Ultrasound sweep evaluation system and method

Abstract

A technique for evaluating sweep acquisitions is disclosed. Ultrasonic data for a sweep acquisition set is received. The received ultrasonic data may include data acquired by non-expert users. A representation of the ultrasonic data in a latent space is generated. The representation of the ultrasonic data is compared with a latent distribution of reference ultrasonic data, and a recommendation is generated based on the difference between the representation of the ultrasonic data and the closest subset of the reference ultrasonic data. The recommendation may be to improve the quality of the received ultrasonic data or to repeat at least a portion of the sweep acquisition set. The reference ultrasonic data may be data acquired by an expert user or data with characteristics of expert-acquired data. A neural network can be used to perform the comparison between the received ultrasonic data and the reference ultrasonic data.

Description

Technical Field

[0001] This disclosure relates to the evaluation of medical image acquisition. More specifically, this application relates to the evaluation of ultrasound image acquisition, such as sweep acquisition. Background Technology

[0002] Various medical imaging modalities, such as magnetic resonance imaging (MRI), ultrasound (US), or computed tomography (CT), can be used for clinical analysis and medical interventions, as well as for the visual representation of organ and tissue function. Implementation of a medical imaging modality can utilize one or more protocols, such as imaging protocols for acquiring ultrasound images using sweep. Sweeping can refer to acquiring multiple image frames at different physical locations during continuous or substantially continuous movement of the position in the imaging plane (e.g., by physically moving a transducer and / or electronic beamforming). Sweeps can be acquired according to patterns and / or grids. Imaging protocols can be used, for example, to evaluate a fetus using an ultrasound imaging system. Summary of the Invention

[0003] Apparatus, systems, and methods for evaluating received ultrasound imaging data are disclosed. In some embodiments, the disclosed techniques receive ultrasound imaging data acquired by a novice user according to a protocol, which may be a protocol for acquiring ultrasound images using sweep. The disclosed techniques can analyze novice user ultrasound imaging data by comparing it with ultrasound imaging datasets acquired by expert users to determine the characteristics of the novice user ultrasound imaging data. For example, the disclosed techniques can identify the closest to an expert acquisition and generate recommendations based on comparing the novice user ultrasound imaging data with the closest expert acquisition. Recommendations may include recommendations for improving data quality or repeating acquisitions. The operation of the disclosed techniques can be performed using a neural network trained to analyze novice user ultrasound imaging data relative to a latent space including representations of expert user ultrasound imaging data.

[0004] An ultrasound imaging system is disclosed according to at least one example disclosed herein. The ultrasound imaging system includes an ultrasound probe configured to acquire ultrasound imaging data and at least one processor in communication with the ultrasound probe. The at least one processor is configured to receive ultrasound imaging data corresponding to a set of sweep acquisitions performed using the ultrasound probe. A representation of the received ultrasound imaging data in a latent space is generated. The representation of the received ultrasound imaging data in the latent space is compared with a latent distribution of a reference ultrasound imaging dataset represented in the latent space to identify the closest subset of the reference ultrasound imaging dataset. At least one recommendation is generated based on the difference between the representation of the received ultrasound imaging data in the latent space and the identified closest subset of the reference ultrasound imaging dataset, the at least one recommendation including a recommendation to improve the quality of the received ultrasound imaging data, a recommendation for at least a portion of the repeated sweep acquisition set, or both.

[0005] In some implementations, a neural network trained using a reference ultrasound imaging dataset is used to compare the representation of the received ultrasound imaging data in the latent space with the latent distribution of the reference ultrasound imaging dataset represented in the latent space.

[0006] In some implementations, at least one processor is configured to access and / or generate a training dataset that includes a reference ultrasound imaging dataset, and to use the training dataset to train a neural network.

[0007] In some implementations, at least one processor is configured to receive updated ultrasound imaging data based at least in part on at least one recommendation, and to evaluate at least one characteristic of the anatomical structures in the updated ultrasound imaging data, such as fetal health characteristics of the fetus.

[0008] In some implementations, the received ultrasonic imaging data includes sensor data captured via at least one of an accelerometer, magnetometer, gyroscope, or electromagnetic positioning sensor.

[0009] In some implementations, at least one processor is also configured to display at least one recommendation via a graphical user interface.

[0010] In some implementations, the difference between the representation of the received ultrasound imaging data in the latent space and the identified closest subset of the reference ultrasound imaging dataset is based at least in part on reconstruction error, Kullback-Leibler divergence, or both.

[0011] In some implementations, the at least one processor is configured to automatically apply the at least one recommendation to the received ultrasound imaging data.

[0012] In some implementations, the reference ultrasound imaging dataset includes data acquired by experts or data with characteristics of data acquired by experts, which may be simulated data.

[0013] In some implementations, the sweep acquisition set is associated with an ultrasound imaging protocol.

[0014] According to at least one example disclosed herein, a non-transient computer-readable medium carrying instructions that, when executed, cause a processor to perform operations to execute at least one method disclosed herein.

[0015] Based on at least one example disclosed herein, this document discloses at least a portion of methods for using the ultrasound imaging system disclosed herein. Attached Figure Description

[0016] Figure 1 It is a block diagram of an ultrasound imaging system arranged according to the principles of this disclosure.

[0017] Figure 2 This is a block diagram illustrating a workflow for training a model to evaluate sweep acquisition based on the principles of this disclosure.

[0018] Figure 3 This is a diagram illustrating the mapping of a sweep acquisition in the latent space according to the principles of this disclosure.

[0019] Figure 4 This is a block diagram illustrating a workflow for evaluating sweep acquisition using a trained model, based on the principles of this disclosure.

[0020] Figure 5 This is a flowchart illustrating a process for evaluating ultrasound imaging data according to the principles of this disclosure. Detailed Implementation

[0021] The following descriptions of certain examples are illustrative in nature and are in no way intended to limit the disclosed technology or its application or use. In the following detailed description of examples of apparatuses, systems, and methods, reference is made to the accompanying drawings, which form part of the invention and illustrate specific examples of the described apparatuses, systems, and methods by way of illustration. These examples are described in sufficient detail to enable those skilled in the art to practice the currently disclosed apparatuses, systems, and methods, and it should be understood that other examples may be utilized and structural and logical changes may be made without departing from the spirit and scope of this disclosure. Furthermore, for clarity, certain features will not be discussed in detail where they are obvious to those skilled in the art, so as not to obscure the description of the system. Therefore, the following detailed description should not be considered limiting, and the scope of the technology is defined only by the appended claims.

