Method and system for quantitative ultrasound imaging for spatio-temporal consistency

Abstract

A method of operation of an imaging device operated by at least one processor comprises the steps of sequentially acquiring ultrasound data of a tissue, generating an initial quantitative image of the initial ultrasound data using a neural network, and iteratively generating a current quantitative image by inferring a quantitative image difference from a relationship between previous ultrasound data and current ultrasound data using the neural network and updating the previous quantitative image using the quantitative image difference.

Description

Technology Field

[0001] The present invention relates to ultrasonic imaging. Background Technology

[0002] Because cancer is difficult to detect early, periodic diagnosis is necessary, and the size and characteristics of lesions must be continuously monitored. Representative imaging equipment for this purpose includes X-ray machines, MRI (magnetic resonance imaging), CT (computed tomography), and ultrasound. While X-ray, MRI, and CT have disadvantages such as the risk of radiation exposure, long measurement times, and high costs, ultrasound equipment is safe, relatively inexpensive, and provides real-time imaging, allowing users to monitor lesions in real time and obtain desired images.

[0003] Ultrasound equipment locates the position of lesions in real time, allowing users to efficiently obtain desired images while monitoring the lesions in real time. It is also safe, relatively inexpensive, and highly accessible. However, it has the disadvantage that image quality cannot be consistently maintained depending on the user's skill level, and it cannot image quantitative characteristics.

[0004] Recently, research is being conducted on reconstructing biomechanical characteristics in real time through quantitative ultrasound imaging. In real-time quantitative imaging, spatio-temporal consistency is a critical requirement because inconsistency between frame-to-frame images can degrade diagnostic accuracy. However, acquired RF signals have a low signal-to-noise ratio due to short detection times. Furthermore, since non-linear neural networks are vulnerable to disturbances such as noise or user hand tremors, temporal consistency is consequently not guaranteed in frame-to-frame reconstruction. (Patent Document 1) KR 10-2490019 B (Patent Document 2) KR 10-2380167 B (Patent Document 3) KR 10-2022-0107912 A The problem to be solved

[0005] The present disclosure provides a quantitative ultrasound imaging method and apparatus for spatiotemporal consistency.

[0006] The present disclosure provides a neural network that reconstructs a quantitative image of an ultrasound sequence in real time. means of solving the problem

[0007] According to one embodiment, a method of operation of an imaging device operated by at least one processor comprises the steps of: sequentially inputting an ultrasound sequence composed of ultrasound data sequentially acquired from a tissue into a neural network; using the neural network to infer a difference image representing the difference between first ultrasound data constituting the ultrasound sequence and second ultrasound data input after the first ultrasound data, and using the difference image to update a first quantitative image for the first ultrasound data to generate a second quantitative image for the second ultrasound data, and repeating a quantitative image generation process; and acquiring time-series quantitative images sequentially generated through the quantitative image generation process for the ultrasound sequence.

[0008] delete

[0009] The above neural network can be implemented to generate an initial quantitative image by encoding the initial ultrasound data constituting the ultrasound sequence and decoding the initial features extracted, and to proceed with the process of generating the quantitative image for ultrasound data sequentially input after the initial ultrasound data by using the initial quantitative image as the first quantitative image.

[0010] The above neural network can be implemented to generate the difference image by decoding the difference between the first feature extracted by encoding the first ultrasound data and the second feature extracted by encoding the second ultrasound data.

[0011] The above neural network can be implemented to generate the difference image by decoding the third feature extracted by encoding the first quantitative image together with the first feature and the second feature.

[0012] The sequentially generated quantitative images may be attenuation coefficient images, sound velocity images, scatterer density images, or scatterer size images.

[0013] The sequentially generated quantitative images may be attenuation coefficient images of breast tissue.

[0014] An imaging device according to one embodiment comprises a memory and a processor that executes instructions stored in the memory. By executing the instructions, the processor sequentially inputs an ultrasound sequence composed of ultrasound data sequentially acquired from a tissue into a neural network. Using the neural network, the processor infers a difference image representing the difference between first ultrasound data constituting the ultrasound sequence and second ultrasound data input after the first ultrasound data. By updating a first quantitative image for the first ultrasound data using the difference image, the processor repeats a quantitative image generation process to generate a second quantitative image for the second ultrasound data. The device is implemented to acquire time-series quantitative images sequentially generated through the quantitative image generation process for the ultrasound sequence.

