WO2021004076A1 - 基于人工智能芯片的适形穿戴式生物信息监测设备及系统 - Google Patents

基于人工智能芯片的适形穿戴式生物信息监测设备及系统 Download PDF

Info

Publication number
WO2021004076A1
WO2021004076A1 PCT/CN2020/077932 CN2020077932W WO2021004076A1 WO 2021004076 A1 WO2021004076 A1 WO 2021004076A1 CN 2020077932 W CN2020077932 W CN 2020077932W WO 2021004076 A1 WO2021004076 A1 WO 2021004076A1
Authority
WO
WIPO (PCT)
Prior art keywords
heart
imaging
ultrasonic
time
signal
Prior art date
Application number
PCT/CN2020/077932
Other languages
English (en)
French (fr)
Inventor
张鹏飞
刘治
Original Assignee
山东大学
山东大学齐鲁医院
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Priority claimed from CN201910605155.0A external-priority patent/CN112168140B/zh
Priority claimed from CN201911282999.2A external-priority patent/CN110974304B/zh
Priority claimed from CN201911283022.2A external-priority patent/CN110974305B/zh
Application filed by 山东大学, 山东大学齐鲁医院 filed Critical 山东大学
Publication of WO2021004076A1 publication Critical patent/WO2021004076A1/zh

Links

Images

Classifications

    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00—Measuring for diagnostic purposes; Identification of persons
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B8/00—Diagnosis using ultrasonic, sonic or infrasonic waves
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B8/00—Diagnosis using ultrasonic, sonic or infrasonic waves
    • A61B8/08—Detecting organic movements or changes, e.g. tumours, cysts, swellings
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F13/00—Interconnection of, or transfer of information or other signals between, memories, input/output devices or central processing units
    • G06F13/14—Handling requests for interconnection or transfer
    • G06F13/20—Handling requests for interconnection or transfer for access to input/output bus
    • G06F13/28—Handling requests for interconnection or transfer for access to input/output bus using burst mode transfer, e.g. direct memory access DMA, cycle steal
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F9/00—Arrangements for program control, e.g. control units
    • G06F9/06—Arrangements for program control, e.g. control units using stored programs, i.e. using an internal store of processing equipment to receive or retain programs
    • G06F9/30—Arrangements for executing machine instructions, e.g. instruction decode
    • G06F9/38—Concurrent instruction execution, e.g. pipeline, look ahead