[0022] Users of medical imaging systems (such as ultrasound imaging systems) face technical challenges related to acquiring images of sufficient quality to evaluate one or more conditions of a subject. For example, medical imaging systems may be used in resource-constrained care settings where trained, experienced, and / or expert users may not be readily available. In these and other settings, acquiring imaging data by non-expert users (such as untrained, minimally trained, and / or novice users) may be necessary or advantageous. In some examples, non-expert users may perform imaging sweeps along a predetermined grid, which can be performed without real-time access to the acquired images (referred to as a “blind scan”) or using another protocol. In other examples, users (e.g., non-expert users) may perform guided sweeps, where the user follows guidance provided via the device or system to reach a location and / or orientation for capturing a specific view. These sweeps contrast with freehand sweeps performed by trained users, during which the user can move the transducer as needed to locate anatomical structures of interest.

[0023] Existing systems may have the capability to use imaging data acquired using blind or guided scans, such as for performing pregnancy estimation (e.g., age, viability, multiple pregnancies) based on blind ultrasound imaging data. Some systems may use one or more artificial intelligence (AI) models to analyze medical imaging data. However, non-expert users may lack the ability to achieve or maintain the quality (e.g., sufficient resolution, non-blurring, object of interest fully within the frame) and consistency (e.g., consistent spacing between image frames, a sufficient number of frames across the entire object of interest) necessary for the reliable and effective application of these existing systems. For example, low-quality acquisition can affect the performance of AI models used for automated estimation of fetal characteristics (e.g., gestational age, etc.). Therefore, further technologies are needed to assist non-expert users.

[0024] This disclosure describes systems and related methods for evaluating medical imaging data, such as for evaluating the quality of sweep acquisitions (e.g., blind or guided sweeps) using an ultrasound imaging system. In some embodiments, the disclosed techniques can be implemented using an ultrasound imaging system capable of operating with imaging presets optimized and / or configured for a specific type of imaging, such as obstetric imaging. As used herein, “sweep” or “sweep acquisition” can refer to one or more operations (which may be pursuant to a protocol) for acquiring imaging data using an ultrasound transducer or probe, such as operations for acquiring imaging data by moving a transducer or probe along one or more predetermined paths relative to the anatomical structure of an object. In some examples, sweep acquisition may follow a grid or other pattern. The disclosed techniques may include one or more models trained to evaluate medical imaging data to determine the quality of the medical imaging data and / or generate one or more recommendations. For example, recommendations may be provided to improve the quality of medical imaging data and / or repeat acquisitions of at least a portion of the medical imaging data. In some embodiments, the disclosed techniques include AI-based models to evaluate the quality of sweeps collected by non-expert users by comparing sweeps collected by non-expert users with sweeps collected by expert and / or trained users. The model can also generate recommendations for improving the collected data and / or redoing at least some parts of the scans based on the results of the quality assessment.

[0025] In some embodiments, the disclosed technology includes at least one processing controller that receives a sequence of two-dimensional (2D) ultrasound frames including images of the fetus and evaluates scan quality by comparing images of the fetus captured by a user using an ultrasound probe with images of the fetus acquired by an expert user. The system makes recommendations for improving the quality of the acquired data or re-capturing the data before using the data for clinical parameter estimation. The processing controller may be a standalone system that receives real-time or offline data or is integrated into an ultrasound imaging system.

[0026] Advantages of the disclosed techniques may include using blind scan protocols or other protocols or sweeps to improve pregnancy estimation and / or other analyses of medical imaging data. Although examples related to obstetric ultrasound screening for novice or minimally trained users are described herein, it should be understood that the disclosed techniques can be applied to other medical imaging implementations (e.g., general imaging, vascular imaging, etc.). Furthermore, while the examples herein relate to evaluating imaging data acquired using blind scan protocols, it should be understood that the disclosed techniques can be applied to other imaging protocols and / or other uses of medical imaging systems.

[0027] Figure 1 This is a block diagram of an ultrasound imaging system 100 arranged according to the principles of this disclosure. Figure 1In the ultrasound imaging system 100, the ultrasound probe 112 includes a transducer array 114 for emitting ultrasound waves and receiving echo information. The transducer array 114 can be implemented as a linear array, a convex array, a phased array, and / or a combination thereof. For example, the transducer array 114 may include a two-dimensional array of transducer elements (as shown), capable of scanning in both elevation and azimuth dimensions for 2D and / or 3D imaging. The transducer array 114 may be coupled to a microwave beamformer 116 in the probe 112, which controls the transmission and reception of signals by the transducer elements in the array. In this example, the microwave beamformer 116 is coupled via a probe cable to a transmit / receive (T / R) switch 118, which switches between transmission and reception and protects the main beamformer 122 from high-energy transmitted signals. In some embodiments, the T / R switch 118 and other components in the system may be included in the ultrasound probe 112, rather than in a separate ultrasound system base. In some embodiments, the ultrasound probe 112 may be coupled to the ultrasound imaging system via a wireless connection (e.g., WiFi, Bluetooth).

[0028] The transmission of the ultrasonic beam from transducer array 114, under the control of microwave beamformer 116, is guided by a transmit controller 120 coupled to T / R switch 118 and beamformer 122, which receives input from user-to-user interface (e.g., control panel, touchscreen, console) 125. User interface 125 may include soft controls and / or hard controls. One of the functions controlled by transmit controller 120 is the direction in which the beam is steered. The beam can be steered directly forward (orthogonal to transducer array 114) from transducer array 114, or at different angles to obtain a wider field of view. The partial beamforming signal generated by microwave beamformer 116 is coupled to main beamformer 122 via channel 115, wherein the partial beamforming signals from individual patches of transducer elements are combined into a fully beamformed signal. In some embodiments, microwave beamformer 116 is omitted, and transducer array 114 is coupled to beamformer 122 via channel 115. In some embodiments, system 100 may be configured (e.g., including a sufficient number of channels 115 and having a transmit / receive controller programmed to drive transducer array 114) to acquire ultrasonic data in response to a plane wave or diverging ultrasonic beam emitted toward an object. In some embodiments, the number of channels 115 from the ultrasonic probe may be less than the number of transducer elements in transducer array 114, and the system may be operable to acquire ultrasonic data encapsulated in a smaller number of channels than the number of transducer elements.

