Ultrasound examination automatic measurement assistance method, system and apparatus based on video analysis

Abstract

The application is suitable for the technical field of ultrasonic detection, and provides an ultrasonic examination automatic measurement auxiliary method, system and device based on video analysis, which is applied to an auxiliary device connected with an ultrasonic imaging device and an ultrasonic report workstation, and the method comprises the following steps: converting an ultrasonic video signal into a video frame sequence in real time; identifying a key measurement image by using a deep learning model; if the key image is determined, positioning an anatomical region and automatically calculating a measurement value and a mark; superimposing the measurement value and the mark to a current frame to generate a processed image frame; transparently transmitting the image frame to the workstation through a simulation acquisition card interface, and synchronously sending a hardware trigger signal to control the workstation to automatically acquire and store. The application solves the problems that in the existing ultrasonic examination process, a doctor manually measures and the operation burden is heavy, oral communication and manual transcription have a high error rate, and the existing automatic solution has a high transformation cost and poor system compatibility when connecting with the ultrasonic report workstation.

Description

Technical Field

[0001] This invention relates to the field of ultrasonic testing technology, and in particular to an automatic measurement auxiliary method, system and device for ultrasonic examination based on video analysis. Background Technology

[0002] Ultrasound examination is one of the most widely used imaging methods in clinical practice. In a typical ultrasound examination workflow, the sonographer operates the ultrasound imaging equipment to scan the patient. When a key image that needs to be recorded is observed (such as the measurement of the diameter, area, or blood flow parameters of a certain anatomical structure), the doctor pauses the scan and freezes the image. Then, the doctor manually operates the trackball or touch panel on the ultrasound equipment to place measuring calipers or markers on the image, and the ultrasound equipment calculates the measurement values. Subsequently, the doctor verbally informs the data entry clerk of the measurement items and values, and the clerk manually enters the corresponding fields into the ultrasound report workstation, ultimately generating a structured report.

[0003] The aforementioned mainstream business processes exhibit the following prominent technical and operational issues: 1. Heavy workload and low scanning efficiency for doctors: During the scanning process, doctors not only need to focus on observing and judging anatomical structures, but also need to frequently interrupt the scanning to manually perform mechanical operations such as placing calipers and selecting measurement items, which greatly distracts doctors' attention and prolongs the time of a single examination.

[0004] 2. Risk of human error in measurement data entry: In the chain of "doctor's dictation → data entry operator listens → manual entry", due to the busy and noisy working environment, errors such as mishearing, misremembering, misattribution, or omission are very likely to occur, which seriously affect the quality of ultrasound reports and medical safety.

[0005] 3. Existing automation and intelligent upgrade solutions are costly and have poor compatibility: Currently, most external AI-assisted devices on the market focus on lesion-assisted diagnosis (benign / malignant analysis) or scan quality control, rather than replacing doctors in routine diameter measurements. Existing automatic data transmission solutions mostly rely on DICOM or other digital interfaces (APIs) to directly establish a data channel between the ultrasound equipment and the reporting workstation. However, different hospitals use different brands and system versions of ultrasound reporting workstations, making the development and adaptation of customized interfaces for each hospital extremely costly and prone to failure due to workstation upgrades. Meanwhile, some ultrasound equipment manufacturers have introduced built-in automatic measurement algorithms, but this requires hospitals to spend exorbitant costs to purchase new machines, which is completely ineffective for the tens of thousands of existing ultrasound devices (covering various brands and older models) already deployed in hospitals.

[0006] In summary, there is an urgent need in this field for an ultrasound examination assistance solution that can achieve plug-and-play functionality, cross-brand compatibility, and automatic identification, measurement, and transparent, seamless data transmission to the workstation without modifying existing ultrasound imaging equipment or the underlying software of existing ultrasound reporting workstations. This would reduce the burden of manual measurement for doctors, lower the risk of errors from manual data entry, and improve the overall efficiency of ultrasound examinations and report generation. Summary of the Invention

[0007] Based on this, the purpose of this invention is to provide an automated measurement assistance method, system, and device for ultrasound examination based on video analysis, so as to fundamentally solve the problems of heavy manual measurement operation burden for doctors, high error rate of verbal communication and manual transcription in the existing ultrasound examination process, and high cost and poor system compatibility of existing automation solutions in connecting to ultrasound report workstations.