[0015] delete

[0016] The above neural network can be implemented to generate the difference image by decoding the difference between the first feature extracted by encoding the first ultrasound data and the second feature extracted by encoding the second ultrasound data.

[0017] The above neural network can be implemented to generate the difference image by decoding the third feature extracted by encoding the first quantitative image together with the first feature and the second feature.

[0018] The above time-series quantitative images may be attenuation coefficient images, sound velocity images, scatterer density images, or scatterer size images.

[0019] The above processor can generate time-series quantitative images for sequential ultrasound data. The sequentially generated quantitative images may be attenuation coefficient images for breast tissue.

[0020] According to one embodiment, an imaging device operated by at least one processor comprises: an encoder that extracts features of input ultrasound data constituting an ultrasound sequence; a decoder that receives from the encoder a first feature of first ultrasound data constituting the ultrasound sequence and a first feature of second ultrasound data input after the first ultrasound data, and decodes the difference between the first feature and the second feature to generate a difference image; and a generation module that generates a second quantitative image for the second ultrasound data by adding the difference image to a first quantitative image previously generated for the first ultrasound data. The second quantitative image is used to generate a quantitative image of third ultrasound data input after the second ultrasound data.

[0021] The imaging device may further include a feedback encoder that extracts features of the first quantitative image. The decoder receives features of the first quantitative image along with the first features and the second features, and can output the difference image by decoding the input features. Effects of the invention

[0022] According to the embodiment, a spatially and temporally consistent quantitative image can be generated in real time, and in particular, a real-time attenuation coefficient image can be generated.

[0023] According to the embodiment, lesions can be identified in real time using ultrasound equipment.

[0024] According to the embodiment, by generating spatiotemporally consistent attenuation coefficient images during breast ultrasound examination, the utility as a biomarker for breast cancer diagnosis can be enhanced. Brief explanation of the drawing

[0025] FIG. 1 is a diagram conceptually illustrating a quantitative ultrasound imaging device according to one embodiment. FIG. 2 is a diagram conceptually illustrating the acquisition of ultrasonic data according to one embodiment. FIG. 3 is a diagram illustrating an open-loop neural network according to one embodiment. FIG. 4 is a diagram illustrating a closed-loop neural network according to one embodiment. FIG. 5 is a diagram illustrating a simulation for quantitative ultrasound imaging according to one embodiment. FIG. 6 is a diagram illustrating the results of quantitative ultrasound imaging according to one embodiment. FIG. 7 is a flowchart of a quantitative ultrasound imaging method according to one embodiment. FIG. 8 is a hardware configuration diagram of an imaging device according to one embodiment. Specific details for implementing the invention

[0026] Embodiments of the present invention are described below with reference to the attached drawings so that those skilled in the art can easily implement them. However, the present invention may be embodied in various different forms and is not limited to the embodiments described herein. Furthermore, in order to clearly explain the present invention in the drawings, parts unrelated to the explanation have been omitted, and similar parts throughout the specification are denoted by similar reference numerals.

[0027] Throughout the specification, when a part is described as "comprising" a certain component, this means that, unless specifically stated otherwise, it does not exclude other components but may include additional components. Furthermore, terms such as "…part," "…unit," and "module" as used in the specification refer to a unit that processes at least one function or operation, and this may be implemented in hardware, software, or a combination of hardware and software.

[0028] The neural network of the present disclosure is an artificial intelligence model that learns at least one task and may be implemented as software / program executed on a computing device. The program is stored on a storage medium (non-transitory storage media) and includes instructions described to execute the operation of the present invention by a processor. The program may be downloaded over a network or sold in the form of a product.

[0029] The present disclosure enables the identification of lesions in various body parts where ultrasound equipment is available through quantitative ultrasound imaging for spatiotemporal consistency, and the description primarily uses breast ultrasound as an example.

[0030] Breast cancer is one of the most common cancers in women and a leading cause of death. Since mortality and prognosis vary significantly depending on early detection, regular breast cancer screening is strongly recommended. While magnetic resonance imaging (MRI) offers excellent sensitivity for detecting breast cancer, it is not suitable for routine monitoring due to limited accessibility and high costs. Mammography has contributed to the early detection of breast cancer and has been recommended, but its sensitivity is limited in cases of dense breasts.