[0029] The beamforming signal is coupled to signal processor 126. Signal processor 126 can process the received echo signal in various ways, such as bandpass filtering, decimation, I and Q component separation, and / or harmonic signal separation. Signal processor 126 can also perform additional signal enhancement, such as speckle reduction, signal recombination, and noise cancellation. The processed signal is coupled to mode-B processor 128, which can use amplitude detection to image structures in the body. The signal generated by mode-B processor 128 is coupled to scan converter 130 and multiplane reformer 132. Scan converter 130 arranges the echo signal in the desired image format according to the spatial relationship of the echo signal received from it. For example, scan converter 130 can arrange the echo signal into a two-dimensional (2D) fan-shaped format or a pyramidal three-dimensional (3D) image. Multiplane reformer 132 can convert echoes received from points in a common plane in a volumetric region of the body into an ultrasound image of that plane, as described in U.S. Patent US 6,443,896 (Detmer).

[0030] Volumetric renderer 134 converts the echo signals of a 3D dataset into a projected 3D image viewed from a given reference point, for example, as described in U.S. Patent US 6,530,885 (Entrekin et al.). 2D or 3D images can be coupled from scan converter 130, multiplane reformer 132, and volumetric renderer 134 to at least one processor 137 for further image processing operations. For example, at least one processor 137 may include an image processor 136 configured to perform further enhancement and / or buffering and temporary storage of imaging data for display on an image display 138. Display 138 may include a display device implemented using various display technologies such as LCD, LED, OLED, or plasma display technologies. At least one processor 137 may include a graphics processor 140 capable of generating graphic overlays for display alongside ultrasound images. These graphic overlays may contain, for example, standard identification information such as patient name, date and time of image, imaging parameters, etc. For these purposes, graphics processor 140 receives input from user interface 125, such as a typed patient name. User interface 125 can also be coupled to multiplane reformer 132 for selecting and controlling the display of multiple multiplane reformulated (MPR) images.

[0031] User interface 125 may include one or more mechanical controls, such as buttons, dial pads, trackballs, physical keyboards, etc., which may also be referred to herein as hard controls. Alternatively or additionally, user interface 125 may include one or more soft controls, such as buttons, menus, soft keyboards, and other user interface control elements implemented using touch-sensitive technology (e.g., resistive, capacitive, or optical touchscreens). One or more of the user controls may coexist on control panel 124. For example, one or more mechanical controls may be provided on the console and / or one or more soft controls may coexist on a touchscreen, which may be attached to or integrated with the console. Display 138 and user interface 125 may be included in I / O components through which system 100 provides outputs and / or system 100 receives inputs.

[0032] In some embodiments, the user interface 125 may provide one or more outputs of the disclosed system, including generated recommended calibration sequences and / or lists for improving the determined sweep quality and / or confidence metrics. Additionally or alternatively, information regarding sweep quality and / or characteristics (e.g., sweep quantity, cause of low-quality data, maximum and minimum probe acceleration / velocity, sweep direction, etc.) may be output to the user. In some embodiments, one or more outputs of the system may be provided via one or more graphical user interfaces (e.g., via display 138).

[0033] At least one processor 137 (e.g., image processor 136, graphics processor 140, or a different processor) may also perform functions associated with evaluating medical imaging data, as described herein. For example, at least one processor 137 may compare received ultrasound imaging data (e.g., non-expert and / or novice acquisitions) with reference ultrasound imaging data (e.g., expert and / or trained user acquisitions) and generate one or more recommendations based on that comparison, such as recommendations to improve the quality of the received ultrasound imaging data and / or recommendations to repeat acquisitions of at least a portion of the received ultrasound imaging data. The quality of the imaging data may include various characteristics such as resolution, blurriness, presence / visibility of objects of interest or landmarks, contrast, noise, scaling, frame count, frame spacing, probe or transducer movement or orientation, etc. In some embodiments, at least one processor 137 trains, provides, and / or accesses one or more models disclosed herein, such as neural networks and / or other AI models.

[0034] Although described as a separate processor, it should be understood that the functionality of any processor described herein may be implemented in a single processor (e.g., a CPU or GPU implementing the functionality of processor 137) or in fewer processors than described in this example. While image processor 136 and graphics processor 140 are described, at least one processor 137 may include more or fewer processors, and the functionality of one or more processors may be combined. In some embodiments, at least one processor 137 may be hardware-based (e.g., including a multi-layer interconnect node implemented in hardware). In some embodiments, at least one processor 137 may be implemented in other processing stages, e.g., prior to processing performed by image processor 136, volumetric renderer 134, multi-plane reformer 132, and / or scan converter 130. In some embodiments, at least one processor 137 may be implemented to process ultrasound data in the channel domain, beamspace domain (e.g., before or after beamformer 122), IQ domain (e.g., before, after, or in conjunction with signal processor 126), and / or k-space domain. As described above, in some embodiments, the functions of two or more processing components (e.g., beamformer 122, signal processor 126, B-mode processor 128, scan converter 130, multi-plane reformer 132, volumetric renderer 134, at least one processor 137, image processor 136, graphics processor 140, etc.) can be combined into a single processing unit and / or partitioned among multiple processing units. The processing unit can be implemented in software, hardware, or a combination thereof. For example, at least one processor 137 may include one or more graphics processing units (GPUs). In another example, beamformer 122 may include an application-specific integrated circuit (ASIC).

[0035] At least one processor 137 may be coupled to one or more computer-readable media (e.g., memory 142) included in system 100, which may be non-transient. One or more computer-readable media may carry instructions and / or a computer program that, when executed, causes at least one processor 137 to perform the operations described herein. The computer program may be stored / distributed on any suitable medium, such as optical storage media or solid-state media provided together with or as part of other hardware, but the computer program may also be distributed in other forms, such as via the Internet or other wired or wireless telecommunications systems. Furthermore, embodiments may take the form of a computer program product accessible from a computer-readable medium that provides program code for use by or in connection with a computer or any device or system executing instructions. For the purposes of this disclosure, a computer-readable medium can generally be any tangible means that can contain, store, deliver, propagate, and / or transmit programs for use by or in connection with an instruction execution device. A computer-readable medium may be, for example, but not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems and / or propagation media. Non-limiting examples of computer-readable media include semiconductor or solid-state memory, magnetic tape, removable computer disk, random access memory (RAM), read-only memory (ROM), hard disk, and / or optical disc. Optical discs may include compact disc read-only memory (CD-ROM), compact disc read / write (CD-R / W), and / or DVD.

[0036] Figure 2 This is a block diagram illustrating a workflow 200 for training a model to evaluate sweep acquisition according to the principles of this disclosure. For example, workflow 200 may train an AI model 210 to evaluate received medical imaging data based on a distribution and / or representation generated from reference medical imaging data 205 compared to reference medical imaging data 205. Based on this evaluation, the trained AI model 210 can be used to generate recommendations, as described herein. It can be used... Figure 1 The system 100 is used to execute workflow 200.