[0008] An automated measurement assistance method for ultrasound examination based on video analysis, according to an embodiment of the present invention, is applied to an automated measurement assistance device for ultrasound examination connected to an ultrasound imaging device and an ultrasound reporting workstation, respectively. The method includes: It receives video signals from ultrasound imaging equipment in real time and converts the video signals into a digital video frame sequence; A pre-trained deep learning model is used to identify key measurement images in the digital video frame sequence to determine whether the current frame is a target key measurement image. If the current frame is determined to be a key measurement image of the target, the measurement area of ​​the target anatomical structure is located in the current frame, and automatic measurement calculation is performed based on the visual features in the current frame to obtain the target measurement value and the corresponding measurement mark. The target measurement value and measurement mark are superimposed and drawn onto the current frame to generate a processed image frame carrying measurement information; The processed image frame is transparently transmitted to the ultrasound reporting workstation through the analog video capture card interface, and a hardware acquisition trigger signal is simultaneously sent to the ultrasound reporting workstation through the analog acquisition control interface to control the ultrasound reporting workstation to automatically acquire and retain the currently transmitted processed image frame.

[0009] In addition, the automatic measurement assistance method for ultrasound examination based on video analysis according to the above embodiments of the present invention may also have the following additional technical features: Furthermore, the step of using a pre-trained deep learning model to perform key measurement image recognition on the digital video frame sequence and determining whether the current frame is a target key measurement image includes: The digital video frame sequence is subjected to state feature detection to determine whether the current frame contains the original frozen state identifier of the ultrasound imaging device and / or whether it contains preset original measurement interface elements of the device. If it is determined that the current frame contains the original frozen state identifier of the ultrasound imaging device or contains preset original measurement interface elements of the device, then the current frame is marked as a candidate measurement frame. Input the current frame, which is marked as a candidate measurement frame, into the pre-trained deep learning model to calculate the confidence score of the current frame belonging to each standard section. Select the slice category corresponding to the highest confidence score. If the highest confidence score is greater than the preset first classification threshold, confirm the standard slice type corresponding to the current frame. The recognition results of the current frame and its adjacent frames are combined for temporal smoothing. If the number of frames that are continuously identified as having the same standard cross-section type within a preset time window exceeds a preset proportion, then the current frame is determined to be the target key measurement image.

[0010] Further, the step of locating the measurement region of the target anatomical structure in the current frame and automatically measuring and calculating based on the visual features within the current frame to obtain the target measurement value and the corresponding measurement marker includes: The scale graphic of the edge of the ultrasound image in the current frame is extracted using a pre-trained image segmentation model; By calculating the number of pixels between adjacent scale marks in the scale graphic, the mapping scale from the pixels of the current frame to the actual physical size is calculated in reverse. The boundary extreme points of the target anatomical structure in the current frame are identified and located using a pre-trained target detection model or semantic segmentation model. Based on the measurement type formula corresponding to the target anatomical structure, the pixel distance, pixel area, or pixel contour geometric parameters between each boundary extreme point are calculated, and combined with the mapping scale, the target measurement values ​​corresponding to the target anatomical structure are calculated and converted.

[0011] Further, the step of locating the measurement region of the target anatomical structure in the current frame and automatically measuring and calculating based on the visual features within the current frame to obtain the target measurement value and the corresponding measurement marker includes: Query a preset knowledge graph or mapping table to obtain multiple measurement indicators associated with the standard cross-section type of the target key measurement image; The pre-trained target detection model and semantic segmentation model are invoked to synchronously and respectively locate multiple anatomical sub-regions or multiple key morphological landmarks corresponding to the multiple measurement indicators in the current frame. Measurement calculations are performed in parallel for each anatomical sub-region or key morphological landmark to obtain a target set of measurement values ​​containing multiple measurement indicators.

