each realized by a single processing circuit (noise reduction processing circuit 122), but the embodiment is not limited to this. For example, the noise reduction processing circuit 122 may be configured by combining multiple independent processors, and each processor may realize each processing function by executing a respective program. Furthermore, each processing function of the noise reduction processing circuit 122 may be realized by being appropriately distributed or integrated into a single or multiple processing circuits.

[0131] 13 is a diagram illustrating an example of processing executed by the noise reduction processing function 122a according to the fourth embodiment. The noise reduction processing function 122a reads out the two-dimensional IQ signal 20 shown in FIG. 13 from the memory 120b, and divides the read out two-dimensional IQ signal 20 into a plurality of regions 25.

[0132] Then, the noise reduction processing function 122a calculates a covariance matrix from the region 25 for each region 25. In this way, multiple covariance matrices corresponding to all of the regions 25 are calculated. Then, the noise reduction processing function 122a adds up the multiple covariance matrices corresponding to all of the regions 25. In this way, one matrix is ​​obtained.

[0133] Then, the noise reduction processing function 122a performs eigenvalue decomposition on one matrix obtained by adding together a plurality of covariance matrices, and calculates a filter matrix that outputs only a predetermined number of top eigenvalues ​​(principal components).

[0134] The noise reduction processing function 122a then applies a filter matrix to each region 25 to acquire only the principal components. The noise reduction processing function 122a then integrates all principal components corresponding to all regions 25 to acquire an overall principal component signal. This principal component signal is a signal from which non-principal component signals have been removed, i.e., a signal from which residual (noise) signals have been removed or reduced. This technique is the same as performing a Karhunen-Loeve orthogonal transform, filtering on the frequency axis, and then performing an inverse transform. That is, this is the same as the third embodiment, in which the ultrasound diagnostic apparatus 1 performs a Karhunen-Loeve transform instead of a discrete cosine transform, filtering on the frequency axis instead of the DnCNN 120d, and performing a Karhunen-Loeve inverse transform instead of the inverse discrete cosine transform. As the orthogonal transform, in addition to the Karhunen-Loeve transform, which adaptively changes depending on the input data, fixed orthogonal transforms such as a discrete Fourier transform, a discrete cosine transform, and a discrete wavelet transform can be used. Inverse transforms of these can also be used. That is, the noise reduction processing function 122a reduces noise using a predetermined orthogonal transform and an inverse transform of the predetermined orthogonal transform.

[0135] The noise reduction processing function 122a according to the fourth embodiment divides an IQ signal 20, which has been subjected to processing for making the noise level constant, into a plurality of regions 21 as a noise reduction process. Then, the noise reduction processing function 122a performs principal component analysis, Karhunen-Loeve transform, or eigenvalue decomposition on each of the plurality of regions 21 to reduce noise. The noise reduction processing function 122a may also perform singular value decomposition on each of the plurality of regions 21 to reduce noise. Alternatively, the noise reduction processing function 122a may also perform processing mathematically equivalent to at least one of principal component analysis, Karhunen-Loeve transform, eigenvalue decomposition, and singular value decomposition on each of the plurality of regions 21 to reduce noise. Alternatively, the noise reduction processing function 122a may also reduce noise contained in each of the plurality of regions 21 by utilizing statistical properties.

[0136] The above has described the ultrasonic diagnostic device 1 according to the fourth embodiment. The ultrasonic diagnostic device 1 according to the fourth embodiment provides the same effects as the ultrasonic diagnostic device 1 according to the first embodiment.

[0137] (Fifth embodiment) In the first to fourth embodiments, the cases where the ultrasound diagnostic apparatus 1 executes various processes have been described, but a medical image processing apparatus may execute processes similar to the various processes executed by the ultrasound diagnostic apparatus 1. Therefore, such an embodiment will be described as the fifth embodiment. Note that in the description of the fifth embodiment, differences from the first embodiment will be mainly described, and descriptions of configurations similar to those of the first embodiment may be omitted.

[0138] 14 is a diagram showing an example of the configuration of a medical image processing apparatus 10 according to the fifth embodiment. The medical image processing apparatus 10 acquires, via a network, from the ultrasound diagnostic apparatus 1, a digital IQ signal converted from an RF signal by a demodulator 112d. This IQ signal is a digital received signal converted from an analog received signal that has been subjected to a gain multiplication process while changing the gain according to depth. The medical image processing apparatus 10 then performs the same process as that performed by the ultrasound diagnostic apparatus 1 on the acquired IQ signal.

[0139] As shown in FIG. 14, the medical image processing apparatus 10 includes a network (NW) interface 11, a storage circuitry 12, an input interface 13, a display 14, and a processing circuitry 15.