[0037] Workflow 200 begins upon receipt of reference medical imaging data 205. Reference medical imaging data 205 may include ultrasound sequences and / or associated data related to scans performed by trained and / or experienced users of the ultrasound system (e.g., expert users). In some embodiments, reference medical imaging data 205 may include any data with characteristics of expert-acquired data, including analog data. In example embodiments, retrospective blind scan data may be received, comprising ultrasound imaging data acquired by a large number of expert users (e.g., 100 or more expert users). Reference medical imaging data 205 may be annotated, tagged, and / or associated with metadata. For example, the received data may include annotations for each scan in the scan set, such as annotations related to the location and / or orientation of one or more scans (e.g., horizontal / vertical, right / left / center, top / bottom center, etc.). Additionally or alternatively, annotations may include gestational age, number of fetuses, demographic data, measurement data, and / or electronic health record (EHR) data.

[0038] In some embodiments, the reference medical imaging data 205 may include supplemental data, such as data acquired using an inertial measurement unit (IMU) and / or using one or more sensors (e.g., accelerometers, magnetometers, gyroscopes, and / or electromagnetic positioning sensors). IMU data can be used, for example, to determine probe position, velocity, acceleration, orientation, location (e.g., starting position, stopping position), orientation, etc. In some embodiments, the reference medical imaging data 205 may include expert and novice datasets. In these and other embodiments, the reference medical imaging data 205 may be labeled and / or categorized based on various characteristics, such as the skill / experience level of the user acquiring the data. In these and other embodiments, the reference medical imaging data 205 may additionally or alternatively include raw and / or corrected datasets, such as medical imaging data including suggested corrections or modifications, annotated fetal pregnancy parameters or other parameters, and / or associated errors. Reference medical imaging data 205 can be used to generate training datasets, such as by determining one or more variable values ​​associated with reference medical imaging data 205 (e.g., contrast, gestational age, probe orientation, probe orientation, probe speed, user skill / experience level, number of fetuses, known medical conditions, fetal measurements, etc.).

[0039] Reference medical imaging data 205 is provided to AI model 210, which may include one or more neural networks. In some embodiments, AI model 210 may be a variational autoencoder (VAE) that can be trained based on the reference medical imaging data to determine the characteristics of the received medical imaging data. In some embodiments, AI model 210 includes encoder 220, decoder 225, and one or more modules 230 that can perform the operations further described herein. For example, AI model 210 may be trained to identify characteristics of the reference medical imaging data. For example, AI model 210 may be trained to identify more useful data for retention and / or analysis, and / or to identify less useful data to be discarded, such as data at the start and end of a sweep (e.g., blank frames, duplicate frames, etc.). Other characteristics that may be identified include cavitation, velocity (e.g., uniform velocity), appropriate start and stop points, orientation (e.g., correct or incorrect orientation), etc. Additionally or alternatively, the AI ​​model 210 can be trained to derive information from ultrasound images via optical flow or similar image-based motion tracking methods, or by using information input to it as the output of another signal processing controller that utilizes additional data streams (e.g., inertial measurement unit data).

[0040] Although this paper describes an example of a VAE, the AI ​​model 210 can include any construct trained using training data to predict or provide probabilities for new data items, regardless of whether the new data item is included in the training data. For example, training data for supervised learning can include items with various parameters and assignments for classification. New data items can have parameters that the model can use to assign classifications to them. As another example, the model can be a probability distribution generated by analysis of the training data, such as the probability of an n-gram appearing in a given language based on analysis of a large corpus from that language. Examples of models and / or related techniques include, but are not limited to: neural networks, support vector machines, decision trees, Parzen windows, Bayesian methods, clustering, reinforcement learning, probability distributions, decision trees, decision tree forests, etc. The model can be configured for various situations, data types, source and output formats.

[0041] In some implementations, the model trained using workflow 200 may include a neural network with multiple input nodes that receive a training dataset. The input nodes may correspond to functions that receive input and produce results. These results may be fed to one or more intermediate nodes at one or more levels, each intermediate node producing a further result based on a combination of results from lower-level nodes. Weighting factors may be applied to the output of each node before passing the results to the next layer of nodes. At the final layer (“output layer”), one or more nodes may produce values ​​that classify the input, which, once the model has been trained, can be used to evaluate one or more properties of the received medical imaging data, such as the coordinates of the medical imaging data in the latent space. In some implementations, such a neural network, referred to as a deep neural network, may have multiple layers of intermediate nodes with different configurations, which may be combinations of models that receive input from other parts of the deep neural network and / or different parts of the input, or convolutions that partially use the output from previous iterations of the applied model as further input to produce results for the current input.

[0042] Supervised learning can be used to train the model. Test data can then be fed to the model to evaluate its accuracy. The test data can be, for example, a portion of the entire dataset (e.g., 10%), which is retained for evaluating the model. The output from the model can be compared with the expected and / or anticipated output for the training data, and based on this comparison, the model can be modified, such as by changing the weights between nodes in the neural network and / or the parameters of the function used at each node in the neural network (e.g., applying a loss function). Based on the results of the model evaluation, and after applying the described modifications, the model can then be retrained to evaluate new data.

[0043] AI model 210 can be trained to generate one or more distributions of medical imaging data, which may include a Gaussian distribution and / or representation of the medical imaging data in a latent space (z) (e.g., a three-dimensional (3D) distribution) and / or other mappings (e.g., the distribution of data in one or more dimensional spaces). For example, AI model 210 may be trained to generate distributions and / or representations of medical imaging data based on various characteristics of the medical imaging data, such as one or more values ​​of the medical imaging data compared to the mean (m), median, mode, standard deviation (σ), and / or other values ​​associated with a medical imaging dataset (e.g., an expert medical imaging dataset). These and other characteristics can be determined using module 230 and based on one or more outputs of reference medical imaging data 205 and / or encoder 220. To generate distributions and / or representations of medical imaging data, AI model 210 may determine and / or identify (e.g., using module 230) one or more values ​​associated with the medical imaging data, such as probe velocity, orientation and / or orientation, image contrast and / or other image characteristics or parameters. AI models 210 can be trained to represent datasets (e.g., sets of images collected from individuals and / or associated data) as distributions (e.g., Figure 3 Points within 300 (of the dataset). AI model 210 may additionally or alternatively generate reconstructed image 215 based on a representation of the dataset, for example, to compare that representation with the overall distribution of reference medical imaging data. Reconstructed image 215 may be provided by and / or determined using one or more outputs of decoder 225. In some examples, AI model 210 generates reconstructed image 215 and determines coordinates in the latent space / mapping / distribution for the corresponding dataset based on the characteristics of reconstructed image 215.