[0012] Further, after the steps of locating the measurement region of the target anatomical structure in the current frame and performing automatic measurement calculations based on visual features within the current frame to obtain the target measurement value and the corresponding measurement marker, the method further includes: The overall confidence score is calculated by weighting the boundary extreme point confidence scores output by the object detection model or semantic segmentation model and the image quality score of the current frame. Determine whether the overall confidence score is greater than a preset security assessment threshold; If the overall confidence score is less than or equal to the preset safety assessment threshold, a processed image frame containing early warning information is generated, and the instruction to generate the hardware acquisition trigger signal is intercepted to block the automatic acquisition and retention action of the ultrasound report workstation. If the overall confidence score is greater than the preset security assessment threshold, then the step of overlaying the target measurement value and measurement mark onto the current frame is executed.

[0013] Further, the step of overlaying the target measurement value and measurement marker onto the current frame to generate a processed image frame carrying measurement information includes: At the location corresponding to the anatomical structure in the current frame, caliper line segments and contour lines are drawn as measurement marks according to preset visual rendering parameters; Identify the edge blank areas of non-ultrasound scan image regions in the current frame, and obtain the coordinate range of the edge blank areas; The target measurement values ​​are plotted in the form of structured text within the coordinate range of the blank area at the edge, and the processed image frame is generated to avoid the target measurement values ​​obscuring the original lesion image of the anatomical structure.

[0014] Further, the step of overlaying the target measurement value and measurement marker onto the current frame includes: The received video signal is analyzed to obtain the original video frame resolution and image aspect ratio parameters output by the ultrasound imaging device; Based on the difference between the original video frame resolution and the resolution calculated by the benchmark model, the corresponding overlay rendering scaling factor and coordinate translation vector are calculated. Based on the overlay rendering scaling factor and coordinate translation vector, the absolute pixel coordinate mapping relationship of the measurement mark in the current frame, the drawing line width of the caliper line segment, and the font size of the structured text of the target measurement value are dynamically adjusted. The rendering was performed using the adjusted mapping relationship and style parameters to ensure that the superimposed measurement marks under the output resolution of different models of ultrasound imaging equipment could be aligned with the edges of anatomical structures without causing visual distortion.

[0015] Furthermore, the step of transparently transmitting the processed image frame to the ultrasound reporting workstation via the analog video acquisition card interface, and simultaneously sending a hardware acquisition trigger signal to the ultrasound reporting workstation via the analog acquisition control interface, includes: The processed image frame carrying measurement information is written into the output frame buffer of the analog video acquisition card module so that a video signal conforming to the standard video protocol is continuously output to the ultrasound reporting workstation. Simultaneously with or within a preset delay, the processed image frame carrying measurement information is written into the output frame buffer, and the hardware acquisition trigger signal is sent to the ultrasound reporting workstation via the analog acquisition control interface. The hardware acquisition trigger signal includes one or more of the following: simulated USB HID keyboard shortcut key commands, TTL level change signals output through specific pins of the VGA interface, serial port COM commands, or Joystick protocol signals.

[0016] Furthermore, prior to the step of performing key measurement image recognition on the digital video frame sequence using a pre-trained deep learning model, the method further includes: Read the pre-configured acquisition area coordinate parameters; Based on the coordinate parameters of the acquisition area, each frame in the digital video frame sequence is cropped in a fixed region to remove the original non-scanning area interface of the ultrasound imaging device, and the pure ultrasound scan image area is extracted as the input image for deep learning model processing.

[0017] Furthermore, prior to the step of performing key measurement image recognition on the digital video frame sequence using a pre-trained deep learning model, the method further includes: Calculate the inter-frame motion vector or optical flow velocity between consecutive digital video frame sequences; When the inter-frame motion vector or optical flow velocity is greater than the preset motion distortion threshold, it is determined that the current image is in a state of rapid probe movement distortion, and the step of using a pre-trained deep learning model to perform key measurement image recognition on the digital video frame sequence is paused until the inter-frame motion vector or optical flow velocity is less than the preset motion distortion threshold.