Definitions

  • the present disclosure relates to the field of biological information monitoring, and in particular to conformal wearable biological information monitoring equipment and systems based on artificial intelligence chips.
  • physiological parameters of the human body from commonly used blood pressure, body temperature, heart rate, to more complex organ activity states, all change under different physiological states or different pathological states.
  • the physiological changes of various parameters are the human body's response to the coordinated work of various organs and tissues in order to adapt to different states, and the changes in the pathological state of each parameter reflect the state of the human body's disease process.
  • these parameters are usually measured under various conditions, such as changes in blood pressure and heart rate after exercise, which should be measured immediately after exercise.
  • wearable biological information monitoring equipment for obtaining heart rate, blood pressure, and ECG signals. These devices either require multiple signal detection lines to cause cumbersome wear and signal detection susceptible to interference, or the detection methods are too simple and poorly accurate.
  • wearable physiological parameter measurement equipment including various types of equipment worn on the wrist, chest, etc., can detect heart rate, blood oxygen saturation, blood pressure, and ECG signals.
  • the photoplethysmograph PhotoPlethysmoGraphy
  • the electrocardiogram signal, pulse wave detection, Korotkoff-Sound detection, or a combination of the above methods are used.
  • the existing solution is to increase the complexity of the device to obtain more comprehensive and accurate biological information, which means that the device must perform more information processing. This is bound to increase the size of the device, which not only increases the power consumption of the device and is not conducive to wearing, but also has a poor wearing experience.
  • hydrodynamic parameters such as tissue blood flow and intra-articular pressure are important parameters for evaluating organs, tissues, and especially micro-joint functions. These indicators are difficult to achieve with existing wearable detection technologies. More important is the visualization of collected biological information, such as the movement of blood vessels and the display of blood flow velocity.
  • the inventor also discovered during the development process that the current portable cardiac ultrasound imaging method mainly uses a 1-dimensional phased array ultrasound transducer, supplemented by an array ultrasound imaging system.
  • This method can only obtain some two-dimensional ultrasound images, but cannot obtain the three-dimensional shape of the heart, and cannot monitor the motion state of the heart in real time.
  • the existing imaging methods can only be performed in the hospital and are limited by the venue, and cannot detect the heart shape of the patient in different motion states.
  • the direction and position of the ultrasound transmitted and received by the transducer must be controlled.
  • a set of array elements performs beam synthesis to generate One beam scans the sound beam and receives the signal, and then the next group generates the next sound beam and receives the signal.
  • an ultrasound image can be synthesized. Increasing the scanning line density can improve the spatial resolution of the image, but it also produces a large amount of data.
  • the current mobile cardiac ultrasound equipment adopts methods to reduce the amount of data such as reducing imaging elements and resolution, and the imaging modalities are mostly limited to one-dimensional, two-dimensional and multi-color. Puller cannot perform three-dimensional reconstruction and three-dimensional information calculation; on the other hand, in order to achieve real-time high frame rate imaging, higher requirements are placed on the hardware system, and the frame frequency and time resolution of existing mobile imaging equipment are affected. limit.
  • the high requirements for the hardware architecture are contrary to the requirements for the miniaturization, integration, and weight reduction of mobile (including but not limited to trolley type, hand-held, palm-top type, etc.) or wearable ultrasound equipment, and some handheld ultrasound equipment and Wearable devices, in order to further reduce the weight, power consumption and volume of the device, part of the beam synthesis and imaging processing are remotely processed and imaged through interactive methods such as cloud computing.
  • This also has high requirements for the data transmission rate of the device, and Mobile ultrasound has more applications in remote areas and poor network coverage.
  • traditional ultrasound equipment architecture and imaging methods are no longer available, and they are used in mobile or wearable ultrasound imaging equipment with limited hardware, data transmission resources, and computing power.
  • the present disclosure provides a conformal wearable biological information monitoring device and system based on an artificial intelligence chip.
  • the technical solution of the conformal wearable biological information monitoring device based on the artificial intelligence chip provided by the present disclosure is:
  • a wearable biological information monitoring equipment based on artificial intelligence chips including analog front-end circuit modules, digital front-end modules and artificial intelligence chips;
  • the analog front-end circuit module is used to generate voltage pulses that excite the ultrasonic transducer, receive the echo electrical signal collected by the ultrasonic array transducer, and perform impedance matching on it, and the echo electrical signal after impedance matching is amplified After the analog-to-digital conversion, input the digital front-end module;
  • the artificial intelligence chip performs different operations on the original biological information data according to the imaging purpose and imaging mode to obtain the imaging region of interest and the points of interest, and output control instructions to the digital front-end module;
  • the digital front-end module is used to collect the echo signal of the desired imaging point after receiving the control instruction output by the artificial intelligence chip, and perform dynamic beam synthesis on it; the signal after beam synthesis is filtered, quadrature demodulated and batched After processing and estimation of the flow rate, ultrasound image reconstruction and real-time imaging are realized.
  • an artificial intelligence chip includes a four-stage composite instruction pipeline and an FPE array convolution calculation unit;
  • the four-level composite instruction pipeline includes a first instruction pipeline for obtaining matrix operation instructions, a second instruction pipeline for processing the monitored raw biological information data, and a matrix for processing the processed biological information raw data.
  • the third instruction pipeline for arithmetic, fixed-point multiplication, and dot multiplication operations, and the fourth instruction pipeline for loading and storing the monitored raw data of biological information;
  • the FPE array convolution calculation unit is used for accumulating the original biological information data processed by the second instruction pipeline and the third instruction pipeline to reconstruct the imaging mode and the imaging region of interest.
  • the ultrasonic transmitting and receiving unit generates a voltage pulse that excites the ultrasonic transducer, and transmits it to the high-voltage pulse chip;
  • the high-voltage pulse chip receives the echo electrical signal collected by the ultrasonic array transducer and transmits it to the ultrasonic transmitter and receiver unit;
  • the ultrasonic transmitting and receiving unit receives the echo electric signal transmitted by the high-voltage pulse chip and performs impedance matching on it;
  • the echo electrical signal after impedance matching is amplified by the front-end receiving unit and converted from analog to digital, and then input to the digital front-end module;
  • the artificial intelligence chip performs different operations according to the imaging purpose and imaging mode to obtain the imaging area and points of interest, and output control instructions to the digital front-end module;
  • the digital front-end module After the digital front-end module receives the control instruction output by the artificial intelligence chip, it collects the echo signal of the desired imaging point and performs dynamic beam synthesis on it;
  • the signal after beam synthesis is filtered by a DC filter, quadrature demodulation by a demodulator, and processed by a processor to realize ultrasound image reconstruction and real-time imaging.
  • the present disclosure provides an ultrasonic beam synthesis system based on a conformal wearable multi-element imaging transducer
  • Ultrasonic beam synthesis system based on conformal wearable multi-element imaging transducer including:
  • a conformal wearable multi-element imaging transducer where the conformal wearable multi-element imaging transducer is set at the position of the chest wall on the body surface corresponding to the heart of the patient when in use;
  • the conformal wearable multi-element imaging transducer receives signals from the ultrasound transmitter and receiver unit of the wearable biological information monitoring equipment based on artificial intelligence chips. Each element of the conformal wearable multi-element imaging transducer will The received signal is transmitted to the patient's heart position after position compensation;
  • Each element of the conformal wearable multi-element imaging transducer receives a feedback signal, and each element of the feedback signal is time-delayed and sent to the ultrasonic emission of a wearable biological information monitoring device based on an artificial intelligence chip
  • the receiving unit, a wearable biological information monitoring device based on an artificial intelligence chip performs beam synthesis on the signal of each element to obtain a three-dimensional ultrasound image of the patient's heart.
  • a conformal wearable multi-element imaging transducer worn on the chest wall of the body surface corresponding to the heart of the patient receives the signal from the ultrasound transmitting and receiving unit of the wearable biological information monitoring device based on the artificial intelligence chip, Each element of the conformal wearable multi-element imaging transducer will transmit the received signal to the patient's heart position after position compensation;
  • Each element of the conformal wearable multi-element imaging transducer receives a feedback signal, and each element of the feedback signal is time-delayed and sent to the ultrasonic emission of a wearable biological information monitoring device based on an artificial intelligence chip
  • the receiving unit, a wearable biological information monitoring device based on an artificial intelligence chip performs beam synthesis on the signal of each element to obtain a three-dimensional ultrasound image of the patient's heart.
  • the present disclosure provides a remote cardiac ultrasound three-dimensional imaging system based on deep learning
  • the remote cardiac ultrasound 3D imaging system based on deep learning includes:
  • the user client is used to control the ultrasound transmitting and receiving unit of the conformal wearable biological information monitoring device based on the artificial intelligence chip, and transmit and select the conformal wearable multi-element imaging transducer that is worn on the chest wall of the human body during use. At the same time, it controls the ultrasonic transmitter and receiver unit of the wearable biological information monitoring device based on artificial intelligence chip to receive the ultrasonic signal fed back by the array element corresponding to the strobe signal command; upload the returned ultrasonic signal to the cloud server;
  • the cloud server is used to process the ultrasound signals uploaded by the user client, and use the pre-trained individual three-dimensional heart model to process the subject’s cardiac ultrasound two-dimensional images to obtain the subject’s real-time heart contour key points, based on Real-time key points of the heart contour to obtain real-time three-dimensional ultrasound imaging of the heart of the subject;
  • the doctor client is used to receive the key points of the heart contour of the subject selected by the doctor, and send the key points of the heart contour of the subject to the user client through the cloud server, which is used to instruct the user client to send a strobe signal instruction.
  • the user client acquires the real-time two-dimensional cardiac ultrasound image of the subject after the subject wears the conformal wearable multi-element imaging transducer during use; the subject will be acquired by the user client
  • the patient’s real-time cardiac ultrasound two-dimensional image is sent to the cloud server, and the cloud server sends the subject’s real-time cardiac ultrasound two-dimensional image to the doctor client; during the acquisition of the real-time cardiac ultrasound two-dimensional image, conformal wearable multi-element imaging
  • the transducer receives the ultrasound signal from the wearable biological information monitoring device based on artificial intelligence chip, and feeds back the feedback signal to the wearable biological information monitoring device based on artificial intelligence chip, forming the subject’s real-time two-dimensional cardiac ultrasound image;
  • the doctor client selects the key points of the subject’s heart contour from the subject’s real-time cardiac ultrasound two-dimensional image; the doctor client sends the selected key points of the subject’s heart contour to the cloud server;
  • the cloud server uses the subject’s real-time cardiac ultrasound two-dimensional image as the input value of the adaptive cardiac neural network model; the cloud server uses the manually selected key points of the subject’s cardiac contour as the output value of the adaptive cardiac neural network model.
  • Real-time imaging process the user client receives the key points of the subject’s heart contour and the subject’s individual three-dimensional heart model;
  • the user client sends a strobe instruction to the ultrasound transmitter and receiver unit of the wearable biological information monitoring device based on artificial intelligence chip, that is, the strobe instruction controls the ultrasonic transmitter and receiver unit to only address the key points of the contour.
  • the array element emits ultrasonic signals, and the array elements corresponding to non-contour key points do not emit ultrasonic signals;
  • the user client acquires the subject’s new real-time cardiac ultrasound two-dimensional image collected by the array element corresponding to the strobe command, and the user client inputs the subject’s new real-time cardiac ultrasound two-dimensional image into the subject’s individual three-dimensional In the heart model, the user client outputs the coordinates of key points of the subject’s heart contour in real time;
  • the artificial intelligence chip of the present disclosure has powerful on-site computing capabilities, and can complete the calculation process that can be completed by traditional large-scale integrated circuit boards on a microchip, making more complex and complete biological information acquisition on wearable devices Process immediately afterwards;
  • the artificial intelligence chip of the present disclosure has the advantage of being reconfigurable, so it is suitable for various artificial intelligence network structures such as RNN, CNN, FCN, and the chip user or application party can easily use it without knowing the physical structure of the chip.
  • Artificial intelligence algorithm instructions to realize the brain-like function that supports both spatial distribution and time long and short memory, better simulate the biological learning model of the human brain, and increase the artificial intelligence chip for a variety of information processing when used on wearable monitoring devices compatibility;
  • the present disclosure has the characteristics of on-chip incremental training. Through adaptive learning capabilities, the calculation accuracy of the chip is continuously improved, and it can cope with different algorithm application environments, adapt to the calculation and processing of different biological information, and adapt to different information visualization processes;
  • the present disclosure has the advantage of a fixed-point adjustable sparse network calculation function, achieves approximate floating-point calculation accuracy, and realizes the key requirements of high-precision calculation and low power consumption of the embedded system on chip.
  • the present disclosure innovatively proposes a technical solution for position compensation.
  • the present disclosure solves the problem that even if the relative position of the array element of the conformal transducer is not fixed and uncertain, an accurate image can be obtained by compensating for the change of the position of the array element during the imaging process. For example, the initial position of the array element is at point A, but since the attachment basis of the array element is conformal, the array element will deviate from point A as the body surface changes due to breathing exercise or muscle movement. Beam synthesis at point A will cause imaging errors, and the compensation problem in beam synthesis must be performed according to the actual position of the array element.
  • This disclosure uploads the complex processing of ultrasound signals to the cloud server, and the mobile or wearable signal collection terminal only needs to perform data collection and transmission, which greatly reduces the hardware and processing of the image collection and processing on the mobile or wearable terminal.
  • the software computing burden reduces the power consumption, heat generation and volume of the mobile terminal or the wearable terminal;
  • the mobile terminal or wearable terminal of the present disclosure uses artificial intelligence technology.
  • the doctor client can interactively determine a small number of key structure points of the heart, and use artificial intelligence technology to track several key structures in real time
  • the point signal greatly reduces the amount of data transmission, uses sparse, reduced-dimensional signals to achieve full-resolution imaging, and improves imaging speed, time and spatial resolution.
  • the present disclosure uses deep learning technology to realize automatic recognition of heart chambers, ventricular walls and valve structures, and complete three-dimensional reconstruction and three-dimensional information acquisition from a series of two-dimensional images of the short axis and long axis of the ventricle;
  • This disclosure uses deep learning technology to not only calculate the above-mentioned two-dimensional parameters such as the diameter of the heart cavity and the thickness of the wall, but also provide three-dimensional parameters such as the volume of the heart cavity, and can automatically calculate the ejection fraction and stroke volume And other functional parameters, can realize automatic report, and automatically alarm for abnormal parameters;
  • the present disclosure uses the interactive access architecture of the cloud server to realize the interactive real-time access of cardiac ultrasound images, which is beneficial to the information sharing of the inspected population by the family, the community and the specialized medical institutions, and the movement of the inspected is adjusted in real time State or treatment plan.
  • Fig. 1 is a structural diagram of an SOC chip in the first embodiment
  • FIG. 2 is a structural diagram of the pipeline structure of the SOC chip according to the first embodiment
  • Fig. 3 is a structural diagram of a FPE array convolution calculation unit in the first embodiment
  • FIG. 4 is a structural diagram of a wearable biological information monitoring device based on an SOC chip in Embodiment 1;
  • Figure 5 is a flow chart of the method of the second embodiment
  • Fig. 6 is a structural diagram of a wearable biological information monitoring device of the second embodiment
  • Fig. 7 is a schematic diagram of wearing a conformal wearable multi-element imaging transducer of the second embodiment
  • Fig. 8 is a detailed enlarged view of the conformal wearable multi-element imaging transducer of the second embodiment
  • 9(a) and 9(b) are schematic diagrams of the principle of beam combining in the second embodiment
  • Fig. 10 is a schematic diagram of the ultrasonic calibration method of the second embodiment
  • FIG. 11 is a schematic diagram of the installation of the fixed reflector in the second embodiment.
  • Fig. 12 is a schematic diagram of hardware connection in the third embodiment.
  • This embodiment provides a wearable biological information monitoring device based on an SOC chip. Please refer to FIG. 4.
  • a wearable biological information monitoring equipment based on artificial intelligence chips including analog front-end circuit modules, digital front-end modules and artificial intelligence chips;
  • the analog front-end circuit module is used to generate voltage pulses that excite the ultrasonic transducer, receive the echo electrical signal collected by the ultrasonic array transducer, and perform impedance matching on it, and the echo electrical signal after impedance matching is amplified After the analog-to-digital conversion, input the digital front-end module;
  • the artificial intelligence chip performs different operations on the original biological information data according to the imaging purpose and imaging mode to obtain the imaging region of interest and key structural points, and output control instructions to the digital front-end module;
  • the digital front-end module is used to collect the echo signal of the desired imaging point after receiving the control instruction output by the artificial intelligence chip, and perform dynamic beam synthesis on it; the signal after beam synthesis is filtered, quadrature demodulated and batched After processing and estimation of the flow rate, ultrasound image reconstruction and real-time imaging are realized.
  • analog front-end circuit module includes a high-voltage pulse chip, an ultrasonic transmitting and receiving unit, and an analog front-end receiving unit;
  • the high-voltage pulse chip is connected to the ultrasonic area array transducer through an interface, receives the echo electric signal collected by the ultrasonic area array transducer, and transmits it to the ultrasonic transmitting and receiving unit;
  • the ultrasonic transmitting and receiving unit includes a transmitting/receiving switch and a signal transmitter.
  • the input end of the signal transmitter is connected to the transmitting channel beam combiner in the digital front-end module, and the output end is connected to the transmitting/receiving switch for generating The voltage pulse that excites the ultrasonic transducer;
  • the transmitting/receiving switch is respectively connected to the high-voltage pulse chip and the analog front-end receiving module, and is used to transmit the electric pulse signal for exciting the ultrasonic transducer to the high-voltage pulse chip, and receive the high-voltage pulse chip sent After impedance matching is performed on the echo electrical signal, the impedance-matched echo signal is transmitted to the analog front-end receiving unit;
  • the analog front-end receiving unit includes a preamplifier and an analog-to-digital converter, and the echo electrical signal after impedance matching is amplified by the preamplifier and converted by the analog-to-digital converter, and then input to the digital front-end module.
  • the digital front-end module includes a transmitting channel beam synthesizer, time gain compensation, a receiving channel beam synthesizer, a DC filter, a demodulator, and a processor;
  • the time gain compensation is connected to the receiving channel beam synthesizer to compensate for the energy attenuation of the echo signal during propagation; the receiving channel beam synthesizer is connected to the analog front-end receiving unit and the artificial intelligence chip to receive the artificial intelligence chip
  • the sent control instructions collect the echo signals of the required imaging key points and perform dynamic beam synthesis; the signals after beam synthesis are filtered by a DC filter, demodulator quadrature demodulation and processor processing to realize ultrasound images Reconstruction and real-time imaging.
  • analog front-end circuit module the digital front-end module and the artificial intelligence chip respectively adopt flexible circuits.
  • the analog front-end circuit module includes a high-voltage pulse chip, an ultrasonic transmitting and receiving unit, and an analog front-end receiving unit.
  • the high-voltage pulse chip is connected to an ultrasonic area array transducer through an interface, and the high-voltage pulse chip is also connected to the ultrasonic transmitting and receiving unit.
  • the transmitting/receiving switch in the unit is connected to receive the echo electric signal collected by the ultrasonic array transducer and transmit it to the ultrasonic transmitting and receiving unit;
  • the ultrasonic transmitting and receiving unit includes a transmitting/receiving switch and a signal transmitter,
  • the input end of the signal transmitter is connected to the transmit channel beam combiner in the digital front-end module, and the output end is connected to the transmit/receive switch for generating a voltage pulse with a peak value of 70V to excite the ultrasonic transducer;
  • the transmit/receive switch They are connected to the high-voltage pulse chip and the preamplifier in the analog front-end receiving module respectively, and are used to transmit the electric pulse signal for exciting the ultrasonic transducer to the high-voltage pulse chip.
  • the impedance The matched echo signal is transmitted to the analog front-end receiving unit to realize flexible switching between the transmitted signal and the received signal;
  • the analog front-end receiving unit includes a preamplifier and an analog-to-digital converter, and the echo electrical signal after impedance matching passes through the front After the amplifier is amplified and converted by the analog-to-digital converter, it enters the receiving channel beam combiner in the digital front-end module for beam combining.
  • the digital front-end module is integrated in a field programmable gate array (FPGA) chip, and includes a transmit channel beam synthesizer, time gain compensation, a receive channel beam synthesizer, a DC filter, a demodulator, and a processor.
  • FPGA field programmable gate array
  • the transmit channel The beam synthesizer is connected to the signal transmitter, and is used to delay the pulse of the signal transmitter, so as to achieve focusing at a specific point in space and stimulate the pulse generation of the signal transmitter;
  • the time gain compensation is connected to the receiving channel beam synthesizer , Used to compensate for the energy attenuation of the echo signal in the propagation process;
  • the receiving channel beam combiner is connected to the analog front-end receiving unit and the SOC chip, and is used to calculate the echo signal delay time at each point in the space during the pre-tuning stage , Perform dynamic beam synthesis on the echo signal after time gain compensation; in the moving stage, receive the instruction of the SOC chip to perform beam synthesis on the echo signal from the key point;
  • the signal after beam synthesis is filtered and decomposed by the DC filter After quadrature demodulation by the modulator, Echo processing by the processor, and blood flow velocity estimation, ultrasonic image reconstruction and real-time imaging are realized.
  • the SOC chip performs different artificial intelligence network operations according to imaging purposes (such as cardiac imaging, blood vessel imaging) and imaging modes (such as cloud architecture, portable integrated imaging platform), and according to imaging regions of interest and heart
  • imaging purposes such as cardiac imaging, blood vessel imaging
  • imaging modes such as cloud architecture, portable integrated imaging platform
  • the receiving channel beam combiner unit is controlled to collect signals at the desired imaging points.
  • the wearable biological information monitoring device proposed in this embodiment uses FPGA to integrate digital front-end integration, and uses SOC chip to realize on-site calculation and control FPGA, and then control analog front-end circuit modules, miniaturizing the functions of the components of the traditional large-scale ultrasound equipment architecture Implemented on the module group. In addition to miniaturization, it achieves low voltage and low power consumption, and also reduces heat generation.
  • the circuit boards and connection lines of each module in the wearable biological information monitoring device proposed in this embodiment all adopt flexible circuits, such as polyimide, polyester, polyester and other polymer materials such as films or bonding sheets. Or copper, aluminum and other metal materials vacuum spraying and other processes to achieve wearable circuits.
  • the flexible circuit includes, but is not limited to, a lead line (lead line), a printed circuit (printed circuit), a connector (connector), and a multifunctional integrated system (integration of function).
  • This embodiment provides a reconfigurable artificial intelligence (SOC) chip that can implement artificial intelligence algorithms and can effectively manage large amounts of data transmission, which not only reduces the transmission of redundant data, and thus This further saves the power consumption of the entire wearable biological information monitoring device, and improves the effectiveness of data processing and transmission.
  • SOC reconfigurable artificial intelligence
  • the SOC chip includes a four-level composite instruction pipeline and an FPE array convolution calculation unit;
  • the four-level composite instruction pipeline includes a first instruction pipeline for obtaining matrix operation instructions, a second instruction pipeline for processing the monitored raw biological information data, and a matrix for processing the processed biological information raw data.
  • the third instruction pipeline for arithmetic, fixed-point multiplication, and dot multiplication operations, and the fourth instruction pipeline for loading and storing the monitored raw data of biological information;
  • the FPE array convolution calculation unit is used for accumulating the original biological information data processed by the second instruction pipeline and the third instruction pipeline to reconstruct the imaging mode, imaging region of interest and key structure points.
  • the first instruction pipeline includes an instruction prefetch buffer for prefetching instruction data received by the instruction interface;
  • the fourth instruction pipeline includes an instruction prefetch buffer for reading the off-chip mass storage through the data interface Loading and storage unit of the original data of biological information.
  • the second instruction pipeline includes a decoder and a general purpose register, the decoder decodes the instruction data stored in the instruction prefetch buffer, and the general purpose register obtains the data after the execution of the third instruction pipeline ,
  • the fourth instruction pipeline stores the raw data of the monitored biological information and the calculation result of the FPE array convolution calculation unit, and performs logical operation processing on it.
  • the data decoded by the decoder and the data processed by the general purpose register are executed Then they are respectively input to the third instruction pipeline and the FPE array convolution calculation unit, and feed back to the first instruction pipeline at the same time.
  • the third instruction pipeline includes a current state register, an arithmetic logic unit, a fixed-point multiplication unit, and a dot product calculation unit; the current state register, arithmetic logic unit, a fixed-point multiplication unit, and a dot product calculation unit respectively respond to the second instruction
  • the raw data of biological information after pipeline processing is processed by matrix operation, logic operation, fixed-point number multiplication and accumulation operation, and point multiplication operation.
  • the current status register is used to receive the processed raw data of the monitored biological information, perform matrix operation processing, and transmit the processed data to the off-chip large-capacity memory through the data interface;
  • the arithmetic logic unit is used to receive the processed raw data of the monitored biological information, perform logical operation processing, and input the processed data into the general purpose register of the second instruction pipeline;
  • the fixed-point multiplication unit is configured to receive the processed raw data of the monitored biological information, perform fixed-point multiplication and accumulation processing, and input the processed data into the general purpose register of the second instruction pipeline;
  • the dot product calculation unit is used to receive the processed raw data of the monitored biological information, perform vector dot product operation processing, and input the processed data into the general purpose register of the second instruction pipeline.
  • the fourth instruction pipeline includes a load storage unit, the load storage unit is used to read the monitored original data of biological information stored in the off-chip large-capacity memory through a data interface, and input it into the general purpose register of the second instruction pipeline .