[0044] In some implementations, encoder 220 and decoder 225 include convolutional neural networks with 3D input (a stack of 2D sequences) or recurrent neural networks with unidirectional or bidirectional long short-term memory (LSTM) architectures. Such networks can be trained to learn a distribution (μ, σ) over a latent space (z) representing frames acquired by an expert (e.g., reference medical imaging data 205). For example, AI model 210 can be trained to generate a set of coordinates within the latent space (z) (e.g., using module 230) to represent a dataset (e.g., a sweep acquisition or a sweep acquisition set).

[0045] In some implementations, the AI ​​model 210 can be trained to output its confidence level in a representation of the input image. Because the autoencoder network reconstructs the input image (e.g., by generating a reconstructed image 215 based on reference medical imaging data 205), the representation of the input image can be based on the quality of the reconstruction (e.g., by using a loss computed between the input and output images). Assigning quality or uncertainty values. Additionally or alternatively, a second loss function of the VAE (e.g., Kolb-Leibler (KL) divergence) can be used to determine uncertainty. That is, the average KL divergence between the representation of the input sequence and the currently trained encoding can indicate whether the input sequence is well represented by the learned encoding (small divergence) or whether it is an outlier (large divergence). These uncertainty or divergence values ​​can be used to estimate confidence for each input. Furthermore, confidence can also be calculated by observing the attention map generated by the network. For example, if the network's attention is on shadows, blank frames, and other features representing novice user acquisition, the network should have higher confidence in its output. However, if the attention map highlights organs or other relevant structures, the network should have lower confidence.

[0046] Workflow 200 can be repeated any number of times. For example, workflow 200 can be repeated to retrain AI model 210 and / or evaluate AI model 210 after training. In some implementations, a portion (e.g., 10%) of the reference medical imaging data 205 can be excluded as test data, and the test data can be used to generate a test dataset. When training using workflow 200, the test dataset can be used to evaluate AI model 210 to determine the accuracy of the trained AI model 210, and AI model 210 can be retrained when the accuracy does not exceed a threshold accuracy (e.g., 70%, 80%, 90%, etc.). Retraining AI model 210 may include repeating workflow 200 using the same reference medical imaging data 205 and / or using an expanded dataset, and / or adjusting one or more weights associated with AI model 210.

[0047] Figure 3 This is a diagram illustrating a mapping 300 of swept acquisitions in a latent space according to the principles of this disclosure. Mapping 300 may be a latent distribution and / or representation of swept acquisitions in the latent space. Mapping 300 may be generated, for example, according to workflow 200, such as using a trained AI model 210 and / or using... Figure 1 System 100. For example, points within map 300 can be located based on a reconstructed image 215 generated using AI model 210. Map 300 includes a set of points located in a latent space, each point being located relative to the remaining points based on a distribution generated according to workflow 200, wherein each point can represent a dataset (e.g., such as...). Figure 2Discrete sweep acquisitions within one or more reconstructed images (215). Although a 2D latent space is illustrated, any dimensional space, such as a 3D / multidimensional latent space and / or multiple latent spaces, can be used. Map 300 can be used to represent and / or determine the distance between a particular sweep acquisition (represented as a point) and one or more expert sweeps (such as the nearest expert sweep and / or the average expert sweep within map 300). This distance can be used to determine the difference between the particular sweep acquisition and one or more expert sweeps, which can indicate one or more relative characteristics of the particular sweep acquisition, such as whether the particular sweep acquisition has too much or too little contrast, whether the sweep direction is correct or incorrect, whether the sweep acquisition has sufficient quality (e.g., compared to a threshold quality), whether the sweep acquisition includes unnecessary frames, etc.

[0048] The center 305 of map 300 represents a sweep acquisition with the characteristics of expert sweep acquisition (e.g., the average of one or more characteristics of a sweep acquisition based on a distribution relative to expert sweep acquisition). The dimensions within the latent space of map 300 can be associated with various characteristics of the acquisition, such as the amount of contrast, quality or confidence measures, probe orientation / orientation / velocity, etc. Therefore, the distance between the center 305 and a point within the latent space corresponds to the difference between the point (e.g., representing a discrete sweep acquisition) and the average expert acquisition.

[0049] The upper left portion 310 of mapping 300 represents a sweep acquisition with low overall quality and / or confidence, such that points within the upper left portion 310 should be removed and / or the corresponding sweep acquisition should be repeated.

[0050] The upper right portion 315 of map 300 represents a sweep acquisition with an incorrect sweep direction. That is, compared to the average expert sweep represented by the center 305 of map 300, a point can be located in the upper right portion 315 to indicate the difference in the direction of movement of the ultrasound probe during the corresponding sweep.

[0051] The lower right portion 320 of map 300 represents a sweep acquisition with insufficient contrast compared to the center 305.

[0052] The lower left portion 325 of mapping 300 represents a sweep acquisition with unnecessary frames (e.g., the first 10 frames acquired) compared to the center 305.

[0053] It should be understood that mapping 300 can represent any number of characteristics acquired by sweep, and the examples provided are merely illustrative. For example, mapping 300 can be used to indicate probe orientation, probe speed, probe direction, image quality, confidence metrics, redundant or unnecessary data (e.g., image frames), insufficient contact with the probe, etc.

[0054] Figure 4 The illustration shows a trained model (e.g., based on the principles of this disclosure) for application according to the principles of this disclosure. Figure 2 The AI ​​model 210 is used to evaluate the flowchart of the sweep acquisition workflow 400. Workflow 400 can be used... Figure 1 System 100, using according to Figure 2 The workflow 200 trains the AI ​​model 210 and / or uses Figure 3 The workflow 400 is executed using mapping 300. For example, workflow 400 can use AI model 210 to determine the location of a non-expert sweep within mapping 300 and / or compare the non-expert sweep with one or more of the closest expert sweeps within mapping 300.

[0055] Workflow 400 begins upon receipt of medical imaging data 405. Medical imaging data 405 may include sweep acquisition performed by a non-expert user (such as an untrained or novice user of an ultrasound imaging system). Medical imaging data 405 may include ultrasound images (e.g., a sequence of images corresponding to the sweep) and / or associated data, such as sensor data (e.g., captured using accelerometers, magnetometers, gyroscopes, and / or electromagnetic positioning sensors).