[0018] Furthermore, while performing the step of transparently transmitting the processed image frame to the ultrasound reporting workstation via the analog video capture card interface, the method also includes: The extracted target measurement values ​​and corresponding examination item identifiers are encapsulated and converted into standard medical structured data format files, including JSON, XML, or DICOM SR formats. The medical structured data format file is sent to the target network address of the ultrasound reporting workstation via a preset network interface. After being parsed by the ultrasound reporting workstation, the target measurement value is directly filled into the corresponding field of the patient report template that the ultrasound reporting workstation is currently in an active receiving state.

[0019] Another embodiment of the present invention aims to provide an automatic measurement auxiliary system for ultrasound examination based on video analysis, applied to an automatic measurement auxiliary device for ultrasound examination connected to an ultrasound imaging device and an ultrasound reporting workstation, respectively, the system comprising: The video acquisition module is used to receive video signals output from the ultrasound imaging device in real time and convert the video signals into a digital video frame sequence; The image recognition module is used to perform key measurement image recognition on the digital video frame sequence using a pre-trained deep learning model, and to determine whether the current frame is a target key measurement image. An automatic measurement module is used to locate the measurement area of ​​the target anatomical structure in the current frame if the image recognition module determines that the current frame is a key measurement image of the target, and to perform automatic measurement calculation based on the visual features in the current frame to obtain the target measurement value and the corresponding measurement mark. The overlay drawing module is used to overlay the target measurement values ​​and measurement marks onto the current frame to generate a processed image frame carrying measurement information. The acquisition trigger module is used to transparently transmit the processed image frame to the ultrasound reporting workstation through the analog video acquisition card interface, and simultaneously send a hardware acquisition trigger signal to the ultrasound reporting workstation through the analog acquisition control interface to control the ultrasound reporting workstation to automatically acquire and retain the currently transmitted processed image frame.

[0020] Another embodiment of the present invention aims to provide an automatic measurement auxiliary device for ultrasound examination, which is connected to an ultrasound imaging device and an ultrasound reporting workstation, respectively. The device includes a memory, a processor, and a program stored in the memory and executable on the processor. When the processor executes the program, it implements the automatic measurement auxiliary method for ultrasound examination based on video analysis as described above.

[0021] The automatic measurement assistance method for ultrasound examination based on video analysis provided in this invention receives video signals output by ultrasound imaging equipment in real time and converts them into a digital video frame sequence. A pre-trained deep learning model is used to identify key measurement images in the digital video frame sequence, achieving fully automatic, continuous, and real-time intelligent perception of key measurement images during ultrasound scanning, eliminating the need for manual pausing or marking of sections by the physician. Furthermore, by automatically locating the measurement area of ​​the target anatomical structure in the current frame identified as the target key measurement image, and automatically performing measurement calculations based on visual features to obtain the target measurement value and corresponding measurement markers, the method achieves fully automatic extraction of morphological and dimensional parameters of the anatomical structure. This completely liberates ultrasound physicians from tedious manual operations such as caliper application and measurement item selection, effectively reducing the physician's workload and significantly improving the overall efficiency of ultrasound scanning. Finally, by directly overlaying the target measurement value and measurement markers onto the current frame to generate a processed image frame carrying measurement information, the method achieves intuitive fusion and visualization of the measurement basis (graphical markers) and measurement results (numerical text) on the original image, ensuring automatic measurement... The system enhances the intuitiveness, traceability, and transparency of measurement results. By transparently transmitting processed image frames to the ultrasound reporting workstation via an analog video capture card interface, it achieves plug-and-play access for various brands and versions of ultrasound reporting workstations without the need for driver installation or software modification. This effectively avoids the high costs and compatibility issues associated with customizing dedicated software interfaces for different workstation systems. Furthermore, by simultaneously sending hardware acquisition trigger signals to the ultrasound reporting workstation via the analog acquisition control interface to control its automatic acquisition and retention while transmitting processed image frames, it achieves time-series synchronization and automatic archiving of key image frames containing measurement information without human intervention. This eliminates human errors such as mishearing, misrecording, and omissions caused by doctors' verbal instructions and manual transcription by clerks in the traditional process. It forms a seamless, automated closed loop from video stream input to intelligent recognition, automatic measurement, layer overlay, and finally hardware-triggered saving. This solves the problems of heavy manual measurement burden for doctors, high error rates in verbal communication and manual transcription during existing ultrasound examinations, and high costs and poor system compatibility of existing automation solutions when integrating with ultrasound reporting workstations. Attached Figure Description