[0031] Medical ultrasound is suitable for routine screening due to its low cost, high accessibility, and ability to capture images in real time; however, it has been used as a supplementary diagnostic tool for breast cancer due to its limitation of providing only morphological information and failing to image quantitative characteristics. Recently, however, research is being conducted to reconstruct biomechanical characteristics in real time through quantitative ultrasound imaging, raising expectations for the utility of ultrasound. Nevertheless, ultrasound equipment has limitations in ensuring spatio-temporal consistency between frame-to-frame images because the user directly manipulates the probe to acquire RF signals, and the signal-to-noise ratio is low due to the short detection time of the RF signal. In this disclosure, a method and apparatus for quantitative ultrasound imaging to ensure spatio-temporal consistency are described in detail.

[0032] FIG. 1 is a diagram conceptually illustrating a quantitative ultrasound imaging device according to one embodiment, and FIG. 2 is a diagram conceptually illustrating ultrasound data acquisition according to one embodiment.

[0033] Referring to FIG. 1, a quantitative ultrasound imaging device (simply referred to as an ‘imaging device’) (100) is a computing device operated by at least one processor, comprising a memory for storing instructions and a processor for executing instructions, and the processor performs the operation of the present disclosure by executing instructions included in a computer program. The imaging device (100) may be implemented to interact with a plurality of ultrasound devices through a communication network or may be implemented by being integrated into an ultrasound device.

[0034] An imaging device (100) can generate a quantitative image of at least one variable representing the characteristics of a tissue by using a neural network (200) that reconstructs the quantitative characteristics of the tissue from ultrasound data. Here, the imaging device (100) can output an image of the quantitative variables of the tissue, such as the attenuation coefficient (AC), speed of sound (SoS), effective scatterer concentration (ESC) representing the density distribution within the tissue, and effective scatterer diameter (ESD) representing the size of cells, etc. In the description, the attenuation coefficient image may be used as an example. The neural network (200) is an artificial intelligence model capable of learning at least one task and can be implemented as software / program executed on a computing device.

[0035] An imaging device (100) receives an ultrasound sequence (300) of tissue obtained through an ultrasound probe. The ultrasound sequence (300) is a temporal sequence composed of ultrasound data obtained at each time point, and can be defined as U(t=0), …, U(t=k-1), U(t=k), …, U(t=N) in order. The ultrasound sequence (300) can be generated sequentially according to the movement of the ultrasound probe.

[0036] The imaging device (100) can generate a quantitative image from ultrasound data at a specific point in time and can generate time-series quantitative images (400, …, 400-1, 400-2, …, 400-3) for an ultrasound sequence (300). The quantitative image at a specific point in time may be called a quantitative frame image or simply a frame image. In the description, represents a quantitative image (400) of the initial ultrasound data U (t=0). represents the quantitative image (400-1) at time point (previous frame) (t-1), and represents the quantitative image (400-2) at time t (current frame). represents the difference between adjacent quantitative images (quantitative image difference) or the difference image (410) representing it. The difference image (410) can be expressed as sparse pixel-wise differences.

[0037] The imaging device (100) sequentially receives ultrasound data of an ultrasound sequence (300) and sequentially generates quantitative images, the difference between adjacent quantitative images Inferring, and the previous quantitative image The difference inferred in By adding, the current quantitative image The process of generating is repeated. The imaging device (100) can improve the spatiotemporal consistency of time-series data by generating quantitative images by considering inter-frame relationships instead of independently generating quantitative images of each frame. The imaging device (100) can generate quantitative images with guaranteed spatiotemporal consistency by utilizing the high spatial correlation between adjacent images. The imaging device (100) can interpret inter-frame relationships through sequential feedback and infer differences between adjacent frames. An imaging device (100) that generates time-series quantitative images using such an iterative image refinement scheme can be called an RQI-net.

[0038] The neural network (200) can be designed with an encoder that extracts quantitative features from ultrasound data and a decoder structure that reconstructs a quantitative image from the quantitative features. The neural network (200) can be composed of an open-loop neural network θ that generates a quantitative image from a single ultrasound data and a closed-loop neural network Ф that infers the difference between adjacent quantitative images. A quantitative image for the first ultrasound data U (t=0) can be generated through the open-loop neural network. The difference between adjacent quantitative images is inferred through the closed-loop neural network, and the current quantitative image can be generated by updating the previous quantitative image using the inferred difference.