[0056] According to workflow 400, medical imaging data 405 is provided to a trained AI model 210 and / or one or more other models provided by the disclosed system. The trained AI model 210 can process the medical imaging data 405 to determine various characteristics of the medical imaging data 405, such as the characteristics of individual sweep acquisitions within the medical imaging data 405. For example, the AI ​​model 210 can generate reconstructed medical imaging data 410 based on the medical imaging data 405, and / or the AI ​​model can determine the coordinates 415 of the medical imaging data 405 and / or reconstructed medical imaging data 410 within one or more distributions 420 and / or one or more latent spaces as described herein. For example, distribution 420 may include and / or be based on, as in reference... Figure 3 Mapping 300 is shown, and coordinate 415 can indicate the characteristics of medical imaging data 405 relative to the sweep represented in mapping 300. Coordinate 415 can include, or be based on, a representation of medical imaging data 405 and / or reconstructed medical imaging data 410 in the latent space. Distribution 420 can include the latent distribution of a reference ultrasound imaging dataset represented in the latent space.

[0057] In some implementations, if acquired by an expert user, AI model 210 generates reconstructed medical imaging data 410 as the closest approximation of the image sequence in medical imaging data 405. According to some embodiments, AI model 210 is trained based on data with the characteristics of expert-collected imaging data, and AI model 210 may not adequately reconstruct all frames generated by a novice user. In some implementations, the closest approximation of the image sequence will be based on coordinates 415 within distribution 420 being classified as statistical outliers (e.g., causing high reconstruction errors and / or large KL divergence), and the approximate location in the latent space (e.g., as illustrated by mapping 300 and / or distribution 420) can be used to suggest actions to improve the underlying data.

[0058] Based on coordinates 415, various characteristics of medical imaging data 405 can be determined by comparing it with expert data (e.g., data from trained and / or experienced users) present in distribution 420. For example, the nearest expert data point within distribution 420 can be determined, and the difference between the location of medical imaging data 405 and / or reconstructed medical imaging data 410 and the nearest expert data point can be determined. Based on the difference between these two points (e.g., the magnitude and direction of the difference), one or more recommendations can be generated. For example, it can be determined that medical imaging data 405 is substantially different from the nearest expert data due to the direction of probe movement, and recommendations can be generated to be used when the corresponding sweep is repeated (e.g., when medical imaging data 405 is positioned at...). Figure 3 (When mapping 300 in the upper right part 315) reverses the direction of probe movement.

[0059] In an implementation where distribution 420 includes mapping 300, distribution 420 may include various dimensions, each representing different characteristics of the sweep acquisition, and workflow 400 may be executed accordingly to generate various recommendations based on various differences between medical imaging data 405 and / or reconstructed medical imaging data 410 and corresponding nearest-expert data. Recommendations may be provided to a user via a user interface such as user interface 125. For example, recommendations may be provided as text on a display such as display 138. In another example, recommendations may be graphical icons (e.g., a thumbs-up sign for a good or expert scan) or color-coded (e.g., a shaded image that should be removed in red).

[0060] Workflow 400 can be executed any number of times, for example, to evaluate different sets of medical imaging data 405, such as individual datasets representing portions of discrete sweep acquisitions and / or sweep acquisitions.

[0061] Figure 5 This is a flowchart illustrating a process 500 for evaluating ultrasound imaging data according to the principles of this disclosure. For example, it can be used... Figure 1 System 100 executes process 500. Alternatively or additionally, process 500 may include... Figure 2 Workflow 200 and / or Figure 4 At least a portion of workflow 400, and / or process 500 may be applied. Figure 3 The mapping 300 is used to determine one or more recommendations and / or to characterize or evaluate medical imaging data in other ways.

[0062] At box 510, ultrasound imaging data is received. In some embodiments, the ultrasound imaging data may correspond to a sweep acquisition set or other imaging data acquired according to a protocol. In some embodiments, ultrasound imaging data is acquired using an ultrasound probe. The ultrasound imaging data may be received in real time (e.g., while it is being acquired during the examination of the object) or after acquisition. The ultrasound imaging data received at box 510 may be data acquired by a non-expert user of the ultrasound imaging system (such as a novice user, an untrained user, or a minimally trained user). The ultrasound imaging data may include images (e.g., image frames), signals (e.g., echo signals), and / or supplementary data associated with the acquisition of the ultrasound images (e.g., setup information or other parameters).

[0063] In some embodiments, the ultrasound imaging data received at block 510 may include sensor data captured via one or more of an accelerometer, magnetometer, gyroscope, electromagnetic positioning sensor, or a combination thereof. The aforementioned sensors and / or other sensors may be included in the ultrasound imaging system, within the ultrasound probe, and / or in separate components or modules. In some embodiments, an inertial measurement unit (IMU) may be used to capture the sensor data.

[0064] In some embodiments, the ultrasound imaging data received at block 510 may include a sequence of frames corresponding to one or more characteristics of the sweep acquisition and / or sweep, such as sweep number or other identifiers (e.g., to identify sweeps within a predetermined sequence or series of sweeps) and / or the intended location of the sweep (e.g., relative to the anatomy of the object). Additionally or alternatively, the ultrasound imaging data may also include approximate gestational age / classification period, EHR data, and / or any other information that may affect the results of one or more calculations to be generated using the data.

[0065] At box 520, a representation of the received ultrasound imaging data is generated in the latent space. For example, a VAE and / or other models (e.g., Figure 2 (210) can process the received ultrasound imaging data to generate a representation of the ultrasound imaging data with respect to the characteristics of the expert data.

[0066] At box 530, the representation of the received ultrasound imaging data generated at box 520 is compared with reference ultrasound imaging data. For example, this representation can be compared with the latent distribution of the reference ultrasound imaging data in the latent space. The comparison can identify the closest subset of the reference ultrasound imaging dataset, which can be a specific sweep acquisition by an expert user that is most similar to a sweep acquisition by a non-expert user. The reference ultrasound imaging data can include sweeps acquired by experts, data from other expert acquisitions, and / or other data with characteristics of expert acquisition data, which can include simulated data.

[0067] The comparison performed at box 530 can be performed using an AI model (e.g., AI model 210) (such as a neural network, machine learning model, and / or VAE). In some implementations, process 500 may include training the AI ​​model using reference ultrasound imaging data, such as based on... Figure 2 The workflow 200 trains the model. For example, process 500 may include generating and / or accessing a training dataset using a reference ultrasound imaging dataset, and training a neural network using the training dataset.