  • the FPE array convolution calculation unit includes a plurality of multiplication and accumulation processing unit groups composed of a plurality of multiplication and accumulation processing units connected in series, and each multiplication and accumulation processing unit group is connected with a buffer, and the buffer is connected to the SRAM through a bus.
  • Memory connection the SRAM memory is connected to the DMA controller;
  • the multiply-accumulate processing unit group processes a plurality of input biological information raw data respectively, and the processed biological information data is input to the DMA controller through the buffer and the SRAM memory, and the DMA controls
  • the device reconstructs the imaging mode, imaging area of interest and key structure points, and stores the imaging mode, imaging area of interest and key structure points in the SRAM memory.
  • the SOC chip of the composite pipeline structure adopts two different pipeline structures at the same time: a short calculation time pipeline and a long calculation time pipeline.
  • the short-time pipeline is the conventional single-cycle pipeline structure.
  • the data handling and data calculation time of each layer is averaged to increase the efficiency of the pipeline and reduce the probability of cavitation, as shown in Figure 2.
  • the FPE array convolution calculation unit includes 128 multiplication and accumulation processing units PE, 32 buffers, SRAM memory and DMA controller, every 4 multiplication and accumulation processing units processing input 4 lines
  • the original data of biological information, and every 4 multiplication and accumulation processing units are connected to a 192-byte buffer
  • the 32 buffers are connected to a 16M byte global SRAM memory through a 64-bit bus
  • the SRAM is connected to a DMA controller
  • the DMA controller is used to obtain all the biological information data processed by the multiply-accumulate processing unit, reconstruct the imaging mode, locate the imaging region of interest and key structural points, and
  • the imaging mode, imaging region of interest and key structure point data are stored in the SRAM memory.
  • the receiving channel beam combiner unit is controlled to collect signals for the desired imaging point.
  • the array convolution calculation unit proposed in this embodiment greatly reduces the data handling and interaction between the large-capacity DDR outside the SOC chip and the inside of the SOC chip, reduces the area of the chip, and greatly reduces the function of the SOC chip for cardiac ultrasound data processing. Consumption.
  • the throughput rate of the SOC chip proposed in this embodiment exceeds 50GOP/s (at a clock frequency of 200MHz), that is, more than 50 ⁇ 109 16-bit multiply-accumulate operations can be performed per second; and the fixed-point number accuracy can be changed to 8 bits for processing as required
  • the speed is increased by 4 times, or even the clock frequency is changed to 400MHz, which increases the processing speed by 8 times compared with the initial state, and the calculation accuracy error does not exceed 2%.
  • it can it be suitable for a variety of artificial intelligence network structures, but also achieves approximate floating-point calculation accuracy and high throughput, and more importantly, maintains low power consumption.
  • the working principle of a wearable biological information monitoring device based on an artificial intelligence chip includes the following steps:
  • Step 301 The signal transmitter of the ultrasonic transmitter and receiver unit receives the excitation signal emitted by the transmitter channel beam combiner in the digital front-end module, generates a voltage pulse for exciting the ultrasonic transducer, and transmits it to the transmitter through the transmitter/receiver switch of the ultrasonic transmitter and receiver unit. High voltage pulse chip.
  • Step 302 The high-voltage pulse chip receives the echo electric signal collected by the ultrasonic array transducer, and transmits it to the transmitting/receiving switch of the ultrasonic transmitting and receiving unit.
  • Step 303 The transmitting/receiving switch of the ultrasonic transmitting and receiving unit receives the echo electric signal transmitted by the high-voltage pulse chip, and performs impedance matching on it, and transmits the echo signal after the impedance matching to the analog front-end receiving unit.
  • step 304 the echo electric signal after impedance matching is amplified by the preamplifier of the front-end receiving unit and converted by the analog-to-digital converter, and then enters the receiving channel beam combiner in the digital front-end module for beam combining.
  • Step 305 The SOC chip performs different artificial intelligence network operations according to the imaging purpose and imaging mode to obtain the imaging region of interest and key structural points, and output control instructions to the digital front-end module.
  • Step 306 the digital front-end module receives the control instruction output by the SOC chip, collects the echo signal of the desired imaging point, and uses the collected signal to perform dynamic beam synthesis; the signal after beam synthesis is filtered and decomposed by a DC filter
  • the quadrature demodulation of the modulator and the processing by the processor Echo and the blood flow velocity estimation are processed by each unit to realize ultrasound image reconstruction and real-time imaging.
  • Conformity refers to: a perfect fit to the target surface.
  • This embodiment provides an ultrasonic beam synthesis system based on a conformal wearable multi-element imaging transducer
  • the ultrasonic beam synthesis system based on the conformal wearable multi-element imaging transducer includes:
  • a conformal wearable multi-element imaging transducer when in use, the conformal wearable multi-element imaging transducer is set at the position of the chest wall on the body surface corresponding to the heart of the patient; as shown in Figure 7;
  • the conformal wearable multi-element imaging transducer receives signals from the ultrasound transmitter and receiver unit of the wearable biological information monitoring equipment based on artificial intelligence chips. Each element of the conformal wearable multi-element imaging transducer will The received signal is transmitted to the patient's heart position after time delay compensation compensation;
  • Each element of the conformal wearable multi-element imaging transducer receives a feedback signal, and each element of the feedback signal is time-delayed and sent to the ultrasonic emission of a wearable biological information monitoring device based on an artificial intelligence chip
  • the receiving unit, a wearable biological information monitoring device based on an artificial intelligence chip performs beam synthesis on the signal of each element to obtain a three-dimensional ultrasound image of the patient's heart. Obtain three-dimensional ultrasound images of the heart, monitor the shape of the heart in different postures and motion states in real time, and provide more abundant data for doctors' diagnosis and treatment.
  • each element of the conformal wearable multi-element imaging transducer transmits the received signal to the patient's heart position after time delay compensation; the specific steps include:
  • the ultrasonic transmitting and receiving unit of the wearable biological information monitoring device based on artificial intelligence chip transmits ultrasonic signals to all array elements at the same time, and each array element receiver receives the ultrasonic signals through any two adjacent array elements. For the time difference to the ultrasonic signal, use one of the array elements as the reference array element to calculate the relative position of the other array element and the reference array element;
  • S402 Using the relative position of another array element and the reference array element, calculate the time delay of the two adjacent array elements transmitting ultrasound to the heart position; compensating the time delay until the two adjacent array elements deviate from themselves On the element with a larger initial position; for the element after time delay compensation, add the time delay on the basis of the set transmission time and then transmit ultrasound to the heart;
  • S403 By setting a time delay for each array element, control the array element to generate a focused sound beam.
  • S401 is allowed to be replaced with:
  • a position sensor is installed on each element, and the position of the element is fed back in real time through the position sensor; the relative displacement of any two adjacent elements is obtained.
  • S401 can also be replaced with:
  • a fixed reflector (as shown in Figure 11) is installed on the back of the conformal transducer (the side away from the human body), and an ultrasonic transducer is also installed on the back of each array element.
  • the time interval of the received echo is calculated to calculate the distance of the array element relative to the radiation plate in the Z direction, and then the relative position of the array element in the Z direction is calculated; the relative displacement of any two adjacent array elements is obtained.
  • ultrasound imaging uses the array elements of a conformal wearable multi-element imaging transducer to transmit ultrasonic waves and then receive the signals back. After the received signal is filtered, amplified, etc., an ultrasound image is formed through analysis and recombination.
  • the ultrasonic waves emitted by all the array elements are usually made to form a focal point.
  • different transducer array elements are added. The time delay allows the ultrasonic waves emitted by the conformal wearable multi-element imaging transducer to reach the focal point at the same time.
  • time delay means that the time of exciting each array element to transmit ultrasonic waves is different, so that the ultrasonic waves emitted by all array elements reach the focal point at the same time.
  • the ultrasonic calibration method when the transducer is fixed on the chest wall of the human body, its deformation is less during the human body movement, and only the change L0 in the Z direction occurs.
  • the z direction refers to the direction perpendicular to the surface of the transducer ;
  • L1 represents the distance between the array element 1 and the ultrasonic transmitting and receiving unit
  • L2 represents the distance between the array element 2 and the ultrasonic transmitting and receiving unit
  • L0 represents the distance between the array element 1 and the ultrasonic transmitting and receiving unit
  • ⁇ t represents the time delay of two adjacent array elements transmitting ultrasound to the heart position.
  • S401 can also be replaced with:
  • the laser calibration method is used to obtain the relative displacement of any two adjacent array elements.
  • the principle of the laser calibration method is the same as that of the ultrasonic calibration method, that is, the ultrasonic transmitting and receiving unit is replaced with a light source transmitter.
  • the conformal wearable multi-element imaging transducer includes:
  • a conformal base, a number of array elements are evenly distributed on the conformal base, and a corresponding array element transmitter and an array element receiver are arranged inside each array element;
  • the array element receiver is used to convert the received electrical signal transmitted by the ultrasonic transmitter and receiver unit into an ultrasonic signal, and transmit the ultrasonic signal to the heart position of the patient after time delay compensation;
  • the array element transmitter is used for converting the feedback ultrasonic signal into an electrical signal after a time delay, and transmitting the electrical signal to the ultrasonic transmitting and receiving unit.
  • the conformal wearable multi-element imaging transducer further includes: a heat dissipating component, the heat dissipating component is arranged in the gap of the element, or on the side of the element away from the human body .
  • the conformal wearable multi-element imaging transducer when the conformal wearable multi-element imaging transducer is in use, it is set at the position of the chest wall on the body surface corresponding to the heart of the patient; it means that the conformal base is attached to the position of the chest wall on the body surface, An ultrasonic coupling medium is arranged between the conformal substrate and the body surface.
  • the material of the conformable substrate is a biocompatible flexible material, such as PDMS (polydimethylsiloxane), soft silica gel, etc.
  • the wearable biological information monitoring device based on an artificial intelligence chip transmits an ultrasonic signal to each element of a conformal wearable multi-element imaging transducer; and receives; The electrical signal fed back by each element of the conformal wearable multi-element imaging transducer; beam synthesis is performed on the received electrical feedback signal to obtain a three-dimensional ultrasound image of the heart.
  • the present disclosure includes a conformal wearable multi-element imaging transducer.
  • the conformal wearable multi-element imaging transducer When in use, the conformal wearable multi-element imaging transducer is worn at the corresponding position of the human heart and fits the body surface of the human body. When a person moves, it can be deformed accordingly.
  • the conformal wearable multi-element imaging transducer includes multiple independent working elements.
  • Beam synthesis includes multiple transmitters, which can be one or more array element receivers with only one in the figure and on the surface of the transducer (preferably, each array element corresponds to an array element receiver), ultrasonic transmitting and receiving unit
  • the laser beam or ultrasound can be emitted and then received by the element receiver.
  • the relative displacement between the transducer elements can be calculated based on the received time interval, and these relative displacements can be used to compensate for the time delay of the transducer element.
  • Wearable biological information monitoring equipment based on artificial intelligence chips can control the array element to generate a focused sound beam by setting a time delay for each element.
  • Figure 8 is a schematic diagram of the structure of a conformal wearable multi-element imaging transducer.
  • the array elements of the conformal wearable multi-element imaging transducer can be in various forms, such as linear array, area array, ring array, and arrangement;
  • the array elements are filled with conformal materials, which can enable the transducer to achieve deformation functions such as bending, compression and stretching.
  • the relative position of the transducer elements is obtained by the previous method, and the distance from each element to the focus can be calculated, and the distance difference can be used to calculate the time delay.
  • a pulse signal is applied to the elements of the conformal wearable multi-element imaging transducer, a digital delay added to each element can make the sound waves emitted by the conformal wearable multi-element imaging transducer be at the focus Convergence, in the same way, adding a delay line at the receiving end can return the echo signal received by the array element to the signal processing unit at the same time.
  • the array element is a part of the transducer, and the conformal wearable multi-element imaging transducer also includes filling materials, cables, etc. between the elements.
  • the wearable multi-element imaging transducer can be a piezoelectric ceramic transducer, a piezoelectric single crystal transducer, or a CMUT (Capacitive Micromachined Ultrasonic Transducers), PMUT (Piezoelectric Micromachined Ultrasound Transducer) or other types of transducers.
  • the ultrasonic beam synthesis method based on the conformal wearable multi-element imaging transducer includes:
  • a conformal wearable multi-element imaging transducer worn on the chest wall of the patient's heart corresponding to the body surface receives the signal from the ultrasound transmitter and receiver unit of the wearable biological information monitoring device based on the artificial intelligence chip.
  • Each element of the wearable multi-element imaging transducer will transmit the received signal to the patient's heart position after time delay compensation;
  • Each element of the conformal wearable multi-element imaging transducer receives a feedback signal, and each element will send the feedback signal to the ultrasound of the wearable biological information monitoring equipment based on artificial intelligence chip after time delay
  • the transmitting and receiving unit, a wearable biological information monitoring device based on an artificial intelligence chip performs beam synthesis on the signal of each element to obtain a three-dimensional ultrasound image of the patient's heart.
  • each element of the conformal wearable multi-element imaging transducer transmits the received signal to the patient's heart position after time delay compensation; the specific steps include:
  • the ultrasonic transmitting and receiving unit transmits ultrasonic signals to all array elements at the same time, and each array element receiver receives the ultrasonic signals.
  • the time difference between receiving the ultrasonic signal through any two adjacent array elements is taken as one of the array elements.
  • Reference element calculate the relative position of another element and the reference element;
  • S5012 Using the relative position of another array element and the reference array element, calculate the time delay of the two adjacent array elements transmitting ultrasound to the heart position; compensate the time delay to the deviation from itself in the two adjacent array elements On the element with a larger initial position; for the element after time delay compensation, add the time delay on the basis of the set transmission time and then transmit ultrasound to the heart;
  • S5013 By setting a time delay for each element, control the element to generate a focused sound beam.
  • the relative displacement of the two adjacent array elements refers to the distance difference between the two projection points, and the two projection points are the projections of the two adjacent array elements on a line perpendicular to the outer tangent plane of the transducer. owned.
  • Embodiment 3 This embodiment also provides a remote cardiac ultrasound three-dimensional imaging system based on deep learning
  • the remote cardiac ultrasound 3D imaging system based on deep learning includes:
  • the user client is used to control the ultrasound transmitting and receiving unit to transmit a strobe signal command to the conformal wearable multi-element imaging transducer worn on the chest wall of the human body during use (as shown in Figure 7); at the same time, control
  • the ultrasound transmitting and receiving unit of the wearable biological information monitoring equipment based on artificial intelligence chip receives the ultrasound signal fed back by the array element corresponding to the strobe signal instruction; uploads the fed back ultrasound signal to the cloud server;
  • the cloud server is used to process the ultrasound signals uploaded by the user client, and use the pre-trained individual three-dimensional heart model to process the subject’s cardiac ultrasound two-dimensional images to obtain the subject’s real-time heart contour key points, based on Real-time key points of the heart contour to obtain real-time 3D ultrasound imaging of the heart of the subject;
  • the doctor client is used to receive the key points of the heart contour of the subject selected by the doctor, and send the key points of the heart contour of the subject to the user client through the cloud server, which is used to instruct the user client to send a strobe signal instruction.
  • the conformal wearable multi-element imaging transducer includes:
  • a conformal base, a number of array elements are evenly distributed on the conformal base, and a corresponding array element transmitter and an array element receiver are arranged inside each array element;
  • the array element receiver is used to convert the received electrical signal transmitted by the ultrasonic transmitter and receiver unit into an ultrasonic signal, and transmit the ultrasonic signal to the heart position of the patient;
  • the array element transmitter is used to convert the feedback ultrasonic signal into an electric signal, and transmit the electric signal to the ultrasonic transmitting and receiving unit.
  • the conformal wearable multi-element imaging transducer is arranged at the position of the chest wall on the body surface directly opposite to the heart of the patient; it means that the conformal substrate is applied to the position of the chest wall on the body surface, and the An ultrasonic coupling medium is arranged between the shaped base and the body surface.
  • the material of the conformable substrate is a biocompatible flexible material, such as PDMS (polydimethylsiloxane), etc.
  • the user client includes: an artificial intelligence chip-based wearable biological information monitoring device and a conformal wearable multi-element imaging transducer connected to each other.
  • the user client includes a display
  • the display is used to display key points of the heart contour of the subject selected by the doctor client and sent by the cloud server.
  • the remote cardiac ultrasound 3D imaging method based on deep learning includes:
  • S701 Pre-tuning the imaging process:
  • the user client acquires the real-time two-dimensional cardiac ultrasound image of the subject after the subject wears the conformal wearable multi-element imaging transducer during use, and the subject will be acquired by the user client
  • the real-time cardiac ultrasound two-dimensional image of the heart is sent to the cloud server, and the cloud server sends the subject’s real-time cardiac ultrasound two-dimensional image to the doctor’s client; in the process of acquiring the real-time cardiac
  • the energy sensor receives the ultrasound signals emitted from the wearable biological information monitoring equipment based on artificial intelligence chips, and feeds the feedback signals to the wearable biological information monitoring equipment based on artificial intelligence chips, forming a real-time two-dimensional ultrasound image of the subject’s heart ;
  • the doctor client selects the key points of the subject’s heart contour from the subject’s real-time cardiac ultrasound two-dimensional image; the doctor client sends the selected key points of the subject’s heart contour to the cloud server;
  • the cloud server uses the subject’s real-time cardiac ultrasound two-dimensional image as the input value of the adaptive cardiac neural network model; the cloud server uses the manually selected key points of the subject’s cardiac contour as the output value of the adaptive cardiac neural network model.
  • S702 Real-time imaging process: the user client receives key points of the subject's heart contour and the subject's individual three-dimensional heart model;
  • the user client sends a gating instruction to the ultrasound transmitting and receiving unit according to the key points of the subject's heart contour, that is, the gating instruction controls the ultrasound transmitting and receiving unit to only transmit ultrasound signals to the elements corresponding to the contour key points, and for non-contour key points corresponding
  • the array element does not emit ultrasonic signals;
  • the user client acquires the subject’s new real-time cardiac ultrasound two-dimensional image collected by the array element corresponding to the strobe command, and the user client inputs the subject’s new real-time cardiac ultrasound two-dimensional image into the subject’s individual three-dimensional In the heart model, the user client outputs the coordinates of key points of the subject’s heart contour in real time;
  • the step of acquiring the adaptive cardiac neural network model includes:
  • the preprocessing of the two-dimensional ultrasound image of a normal person includes:
  • Interpolate the tomographic image on the fused image add a virtual slice layer between different layers of the two-dimensional image to obtain a three-dimensional image of the outer contour of the heart.
  • the images obtained under the different two-dimensional scanning slices include: images of the long axis of the heart at multiple angles and a series of short axis images from the bottom of the heart to the apex of the heart.
  • cardiac anatomical information refers to anatomical information such as papillary muscles, valves, endocardium, and apex.
  • the preprocessing of the two-dimensional ultrasound image of a normal person further includes:
  • the chamber area of the heart section includes: left ventricle, left atrium, right ventricle and right atrium;
  • the short-axis view refers to the standard view of the cardiac ultrasound scan, which is a cross-sectional image obtained by the ultrasound probe by making the beam section perpendicular to the long axis of the left ventricle next to the sternum. From the base of the left ventricle to the apex, at least three images can be obtained. A standard short axis section.
  • the processed image is input into the pre-trained intracardiac contour segmentation neural network, and the segmented intracardiac contour image is output.
  • the training set in the training phase is the intracardiac contour image marked by the doctor; during the training process, the input value of the neural network is the intracardiac contour image, and the output value of the neural network The coordinate position marked by the doctor, and the trained intracardiac contour segmentation neural network is obtained after the training.
  • the real-time cardiac ultrasound three-dimensional imaging of the subject is obtained based on the real-time heart contour key point coordinate position; the specific steps include:
  • the training steps include:
  • the training set is a three-dimensional cardiac ultrasound image with known key point coordinate positions;
  • the known key point coordinates of the training set are used as the input value of the neural network, and the three-dimensional cardiac ultrasound image of the training set is used as the output value of the neural network; the neural network is trained to obtain a trained three-dimensional imaging neural network model.
  • the method further includes:
  • the subject’s real-time heart contour key point coordinate position After obtaining the subject’s real-time heart contour key point coordinate position, compare the subject’s real-time heart contour key point coordinate position with the set coordinate range. If it is within the set coordinate range, it means the current The obtained coordinates of the key points of the subject’s real-time heart contour are correct. Based on the real-time key points of the heart contour, the subject’s real-time cardiac ultrasound three-dimensional imaging is obtained;
  • the individual three-dimensional heart model; specific steps include:
  • B i represents the feature vector of the user's heart motion dimension
  • i represents the number of selected feature vectors, which is the same as the number of key structural points of the heart
  • a i and w i constitute the parameterized representation of the heart, through adaptive The heart neural network model is calculated.
  • the real-time ultrasound images of each heart chamber are summarized and synthesized, and the synthesized real-time three-dimensional dynamic image of the whole heart is output.
  • the determination and division of the various chambers of the heart facilitate the adjustment of the adaptive cardiac neural network model according to the real-time ultrasound image, so as to form a real-time three-dimensional dynamic image of the user's heart.
  • the pre-adjusted imaging process includes: after the conformal wearable multi-element imaging transducer of the user client obtains the initial ultrasound signal, it passes the analog-to-digital conversion and sends it to the cloud via the wireless transmission module.
  • cloud server performs echo signal processing on the received data to form the initial two-dimensional image, and receives the key structural points (valve structure, apex) of the heart cavity, ventricular wall and valve area uploaded by the doctor’s client on the initial two-dimensional image Position, ventricular wall) marking results, and input the marked key structural point images of the heart into the adaptive cardiac neural network model to establish an individual three-dimensional heart model;
  • the gating instruction is sent to the user client, and the ultrasonic signal is transmitted and collected on the key structure to realize beam synthesis.
  • Optional implementation methods include but are not limited to the following methods (take left ventricular imaging as an example):
  • the pre-adjusted imaging process can also adopt another embodiment: the user client's conformal wearable multi-element imaging transducer obtains the initial ultrasound signal, and after analog-to-digital conversion, time gain compensation adjustment, beam synthesis, After filtering and demodulation, it is sent to the cloud server through the wireless transmission module; the cloud server performs echo signal processing on the received data to form an initial two-dimensional image, and receives the heart cavity, ventricular wall and ventricular wall uploaded by the doctor client on the initial two-dimensional image. The result of identifying key structural points of the heart in the valve area, and inputting the image of the identified key structural points of the heart into the individual 3D heart model of the subject to establish an individualized accurate heart model;
  • the network coordinate points of the individual three-dimensional heart model are transmitted to the ultrasound controller of the user client to realize the transmission and collection of ultrasound signals at key structural points.
  • the real-time imaging process includes: the conformal wearable multi-element imaging transducer of the user client performs gating instructions to issue and collect ultrasound signals according to the key coordinate points of the subject's individual three-dimensional heart model, through analog-to-digital conversion After that, it is sent to the cloud server via the wireless transceiver module, and the cloud server performs echo signal processing on the received data to form a real-time two-dimensional image and perform automatic gain adjustment.
  • the sphere whose initial position of the key point of the heart contour is the center of the circle (preferably with a radius of 2 cm, determined according to the range of movement of the diaphragm up and down when breathing calmly and the range of movement of the mediastinum when the body is laterally positioned), follow the method described in the preset imaging process
  • the echo signal is acquired and transmitted to the individual three-dimensional heart model, and the echo signal in the spatial range is compared with the initial signal of the key structure points of the heart in real time, and the cross-correlation algorithm is used, but not limited to, to determine the real-time coordinates of the key structure points.
  • the actual change range and change trajectory of the key points of the heart contour are input into the subject's individual three-dimensional heart model to realize the incremental learning and adaptive learning of the individual three-dimensional heart model deep learning network .
  • the initial distance between the key points of the heart contour and the ultrasound array element is obtained on two orthogonal cut planes or multiple intersecting cut planes and converted into digital delayed.
  • the key points include but are not limited to the key points of the anatomical structure of the heart model such as the ventricular septum, apex, heart cavity, valve leaflet, and valve annulus.
  • the automatic positioning of key points requires deep learning of different cardiac anatomical features, and the use of deep convolutional network models to extract key points of the heart on the individualized heart model. All deep learning methods are applicable to this implementation method, and the preferred implementation is to use data augmentation methods (horizontal, vertical flip, random rotation, random scaling) to increase the data set, and use the attention-based CNN reinforcement learning model to automatically learn the heart Key point location. Return the learned key point coordinates to the mobile terminal or realize distance gating, and only select key point echo signal collection and transmission.
  • the acquisition of the key point signal in the cardiac ultrasound image needs to locate the key point position, which is related to the anatomical shape of the heart, and the three-dimensional model of the heart reconstructed by the ultrasound data can be used for positioning calculation.
  • the initial ultrasound echo signal obtained by the user client is processed to obtain a B-mode initial image containing the whole heart.
  • a series of image preprocessing techniques such as denoising and enhancement are performed on the original image, and the region of interest is selected to complete the heart cavity
  • image segmentation refer to the general three-dimensional heart model of normal people and the individual three-dimensional heart model of the subject.
  • the user client is a mobile terminal, which can be in various forms such as portable and wearable, and the portable can be in various forms such as a tablet and palmtop.
  • the transducer can be a portable, wearable transducer, and the type of transducer can be linear array, convex array, area array, phased array and other forms.
  • the high-voltage pulse chip is optional.
  • the beam combiner and time gain compensation are also optional modules, and their functions are completed by the cloud server.
  • the cloud server can be an edge type, a decentralized type, etc., and can be of various types such as cloud computing, fog computing, and ocean computing according to the distribution of data calculations between the user client and the cloud server.