[0140] The NW interface 11 controls the transmission and communication of various information and data transmitted and received between the medical image processing device 10 and the ultrasound diagnostic device 1. The NW interface 11 is connected to the processing circuit 15. The NW interface 11 receives the IQ signal transmitted by the ultrasound diagnostic device 1 via the network. In this case, the NW interface 11 transmits the received IQ signal to the processing circuit 15. For example, the NW interface 11 is realized by a network card, a network adapter, a NIC (Network Interface Controller), or the like.

[0141] The memory circuitry 12 is connected to the processing circuitry 15 and stores various data. For example, the memory circuitry 12 is realized by a semiconductor memory element such as a RAM (Random Access Memory), a flash memory, a hard disk, or an optical disk. The memory circuitry 12 is an example of a memory unit.

[0142] The memory circuitry 12 also stores various information used in the processing of the processing circuitry 15, the processing results of the processing circuitry 15, and the like.

[0143] The input interface 13 is connected to the processing circuitry 15, converts input operations received from an operator into electrical signals, and outputs the signals to the processing circuitry 15. In this specification, the input interface 13 is not limited to an interface equipped with physical operation components such as a mouse and a keyboard. For example, a processing circuit that receives electrical signals corresponding to input operations from an external input device provided separately from the device and outputs the electrical signals to the processing circuitry 15 is also included as an example of the input interface.

[0144] For example, the input interface 13 may be realized by a trackball for making various settings, a switch button, a mouse, a keyboard, a touchpad for performing input operations by touching the operation surface, a touchscreen in which the display screen and touchpad are integrated, a non-contact input interface using an optical sensor, or a voice input interface.

[0145] The display 14 is connected to the processing circuit 15 and displays various information and images output from the processing circuit 15. For example, the display 14 is realized by a liquid crystal monitor, a CRT (Cathode Ray Tube) monitor, or a touch panel. For example, the display 14 displays a GUI (Graphical User Interface) for receiving instructions from an operator, various display images, and various processing results by the processing circuit 15. The display 14 is an example of a display unit.

[0146] The processing circuit 15 is realized by a processor. The processing circuit 15 executes an acquisition function 15a, a noise reduction processing function 15b, and an image generation function 15c. Here, for example, the acquisition function 15a, the noise reduction processing function 15b, and the image generation function 15c, which are components of the processing circuit 15 shown in FIG. 14, are recorded in the storage circuit 12 in the form of a program executable by a computer. The processing circuit 15 reads each program from the storage circuit 12 and executes the read program to realize the function corresponding to each program. In other words, the processing circuit 15 in a state in which each control program has been read has each function shown in the processing circuit 15 of FIG. 14.

[0147] 14 shows a case where the acquisition function 15a, noise reduction processing function 15b, and image generation function 15c are each realized by a single processing circuit 15, but the embodiment is not limited to this. For example, the processing circuit 15 may be configured by combining multiple independent processors, and each processor may realize each processing function by executing a respective program. Furthermore, each processing function of the processing circuit 15 may be realized by being appropriately distributed or integrated into a single or multiple processing circuits.

[0148] The acquisition function 15a acquires the IQ signals from the ultrasound diagnostic device 1 via the NW interface 11. Then, the acquisition function 15a stores the acquired IQ signals in the storage circuitry 12. The acquisition function 15a is an example of an acquisition unit.

[0149] The noise reduction processing function 15b performs processing similar to that performed by the noise reduction processing circuit 120, the noise reduction processing circuit 121, or the noise reduction processing circuit 122. For example, the noise reduction processing function 15b reads out an IQ signal stored in the storage circuit 12, and performs processing similar to that performed by the noise reduction processing circuit 120, the noise reduction processing circuit 121, or the noise reduction processing circuit 122 on the read IQ signal. To explain this as an example, the noise reduction processing function 15b performs processing on the IQ signal to make the level of noise contained in the IQ signal constant, and then performs processing to reduce the noise contained in the IQ signal after the processing to make the noise level constant has been performed. The noise reduction processing function 15b is an example of a noise reduction processing unit.

[0150] The image generation function 15c executes at least a part of the same processes as those executed by the beamformer 130, the signal processing circuit 140, the image generation circuit 150, and the control circuit 180 according to the first embodiment.

[0151] However, in the fifth embodiment, when the noise reduction processing function 15b and the image generation function 15c perform processing, an input interface 13, a display 14 and a memory circuit 12 are used instead of the input device 102, the display 103, the image memory 160 and the memory circuit 170 of the first embodiment.

[0152] The medical image processing apparatus 10 according to the fifth embodiment has been described above. The medical image processing apparatus 10 according to the fifth embodiment can provide the same effects as the ultrasound diagnostic apparatus 1 according to any one of the first to fourth embodiments.