[0068] In some implementations, the comparison performed at block 530 may include evaluating the representation of the received ultrasound imaging data using a mapping of a reference ultrasound imaging dataset in the latent space. For example, the comparison may include accessing or generating... Figure 3 A mapping 300 (which may be a mapping of reference ultrasound imaging data in a latent space) is used, and the coordinates of the received ultrasound imaging data within mapping 300 are determined (e.g., represented as points within mapping 300). The locations of the received ultrasound imaging data can then be compared with one or more additional points (such as points representing the closest subset of the reference ultrasound imaging data). For example, mapping 300 can be used to determine the non-expert acquisition in the reference ultrasound imaging data that is closest to the expert acquisition in the received ultrasound imaging data, and the difference between these two points within the mapping can be used to make a recommendation.

[0069] At box 540, one or more recommendations are generated based on the comparison performed at box 530. The one or more recommendations may be based on the difference between the received ultrasound imaging data and the identified closest subset of a reference ultrasound imaging dataset. The difference between the received ultrasound imaging data and the identified closest subset of the reference ultrasound imaging dataset may be based at least in part on reconstruction error, Korbeck-Leibler divergence, or both. For example, reconstruction error and / or Korbeck-Leibler divergence may indicate how “far” the input image sequence is from the closest expert sweep. Recommendations may include recommendations to improve the quality of the received ultrasound imaging data, recommendations to repeat at least a portion of the sweep acquisition set, or combinations thereof. For example, recommendations may be adjusting one or more settings or parameters (e.g., to improve contrast, resolution, or scaling), discarding one or more frames of ultrasound imaging data, repeating at least a portion of the sweep acquisition, changing the orientation, position, and / or orientation of the ultrasound probe, etc. In some embodiments, recommendations may be applied automatically (e.g., to automatically remove unnecessary frames from the received ultrasound imaging data). In these and other embodiments, it may be determined on a feature-by-feature basis whether recommendations can be applied automatically or whether additional user input is required. In some implementations, recommendations may be provided to the user (e.g., via a graphical user interface).

[0070] Process 500 may also include receiving updated ultrasound imaging data based on one or more recommendations generated at box 540, and evaluating the updated ultrasound imaging data (e.g., by repeating at least a portion of process 500). For example, the recommendation may be to reverse the probe orientation during sweep acquisition, and the updated ultrasound imaging data may correspond to a repetition of sweep acquisition during which the user reverses the probe orientation in response to the recommendation.

[0071] In some implementations, process 500 may include evaluating one or more characteristics of an anatomical structure based on received and / or updated ultrasound imaging data, such as using AI technology or other techniques. For example, one or more fetal health characteristics may be determined. In some implementations, artificial intelligence techniques may be used to automatically extract pregnancy estimates (e.g., age, fetal viability, multiple pregnancies, etc.) from the received and / or updated ultrasound imaging data. For example, pregnancy estimates and / or other operations may be performed after (e.g., automatically and / or by a user) one or more recommendations generated at box 540.

[0072] Process 500 can be executed in any order, including performing one or more operations in parallel and / or repeating one or more operations. Additionally, operations can be added to or removed from process 500 without departing from the teachings of this disclosure.

[0073] Advantageously, the systems and related methods disclosed herein can evaluate medical imaging data associated with a sweep acquisition (e.g., performed by a novice user) by comparing it with reference medical imaging data (e.g., sweep acquisition performed by an expert user), and generate various recommendations based on this comparison, such as recommendations for improving the quality of sweep acquisitions. That is, by evaluating the difference between a sweep acquisition and the closest possible expert sweep, the disclosed techniques can be used by minimally trained and / or inexperienced users to improve the quality of sweep acquisitions, enabling novice users to evaluate objects more accurately and effectively. The disclosed embodiments can be used, for example, in resource-constrained settings and / or when trained and / or experienced users (e.g., expert users) are unavailable or substantially unavailable. Using the disclosed techniques, novice users can be instructed to more accurately simulate the actions of expert users. Although examples related to evaluating sweeps performed by novice users are described herein, the disclosed techniques can be additionally or alternatively applied to guiding sweep quality control and / or sweeps acquired by trained and / or experienced users of ultrasound imaging systems.

[0074] In various examples of implementing components, systems, and / or methods using computer-based systems or programmable devices with programmable logic, it should be appreciated that the aforementioned systems and methods can be implemented using various known or later-developed programming languages ​​such as "C", "C++", "FORTRAN", "Pascal", "VHDL", etc. Accordingly, various storage media, such as magnetic computer disks, optical disks, electronic storage devices, etc., can be prepared, which can contain information that can boot devices such as computers to implement the aforementioned systems and / or methods. Once a suitable device has access to the information and programs contained on the storage medium, the storage medium can provide the information and programs to the device, thereby enabling the device to perform the functions of the systems and / or methods described herein. For example, if a computer disk containing appropriate materials (such as source files, object files, executable files, etc.) is provided to a computer, the computer can receive this information, appropriately configure itself, and perform the functions of the various systems and methods outlined in the diagrams and flowcharts above to implement various functions. That is, a computer can receive various parts of information relating to different elements of the aforementioned systems and / or methods from the disk, implement individual systems and / or methods, and coordinate the functions of the individual systems and / or methods described above.

[0075] In view of this disclosure, it should be noted that the various methods and apparatuses described herein can be implemented in hardware, software, and / or firmware. Furthermore, the various methods and parameters are included by way of example only and not in any limiting sense. In view of this disclosure, those skilled in the art can implement the teachings to determine their own techniques and the apparatus required to implement these techniques, while remaining within the scope of the invention. The functionality of one or more of the processors described herein can be incorporated into a smaller number or a single processing unit (e.g., a CPU) and can be implemented using application-specific integrated circuits (ASICs) or general-purpose processing circuitry programmed in response to executable instructions that perform the functions described herein.

[0076] Although this system may have been described with particular reference to ultrasound imaging systems, it is also contemplated that this system can be extended to other medical imaging systems in which one or more images are acquired systematically. Therefore, this system can be used to acquire and / or record image information relating to, but not limited to, the kidneys, testes, breasts, ovaries, uterus, thyroid gland, liver, lungs, musculoskeletal system, spleen, heart, arteries, and vascular system, as well as other imaging applications related to ultrasound-guided interventions. Furthermore, this system may include one or more procedures that can be used with conventional imaging systems, such that they can provide the features and advantages of this system. Certain additional advantages and features of this disclosure will be apparent to those skilled in the art upon studying this disclosure, or may be experienced by those employing the novel systems and methods of this disclosure. Another advantage of this system and method is that conventional medical imaging systems can be easily upgraded to incorporate the features and advantages of this system, device, and method.