Abstract

本公开公开了基于人工智能芯片的适形穿戴式生物信息监测设备及系统，该设备包括模拟前端电路模块、数字前端模块以及SOC芯片；模拟前端电路模块产生激励超声换能器的电压脉冲，接收超声面阵换能器采集的回波电信号，并对其进行阻抗匹配，阻抗匹配后的回波电信号经过放大和模数转换后，输入数字前端模块；SOC芯片根据成像目的和成像模式对生物信息原始数据进行不同运算，输出控制指令至数字前端模块；数字前端模块接收到SOC芯片输出的控制指令后，采集所需成像点的回波信号，并对其进行动态的波束合成；波束合成后的信号经过滤波、正交解调以及批处理和流速估计后，实现超声图像重建与实时成像。

Description

基于人工智能芯片的适形穿戴式生物信息监测设备及系统 技术领域
本公开涉及生物信息监测领域，具体涉及基于人工智能芯片的适形穿戴式生物信息监测设备及系统。
背景技术
人体的各种生理参数，从常用的血压、体温、心率，到更为复杂的脏器活动状态，都是在不同生理状态或不同病理状态下发生变化的。各种参数的生理变化是人体为了适应不同的状态各器官组织协同工作的反应，而各参数的病理状态下变化，反应了人体疾病进程的状态。临床上对于这些参数的获得，通常是在各种状态已然的情况下进行测量，比如运动后血压、心率的变化，要在运动后即刻进行测量，存在特定的检查体位和检查时间延迟，不利于准确反应特定状态下生理参数。而罹患某种疾病后，这些生理参数的检测对于诊断和疾病过程的判断具有重要价值，如能早期发现这些参数的变化更有意义，然而目前大都是在已然出现疾病表现后再去检测，失去早期检测的时机，如心力衰竭早期，心脏功能在静息状态下处于代偿状态而表现为正常，但在活动后方表现出异常，而目前最为简便最为准确判定心衰的技术——心脏超声检查，需要特定的静息、卧位检查，往往不能及时发现这种活动后才可以表现出的心功能异常。
为解决这种问题，目前已有穿戴式生物信息监测设备，用于获取心率、血压、心电信号。这些设备或者需要多个信号检测线造成穿戴的繁琐和信号检测易受到干扰，或者检测方式过于简单而准确性欠佳。目前穿戴式生理参数测量设备，包括佩戴于手腕、胸前等部位的各类形式设备，可以检测心率、血氧饱和度、血压、心电信号等。一般所采用的是光电容积脉搏波描记法(PhotoPlethysmoGraphy)，或者采用心电信号、脉搏波检测、柯氏音检测，或者上述几种方法结合。
发明人在研发过程中发现，上述方案普遍存在的问题就是准确率低，现有解决方案是增加设备复杂性来获取更全面和更为准确的生物信息，意味着设备必须进行更多的信息处理，这势必造成设备体积增大，不仅增加设备功耗不利于穿戴，而且佩戴体验不佳。此外，组织血流量、关节腔内压力等流体力学参数是评估器官、组织、尤其微小关节功能的重要参数，这些指标通过目前已有的穿戴式检测技术难以实现。更为重要的是所采集生物信息的可视化，如血管的运动、血流速度显示，这不仅需要对采集的信息进行计算，还需要信号处理后进行图像重建，这需要更多的集成线路、板卡等硬件设备，且由于计算量大，功耗大，即使有些系统将采集到的信息上传至云端或网络，利用云计算或网络平台进行信号处理，仍面临大量数据实时传输效率问题。
另外，发明人在研发过程中还发现，现今用于便携心脏超声成像的方法，主要是使用一个1维相控阵超声换能器，辅以阵列超声成像系统。这种方法只能获得一些二维的超声图像，不能获得心脏的三维形貌，也不能实时监测心脏的运动状态。由于人体肋骨，肺，以及其他器官的遮挡，现有的1维相控阵超声成像技术需要医生具有高超的技术和丰富的经验，找寻合适的角度进行成像，并且需要将换能器对患者胸壁进行挤压也给患者带来很多痛苦，另外现有的成像方法只能在医院进行检查受限 于场地，也不能检测患者不同运动状态下的心脏形貌。
现有的超声诊断设备为了得到具有诊断信息的超声回波，换能器所发射和接收超声的方向和位置必须进行控制，每次发射和接收声波时，由一组阵元进行波束合成，产生一束扫描声束，并接收信号，然后由下一组产生下一次发射声束，并接收信号。把每次接收到的回波信号经放大和后处理后，就可以合成一幅超声图像。提高扫查线密度，可以提高图像的空间分辨力，但也产生了很大的数据量。
为了在有限的硬件条件下进行成像相关处理，目前的移动式心脏超声设备采用减少成像阵元、降低分辨率等减少数据量的方法，成像模态也多局限于一维、二维和彩色多普勒，无法进行三维重建和三维信息计算；另一方面，为达到实时高帧频成像，对硬件系统提出了更高的要求，现有移动式成像设备的帧频率和时间分辨率都受到了限制。对硬件架构的高要求，与移动式(包含但不限于手推车式、手提式、掌上型等)或穿戴式超声设备小型化、集成化、轻量化的要求相悖，而且某些掌上型超声设备和穿戴式设备，为进一步缩小设备的重量、功耗和体积，将部分波束合成和成像处理通过云计算等交互式方式进行远程处理和成像，这对于设备的数据传输速率也有很高的要求，且移动式超声有较多边远地区和网络覆盖差的应用场合。显然，传统的超声设备架构和成像方法已经不能，用于硬件、数据传输资源和算力有限的移动式或可穿戴超声成像设备。
发明内容
为了克服上述现有技术的不足，本公开提供了基于人工智能芯片的适形穿戴式生物信息监测设备及系统。
第一方面，本公开提供的基于人工智能芯片的适形穿戴式生物信息监测设备的技术方案是：
一种基于人工智能芯片的穿戴式生物信息监测设备，包括模拟前端电路模块、数字前端模块以及人工智能芯片；
所述模拟前端电路模块，用于产生激励超声换能器的电压脉冲，接收超声面阵换能器采集的回波电信号，并对其进行阻抗匹配，阻抗匹配后的回波电信号经过放大和模数转换后，输入数字前端模块；
所述人工智能芯片根据成像目的和成像模式对生物信息原始数据进行不同运算，得到成像感兴趣区域和感兴趣点，输出控制指令至数字前端模块；
数字前端模块，用于接收到人工智能芯片输出的控制指令后，采集所需成像点的回波信号，并对其进行动态的波束合成；波束合成后的信号经过滤波、正交解调以及批处理和流速估计后，实现超声图像重建与实时成像。
进一步地，一种人工智能芯片，包括四级复合指令流水线和FPE阵列卷积计算单元；
所述四级复合指令流水线包括用于获取矩阵运算指令的第一指令流水线、用于处理所监测到的生物信息原始数据的第二指令流水线、用于对处理后的生物信息原始数据进行矩阵、算数、定点数乘以及点乘运算的第三指令流水线以及加载和存储所监测到的生物信息原始数据的第四指令流水线；
所述FPE阵列卷积计算单元，用于对第二指令流水线和第三指令流水线处理后的生物信息原始数据进行累加处理，重建成像模式和成像感兴趣区域。
进一步地，超声发射接收单元产生激励超声换能器的电压脉冲，并发射至高压脉冲芯片；
高压脉冲芯片接收超声面阵换能器采集的回波电信号，并传送给超声发射接收单元；
超声发射接收单元接收高压脉冲芯片传送的回波电信号，并对其进行阻抗匹配；
阻抗匹配后的回波电信号经过前端接收单元放大和模数转换后，输入数字前端模块；
人工智能芯片根据成像目的和成像模式进行不同运算，得到成像感兴趣区域和感兴趣点，输出控制指令至数字前端模块；
数字前端模块接收到人工智能芯片输出的控制指令后，采集所需成像点的回波信号，并对其进行动态的波束合成；
波束合成后的信号经过直流滤波器滤波、解调器正交解调和处理器处理后，实现超声图像重建与实时成像。
第二方面，本公开提供了基于适形穿戴式多阵元成像换能器的超声波束合成系统；
基于适形穿戴式多阵元成像换能器的超声波束合成系统，包括：
适形穿戴式多阵元成像换能器，所述适形穿戴式多阵元成像换能器在使用时，设置在患者心脏对应的体表胸壁位置；
适形穿戴式多阵元成像换能器接收基于人工智能芯片的穿戴式生物信息监测设备的超声发射接收单元发出的信号，适形穿戴式多阵元成像换能器的每个阵元均将接收到的信号进行位置补偿后发射到患者心脏位置；
适形穿戴式多阵元成像换能器的每个阵元均接收反馈信号，每个阵元均将反馈信号进行时间延迟换算后发送给基于人工智能芯片的穿戴式生物信息监测设备的超声发射接收单元，基于人工智能芯片的穿戴式生物信息监测设备对每个阵元的信号进行波束合成，得到患者心脏的三维超声图像。
进一步地，使用时，佩戴在患者心脏对应的体表胸壁位置的适形穿戴式多阵元成像换能器，接收基于人工智能芯片的穿戴式生物信息监测设备的超声发射接收单元发出的信号，适形穿戴式多阵元成像换能器的每个阵元均将接收到的信号进行位置补偿后发射到患者心脏位置；
适形穿戴式多阵元成像换能器的每个阵元均接收反馈信号，每个阵元均将反馈信号进行时间延迟换算后发送给基于人工智能芯片的穿戴式生物信息监测设备的超声发射接收单元，基于人工智能芯片的穿戴式生物信息监测设备对每个阵元的信号进行波束合成，得到患者心脏的三维超声图像。
第三方面，本公开提供了基于深度学习的远程心脏超声三维成像系统；
基于深度学习的远程心脏超声三维成像系统，包括：
用户客户端，用于控制基于人工智能芯片的适形穿戴式生物信息监测设备的超声发射接收单元，向使用时穿戴在人体体表胸壁位置的适形穿戴式多阵元成像换能器发射选通信号指令；同时，控制基于人工智能芯片的穿戴式生物信息监测设备的超声发射接收单元接收选通信号指令对应的阵元反馈回来的超声信号；将反馈回来的超声信号上传给云服务器；
云服务器，用于对用户客户端上传的超声信号进行处理，利用预训练的个体三维心脏模型，对受试者的心脏超声二维图像进行处理，得到受试者实时的心脏轮廓关键点，基于实时的心脏轮廓关键点， 得到受试者实时的心脏超声三维成像；
医生客户端，用于接收医生选取的受试者心脏轮廓关键点，将受试者心脏轮廓关键点通过云服务器发送给用户客户端，用于指导用户客户端发送选通信号指令。
进一步地，预调成像过程：用户客户端获取使用时受试者佩戴适形穿戴式多阵元成像换能器后，受试者的实时心脏超声二维图像；用户客户端将获取的受试者的实时心脏超声二维图像发送给云服务器，云服务器将受试者的实时心脏超声二维图像发送给医生客户端；实时心脏超声二维图像获取过程中，适形穿戴式多阵元成像换能器接收来自基于人工智能芯片的穿戴式生物信息监测设备发射的超声信号，并将反馈的信号反馈给基于人工智能芯片的穿戴式生物信息监测设备，形成受试者的实时心脏超声二维图像；
医生客户端从受试者的实时心脏超声二维图像中选取受试者心脏轮廓关键点；医生客户端将选取的受试者心脏轮廓关键点发送给云服务器；
云服务器将受试者的实时心脏超声二维图像作为自适应心脏神经网络模型的输入值；云服务器将人工选取的受试者心脏轮廓关键点作为自适应心脏神经网络模型的输出值，对自适应心脏神经网络模型进行训练，得到受试者的个体三维心脏模型；云服务器将受试者心脏轮廓关键点和受试者的个体三维心脏模型均发送给用户客户端；
实时成像过程：用户客户端接收受试者心脏轮廓关键点和受试者的个体三维心脏模型；
用户客户端根据受试者心脏轮廓关键点，向基于人工智能芯片的穿戴式生物信息监测设备的超声发射接收单元发出选通指令，即选通指令控制超声发射接收单元只向轮廓关键点对应的阵元发射超声信号，对于非轮廓关键点对应的阵元不发射超声信号；
用户客户端获取由选通指令对应的阵元所采集的受试者新的实时心脏超声二维图像，用户客户端将受试者新的实时心脏超声二维图像输入到受试者的个体三维心脏模型中，用户客户端输出受试者实时的心脏轮廓关键点的坐标位置；
基于实时的心脏轮廓关键点坐标位置，得到受试者实时的心脏超声三维成像。
通过上述技术方案，本公开的有益效果是：
(1)本公开的人工智能芯片具有强大的现场计算能力，可在微型芯片上完成传统大规模集成电路板卡方能完成的计算过程，使得更加复杂和完整的生物信息获取在穿戴设备上获取后即刻进行处理；
(2)本公开的人工智能芯片具有可重构的优点，从而适用于RNN、CNN、FCN等多种人工智能网络结构，且芯片使用者或应用方不需要了解芯片的物理结构就可以轻易使用人工智能算法指令来实现同时支持空间分布和时间长短记忆的类脑功能，更好地模拟人脑的生物学习模型，增加了人工智能芯片在穿戴式监测设备上使用时，对多种信息处理的兼容性；
(3)本公开具有片上增量训练特点，通过自适应学习能力，不断提升芯片的计算精度，并能应对不同的算法应用环境，适应不同生物信息的计算与处理，适应不同的信息可视化过程；
(4)本公开具有定点数可调稀疏网络计算功能的优点，达到近似浮点数计算精度，实现嵌入式片上系统高精度计算和低功耗的关键需求。
(5)因为人体在活动的过程中适形穿戴式多阵元成像换能器的阵元之间的相对位置会实时变化，所以本公开创新性的提出了位置补偿的技术方案。本公开解决了即使适形换能器阵元相对位置不固定、不确定，也可以通过成像过程中，补偿阵元位置变化，来获得准确的图像。比如，阵元初始位置在A点，但由于阵元的附着基础是适形的，随着呼吸运动或肌肉运动造成的体表形态变化，阵元就会偏离A点，如果仍然按照阵元在A点进行波束合成，就会造成成像的错误，就必须根据阵元实际位置进行波束合成中的补偿问题。
(6)、本公开将超声信号的复杂处理上传至云服务器，而信号采集的移动端或穿戴端仅需进行数据采集和传输，极大降低了移动端或穿戴端图像采集与处理的硬件和软件运算负担，降低了移动端或穿戴端的功耗、发热、体积；
(7)、本公开利用人工智能技术的移动端或穿戴端，可在预成像阶段人机交互式由医生客户端确定少量若干个心脏关键结构点，并利用人工智能技术实时追踪若干个关键结构点信号，极大降低了数据传输量，利用稀疏的、降维的信号实现全分辨率的成像，提高了成像速度、时间与空间分辨率。
(8)、本公开利用深度学习技术，实现了心脏心腔、室壁和瓣膜结构的自动识别，从系列心室短轴和长轴的二维图像，完成三维重建和三维信息的获取；
(9)、本公开利用深度学习技术，不仅可以计算上述的心腔径线、室壁厚度等二维参数，还可提供心腔容积等三维参数，且可以自动计算射血分数、每搏量等功能参数，可实现自动报告，并对出现异常的参数自动警报；
(10)、本公开利用云服务器的交互式访问架构，实现了心脏超声图像的交互式实时访问，有利于家庭、社区和专科医疗机构对受检测人群的信息共享，实时调整受检测者的运动状态或治疗方案。
附图说明
构成本公开的一部分的说明书附图用来提供对本公开的进一步理解，本公开的示意性实施例及其说明用于解释本申请，并不构成对本公开的不当限定。
图1是实施例一SOC芯片的结构图；
图2是实施例一SOC芯片流水线结构的结构图；
图3是实施例一FPE阵列卷积计算单元的结构图；
图4是实施例一基于SOC芯片的穿戴式生物信息监测设备的结构图；
图5为实施例二的方法流程图；
图6为实施例二的穿戴式生物信息监测设备结构图；
图7为实施例二的适形穿戴式多阵元成像换能器佩戴示意图；
图8为实施例二的适形穿戴式多阵元成像换能器细节放大图；
图9(a)和图9(b)为实施例二的波束合成原理示意图；
图10为实施例二的超声标定法示意图；
图11为实施例二的固定反射板安装示意图。
图12为实施例三的硬件连接示意图。
具体实施方式
下面结合附图与实施例对本公开作进一步说明。
实施例一
本实施例提供一种基于SOC芯片的穿戴式生物信息监测设备，请参阅附图4。
一种基于人工智能芯片的穿戴式生物信息监测设备，包括模拟前端电路模块、数字前端模块以及人工智能芯片；
所述模拟前端电路模块，用于产生激励超声换能器的电压脉冲，接收超声面阵换能器采集的回波电信号，并对其进行阻抗匹配，阻抗匹配后的回波电信号经过放大和模数转换后，输入数字前端模块；
所述人工智能芯片根据成像目的和成像模式对生物信息原始数据进行不同运算，得到成像感兴趣区域和关键结构点，输出控制指令至数字前端模块；
数字前端模块，用于接收到人工智能芯片输出的控制指令后，采集所需成像点的回波信号，并对其进行动态的波束合成；波束合成后的信号经过滤波、正交解调以及批处理和流速估计后，实现超声图像重建与实时成像。
进一步的，所述模拟前端电路模块包括高压脉冲芯片、超声发射接收单元和模拟前端接收单元；
所述高压脉冲芯片通过接口与超声面阵换能器连接，接收超声面阵换能器采集的回波电信号，并传输至超声发射接收单元；
所述超声发射接收单元包括发射/接收转换开关和信号发射器，所述信号发射器的输入端与数字前端模块中发射通道波束合成器连接，输出端与发射/接收转换开关连接，用于产生激励超声换能器的电压脉冲；所述发射/接收转换开关分别与高压脉冲芯片、模拟前端接收模块连接，用于向高压脉冲芯片发射激励超声换能器的电脉冲信号，接收高压脉冲芯片发送的回波电信号进行阻抗匹配后，将阻抗匹配后的回波信号发射给模拟前端接收单元；