[0022] Figure 1 This is a flowchart illustrating the automatic measurement assistance method for ultrasound examination based on video analysis in the first embodiment of the present invention. Figure 2 This is a schematic diagram of the automatic measurement auxiliary system for ultrasound examination based on video analysis according to the second embodiment of the present invention; Figure 3 This is a schematic diagram of the automatic measurement auxiliary device for ultrasound examination according to the third embodiment of the present invention; The following detailed description of the embodiments will further illustrate the present invention in conjunction with the above-described accompanying drawings. Detailed Implementation

[0023] To facilitate understanding of the present invention, a more complete description will be given below with reference to the accompanying drawings. Several embodiments of the invention are illustrated in the drawings. However, the invention can be implemented in many different forms and is not limited to the embodiments described herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete.

[0024] It should be noted that when a component is said to be "fixed to" another component, it can be directly on the other component or there may be an intervening component. When a component is said to be "connected to" another component, it can be directly connected to the other component or there may be an intervening component. The terms "vertical," "horizontal," "left," "right," and similar expressions used in this document are for illustrative purposes only.

[0025] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. The terminology used herein in the description of the invention is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0026] Example 1 Please see Figure 1 The image shows an automated ultrasound examination measurement assistance method based on video analysis according to the first embodiment of the present invention. For ease of explanation, only the parts relevant to the embodiment of the present invention are shown. The automated ultrasound examination measurement assistance method based on video analysis provided by the embodiment of the present invention is applied to an automated ultrasound examination measurement assistance device connected to an ultrasound imaging device and an ultrasound reporting workstation, respectively. The method includes: Step S10: Receive video signals output from the ultrasound imaging device in real time and convert the video signals into a digital video frame sequence; In one embodiment of the present invention, the method is applied to an automated measurement auxiliary device for ultrasound examination, which is connected to both an ultrasound imaging device and an ultrasound reporting workstation. This automated measurement auxiliary device includes an externally mounted, independently operating image processing host or edge computing box. Specifically, the automated measurement auxiliary device in this embodiment is actually a small, independent intermediate computing device, primarily connected in series between the ultrasound imaging device and the ultrasound reporting workstation to achieve transparent data transmission and intelligent analysis. Furthermore, the hardware architecture of this automated measurement auxiliary device includes, but is not limited to, a central processing unit (CPU), a neural network accelerator, a video acquisition and processing module, an image overlay rendering module, and an analog data output module. The CPU serves as the core control unit connecting all modules. The neural network accelerator runs complex artificial intelligence deep learning models based on existing technologies. The video acquisition and processing module has a high-definition digital video input interface or an analog video input interface compatible with multiple protocols, used to acquire the original image from the ultrasound imaging device in real time. The image overlay rendering module plots the calculated measurement data and markers onto the corresponding image frames. The analog data output module includes a universal serial bus controller and a video output control unit, used to send hardware trigger commands and the processed overlay image to the ultrasound reporting workstation.

[0027] In one embodiment of the present invention, the step of receiving a video signal output from an ultrasound imaging device in real time and converting the video signal into a digital video frame sequence includes: The video acquisition and processing module is physically connected to the video output terminal of the ultrasound imaging equipment to acquire and decode continuous video signals in real time. The decoded video image data is preprocessed and resolution adapted, and the formatted data is organized in chronological order to generate a standard digital video frame sequence.

[0028] Specifically, the central processing unit (CPU) first sends an initialization configuration command to the video acquisition and processing module. Based on the video interface type of the currently connected ultrasound imaging equipment, it automatically identifies and matches basic parameters such as the resolution, color format, and refresh rate of the input video signal, and configures the receiving timing of the video acquisition and processing module to be fully synchronized with the ultrasound imaging equipment. Simultaneously, the CPU allocates and establishes a circular frame buffer in system memory as a unified data inflow buffer pool to ensure stable reception and temporary storage of the ultrasound video stream.