[0039] Referring to FIG. 2, the ultrasound probe (10) can output an ultrasound signal through a transducer and receive an RF (Radio Frequency) signal reflected from the tissue. Ultrasound data can be generated by processing the RF signal. The transducer of the ultrasound probe (10) is composed of arrayed elements, and the array can be varied. The transducer can be implemented with piezoelectric elements. Here, a channel for transmitting and receiving signals can be assigned to each transducer element. Ultrasound can be output for each channel of the transducer, and echo data reflected from the tissue can be obtained.

[0040] The ultrasound probe (10) can output beams of different angles (TX beam #1 to #k) and receive data reflected back from the tissue. The ultrasound data used for quantitative ultrasound imaging may be RF data acquired for multi-angle beam output. The k beams may be set, for example, to beam angles of -15°, -10°, -5°, 0°, 5°, 10°, and 15°. The ultrasound output from the channels with a constant delay may form a plane wave.

[0041] For example, ultrasound data U 1:7 (t) can be generated from RF data obtained using seven different plane waves at time t. The ultrasonic data includes delay time information for receiving the reflected ultrasonic signal for each element of the ultrasonic probe (10). Accordingly, the ultrasonic data can be represented as an image showing the delay time information for each element.

[0042] FIG. 3 is a diagram illustrating an open-loop neural network according to one embodiment, and FIG. 4 is a diagram illustrating a closed-loop neural network according to one embodiment.

[0043] Referring to FIG. 3, the open-loop neural network θ (200A) is a quantitative image for the first ultrasound data U (t=0) of the ultrasound sequence. It can be configured to generate. The structure of the open-loop neural network (200A) can be designed in various ways, for example, quantitative features ENC from ultrasound data t=0 An encoder (210A) for extracting, and quantitative feature ENC t=0 Quantitative image from It can be composed of a decoder (230A) that reconstructs it.

[0044] The encoder (210A) analyzes the time-varying amplitude of the ultrasound data U (t=0) to obtain quantitative features ENC t=0 ~R 512x16x16The following can be extracted. The encoder (210A) may be composed of 2D convolution, batch normalization, activation (e.g., ReLU), etc. RF data U--1 to U7 for each beam angle of ultrasound data U (t=0) may be designed to be filtered individually and connected on a channel basis in the middle of the encoder.

[0045] The decoder (230A) is a quantitative feature ENC t=0 Quantitative image from It can be reconstructed. The decoder (230A) can be configured with a decoding network using various network models / network blocks. For example, the decoder (230A) can generate a high-resolution quantitative image using an upsampling method. Alternatively, the decoder (230A) can generate a high-resolution quantitative image based on a High-Resolution Network (HRNet). The decoder (230A), composed of parallel multi-resolution subnets, can generate a high-resolution quantitative image by sequentially performing multi-resolution convolution starting from a low-resolution subnet to increase the image resolution.

[0046] The objective function θ used for training the open-loop neural network θ(200A). * can be defined as in Equation 1. The open-loop neural network θ(200A) is the ground truth y and the reconstructed quantitative image It can be trained to minimize the difference between them. L2 minimization can be used to prevent overfitting.

[0047]

[0048] Referring to Fig. 4, the closed-loop neural network Ф (200B) quantitatively analyzes the image difference based on the relationship between the previous ultrasound data U (t-1) and the current ultrasound data U (t). Inferring, and the inferred difference Previous quantitative image using By updating, the current quantitative image It can be configured to generate an initial quantitative image for the first ultrasound data U (t=0) through an open-loop neural network (200A). After generating, quantitative image difference through a closed-loop neural network (200B) The procedure for obtaining and generating the current quantitative image by adding the difference between the previous quantitative image and the inferred quantitative image can be carried out sequentially.

[0049] The structure of the closed-loop neural network (200B) can be designed in various ways; for example, the same encoder (210B) and decoder (230B) as the open-loop neural network (200A) can be used, but the features input to the decoder (230B) can be designed differently from those of the decoder (230A). Additionally, the closed-loop neural network (200B) is the previous quantitative image It may further include a feedback encoder (250B) and a time delay module (270B) for inputting the features of to the decoder (230B). Quantitative image difference When adding sequentially, reconstruction errors may accumulate, so a feedback encoder (250B) may be used to prevent the accumulation of reconstruction errors.

[0050] The encoder (210B), like the encoder (210A), can extract quantitative features by analyzing the time-varying amplitude of the input ultrasound data. Features of adjacent frames extracted by the encoder (210B), ENC t-1 and ENC t It can be input to the decoder (230B).