所述模拟前端接收单元包括前置放大器和模数转换器，阻抗匹配后的回波电信号经过前置放大器放大、模数转换器转换后，输入数字前端模块。
进一步的，所述数字前端模块包括发射通道波束合成器、时间增益补偿、接收通道波束合成器、直流滤波器、解调器和处理器；
所述时间增益补偿与接收通道波束合成器连接，用于补偿传播过程中回波信号的能量衰减；所述接收通道波束合成器与模拟前端接收单元、人工智能芯片连接，用于接收人工智能芯片发送的控制指令，采集所需成像关键点的回波信号，并进行动态的波束合成；波束合成后的信号经过直流滤波器滤波、解调器正交解调以及处理器处理后，实现超声图像重建与实时成像。
进一步的，所述模拟前端电路模块、数字前端模块以及人工智能芯片分别采用柔性电路。
具体地，所述模拟前端电路模块包括高压脉冲芯片、超声发射接收单元和模拟前端接收单元，所述高压脉冲芯片通过接口与超声面阵换能器连接，所述高压脉冲芯片还与超声发射接收单元中发射/接收转换开关连接，用于接收超声面阵换能器采集的回波电信号，并传输至超声发射接收单元；所述 超声发射接收单元包括发射/接收转换开关和信号发射器，信号发射器的输入端与数字前端模块中发射通道波束合成器连接，输出端与发射/接收转换开关连接，用于产生峰值70V的激励超声换能器的电压脉冲；所述发射/接收转换开关分别与高压脉冲芯片、模拟前端接收模块中前置放大器连接，用于向高压脉冲芯片发射激励超声换能器的电脉冲信号，接收高压脉冲芯片发送的回波电信号进行阻抗匹配后，将阻抗匹配后的回波信号发射给模拟前端接收单元，实现发射信号和接收信号之间灵活切换；所述模拟前端接收单元包括前置放大器和模数转换器，阻抗匹配后的回波电信号经过前置放大器放大、模数转换器转换后，进入数字前端模块中接收通道波束合成器进行波束合成。
所述数字前端模块整合于现场可编程门阵列(FPGA)芯片中，包括发射通道波束合成器、时间增益补偿、接收通道波束合成器、直流滤波器、解调器和处理器，所述发射通道波束合成器与信号发射器连接，用于对信号发射器的脉冲进行延时，从而实现在空间特定点的聚焦，激发信号发射器的脉冲产生；所述时间增益补偿与接收通道波束合成器连接，用于补偿传播过程中回波信号的能量衰减；所述接收通道波束合成器与模拟前端接收单元、SOC芯片连接，用于在预调阶段，计算空间中每个点的回波信号延迟时间，对时间增益补偿后的回波信号进行动态的波束合成；在移动阶段，接收SOC芯片的指令仅对来自关键点的回波信号进行波束合成；波束合成后的信号经过直流滤波器滤波、解调器正交解调以及处理器Echo处理和血流速度估计等处理后，实现超声图像重建与实时成像。
在本实施例中，SOC芯片根据成像目的(如心脏成像、血管成像)和成像模式(如利用云架构、便携式一体化成像平台)执行不同的人工智能网络的运算，根据成像感兴趣区域和心脏关键结构点(如瓣环、心尖、室壁)，控制接收通道波束合成器单元对所需成像点进行信号采集。
本实施例提出的穿戴式生物信息监测设备利用FPGA整合数字前端整合，并利用SOC芯片实现现场计算并控制FPGA，继而控制模拟前端电路模块，将传统的大型超声设备架构的各部件功能在小型化的模块组上实现。除了小型化特点外，实现了低电压低功耗，还减少发热。
本实施例提出的穿戴式生物信息监测设备中各模块的电路板和连接线路均采用柔性电路，如聚酰亚胺、聚酯、涤纶等高分子材料的薄膜或粘结片等工艺的电路，或铜、铝等金属材料真空喷镀等工艺的电路，实现可穿戴。其中所述柔性电路包含但不限于引线路(Lead Line)、印刷电路(Printed Circuit)、连接器(Connector)以及多功能整合系统(Integratioon of Function)。
为解决适型可穿戴式生物信息监测设备的低功耗、高吞吐率的难题，需要克服传统人工智能处理硬件系统中存在的通用性弱、吞吐率较低、功耗比较大、不能同时支持多种不同网络类型的缺陷，本实施例提供一种可重构的人工智能(SOC)芯片，能够实现人工智能算法还能够对大量数据传输进行有效管理，不仅减少了冗余数据的传输，从而进一步节省了整个可穿戴式生物信息监测设备的功耗，提高了数据处理和传输的有效性。
请参阅附图1，所述SOC芯片包括四级复合指令流水线和FPE阵列卷积计算单元；
所述四级复合指令流水线包括用于获取矩阵运算指令的第一指令流水线、用于处理所监测到的生物信息原始数据的第二指令流水线、用于对处理后的生物信息原始数据进行矩阵、算数、定点数乘以及点乘运算的第三指令流水线以及加载和存储所监测到的生物信息原始数据的第四指令流水线；
所述FPE阵列卷积计算单元，用于对第二指令流水线和第三指令流水线处理后的生物信息原始数据进行累加处理，重建成像模式、成像感兴趣区域和关键结构点。
进一步的，所述第一指令流水线包括用于预取指令接口接收的指令数据的指令预取缓存器；所述第四指令流水线包括用于通过数据接口读取片外大容量存储器存储的所监测到的生物信息原始数据的加载存储单元。
进一步的，所述第二指令流水线包括译码器和通用目的寄存器，所述译码器对指令预取缓存器存储的指令数据进行译码，所述通用目的寄存器获取第三指令流水线执行后数据、第四指令流水线存储的所监测生物信息原始数据以及FPE阵列卷积计算单元的计算结果，并对其进行逻辑运算处理，译码器译码后的数据、通用目的寄存器处理后的数据经过执行后分别输入到第三指令流水线、FPE阵列卷积计算单元，同时反馈给第一指令流水线。
进一步的，所述第三指令流水线包括当前状态寄存器、算数逻辑单元、定点乘法单元和点乘计算单元；所述当前状态寄存器、算数逻辑单元、定点乘法单元和点乘计算单元分别对第二指令流水线处理后的生物信息原始数据进行矩阵运算、逻辑运算、定点数乘累加运算以及点乘运算处理。
所述当前状态寄存器，用于接收处理后的所监测到的生物信息原始数据，进行矩阵运算处理，将处理后的数据通过数据接口传输至片外大容量存储器；
所述算数逻辑单元，用于接收处理后的所监测到的生物信息原始数据，进行逻辑运算处理，将处理后的数据输入第二指令流水线的通用目的寄存器；
所述定点乘法单元，用于接收处理后的所监测到的生物信息原始数据，进行定点数乘累加运算处理，将处理后的数据输入第二指令流水线的通用目的寄存器；
所述点乘计算单元，用于接收处理后的所监测到的生物信息原始数据，进行向量的点乘运算处理，将处理后的数据输入第二指令流水线的通用目的寄存器。
第四指令流水线包括加载存储单元，所述加载存储单元，用于通过数据接口读取片外大容量存储器存储的所监测到的生物信息原始数据，并输入到第二指令流水线的通用目的寄存器中。
进一步的，所述FPE阵列卷积计算单元包括由多个串联的乘累加处理单元组成的若干乘累加处理单元组，每个乘累加处理单元组连接有缓存器，所述缓存器通过总线与SRAM存储器连接，所述SRAM存储器连接DMA控制器；所述乘累加处理单元组分别处理输入的多个生物信息原始数据，处理后的生物信息数据经过缓存器和SRAM存储器输入至DMA控制器，DMA控制器根据处理后的生物信息数据，重建成像模式、成像感兴趣区域和关键结构点，并将成像模式、成像感兴趣区域和关键结构点数据存储到SRAM存储器中。
由于FPE阵列卷积计算单元的数据搬运以及计算时间都远远超过了单个时钟周期，甚至达到上千个时钟周期，因此不能采用常规的流水线结构，而采用了复合流水线结构的概念来解决流水线的平衡问题。复合流水线结构的SOC芯片同时采用短计算时间流水线和长计算时间流水线两套不同的流水线结构，短时间流水线即常规的单周期流水线结构，长时间流水线结构是根据人工智能深度学习算法的特点，将每一层的数据搬运和数据计算时间平均化，增加流水线的效率，降低空泡产生的几率， 如图2所示。
请参阅附图3，所述FPE阵列卷积计算单元包括128个乘累加处理单元PE、32个缓存器、SRAM存储器和DMA控制器，每4个乘累加处理单元处理输入的4行所监测到的生物信息原始数据，而且每4个乘累加处理单元连接一个192字节的缓存器，所述32个缓存器通过64位总线与16M字节的全局SRAM存储器连接，该SRAM与DMA控制器连接，对SRAM的寻址和读写操作进行控制，所述DMA控制器，用于获取所有乘累加处理单元处理后的生物信息数据，重建成像模式，定位成像感兴趣区域和关键结构点，并将成像模式、成像感兴趣区域和关键结构点数据存储到SRAM存储器中，根据成像感兴趣区域和关键结构点，控制接收通道波束合成器单元对所需成像点进行信号采集。
本实施例提出的阵列卷积计算单元大大减少对SOC芯片外部大容量DDR与SOC芯片内部之间的数据搬运和交互，减少了芯片的面积，同时大大降低了SOC芯片对心脏超声数据处理的功耗。
本实施例提出的SOC芯片吞吐率超过50GOP/s(200MHz时钟频率下)，即每秒可进行超过50×109次16位乘累加运算；并可以按需将定点数精度变成8位使处理速度增加4倍，甚或将时钟频率变为400MHz，使处理速度比初始状态增加8倍，而计算精度误差不超过2％。不仅可适于多种人工智能网络结构，而且实现了近似于浮点数计算精度和高吞吐率，更为重要的是保持了低功耗。
一种基于人工智能芯片的穿戴式生物信息监测设备的工作原理，包括以下步骤：
步骤301，超声发射接收单元的信号发射器接收数字前端模块中发射通道波束合成器发射的激发信号，产生激励超声换能器的电压脉冲，并通过超声发射接收单元的发射/接收转换开关发射至高压脉冲芯片。
步骤302，高压脉冲芯片接收超声面阵换能器采集的回波电信号，并传送给超声发射接收单元的发射/接收转换开关。
步骤303，超声发射接收单元的发射/接收转换开关接收高压脉冲芯片传送的回波电信号，并对其进行阻抗匹配，将阻抗匹配后的回波信号发射给模拟前端接收单元。
步骤304，阻抗匹配后的回波电信号经过前端接收单元的前置放大器放大、模数转换器转换后，进入数字前端模块中接收通道波束合成器进行波束合成。
步骤305，SOC芯片根据成像目的和成像模式进行不同的人工智能网络的运算，得到成像感兴趣区域和关键结构点，输出控制指令至数字前端模块。
步骤306，数字前端模块接收到SOC芯片输出的控制指令，对所需成像点的回波信号进行采集，利用采集到的信号进行动态的波束合成；波束合成后的信号经过直流滤波器滤波、解调器正交解调和处理器Echo处理和血流速度估计各单元处理后，实现超声图像重建与实时成像。
适形，是指：完好的贴合到目标表面。
实施例二
本实施例提供了基于适形穿戴式多阵元成像换能器的超声波束合成系统；
如图5所示，基于适形穿戴式多阵元成像换能器的超声波束合成系统，包括：
适形穿戴式多阵元成像换能器，所述适形穿戴式多阵元成像换能器在使用时，设置在患者心脏对 应的体表胸壁位置；如图7所示；
适形穿戴式多阵元成像换能器接收基于人工智能芯片的穿戴式生物信息监测设备的超声发射接收单元发出的信号，适形穿戴式多阵元成像换能器的每个阵元均将接收到的信号进行时间延迟补偿补偿后发射到患者心脏位置；
适形穿戴式多阵元成像换能器的每个阵元均接收反馈信号，每个阵元均将反馈信号进行时间延迟换算后发送给基于人工智能芯片的穿戴式生物信息监测设备的超声发射接收单元，基于人工智能芯片的穿戴式生物信息监测设备对每个阵元的信号进行波束合成，得到患者心脏的三维超声图像。获得心脏三维超声图像，实时监测患者不同姿势和运动状态下心脏形貌，为医生的诊疗提供更丰富的数据。
作为一个或多个实施例，所述适形穿戴式多阵元成像换能器的每个阵元均将接收到的信号进行时间延迟补偿后发射到患者心脏位置；具体步骤包括：
S401：基于人工智能芯片的穿戴式生物信息监测设备的超声发射接收单元同时向所有的阵元发射超声信号，每个阵元接收器均对超声信号进行接收，通过任意相邻两个阵元接收到超声信号的时间差，以其中一个阵元为参考阵元，计算出另外一个阵元与参考阵元的相对位置；
S402：用另外一个阵元与参考阵元的相对位置，计算出所述相邻两个阵元向心脏位置发射超声波的时间延迟；将时间延迟补偿到所述相邻两个阵元中偏离自身初始位置较大的阵元上；得到时间延迟补偿后的阵元，在设定发射时间的基础上增加时间延迟后再向心脏发射超声波；
S403：通过对每个阵元设定时间延迟，控制阵元产生聚焦声束。
进一步地，所述S401允许被替换为：
在每个阵元上安装位置传感器，通过位置传感器实时反馈阵元的位置；获取任意相邻两个阵元的相对位移。
进一步地，所述S401还允许被替换为：
使用三维超声实时成像扫描监测阵元的实时位置；获取任意相邻两个阵元的相对位移；或者，
使用三维CT实时成像扫描监测阵元的实时位置；获取任意相邻两个阵元的相对位移；或者，
使用高清摄像机实时拍摄换能器在人体活动过程中的位置变换情况，实时记录换能器的相对位置；获取任意相邻两个阵元的相对位移；或者，
使用大数据分析方法，记录阵元在人体每一个动作时的相对位置，通过摄像观测人体的动作推算出阵元的相对位置，(只需要知道阵元之间的相对位置就可以计算出延迟时间，比如图9(a)和图9(b)，在某一时刻几个阵元排成一条直线，中间到焦点的距离短一点，延迟时间就长一些，边上阵元到焦点距离长一点，延迟时间就短一些，这样他们发射的超声波就能同时到达焦点)；获取任意相邻两个阵元的相对位移；或者，
在适形换能器的背面(远离人体一侧)安装一个固定反射板(如图11所示)，在每个阵元的背面也安装一个超声换能器，通过计算超声换能器发射和接收到回波的时间间距来计算出阵元相对放射板的Z方向的距离，进而推算出阵元间Z方向的相对位置；获取任意相邻两个阵元的相对位移。
应理解的，超声成像是利用适形穿戴式多阵元成像换能器的阵元发射超声波，然后接收回来的信 号。接收到的信号经过滤波、放大等处理后，通过分析重组形成超声图像。为了使信号强度最大通常会让所有的阵元发射的超声波形成一个焦点，为了形成焦点，在激励适形穿戴式多阵元成像换能器的时候会给不同的换能器阵元加不同的时间延迟，让适形穿戴式多阵元成像换能器发射的超声波能同时到达焦点处。
应理解的，所述时间延迟就是激励每个阵元的发射超声波的时间不一样，让所有阵元发射的超声波同时到达焦点处。
如图10所示，超声标定法：换能器固定在人体胸壁时，在人体运动过程中它的形变少，只产生Z方向的变化L0，z方向指的是垂直于换能器表面的方向；
Figure PCTCN2020077932-appb-000001
元2接收到超声发射接收单元发出的超声信号的时间；L1表示阵元1距离超声发射接收单元的距离；L2表示阵元2距离超声发射接收单元的距离；L0表示阵元1距离超声发射接收单元的距离与阵元2距离超声发射接收单元的距离的差值；Δt表示相邻两个阵元向心脏位置发射超声波的时间延迟。
进一步地，所述S401还允许被替换为：
采用激光标定法来获取任意相邻两个阵元的相对位移，所述激光标定法与超声标定法的原理相同，即将超声发射接收单元更换为光源发射器。
作为一个或多个实施例，所述适形穿戴式多阵元成像换能器，包括：
适形基底，所述适形基底上均匀分布若干个阵元，每个阵元内部均设有对应的阵元发射器和阵元接收器；
所述阵元接收器，用于将接收到的超声发射接收单元发射的电信号，将电信号转换为超声信号，并对超声信号进行时间延迟补偿后发射到患者心脏位置；
所述阵元发射器，用于将反馈的超声信号，进行时间延迟后，转换为电信号，并将电信号发射给超声发射接收单元。
作为一个或多个实施例，所述适形穿戴式多阵元成像换能器，还包括：散热部件，所述散热部件设置在阵元间隙中，或者设置在阵元的远离人体的一侧。
作为一个或多个实施例，所述适形穿戴式多阵元成像换能器在使用时，设置在患者心脏对应的体表胸壁位置；是指，适形基底贴敷在体表胸壁位置，所述适形基底与体表之间设有超声耦合介质。
所述适形基底的材质为生物相容性柔性材料，如PDMS(聚二甲基硅氧烷)、软硅胶等。
作为一个或多个实施例，如图6所示，所述基于人工智能芯片的穿戴式生物信息监测设备，向适形穿戴式多阵元成像换能器的每个阵元发射超声信号；接收适形穿戴式多阵元成像换能器的每个阵元反馈回来的电信号；对接收的反馈电信号进行波束合成，得到心脏的三维超声图像。
如图7所示，本公开包括适形穿戴式多阵元成像换能器，使用时将该适形穿戴式多阵元成像换能器佩戴在人体心脏对应位置，与人体体表贴合，在人运动的时候能够随之产生形变，适形穿戴式多阵元成像换能器包括有多个独立工作的阵元。