[0077] Of course, it should be recognized that any of the examples, paradigms, or processes described herein may be combined with one or more other examples, paradigms, and / or processes, or may be separated and / or performed within a separate device or device portion according to this system, device, and method.

[0078] Finally, the foregoing discussion is intended merely to illustrate the system and method and should not be construed as limiting the claims to any particular example or set of examples. Therefore, although the system has been described in particular detail with reference to exemplary examples, it should be appreciated that many modifications and alternative examples can be devised by those skilled in the art without departing from the broader and contemplated spirit and scope of the system and method as set forth in the following claims. Thus, the specification and drawings are to be viewed in an illustrative manner and not intended to limit the scope of the claims.

Claims

1. An ultrasound imaging system, comprising: An ultrasound probe, configured to acquire ultrasound imaging data; and At least one processor, which communicates with the ultrasound probe, is configured to: Receive the ultrasound imaging data, wherein the ultrasound imaging data corresponds to a sweep acquisition set performed using the ultrasound probe; Generate a representation of the received ultrasound imaging data in the latent space; The representation of the received ultrasound imaging data in the latent space is compared with the latent distribution of a reference ultrasound imaging dataset represented in the latent space to identify the closest subset of the reference ultrasound imaging dataset; and At least one recommendation is generated based on the difference between the representation of the received ultrasound imaging data in the latent space and the identified closest subset of the reference ultrasound imaging dataset, wherein the at least one recommendation includes a recommendation to improve the quality of the received ultrasound imaging data, a recommendation to repeat at least a portion of the sweep acquisition set, or both.

2. The ultrasound imaging system according to claim 1, wherein, The received ultrasound imaging data representation in the latent space is compared with the latent distribution of the reference ultrasound imaging dataset represented in the latent space using a neural network trained on the reference ultrasound imaging dataset.

3. The ultrasound imaging system according to claim 2, wherein, The at least one processor is further configured to: Access the training dataset that includes the reference ultrasound imaging dataset; and The neural network is trained using the training dataset.

4. The ultrasound imaging system according to claim 1, wherein, The at least one processor is further configured to: The updated ultrasound imaging data are received based at least in part on the at least one recommendation; and Evaluate at least one characteristic of the anatomical structures in the updated ultrasound imaging data.

5. The ultrasound imaging system according to claim 1, wherein, The received ultrasonic imaging data includes sensor data captured via at least one of the following: accelerometer, magnetometer, gyroscope, or electromagnetic positioning sensor.

6. The ultrasound imaging system according to claim 1, wherein, The at least one processor is further configured to: This enables the display of the at least one recommendation via a graphical user interface.

7. The ultrasound imaging system according to claim 1, wherein, The difference between the representation of the received ultrasound imaging data in the latent space and the identified closest subset of the reference ultrasound imaging dataset is based at least in part on reconstruction error, Kohlbek-Leibler divergence, or both.

8. The ultrasound imaging system according to claim 1, wherein, The at least one processor is further configured to: The at least one recommendation is automatically applied to the received ultrasound imaging data.

9. A computer-implemented method for evaluating ultrasonic scanning acquisition, the method comprising: An ultrasonic probe is used to acquire ultrasonic imaging data, wherein the ultrasonic imaging data corresponds to a sweep acquisition set; A processor is used to generate a representation of the acquired ultrasound imaging data in the latent space; The processor is used to compare the representation of the acquired ultrasound imaging data in the latent space with the latent distribution of a reference ultrasound imaging dataset represented in the latent space to identify the closest subset of the reference ultrasound imaging dataset; and The processor is used to generate at least one recommendation based on the difference between the representation of the acquired ultrasound imaging data in the latent space and the identified closest subset of the reference ultrasound imaging dataset, wherein the at least one recommendation includes a recommendation to improve the quality of the acquired ultrasound imaging data, a recommendation to repeat at least a portion of the sweep acquisition set, or both.

10. The computer-implemented method according to claim 9, wherein, The neural network is used to compare the representation of the acquired ultrasound imaging data in the latent space with the latent distribution of the reference ultrasound imaging dataset represented in the latent space, the neural network being trained using the reference ultrasound imaging dataset.

11. The computer-implemented method according to claim 10, further comprising: The reference ultrasound imaging dataset is used to generate the training dataset; and The neural network is trained using the generated training dataset.

12. The computer-implemented method according to claim 9, wherein, The reference ultrasound imaging dataset includes ultrasound images acquired by expert users or imaging data with characteristics of expert-acquired images.

13. The computer-implemented method according to claim 9, further comprising: Updated ultrasound imaging data are acquired based at least in part on the at least one of the recommendations; and Evaluate at least one characteristic of the anatomical structures in the updated ultrasound imaging data.

14. The computer-implemented method according to claim 13, wherein, The anatomical structure includes the fetus, and the at least one characteristic includes fetal health characteristics.

15. The computer-implemented method according to claim 9, wherein, The acquired ultrasound imaging data includes sensor data captured via at least one of the following: accelerometer, magnetometer, gyroscope, or electromagnetic positioning sensor.

16. The computer-implemented method according to claim 9, further comprising: The at least one recommendation is displayed via a graphical user interface.

17. The computer-implemented method according to claim 9, wherein, The sweep acquisition set is associated with an ultrasound imaging protocol.

18. The computer-implemented method according to claim 9, wherein, The difference between the representation of the acquired ultrasound imaging data in the latent space and the identified closest subset of the reference ultrasound imaging dataset is based at least in part on reconstruction error, Kohlbek-Leibler divergence, or both.

19. The computer-implemented method according to claim 9, further comprising: The at least one recommendation is automatically applied to the acquired ultrasound imaging data.

20. A non-transitory computer-readable medium carrying instructions that, when executed by a processor, cause the processor to perform operations, the operations including: Receive ultrasound imaging data, wherein the ultrasound imaging data corresponds to a sweep acquisition set performed using an ultrasound probe; Generate a representation of the received ultrasound imaging data in the latent space; The representation of the received ultrasound imaging data in the latent space is compared with the latent distribution of a reference ultrasound imaging dataset represented in the latent space to identify the closest subset of the reference ultrasound imaging dataset; and At least one recommendation is generated based on the difference between the representation of the received ultrasound imaging data in the latent space and the identified closest subset of the reference ultrasound imaging dataset, wherein the at least one recommendation includes a recommendation to improve the quality of the received ultrasound imaging data, a recommendation to repeat at least a portion of the sweep acquisition set, or both.

Images