[0051] The decoder (230B) is ENC t-1 and ENC tQuantitative image difference (y) based on information included in the relationship between t -y t-1 Infers ). Here, y represents the correct answer value at that time.

[0052] Meanwhile, the difference When sequentially added, even the reconstruction error of the previous frame may be included. Therefore, the decoder (230B) [requires] the feature FB of the previous quantitative image output from the feedback encoder (250B). t-1 It can receive additional input and use it to resolve error accumulation. That is, the decoder (230B) receives the feature FB of the previous quantitative image. t-1 By using this, the desired quantitative image difference can be inferred by capturing the reconstruction error of the previous frame and correcting (removing) the reconstruction error of the previous frame. Here, the reconstruction error of the previous frame is the ground truth value y t-1 and reconstructed previous quantitative image It can be expressed as a difference.

[0053] The decoder (230B) receives the input ENC t-1 , ENC t , and FB t-1 Using, the correct answer value (y t ) and previous quantitative image Quantitative image difference, which is the difference It can output.

[0054] The generation module (290B) is the quantitative image difference which is the output of the decoder (230B). and previous quantitative image Adding them element by element to finally the current quantitative image It can generate.

[0055] Objective function Ф used for training the closed-loop neural network Ф (200B) * can be defined as in Equation 2. The closed-loop neural network Ф (200B) is the output of the decoder (230B). a is the correct answer value y tand reconstructed previous quantitative image The difference (y t - You can train to get closer to ).

[0056]

[0057] Here, (y t - ) is the difference in correct values ​​between adjacent frames (y t - y t-1 ) and reconstruction error of the previous frame (y t-1 - It can be expressed as the sum of ). Therefore, (y t - Inferring ) is the reconstruction error of the previous frame (y t-1 - While correcting for ), the image difference (y) caused by probe manipulation t - y t-1 It can be intended to update ).

[0058] FIG. 5 is a diagram illustrating a simulation for quantitative ultrasound imaging according to one embodiment.

[0059] Referring to FIG. 5, the training data of the neural network (200) can be composed of ultrasound data obtained from various human body environments and can be collected using an ultrasound simulation tool. Simulation phantoms representing organs and lesions can have sound velocity distributions, attenuation coefficient distributions, and density distributions modeled so as to mimic the human body without losing simplicity and generality. The acquisition of ultrasound data through breast phantom simulation is explained as an example.

[0060] A simulation device (not shown) acquires MRI data related to breast tissue (S110).

[0061] The simulation device models a simulation phantom containing a lesion using MRI data (S120).

[0062] The simulation device maps ultrasound features (S130). The simulation device can map quantitative features of breast tissue (SoS, AC, Density, Scatter concentration, Scatter diameter).

[0063] The simulation device generates a Region of Interest (ROI) while varying the probe position and acquires ultrasound data by repeatedly simulating ultrasound propagation and predicting measurements (S140). The probe is positioned along the breast skin layer, and the region of interest can be established accordingly. To model surface tissue deformation caused by the pressure of the probe, the axial grid can be distorted according to the curvature of the skin layer.

[0064] The simulation uses data configured for the generated phantom and a single time frame (U 1:7 ,Y)=D t=0 It is executed by. A time-series data set can be generated by considering the movement of the probe. For sequential data generation, the probe can be positioned by assuming a movement speed of 0–50 mm / s. By repeating phantom deformation and simulation, D t=1 It can be constructed. The probe movement procedure is repeated a certain number of times, and a series of spacetime data{D t=0 , D t=1 , ..., D t=T} can be obtained.

[0065] FIG. 6 is a diagram illustrating the results of quantitative ultrasound imaging according to one embodiment.

[0066] Referring to FIG. 6, when comparing the imaging device (RQI-Net) (100) of the present disclosure with time-series AC images reconstructed by the prior art, it can be seen that the imaging device (100) clearly depicts a cyst mass with improved spatiotemporal consistency compared to the prior art. The attenuation coefficients of the cyst and benign mass are, on average, 0.294±0.096 and 0.580±0.161 dB / cm / MHz, respectively. The malignant lesion shows a higher attenuation coefficient value (0.771±0.131) than the cyst and benign mass. Thus, cysts, benign and malignant lesions of the breast can be precisely distinguished through the AC images produced by the imaging device (100).