波束合成包括多个发射器可以是一个或多个图中只有一个和位于换能器表面的多个阵元接收器(优选情况为每个阵元对应一个阵元接收器)，超声发射接收单元可以发射激光束或者超声，然后被阵元接收器接收，通过接收的时间间距可以计算出换能器阵元之间的相对位移，并用这些相对位移补偿换能器阵元的时间延迟。基于人工智能芯片的穿戴式生物信息监测设备能够通过对每个阵元设定时间延迟控制阵元产生聚焦声束。
图8为适形穿戴式多阵元成像换能器结构示意图，适形穿戴式多阵元成像换能器的阵元可以是多种形式，比如线阵，面阵，环阵，排列；在阵元之间填充有适形材料，可以使换能器实现弯曲，压缩和拉伸等变形功能。
如图9(a)和图9(b)所示，通过前面方法获得换能器阵元的相对位置，可以计算出每个阵元到焦点的距离，距离差可以计算出时间的延迟。在给适形穿戴式多阵元成像换能器的阵元施加脉冲信号的时候给每个阵元增加的一个数字延迟可以让适形穿戴式多阵元成像换能器发射的声波在焦点处汇聚，同样的，在接收端增加一个延迟线可以将阵元接收到的回波信号同时回到信号处理单元。形成超声图像。
所述适形穿戴式多阵元成像换能器，阵元是换能器的一部分，适形穿戴式多阵元成像换能器还包括阵元之间的填充材料，电缆等，所述适形穿戴式多阵元成像换能器可以是压电陶瓷换能器，压电单晶换能器，也可以是CMUT(Capacitive Micromachined Ultrasonic Transducers)，PMUT(Piezoelectric MicromachinedUtrasoundTransducer)或其他类型换能器。
基于适形穿戴式多阵元成像换能器的超声波束合成方法，包括：
S501：使用时，佩戴在患者心脏对应的体表胸壁位置的适形穿戴式多阵元成像换能器，接收基于人工智能芯片的穿戴式生物信息监测设备的超声发射接收单元发出的信号，适形穿戴式多阵元成像换能器的每个阵元均将接收到的信号进行时间延迟补偿后发射到患者心脏位置；
S502：适形穿戴式多阵元成像换能器的每个阵元均接收反馈信号，每个阵元均将反馈信号进行时间延迟后发送给基于人工智能芯片的穿戴式生物信息监测设备的超声发射接收单元，基于人工智能芯片的穿戴式生物信息监测设备对每个阵元的信号进行波束合成，得到患者心脏的三维超声图像。
作为一个或多个实施例，所述适形穿戴式多阵元成像换能器的每个阵元均将接收到的信号进行时间延迟补偿后发射到患者心脏位置；具体步骤包括：
S5011：超声发射接收单元同时向所有的阵元发射超声信号，每个阵元接收器均对超声信号进行接收，通过任意相邻两个阵元接收到超声信号的时间差，以其中一个阵元为参考阵元，计算出另外一个阵元与参考阵元的相对位置；
S5012：用另外一个阵元与参考阵元的相对位置，计算出所述相邻两个阵元向心脏位置发射超声波的时间延迟；将时间延迟补偿到所述相邻两个阵元中偏离自身初始位置较大的阵元上；得到时间延迟补偿后的阵元，在设定发射时间的基础上增加时间延迟后再向心脏发射超声波；
S5013：通过对每个阵元设定时间延迟，控制阵元产生聚焦声束。
所述相邻两个阵元的相对位移，是指两个投影点之间的距离差，所述两个投影点是相邻两个阵元 分别向垂直于换能器外切面的直线上投影得到的。
实施例三本实施例还提供了基于深度学习的远程心脏超声三维成像系统；
如图12所示，基于深度学习的远程心脏超声三维成像系统，包括：
用户客户端，用于控制超声发射接收单元，向使用时穿戴在人体体表胸壁位置的适形穿戴式多阵元成像换能器发射选通信号指令(如图7所示)；同时，控制基于人工智能芯片的穿戴式生物信息监测设备的超声发射接收单元接收选通信号指令对应的阵元反馈回来的超声信号；将反馈回来的超声信号上传给云服务器；
云服务器，用于对用户客户端上传的超声信号进行处理，利用预训练的个体三维心脏模型，对受试者的心脏超声二维图像进行处理，得到受试者实时的心脏轮廓关键点，基于实时的心脏轮廓关键点，得到受试者实时的心脏超声三维成像；
医生客户端，用于接收医生选取的受试者心脏轮廓关键点，将受试者心脏轮廓关键点通过云服务器发送给用户客户端，用于指导用户客户端发送选通信号指令。
如图8所示，作为一个或多个实施例，所述适形穿戴式多阵元成像换能器，包括：
适形基底，所述适形基底上均匀分布若干个阵元，每个阵元内部均设有对应的阵元发射器和阵元接收器；
所述阵元接收器，用于将接收到的超声发射接收单元发射的电信号，将电信号转换为超声信号，并将超声信号发射到患者心脏位置；
所述阵元发射器，用于将反馈的超声信号，转换为电信号，并将电信号发射给超声发射接收单元。
作为一个或多个实施例，所述适形穿戴式多阵元成像换能器设置在患者心脏正对的体表胸壁位置；是指，适形基底贴敷在体表胸壁位置，所述适形基底与体表之间设有超声耦合介质。
所述适形基底的材质为生物相容性柔性材料，如PDMS(聚二甲基硅氧烷)、等。
作为一个或多个实施例，所述用户客户端，包括：彼此连接的基于人工智能芯片的穿戴式生物信息监测设备和适形穿戴式多阵元成像换能器。
作为一个或多个实施例，所述用户客户端，包括：显示器，显示器用于对云服务器下发的由医生客户端选取的受试者心脏轮廓关键点，进行显示。
基于深度学习的远程心脏超声三维成像方法，包括：
S701：预调成像过程：用户客户端获取使用时受试者佩戴适形穿戴式多阵元成像换能器后，受试者的实时心脏超声二维图像，用户客户端将获取的受试者的实时心脏超声二维图像发送给云服务器，云服务器将受试者的实时心脏超声二维图像发送给医生客户端；实时心脏超声二维图像获取过程中，适形穿戴式多阵元成像换能器接收来自基于人工智能芯片的穿戴式生物信息监测设备发射的超声信号，并将反馈的信号反馈给基于人工智能芯片的穿戴式生物信息监测设备，形成受试者的实时心脏超声二维图像；
医生客户端从受试者的实时心脏超声二维图像中选取受试者心脏轮廓关键点；医生客户端将选取的受试者心脏轮廓关键点发送给云服务器；
云服务器将受试者的实时心脏超声二维图像作为自适应心脏神经网络模型的输入值；云服务器将人工选取的受试者心脏轮廓关键点作为自适应心脏神经网络模型的输出值，对自适应心脏神经网络模型进行训练，得到受试者的个体三维心脏模型；云服务器将受试者心脏轮廓关键点和受试者的个体三维心脏模型均发送给用户客户端；
S702：实时成像过程：用户客户端接收受试者心脏轮廓关键点和受试者的个体三维心脏模型；
用户客户端根据受试者心脏轮廓关键点，向超声发射接收单元发出选通指令，即选通指令控制超声发射接收单元只向轮廓关键点对应的阵元发射超声信号，对于非轮廓关键点对应的阵元不发射超声信号；
用户客户端获取由选通指令对应的阵元所采集的受试者新的实时心脏超声二维图像，用户客户端将受试者新的实时心脏超声二维图像输入到受试者的个体三维心脏模型中，用户客户端输出受试者实时的心脏轮廓关键点的坐标位置；
基于实时的心脏轮廓关键点坐标位置，得到受试者实时的心脏超声三维成像。
作为一个或多个实施例，所述自适应心脏神经网络模型的获取步骤包括：
获取正常人的心脏超声二维图像；对正常人的心脏超声二维图像进行预处理，预处理后图像作为神经网络的输入值，将心脏三维图像作为神经网络的输出值，对神经网络进行训练，得到自适应心脏神经网络模型。
作为一个或多个实施例，所述对正常人的超声二维图像进行预处理，包括：
对正常人的超声二维图像进行格式转化；
对格式转化的图像进行归一化处理；
对归一化处理后的图像进行图像滤波，滤除图像传输过程中的随机扰动、噪声和失真；
对图像滤波后的图像进行图像增强处理，增强组织边界；
对图像增强后的图像进行图像配准处理，将不同二维扫查切面下获得的图像，基于心脏解剖信息、图像灰阶和图像纹理特征进行图像配准；
并将各二维切面图像进行图像融合；
对融合后的图像进行断层图像的插值处理：在二维图像的不同层间加入虚拟切面层，得到心脏外轮廓三维图像。
保证重建的三维图像不畸变、不失真。
应理解的，所述不同二维扫查切面下获得的图像，包括：多个角度心脏长轴和自心底部至心尖部的系列短轴的图像。
应理解的，所述心脏解剖信息是指：乳头肌、瓣膜、心内膜、心尖等解剖信息。
作为一个或多个实施例，所述对正常人的超声二维图像进行预处理，还包括：
对插值处理后的若干幅超声二维图像进行分割，分割出心脏切面的腔室区域；所述心脏切面的腔 室区域包括：左心室、左心房、右心室和右心房；
从心尖四腔心和心尖两腔心切面获取心腔轮廓，再以短轴切面为约束将二维超声图像配准到各心脏腔室区域；
所述短轴切面指心脏超声扫查的标准切面，是超声探头在胸骨旁使声束切面垂直于左心室的长轴而获取的横切面图像，从左心室基底部到心尖，可获得至少三个标准短轴切面。
将处理后的图像输入到预训练的心腔内轮廓分割神经网络中，输出分割后的心脏内轮廓图像。
将心脏外轮廓三维图像和心脏内轮廓图像整合，得到预处理后的心脏三维图像
所述预训练的心腔内轮廓分割神经网络，训练阶段的训练集为有医生标注的心腔内轮廓图像；训练过程中，神经网络的输入值为心腔内轮廓图像，神经网络的输出值为医生标注的坐标位置，训练结束后得到训练好的心腔内轮廓分割神经网络。
作为一个或多个实施例，所述基于实时的心脏轮廓关键点坐标位置，得到受试者实时的心脏超声三维成像；具体步骤包括：
将实时心脏轮廓关键点坐标位置，输入到预训练的三维成像神经网络模型中，输出受试者实时的心脏超声三维成像。
进一步地，所述预训练的三维成像神经网络模型，训练步骤包括：
构建神经网络模型，构建训练集；所述训练集为已知关键点坐标位置的心脏超声三维图像；
训练阶段，将训练集的已知关键点坐标作为神经网络的输入值，将训练集的心脏超声三维图像作为神经网络的输出值；对神经网络进行训练，得到训练好的三维成像神经网络模型。
作为一个或多个实施例，所述方法还包括：
获取到受试者实时的心脏轮廓关键点的坐标位置后，对受试者实时的心脏轮廓关键点的坐标位置与设定的坐标范围进行比较，如果在设定的坐标范围内，则表示当前获取的受试者实时的心脏轮廓关键点的坐标位置是正确的，基于实时的心脏轮廓关键点坐标位置，得到受试者实时的心脏超声三维成像；
如果超出设定的坐标范围，则表示当前获取的受试者实时的心脏轮廓关键点的坐标位置是无效的坐标范围，返回预调成像过程，重新进行关键点坐标位置的选取。
作为一个或多个实施例，所述个体三维心脏模型；具体步骤包括：
Figure PCTCN2020077932-appb-000002
状的特征向量，B i表示用户心脏运动维度的特征向量，i表示所选择的特征向量个数，与心脏关键结构点个数相同，a i和w i组成心脏的参数化表示，通过自适应心脏神经网络模型计算得到。
将各心脏腔室实时超声图像汇总合成，输出合成的整体心脏的实时三维立体动态图像。该实施过程中，对心脏各腔室的确定和划分，利于根据实时超声图像调整自适应心脏神经网络模型，以便形成针对用户心脏的实时三维立体动态图像。
作为一个或多个实施例，所述预调成像过程，包括：用户客户端的适形穿戴式多阵元成像换能器 获得初始超声信号后，通过模数转换后，经无线传输模块发送至云服务器，云服务器对接收的数据进行回波信号处理形成初始二维图像，在初始二维图像上接收医生客户端上传的对心腔、室壁和瓣膜区域的心脏关键结构点(瓣膜结构、心尖位置、心室壁)标识结果，将标识的心脏关键结构点图像输入自适应心脏神经网络模型，建立个体三维心脏模型；
根据个体三维心脏模型的网络坐标点，发送选通指令至用户客户端，对关键结构进行超声信号发射与采集，实现波束合成。
可选的实现方式包括但不局限于以下方式(以左心室成像为例)：
对比个体三维心脏模型与初始左心室长轴和心尖四腔心切面二维图像，在两个正交切面或多个相交切面上获得关键结构点距离超声阵元的初始距离，并转换为数字延迟；在给适形穿戴式多阵元成像换能器的阵元施加脉冲信号的时候给每个阵元增加的一个数字延迟可以让适形穿戴式多阵元成像换能器发射的声波在焦点处汇聚，同样的，在超声发射接收单元增加一个延迟线可以将阵元接收到的回波信号同时回到信号处理单元，实现距离选通。
所述预调成像过程亦可采取另一种实施例：用户客户端适形穿戴式多阵元成像换能器获得初始超声信号，通过模数转换后，再通过时间增益补偿调节、波束合成、滤波与解调后经无线传输模块发送至云服务器；云服务器对接收的数据进行回波信号处理形成初始二维图像，在初始二维图像上接收医生客户端上传的对心腔、室壁和瓣膜区域的心脏关键结构点标识结果，将标识的心脏关键结构点图像输入受试者的个体三维心脏模型，建立个体化精准心脏模型；
将个体三维心脏模型的网络坐标点，传输至用户客户端的超声控制器，实现对关键结构点进行超声信号发射与采集。
在预调成像过程中，对图像采取自动增益调节，具体步骤包括：
(1)线性地将初始图像强度范围转换为0-255；
(2)计算灰度直方图；
(3)根据灰度直方图对γ取值，按照s＝cr对图像进行γ变换，其中γ为常数，对于不同的γ值，有不同的曲线。c为正常数，一般取值不大于1，，r为输入灰度级、s为输出灰度级；如果灰度直方图中出现频率最高的值大于128，则γ取1-25之间的随机值，否则取0-1之间的随机值。
所述实时成像过程，包括：用户客户端的适形穿戴式多阵元成像换能器根据受试者的个体三维心脏模型的关键坐标点，进行选通指令发放和采集超声信号，通过模数转换后，经无线收发模块发送至云服务器，云服务器对接收的数据进行回波信号处理形成实时二维图像，进行自动增益调节。
在所述心脏轮廓关键点初始位置为圆心的球体(优选半径为2厘米，根据平静呼吸时膈肌上下移动范围及躯体侧位时纵隔移动范围确定)空间范围内，按照预调成像过程所述方法获取回波信号，并传送至个体三维心脏模型，实时比对所述空间范围内回波信号与心脏关键结构点初始信号，利用但不限定于互相关算法以确定关键结构点实时坐标。
将受试者在佩戴本系统过程中，所述心脏轮廓关键点位置实际变化范围与变化轨迹输入受试者的个体三维心脏模型，实现个体三维心脏模型深度学习网络的增量学习和自适应学习，获得心脏轮廓关键点位置变化的预判，并根据该预判位置，利用在两个正交切面或多个相交切面上获得心脏轮廓关键点距离超声阵元的初始距离并转换为数字延迟的优选技术方案，预判性调整距离选通以实现动态波束合成。
为了降低传输的数据量，只需传输心脏超声图像中的关键点信号。该关键点包括但不局限于室间隔、心尖、心腔、瓣叶、瓣环等心脏模型解剖结构关键点。
关键点的自动定位需要深度学习不同的心脏解剖结构特征，利用深度卷积网络模型在个体化心脏模型上提取心脏关键点。所有深度学习方法都适用于本实现方法，其中优选的实施方案为采用数据扩增方法(水平、垂直翻转，随机旋转，随机缩放)增加数据集，用基于注意力的CNN强化学习模型自动学习心脏关键点位置。将学习到的关键点坐标返回到移动端或实现距离选通，只选择关键点回波信号采集与传输。
心脏超声图像中关键点信号的获取需定位关键点位置，其与心脏的解剖形态有关，可利用超声数据重构的心脏三维模型进行定位计算。这包含两个关键技术，一是个体化心脏模型建立，二是关键点的自动定位。通过用户客户端获得的初始超声回波信号，经处理后得到包含有整体心脏的B-mode初始图像，对原始图像进行去噪、增强等一系列图像预处理技术，选择感兴趣区域完成心腔图像分割，参考正常人通用三维心脏模型可受试者的个体三维心脏模型。
为保证数据的传输处理速度，利用深度网络的特征提取技术和优选地人工智能芯片，计算心脏模型中的关键点特征，并对超声振元信号接收器进行开关调制，达到超声数据降采样传输要求，不仅满足人工智能芯片的处理速度，且由于去掉了原始数据中的大量的冗余数据，使得系统的数据传输量大大减少，从而降低了功耗，并可进行实时成像。
所述用户客户端为移动端，可为便携式、穿戴式等各种形式，所述便携式可为平板、掌上型等各种形式。换能器可为便携式、穿戴式换能器，换能器类型可为线阵、凸阵、面阵、相控阵等各种形式。所述的用户客户端中，根据换能器的类型，高压脉冲芯片为可选项。在所述的移动端中，根据移动端的体积和类型，其中波束合成器和时间增益补偿亦为可选模块，其功能由云服务器完成。
所述云服务器可为边缘式、分散式等架构，根据用户客户端与云服务器数据运算量的分布，可为云计算、雾计算、海计算等各种类型。
上述虽然结合附图对本公开的具体实施方式进行了描述，但并非对本公开保护范围的限制，所属领域技术人员应该明白，在本公开的技术方案的基础上，本领域技术人员不需要付出创造性劳动即可做出的各种修改或变形仍在本公开的保护范围以内。