[0029] Next, the video acquisition and processing module continuously receives video signals from the ultrasound imaging equipment. For different types of video signals, the module decodes the continuously input video stream into a standard digital pixel matrix according to pre-matched basic parameters. Specifically, if the current input is an analog video signal, the analog-to-digital converter inside the video acquisition and processing module performs high-frequency sampling and discretization of the continuously changing analog level signal at a set sampling frequency, based on the previously configured synchronization clock signal, converting it into a quantized digital pixel data stream. If the current input is a high-definition digital video signal, the protocol decoding circuit directly strips away the digital audio signal, padding bytes, and hardware synchronization control data packets from the transport layer protocol, accurately extracting the clean and effective video payload data. This process automatically eliminates irrelevant data such as audio and handshake control in the video transmission channel, extracting the original video image containing the complete visible image of the ultrasound screen.

[0030] Furthermore, after extracting the original video images, the video acquisition and processing module preprocesses and formats the extracted images. Specifically, the system converts the original images into a unified color space format suitable for artificial intelligence model calculations. Simultaneously, to eliminate interference from the native system menu bars, patient information areas, or black borders around ultrasound imaging devices from different brands, the central processing unit physically crops a fixed area of ​​the current frame based on pre-read acquisition area coordinate parameters, retaining only the clean ultrasound scan image area containing characteristic echo signals.

[0031] Finally, the central processing unit (CPU) converts and trims the clean ultrasound scan image data and writes it sequentially into the circular frame buffer according to the order of reception. Upon completion of each image write, the CPU appends a system timestamp and an incrementing sequence number to the current image data block, thereby converting the originally continuous external video signal, which is interspersed with interference from various brand interfaces, into a sequence of independent, ordered, and clean digital video frames, which are then pushed to the neural network acceleration processor.

[0032] In one embodiment of the present invention, the step of physically cropping a fixed region of the current frame based on pre-read acquisition area coordinate parameters, retaining only the clean ultrasound scan image region containing characteristic echo signals, includes: Read the pre-configured acquisition area coordinate parameters; Based on the coordinate parameters of the acquisition area, each frame in the digital video frame sequence is cropped to a fixed region to remove the original non-scanning area interface of the ultrasound imaging device, and the pure ultrasound scan image region is extracted as the input image for deep learning model processing.

[0033] Specifically, after system startup or establishing a video connection, the central processing unit (CPU) automatically acquires the parameter information of the currently connected ultrasound imaging device, or determines the model of the target device through the user's initial configuration commands. Addressing the issue of different brands and models of ultrasound devices having different interface layouts, the CPU extracts the corresponding acquisition area coordinate parameters from its internally stored configuration database. These coordinate parameters are typically presented as vertex coordinates or relative proportions on a two-dimensional plane, used to precisely delineate a fan-shaped or rectangular core area on the screen that purely displays tissue echoes, avoiding the surrounding parameter text and control menus.

[0034] Next, as the video acquisition and processing module continuously outputs a sequence of digital video frames, the central processing unit, in conjunction with the module's image processing capabilities, strictly adheres to the read coordinate parameters of the acquisition area and performs a fixed-region cropping operation on each full-screen image frame that comes in real time. This operation, through coordinate mapping, precisely preserves pixel data within the preset boundaries while simultaneously masking or discarding pixel data outside the boundaries. This step completely eliminates the non-scanning interface elements native to the ultrasound imaging device, such as the patient information panel, manufacturer logos, side depth scales, and various system control controls, effectively eliminating visual background interference caused by these complex, variable, and diagnostically meaningless interface elements to subsequent algorithms.

[0035] Furthermore, after successfully extracting the clean ultrasound scan image region, the central processing unit (CPU) needs to further format the clean image region because deep learning models typically have strict requirements on the resolution, aspect ratio, and channel distribution of the input image. Specifically, the system scales the cropped clean image proportionally. If there are differences in length and width after scaling, black pixels are used to fill the edges. At the same time, the pixel grayscale values ​​of the image are normalized, ultimately generating a standard input image that fully conforms to the specifications of the pre-trained deep learning model.