[0067] AC images reconstructed by the imaging device (100) show 79% improved temporal stability compared to conventional technology. Spatially and temporally consistent AC images can provide accurate AC values ​​to medical personnel and reduce the possibility of misdiagnosis, thereby contributing to the early detection of breast cancer.

[0068] FIG. 7 is a flowchart of a quantitative ultrasound imaging method according to one embodiment.

[0069] Referring to FIG. 7, the imaging device (100) sequentially acquires ultrasound data of the tissue (S210). The imaging device (100) can acquire an ultrasound sequence (300) composed of ultrasound data acquired at each time point. Each ultrasound data can be generated from RF data acquired using plane waves having k different angles of incidence.

[0070] The imaging device (100) generates a quantitative image of the initial ultrasound data using a neural network (200) (S220). The imaging device (100) uses an open-loop neural network (200A) to generate a quantitative image of the first ultrasound data U (t=0) of the ultrasound sequence. It can generate.

[0071] The imaging device (100) uses a neural network (200) to infer a quantitative image difference from the relationship between previous ultrasound data and current ultrasound data, and repeats the procedure of generating a current quantitative image by updating the previous quantitative image using the inferred difference (S230). The imaging device (100) can implement the decoder (230B) of the closed-loop neural network (200B) as a feedback structure to capture reconstruction errors included in the previous quantitative image so that reconstruction errors do not accumulate in the quantitative image difference. The decoder (230B) of the closed-loop neural network (200B) [requires] the features of the previous ultrasound data. t-1 , characteristics of current ultrasound data ENC t , and the features of the previous quantitative image FB t-1 Using, the quantitative image difference with reconstruction errors included in the previous quantitative image removed It can output the answer value (y t ) and previous quantitative image It can be trained to get closer to the difference. The imaging device (100) can generate a current quantitative image by adding a difference image representing a difference in pixels and a previous quantitative image.

[0072] FIG. 8 is a hardware configuration diagram of an imaging device according to one embodiment.

[0073] Referring to FIG. 8, the imaging device (100) may be a computing device (500) operated by at least one processor and is connected to an ultrasonic probe (10) or a device that provides data obtained from the ultrasonic probe (10).

[0074] A computing device (500) may include one or more processors (510), a memory (530) for loading a program executed by the processor (510), a storage device (550) for storing the program and various data, a communication interface (570), and a bus (590) connecting them. In addition, the computing device (500) may include various additional components. The program may include instructions that cause the processor (510) to perform a method / operation according to various embodiments of the present disclosure when loaded into the memory (530). That is, the processor (510) may perform a method / operation according to various embodiments of the present disclosure by executing the instructions. An instruction refers to a series of computer-readable instructions grouped by function, which are components of a computer program and are executed by a processor.

[0075] The processor (510) controls the overall operation of each component of the computing device (500). The processor (510) may be configured to include at least one of a CPU (Central Processing Unit), MPU (Micro Processor Unit), MCU (Micro Controller Unit), GPU (Graphic Processing Unit), or any form of processor well known in the art of the present disclosure. Additionally, the processor (510) may perform operations for at least one application or program for executing a method / operation according to various embodiments of the present disclosure.

[0076] The memory (530) stores various data, commands and / or information. The memory (530) may load one or more programs from a storage device (550) to execute a method / operation according to various embodiments of the present disclosure. The memory (530) may be implemented as a volatile memory such as RAM, but the technical scope of the present disclosure is not limited thereto.

[0077] The storage device (550) may store a program non-temporarily. The storage device (550) may be configured to include non-volatile memory such as ROM (Read Only Memory), EPROM (Erasable Programmable ROM), EEPROM (Electrically Erasable Programmable ROM), flash memory, a hard disk, a removable disk, or any form of computer-readable recording medium well known in the art to which this disclosure belongs.

[0078] The communication interface (570) supports wired and wireless communication of the computing device (500). To this end, the communication interface (570) may be configured to include a communication module well known in the art of the present disclosure.

[0079] The bus (590) provides communication functions between components of the computing device (500). The bus (690) can be implemented as various types of buses, such as an address bus, a data bus, and a control bus.

[0080] As such, according to the embodiment, a spatially and temporally consistent quantitative ultrasound image can be generated in real time, and in particular, a real-time attenuation coefficient image (real-time AC image) can be generated.