Claims (20)

  1. 一种基于人工智能芯片的适形穿戴式生物信息监测设备，其特征是，包括模拟前端电路模块、数字前端模块以及人工智能芯片；
    所述模拟前端电路模块，用于产生激励超声换能器的电压脉冲，接收超声面阵换能器采集的回波电信号，并对其进行阻抗匹配，阻抗匹配后的回波电信号经过放大和模数转换后，输入数字前端模块；
    所述人工智能芯片，用于根据成像目的和成像模式对生物信息原始数据进行不同运算，得到成像感兴趣区域和关键结构点，输出控制指令至数字前端模块；
    数字前端模块，用于接收到人工智能芯片输出的控制指令后，采集所需成像点的回波信号，并对其进行动态的波束合成；波束合成后的信号经过滤波、正交解调以及批处理和流速估计后，实现超声图像重建与实时成像。
  2. 根据权利要求1所述的基于人工智能芯片的适形穿戴式生物信息监测设备，其特征是，所述模拟前端电路模块包括高压脉冲芯片、超声发射接收单元和模拟前端接收单元；
    所述高压脉冲芯片通过接口与超声面阵换能器连接，接收超声面阵换能器采集的回波电信号，并传输至超声发射接收单元；
    所述超声发射接收单元包括发射/接收转换开关和信号发射器，所述信号发射器的输入端与数字前端模块中发射通道波束合成器连接，输出端与发射/接收转换开关连接，用于产生激励超声换能器的电压脉冲；所述发射/接收转换开关分别与高压脉冲芯片、模拟前端接收模块连接，用于向高压脉冲芯片发射激励超声换能器的电脉冲信号，接收高压脉冲芯片发送的回波电信号进行阻抗匹配后，将阻抗匹配后的回波信号发射给模拟前端接收单元；
    所述模拟前端接收单元包括前置放大器和模数转换器，阻抗匹配后的回波电信号经过前置放大器放大、模数转换器转换后，输入数字前端模块。
  3. 根据权利要求1所述的基于人工智能芯片的适形穿戴式生物信息监测设备，其特征是，所述数字前端模块包括发射通道波束合成器、时间增益补偿、接收通道波束合成器、直流滤波器、解调器和处理器；
    所述时间增益补偿与接收通道波束合成器连接，用于补偿传播过程中回波信号的能量衰减；所述接收通道波束合成器与模拟前端接收单元、人工智能芯片连接，用于接收人工智能芯片发送的控制指令，采集所需成像关键点的回波信号，并进行动态的波束合成；波束合成后的信号经过直流滤波器滤波、解调器正交解调以及处理器处理后，实现超声图像重建与实时成像。
  4. 根据权利要求1所述的基于人工智能芯片的适形穿戴式生物信息监测设备，其特征是，所述模拟前端电路模块、数字前端模块以及人工智能芯片分别采用柔性电路。
  5. 根据权利要求1所述的基于人工智能芯片的适形穿戴式生物信息监测设备，其特征是，所述人工智能芯片，包括四级复合指令流水线和FPE阵列卷积计算单元；
    所述四级复合指令流水线包括用于获取矩阵运算指令的第一指令流水线、用于处理所监测到的 生物信息原始数据的第二指令流水线、用于对处理后的生物信息原始数据进行矩阵、算数、定点数乘以及点乘运算的第三指令流水线以及加载和存储所监测到的生物信息原始数据的第四指令流水线；
    所述FPE阵列卷积计算单元，用于对第二指令流水线和第三指令流水线处理后的生物信息原始数据进行累加处理，重建成像模式、成像感兴趣区域和关键结构点。
  6. 根据权利要求5所述的基于人工智能芯片的适形穿戴式生物信息监测设备，其特征是，所述第一指令流水线包括用于预取指令接口接收的指令数据的指令预取缓存器；所述第四指令流水线包括用于通过数据接口读取片外大容量存储器存储的所监测到的生物信息原始数据的加载存储单元。
  7. 根据权利要求6所述的基于人工智能芯片的适形穿戴式生物信息监测设备，其特征是，所述第二指令流水线包括译码器和通用目的寄存器，所述译码器对指令预取缓存器存储的指令数据进行译码，所述通用目的寄存器获取第三指令流水线执行后数据、第四指令流水线存储的所监测生物信息原始数据以及FPE阵列卷积计算单元的计算结果，并对其进行逻辑运算处理，译码器译码后的数据、通用目的寄存器处理后的数据经过执行后分别输入到第三指令流水线、FPE阵列卷积计算单元，同时反馈给第一指令流水线。
  8. 根据权利要求5所述的基于人工智能芯片的适形穿戴式生物信息监测设备，其特征是，所述第三指令流水线包括当前状态寄存器、算数逻辑单元、定点乘法单元和点乘计算单元；所述当前状态寄存器、算数逻辑单元、定点乘法单元和点乘计算单元分别对第二指令流水线处理后的生物信息原始数据进行矩阵运算、逻辑运算、定点数乘累加运算以及点乘运算处理。
  9. 根据权利要求5所述的基于人工智能芯片的适形穿戴式生物信息监测设备，其特征是，所述FPE阵列卷积计算单元包括由多个串联的乘累加处理单元组成的若干乘累加处理单元组，每个乘累加处理单元组连接有缓存器，所述缓存器通过总线与SRAM存储器连接，所述SRAM存储器连接DMA控制器；所述乘累加处理单元组分别处理输入的多个生物信息原始数据，处理后的生物信息数据经过缓存器和SRAM存储器输入至DMA控制器，DMA控制器根据处理后的生物信息数据，重建成像模式、成像感兴趣区域和关键结构点，并将成像模式、成像感兴趣区域和关键结构点数据存储到SRAM存储器中。
  10. 根据权利要求1-4任一项所述的基于人工智能芯片的适形穿戴式生物信息监测设备，其特征是，超声发射接收单元产生激励超声换能器的电压脉冲，并发射至高压脉冲芯片；
    高压脉冲芯片接收超声面阵换能器采集的回波电信号，并传送给超声发射接收单元；
    超声发射接收单元接收高压脉冲芯片传送的回波电信号，并对其进行阻抗匹配；
    阻抗匹配后的回波电信号经过前端接收单元放大和模数转换后，输入数字前端模块；
    人工智能芯片根据成像目的和成像模式进行不同运算，得到成像感兴趣区域和关键结构点，输出控制指令至数字前端模块；
    数字前端模块接收到人工智能芯片输出的控制指令后，采集所需成像点的回波信号，并对其进行动态的波束合成；
    波束合成后的信号经过直流滤波器滤波、解调器正交解调和处理器处理后，实现超声图像重建与 实时成像。
  11. 基于适形穿戴式多阵元成像换能器的超声波束合成系统，其特征是，包括：
    适形穿戴式多阵元成像换能器，所述适形穿戴式多阵元成像换能器在使用时，设置在患者心脏对应的体表胸壁位置；
    适形穿戴式多阵元成像换能器接收穿戴式生物信息监测设备的超声发射接收单元发出的信号，适形穿戴式多阵元成像换能器的每个阵元均将接收到的信号进行时间延迟补偿补偿后发射到患者心脏位置；
    适形穿戴式多阵元成像换能器的每个阵元均接收反馈信号，每个阵元均将反馈信号进行时间延迟换算后发送给穿戴式生物信息监测设备的超声发射接收单元，穿戴式生物信息监测设备对每个阵元的信号进行波束合成，得到患者心脏的三维超声图像。
  12. 如权利要求11所述的系统，其特征是，所述适形穿戴式多阵元成像换能器的每个阵元均将接收到的信号进行时间延迟补偿后发射到患者心脏位置；具体步骤包括：
    S401：超声发射接收单元同时向所有的阵元发射超声信号，每个阵元接收器均对超声信号进行接收，通过任意相邻两个阵元接收到超声信号的时间差，以其中一个阵元为参考阵元，计算出另外一个阵元与参考阵元的相对位置；
    S402：用另外一个阵元与参考阵元的相对位置，计算出所述相邻两个阵元向心脏位置发射超声波的时间延迟；将时间延迟补偿到所述相邻两个阵元中偏离自身初始位置较大的阵元上；得到时间延迟补偿后的阵元，在设定发射时间的基础上增加时间延迟后再向心脏发射超声波；
    S403：通过对每个阵元设定时间延迟，控制阵元产生聚焦声束。
  13. 如权利要求11所述的系统，其特征是，所述适形穿戴式多阵元成像换能器，包括：
    适形基底，所述适形基底上均匀分布若干个阵元，每个阵元内部均设有对应的阵元发射器和阵元接收器；
    所述阵元接收器，用于将接收到的超声发射接收单元发射的电信号，将电信号转换为超声信号，并对超声信号进行时间延迟补偿后发射到患者心脏位置；
    所述阵元发射器，用于将反馈的超声信号，进行时间延迟后，转换为电信号，并将电信号传输给超声发射接收单元。
  14. 如权利要求11所述的系统，其特征是，包括：
    使用时，佩戴在患者心脏对应的体表胸壁位置的适形穿戴式多阵元成像换能器，接收穿戴式生物信息监测设备的超声发射接收单元发出的信号，适形穿戴式多阵元成像换能器的每个阵元均将接收到的信号进行时间延迟补偿后发射到患者心脏位置；
    适形穿戴式多阵元成像换能器的每个阵元均接收反馈信号，每个阵元均将反馈信号进行时间延迟换算后发送给穿戴式生物信息监测设备的超声发射接收单元，穿戴式生物信息监测设备对每个阵元的信号进行波束合成，得到患者心脏的三维超声图像。
    所述适形穿戴式多阵元成像换能器的每个阵元均将接收到的信号进行时间延迟补偿后发射到患 者心脏位置；具体步骤包括：
  15. 如权利要求14所述的系统，其特征是，所述适形穿戴式多阵元成像换能器的每个阵元均将接收到的信号进行时间延迟补偿后发射到患者心脏位置；具体步骤包括：
    超声发射接收单元同时向所有的阵元发射超声信号，每个阵元接收器均对超声信号进行接收，通过任意相邻两个阵元接收到超声信号的时间差，以其中一个阵元为参考阵元，计算出另外一个阵元与参考阵元的相对位置；
    用另外一个阵元与参考阵元的相对位置，计算出所述相邻两个阵元向心脏位置发射超声波的时间延迟；将时间延迟补偿到所述相邻两个阵元中偏离自身初始位置较大的阵元上；得到时间延迟补偿后的阵元，在设定发射时间的基础上增加时间延迟后再向心脏发射超声波；
    通过对每个阵元设定时间延迟，控制阵元产生聚焦声束。
  16. 基于深度学习的远程心脏超声三维成像系统，其特征是，包括：
    用户客户端，用于控制超声发射接收单元，向使用时穿戴在人体体表胸壁位置的适形穿戴式多阵元成像换能器发射选通信号指令；同时，控制超声发射接收单元接收选通信号指令对应的阵元反馈回来的超声信号；将反馈回来的超声信号上传给云服务器；
    云服务器，用于对用户客户端上传的超声信号进行处理，利用预训练的个体三维心脏模型，对受试者的心脏超声二维图像进行处理，得到受试者实时的心脏轮廓关键点，基于实时的心脏轮廓关键点，得到受试者实时的心脏超声三维成像；
    医生客户端，用于接收医生选取的受试者心脏轮廓关键点，将受试者心脏轮廓关键点通过云服务器发送给用户客户端，用于指导用户客户端发送选通信号指令。
  17. 如权利要求16所述的系统，其特征是，包括：
    预调成像过程：用户客户端获取使用时受试者佩戴适形穿戴式多阵元成像换能器后，受试者的实时心脏超声二维图像，用户客户端将获取的受试者的实时心脏超声二维图像发送给云服务器，云服务器将受试者的实时心脏超声二维图像发送给医生客户端；
    医生客户端从受试者的实时心脏超声二维图像中选取受试者心脏轮廓关键点；医生客户端将选取的受试者心脏轮廓关键点发送给云服务器；
    云服务器将受试者的实时心脏超声二维图像作为自适应心脏神经网络模型的输入值；云服务器将人工选取的受试者心脏轮廓关键点作为自适应心脏神经网络模型的输出值，对自适应心脏神经网络模型进行训练，得到受试者的个体三维心脏模型；云服务器将受试者心脏轮廓关键点和受试者的个体三维心脏模型均发送给用户客户端；
    实时成像过程：用户客户端接收受试者心脏轮廓关键点和受试者的个体三维心脏模型；
    用户客户端根据受试者心脏轮廓关键点，向超声发射接收单元发出选通指令，即选通指令控制超声发射接收单元只向轮廓关键点对应的阵元发射超声信号，对于非轮廓关键点对应的阵元不发射超声信号；
    用户客户端获取由选通指令对应的阵元所采集的受试者新的实时心脏超声二维图像，用户客户端将受试者新的实时心脏超声二维图像输入到受试者的个体三维心脏模型中，用户客户端输出受试者实时的心脏轮廓关键点的坐标位置；
    基于实时的心脏轮廓关键点坐标位置，得到受试者实时的心脏超声三维成像。
  18. 如权利要求16所述的系统，其特征是，所述自适应心脏神经网络模型的获取步骤包括：获取正常人的心脏超声二维图像；对正常人的心脏超声二维图像进行预处理，预处理后得到心脏三维图像；将正常人的心脏超声二维图像作为神经网络的输入值，将心脏三维图像作为神经网络的输出值，对神经网络进行训练，得到自适应心脏神经网络模型。
  19. 如权利要求16所述的系统，其特征是，所述对正常人的超声二维图像进行预处理，包括：对正常人的超声二维图像进行格式转化；
    对格式转化的图像进行归一化处理；
    对归一化处理后的图像进行图像滤波，滤除图像传输过程中的随机扰动、噪声和失真；
    对图像滤波后的图像进行图像增强处理，增强组织边界；
    对图像增强后的图像进行图像配准处理，将不同二维扫查切面下获得的图像，基于心脏解剖信息、图像灰阶和图像纹理特征进行图像配准；
    并将各二维切面图像进行图像融合；
    对融合后的图像进行断层图像的插值处理：在二维图像的不同层间加入虚拟切面层，得到心脏外轮廓三维图像。
  20. 如权利要求16所述的系统，其特征是，所述系统还包括：
    获取到受试者实时的心脏轮廓关键点的坐标位置后，对受试者实时的心脏轮廓关键点的坐标位置与设定的坐标范围进行比较，如果在设定的坐标范围内，则表示当前获取的受试者实时的心脏轮廓关键点的坐标位置是正确的，基于实时的心脏轮廓关键点坐标位置，得到受试者实时的心脏超声三维成像；
    如果超出设定的坐标范围，则表示当前获取的受试者实时的心脏轮廓关键点的坐标位置是无效的坐标范围，返回预调成像过程，重新进行关键点坐标位置的选取。
PCT/CN2020/077932 2019-07-05 2020-03-05 基于人工智能芯片的适形穿戴式生物信息监测设备及系统 WO2021004076A1 (zh)