[0036] Therefore, the above embodiments, by reading preset coordinate parameters to crop a fixed area of ​​the input image to eliminate non-scanning interfaces, achieve pre-physical stripping of irrelevant interference elements such as native menus, parameters and icons from various manufacturers, providing a clean image input for the deep learning model, effectively reducing the algorithm's misjudgment rate and improving the model's inference efficiency.

[0037] Step S20: Use a pre-trained deep learning model to identify key measurement images in the digital video frame sequence to determine whether the current frame is the target key measurement image. In one embodiment of the present invention, prior to the step of performing key measurement image recognition on a digital video frame sequence using a pre-trained deep learning model, the method further includes: Calculate the inter-frame motion vector or optical flow velocity between consecutive digital video frame sequences; When the inter-frame motion vector or optical flow velocity exceeds the preset motion distortion threshold, the current image is determined to be in a state of rapid probe movement distortion, and the execution of the step of using a pre-trained deep learning model to perform key measurement image recognition on the digital video frame sequence is suspended until the inter-frame motion vector or optical flow velocity is less than the preset motion distortion threshold.

[0038] Specifically, when the video acquisition and processing module continuously writes a sequence of digital video frames into the system's memory, before sending the images to the deep learning model, the central processing unit (CPU) sequentially extracts the video frame at the current time point and the video frame at the previous adjacent time point from the memory to construct a continuous frame image pair. Subsequently, the CPU uses the image pyramid hierarchical structure and the local feature point optical flow tracing algorithm to calculate the spatial displacement distribution of each feature pixel block within the ultrasonic scanning image region of two adjacent frames, and then statistically calculates and solves for the inter-frame motion vector or global optical flow velocity that characterizes the overall dynamic displacement of the entire image.

[0039] Next, the central processing unit (CPU) performs a high-frequency, real-time comparison of the calculated inter-frame motion vector magnitude or optical flow velocity with a motion distortion threshold pre-stored in memory. During clinical ultrasound scans, when doctors rapidly slide or significantly rotate the handheld probe to locate target organs, severe motion blur, tissue edge ghosting, and acoustic beam shearing distortion occur in the image. If the CPU detects that the current inter-frame motion vector or optical flow velocity exceeds the preset motion distortion threshold, it accurately determines that the current image is in a state of severe distortion caused by rapid probe movement, at which point the anatomical boundary information in the image no longer has a reliable recognition benchmark.

[0040] Furthermore, upon determining that the current state is one of rapid probe movement distortion, the central processing unit (CPU) sends a suspension interrupt signal to the neural network accelerator. The CPU immediately suspends the pushing of subsequent incoming blurred video frames to the neural network accelerator and temporarily blocks the feature extraction and forward inference processes of the deep learning model. Through the above-mentioned active hibernation and blocking mechanisms, not only is invalid computation by the deep learning model on motion-blurred interference images, leading to misidentification or false triggering of cross-sections, completely avoided, but the computational power consumption and heat generation of the device during the rapid scanning phase are also significantly reduced. At this time, the system only maintains basic pass-through transmission of the video stream.

[0041] Finally, during the pause in deep learning model recognition, the central processing unit (CPU) continuously monitors the inter-frame motion vectors or optical flow velocities of the subsequent video frames received in real time in the background. When the doctor's hand movements slow down and the probe is stably positioned near the target cut surface, the sharp pixel displacements in the image rapidly decay, and the inter-frame motion vectors or optical flow velocities calculated by the CPU subsequently fall below the preset motion distortion threshold. At this point, the CPU determines that the image has returned to stability and clarity, immediately lifts the suspension, activates and resumes the process of using the pre-trained deep learning model to perform key measurement image recognition on the digital video frame sequence, thereby ensuring that the artificial intelligence recognition calculations always run on high-quality, stable images.

[0042] Therefore, the above embodiments achieve pre-sensing and dynamic throttling of computing power for severe motion distortion during rapid scanning of the probe by calculating inter-frame motion vectors or optical flow velocity and automatically pausing deep learning recognition when the velocity exceeds a threshold, effectively eliminating invalid calculations and misjudgments caused by motion blur.