[0153] The term "processor" used in the above description refers to a circuit such as a central processing unit (CPU), a graphics processing unit (GPU), an application specific integrated circuit (ASIC), or a programmable logic device (e.g., a simple programmable logic device (SPLD), a complex programmable logic device (CPLD), or a field programmable gate array (FPGA)). The processor reads a program stored in the memory circuit 170 and executes the read program to realize its function. Note that instead of storing the program in the memory circuit 170, the program may be directly embedded in the processor circuit. In this case, the processor reads and executes the program embedded in the circuit to realize its function. Note that each processor in this embodiment is not limited to being configured as a single circuit, but may be configured as a single processor by combining multiple independent circuits to realize its function. Furthermore, the multiple circuits in FIGS. 4 and 5 may be integrated into a single processor to realize its function.

[0154] The control program may be provided by being recorded in a computer-readable, non-transitory storage medium such as a CD (Compact Disk)-ROM, a FD (Flexible Disk), a CD-R (Recordable), or a DVD (Digital Versatile Disk) in a computer-installable or executable file format. The control program may also be stored on a computer connected to a network such as the Internet and provided or distributed by being downloaded via the network. For example, the control program may be composed of modules including the above-described processing functions. In actual hardware, a processor reads and executes the program from a storage medium such as a ROM, whereby each module is loaded into a main memory device and generated on the main memory device.

[0155] According to at least one of the embodiments or modifications described above, noise can be effectively reduced.

[0156] Although several embodiments have been described, these embodiments are presented as examples and are not intended to limit the scope of the invention. These embodiments can be implemented in various other forms, and various omissions, substitutions, and modifications can be made without departing from the spirit of the invention. These embodiments and their modifications are included within the scope and spirit of the invention, as well as within the scope of the invention and its equivalents as defined in the claims. [Explanation of symbols]

[0157] 1. Ultrasound diagnostic equipment 112 receiving circuit 120, 121, 122 Noise reduction processing circuit

Claims

1. a receiving circuit that performs a first process of multiplying an analog received signal by a gain that varies depending on the depth, and a second process of converting the first-processed received signal into a digital received signal; a processing circuit that performs third processing on the digital reception signal by multiplying the digital reception signal by a gain that varies depending on depth and is the opposite of the gain multiplied in the first processing so that the level of noise contained in the digital reception signal becomes spatially uniform; performs fourth processing to spatially uniformly reduce the level of noise contained in the digital reception signal after the third processing; and performs fifth processing to multiply the reception signal after the noise has been reduced by a gain; Equipped with As the fourth processing, the processing circuit inputs an added signal obtained by adding a noise signal having a spatially uniform noise level to a signal without spatially uniform noise, inputs the digital received signal after the third processing to a neural network trained to output a signal equivalent to the noise signal, acquires a signal equivalent to the noise signal from the neural network, and subtracts the acquired signal from the digital received signal, thereby spatially uniformly reducing the noise level included in the received signal.

2. 2. The ultrasonic diagnostic apparatus according to claim 1, wherein the processing circuitry further performs, as the third processing, processing for making a noise level included in the digital received signal constant across all channels and all depths based on variations in sensitivity between channels.

3. The ultrasonic diagnostic apparatus according to claim 1 , further comprising a beamformer that performs beamforming on the received signals that have been subjected to the fifth processing.

4. a beam former that performs beamforming on the digital reception signal converted by the second processing, changing the number of channels to be added according to depth; 2. The ultrasound diagnostic apparatus according to claim 1, wherein the processing circuitry performs, as the third processing, processing to make constant a level of noise contained in the data obtained by the beamforming by the beamformer, based on the number of channels to be added, which is changed according to the depth.

5. 2. The ultrasound diagnostic apparatus of claim 1, wherein the processing circuitry, as the fourth processing, divides the digital received signal that has been subjected to the third processing into a plurality of regions, performs a predetermined orthogonal transform on each of the plurality of regions to acquire a plurality of frequencies corresponding to each of the plurality of regions, inputs the plurality of frequencies corresponding to each of the plurality of regions to the neural network, and performs an inverse transform of the predetermined orthogonal transform on an output of the neural network.

6. an acquisition unit that acquires a digital reception signal that has been processed by multiplying the signal by a gain that varies depending on the depth; a noise reduction processing unit that performs a process of multiplying the digital reception signal by a gain that varies depending on depth and is the inverse of the gain multiplied in the process so that the level of noise included in the digital reception signal becomes spatially uniform, performs a process of spatially uniformly reducing the level of noise included in the digital reception signal after the process of multiplying by the inverse gain, and performs a process of multiplying the noise-reduced reception signal by a gain; Equipped with The noise reduction processing unit receives an added signal in which a noise signal with a spatially uniform noise level is added to a spatially uniform noise-free signal, receives the digital received signal after being processed by multiplying the gain by the inverse of the input signal to a neural network trained to output a signal equivalent to the noise signal, obtains a signal equivalent to the noise signal from the neural network, and subtracts the obtained signal from the digital received signal, thereby spatially uniformly reducing the noise level contained in the received signal.

Images