Applications Claiming Priority (6)

Application Number Priority Date Filing Date Title
CN201910605155.0A CN112168140B (zh) 2019-07-05 2019-07-05 基于人工智能芯片的穿戴式生物信息监测设备及方法
CN201910605155.0 2019-07-05
CN201911282999.2 2019-12-13
CN201911282999.2A CN110974304B (zh) 2019-12-13 2019-12-13 基于穿戴式柔性超声换能器的超声波束合成系统及方法
CN201911283022.2A CN110974305B (zh) 2019-12-13 2019-12-13 基于深度学习的远程心脏超声三维成像系统及方法
CN201911283022.2 2019-12-13

Publications (1)

Publication Number Publication Date
WO2021004076A1 true WO2021004076A1 (zh) 2021-01-14

Family

ID=74114932

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2020/077932 WO2021004076A1 (zh) 2019-07-05 2020-03-05 基于人工智能芯片的适形穿戴式生物信息监测设备及系统

Country Status (1)

Country Link
WO (1) WO2021004076A1 (zh)

Cited By (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113008154A (zh) * 2021-02-26 2021-06-22 中煤科工集团重庆研究院有限公司 一种桥梁安全监测的光纤传感系统
CN113096826A (zh) * 2021-03-27 2021-07-09 浙江大学 人工肝数据采集智能分析系统及方法
CN113124795A (zh) * 2021-03-19 2021-07-16 夸克云智科技(深圳)有限公司 一种用于金属壁厚监测的微型可视化终端
CN113552573A (zh) * 2021-06-29 2021-10-26 复旦大学 一种基于超声环阵合成孔径接收的快速成像算法
CN113936069A (zh) * 2021-09-29 2022-01-14 之江实验室 一种用于光声断层成像的阵元虚拟插值方法
CN114098799A (zh) * 2021-10-27 2022-03-01 西安交通大学 一种单脉冲内超声空化的快速低伪影实时动态成像方法与系统
CN116399379A (zh) * 2023-06-07 2023-07-07 山东省科学院激光研究所 分布式光纤声波传感系统及其测量方法
CN116458925A (zh) * 2023-06-15 2023-07-21 山东百多安医疗器械股份有限公司 一种便携式无盲区多模态超声心电系统

Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101152646A (zh) * 2006-09-27 2008-04-02 香港理工大学 柔性超声换能器阵列及其应用装置
CN102640012A (zh) * 2009-09-30 2012-08-15 垓技术公司 超声波3d成像系统
CN103837608A (zh) * 2014-03-12 2014-06-04 深圳市神视检验有限公司 一种相控阵接收动态聚焦补偿方法及系统
WO2014208977A1 (en) * 2013-06-25 2014-12-31 Samsung Electronics Co., Ltd. Ultrasonic imaging apparatus and control method thereof
CN107280707A (zh) * 2017-06-20 2017-10-24 天津大学 用于声电成像的相控阵超声聚焦系统
CN107789006A (zh) * 2017-10-30 2018-03-13 武汉互创科技有限公司 一种记录心脏超声操作手法的系统
CN107992329A (zh) * 2017-07-20 2018-05-04 上海寒武纪信息科技有限公司 一种计算方法及相关产品

Patent Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101152646A (zh) * 2006-09-27 2008-04-02 香港理工大学 柔性超声换能器阵列及其应用装置
CN102640012A (zh) * 2009-09-30 2012-08-15 垓技术公司 超声波3d成像系统
WO2014208977A1 (en) * 2013-06-25 2014-12-31 Samsung Electronics Co., Ltd. Ultrasonic imaging apparatus and control method thereof
CN103837608A (zh) * 2014-03-12 2014-06-04 深圳市神视检验有限公司 一种相控阵接收动态聚焦补偿方法及系统
CN107280707A (zh) * 2017-06-20 2017-10-24 天津大学 用于声电成像的相控阵超声聚焦系统
CN107992329A (zh) * 2017-07-20 2018-05-04 上海寒武纪信息科技有限公司 一种计算方法及相关产品
CN107789006A (zh) * 2017-10-30 2018-03-13 武汉互创科技有限公司 一种记录心脏超声操作手法的系统

Cited By (14)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113008154A (zh) * 2021-02-26 2021-06-22 中煤科工集团重庆研究院有限公司 一种桥梁安全监测的光纤传感系统
CN113124795A (zh) * 2021-03-19 2021-07-16 夸克云智科技(深圳)有限公司 一种用于金属壁厚监测的微型可视化终端
CN113124795B (zh) * 2021-03-19 2023-06-20 夸克云智科技(深圳)有限公司 一种用于金属壁厚监测的微型可视化终端
CN113096826A (zh) * 2021-03-27 2021-07-09 浙江大学 人工肝数据采集智能分析系统及方法
CN113096826B (zh) * 2021-03-27 2023-11-24 浙江大学 人工肝数据采集智能分析系统及方法
CN113552573B (zh) * 2021-06-29 2022-07-29 复旦大学 一种基于超声环阵合成孔径接收的快速成像算法
CN113552573A (zh) * 2021-06-29 2021-10-26 复旦大学 一种基于超声环阵合成孔径接收的快速成像算法
CN113936069A (zh) * 2021-09-29 2022-01-14 之江实验室 一种用于光声断层成像的阵元虚拟插值方法
CN114098799A (zh) * 2021-10-27 2022-03-01 西安交通大学 一种单脉冲内超声空化的快速低伪影实时动态成像方法与系统
CN114098799B (zh) * 2021-10-27 2023-06-27 西安交通大学 一种单脉冲内超声空化的快速低伪影实时动态成像方法与系统
CN116399379A (zh) * 2023-06-07 2023-07-07 山东省科学院激光研究所 分布式光纤声波传感系统及其测量方法
CN116399379B (zh) * 2023-06-07 2023-11-03 山东省科学院激光研究所 分布式光纤声波传感系统及其测量方法
CN116458925A (zh) * 2023-06-15 2023-07-21 山东百多安医疗器械股份有限公司 一种便携式无盲区多模态超声心电系统
CN116458925B (zh) * 2023-06-15 2023-09-01 山东百多安医疗器械股份有限公司 一种便携式无盲区多模态超声心电系统

Similar Documents

Publication Publication Date Title
WO2021004076A1 (zh) 基于人工智能芯片的适形穿戴式生物信息监测设备及系统
CN110974305B (zh) 基于深度学习的远程心脏超声三维成像系统及方法
Lee et al. Theoretical quality assessment of myocardial elastography with in vivo validation
CN104272134B (zh) 超声成像系统中的杂波抑制
US20140046188A1 (en) System and Method for Ultrasonic Diagnostics
US11308609B2 (en) System and methods for sequential scan parameter selection
US20140180111A1 (en) Remote controlled telemedical ultrasonic diagnostic device
US11819363B2 (en) Systems and methods to improve resolution of ultrasound images with a neural network
US20210169455A1 (en) System and methods for joint scan parameter selection
EP4061231B1 (en) Intelligent measurement assistance for ultrasound imaging and associated devices, systems, and methods
JP2021536276A (ja) 超音波画像による脂肪層の識別
WO2020016449A1 (en) Ultrasound imaging by deep learning and associated devices, systems, and methods
Daft Conformable transducers for large-volume, operator-independent imaging
Takuma et al. Real-time, 3-dimensional echocardiography acquires all standard 2-dimensional images from 2 volume sets: a clinical demonstration in 45 patients
US20230134503A1 (en) Systems and methods for non-invasive pressure measurements
Rabben Technical principles of transthoracic three-dimensional echocardiography
Duan et al. Validation of optical-flow for quantification of myocardial deformations on simulated RT3D ultrasound
CN113382685A (zh) 用于研究血管特性的方法和系统
CN115813434A (zh) 用于由胎儿超声扫描自动评估分数肢体体积和脂肪瘦体块的方法和系统
US11890142B2 (en) System and methods for automatic lesion characterization
Ibrahim et al. Apodization scheme for hardware-efficient beamformer
CN114554969A (zh) 用于基于深度学习的超声波束形成的方法和装置
US11766239B2 (en) Ultrasound imaging system and method for low-resolution background volume acquisition
Bourbakis et al. A 3-D Ultrasound Wearable Array Prognosis System With Advanced Imaging Capabilities
JP2024512469A (ja) 超音波システム

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 20837686

Country of ref document: EP

Kind code of ref document: A1

122 Ep: pct application non-entry in european phase

Ref document number: 20837686

Country of ref document: EP

Kind code of ref document: A1

122 Ep: pct application non-entry in european phase

Ref document number: 20837686

Country of ref document: EP

Kind code of ref document: A1

32PN Ep: public notification in the ep bulletin as address of the adressee cannot be established

Free format text: NOTING OF LOSS OF RIGHTS PURSUANT TO RULE 112(1) EPC (EPO FORM 1205A DATED 09.12.2022)

122 Ep: pct application non-entry in european phase

Ref document number: 20837686

Country of ref document: EP

Kind code of ref document: A1