[0081] According to the embodiment, lesions can be identified in real time using ultrasound equipment.

[0082] According to the embodiment, by generating spatiotemporally consistent attenuation coefficient images during breast ultrasound examination, the utility as a biomarker for breast cancer diagnosis can be enhanced.

[0083] The embodiments of the present invention described above are not implemented only through devices and methods, but may also be implemented through a program that realizes a function corresponding to the configuration of the embodiments of the present invention or a recording medium on which such program is recorded.

[0084] Although embodiments of the present invention have been described in detail above, the scope of the present invention is not limited thereto, and various modifications and improvements by those skilled in the art using the basic concept of the present invention as defined in the following claims also fall within the scope of the present invention.

Claims

Claim 1 A method of operation of an imaging device operated by at least one processor, comprising the steps of: sequentially inputting an ultrasound sequence composed of ultrasound data sequentially acquired from a tissue into a neural network; using the neural network to infer a difference image representing the difference between first ultrasound data constituting the ultrasound sequence and second ultrasound data input after the first ultrasound data, and using the difference image to update a first quantitative image for the first ultrasound data to generate a second quantitative image for the second ultrasound data, and repeating a quantitative image generation process; and acquiring time-series quantitative images sequentially generated through the quantitative image generation process for the ultrasound sequence. Claim 2 delete Claim 3 A method of operation according to claim 1, wherein the neural network generates an initial quantitative image by encoding the initial ultrasound data constituting the ultrasound sequence and decoding the initial features extracted therefrom, and uses the initial quantitative image as the first quantitative image to proceed with the process of generating the quantitative image for ultrasound data sequentially input after the initial ultrasound data. Claim 4 A method of operation according to claim 1, wherein the neural network is implemented to generate the difference image by decoding the difference between the first feature extracted by encoding the first ultrasound data and the second feature extracted by encoding the second ultrasound data. Claim 5 A method of operation in which, in paragraph 4, the neural network is implemented to generate the difference image by decoding the third feature extracted by encoding the first quantitative image together with the first feature and the second feature. Claim 6 A method of operation in which, in claim 1, the time-series quantitative images are attenuation coefficient images, sound velocity images, scatterer density images, or scatterer size images. Claim 7 In claim 1, the time-series quantitative images are quantitative images of breast tissue, a method of operation. Claim 8 An imaging device comprising a memory and a processor that executes instructions stored in the memory, wherein the processor, by executing the instructions, sequentially inputs an ultrasound sequence composed of ultrasound data sequentially acquired from an organization into a neural network, uses the neural network to infer a difference image representing the difference between first ultrasound data constituting the ultrasound sequence and second ultrasound data input after the first ultrasound data, and repeats a quantitative image generation process to generate a second quantitative image for the second ultrasound data by updating a first quantitative image for the first ultrasound data using the difference image, and is implemented to acquire time-series quantitative images sequentially generated through the quantitative image generation process for the ultrasound sequence. Claim 9 delete Claim 10 An imaging device according to claim 8, wherein the neural network is implemented to generate the difference image by decoding the difference between the first feature extracted by encoding the first ultrasound data and the second feature extracted by encoding the second ultrasound data. Claim 11 An imaging device according to claim 10, wherein the neural network is implemented to generate the difference image by decoding a third feature extracted by encoding the first quantitative image together with the first feature and the second feature. Claim 12 In paragraph 8, the imaging device wherein the time-series quantitative images are attenuation coefficient images, sound velocity images, scatterer density images, or scatterer size images. Claim 13 In paragraph 8, the imaging device, wherein the time-series quantitative images are quantitative images of breast tissue. Claim 14 An imaging device operated by at least one processor comprises: an encoder that extracts features of input ultrasound data constituting an ultrasound sequence; a decoder that receives from the encoder a first feature of first ultrasound data constituting the ultrasound sequence and a second feature of second ultrasound data input after the first ultrasound data, and decodes the difference between the first feature and the second feature to generate a difference image; and a generation module that adds the difference image to a first quantitative image previously generated for the first ultrasound data to generate a second quantitative image for the second ultrasound data, wherein the second quantitative image is used to generate a quantitative image of third ultrasound data input after the second ultrasound data. Claim 15 An imaging device according to claim 14, further comprising a feedback encoder for extracting features of the first quantitative image, wherein the decoder receives features of the first quantitative image together with the first features and the second features, decodes the input features, and outputs the difference image.

Images