[0043] Furthermore, in one embodiment of the present invention, the step of using a pre-trained deep learning model to perform key measurement image recognition on a digital video frame sequence and determining whether the current frame is a target key measurement image includes: Perform state feature detection on digital video frame sequences to determine whether the current frame contains native freeze state identifiers of the ultrasound imaging device and / or whether it contains preset native measurement interface elements of the device. If it is determined that the current frame contains the original frozen state identifier of the ultrasound imaging device or contains preset original measurement interface elements of the device, then the current frame is marked as a candidate measurement frame. Input the current frame, which is marked as a candidate measurement frame, into the pre-trained deep learning model to calculate the confidence score of the current frame belonging to each standard section. Select the slice category corresponding to the highest confidence score. If the highest confidence score is greater than the preset first classification threshold, confirm the standard slice type corresponding to the current frame. The recognition results of the current frame and its adjacent frames are combined for temporal smoothing. If the number of frames that are continuously identified as having the same standard cross-section type within a preset time window exceeds a preset proportion, then the current frame is determined to be the target key measurement image.

[0044] Specifically, the central processing unit (CPU) performs real-time state feature detection on the received digital video frame sequence. Using lightweight template matching or edge detection algorithms, the CPU quickly scans a specific region of the current frame to determine if the image contains native freeze status indicators such as the snowflake icon from the ultrasound imaging device, or native measurement interface elements such as caliper marks and measurement menus. Since doctors perform a "freeze image" operation when measurements are needed, the ultrasound imaging device's screen changes from a dynamic scan to a static frozen frame, and measurement interface elements (such as caliper marks and measurement menus) typically appear on the screen. Therefore, once these indicators are detected, it means the doctor may have completed the scan of the target section and pressed the freeze button. Following this, to prevent accidental triggering caused by doctors using the device's trackball to rewind and review historical frames, the CPU continuously calculates the rate of change of image content features between the current frame and subsequent consecutive frames. If the feature change rate is higher than the preset fluctuation threshold, the central processing unit determines that the current device is in a dynamic viewing state of trackball playback, and at this time the automatic measurement trigger logic will be temporarily suspended; only when the feature change rate is lower than the preset fluctuation threshold, and the duration of this stable state exceeds the preset static waiting time, will the central processing unit confirm that the doctor has selected the best section and officially mark the final static image as a candidate measurement frame.

[0045] Next, the central processing unit (CPU) pushes the labeled candidate measurement frames to the neural network accelerator. The neural network accelerator is equipped with a pre-trained deep learning multi-classification model, which has been trained on a large amount of image data specifically for standard ultrasound sections of different human body parts. The neural network accelerator initiates forward inference computation, extracts the deep visual features of the candidate measurement frames, and outputs a probability vector containing multiple categories. Each probability value represents the confidence score of the current frame belonging to the corresponding standard section type (e.g., four-chamber view, head circumference view, etc.). Key measurement image categories include, but are not limited to, the following types of standard ultrasound images: For abdominal examinations, standard ultrasound images include: the maximum oblique diameter measurement section of the right lobe of the liver, the thickness / length measurement section of the left lobe of the liver, the length / transverse diameter / wall thickness measurement section of the gallbladder, the thickness measurement section of the pancreatic head / body / tail, the thickness / length measurement section of the spleen, the length / width / thickness measurement section and renal parenchyma thickness measurement section of both kidneys, and the abdominal aortic diameter measurement section, etc.; For cardiac examinations, standard ultrasound images include: the left ventricular end-diastolic / end-systolic diameter measurement section, the interventricular septum / left ventricular posterior wall thickness measurement section, the left atrial anteroposterior diameter measurement section, the right ventricular diameter measurement section, the aortic root diameter measurement section, and the blood flow spectrum measurement section of each valve orifice, etc.; For thyroid examinations, standard ultrasound images include... Standard ultrasound images include: sections measuring the diameters of the left / right lobe / isthmus of the thyroid gland, sections measuring the diameters of thyroid nodules, etc.; for breast examinations, standard ultrasound images include: sections measuring the diameters of breast nodules / masses, sections measuring the internal diameter of mammary ducts, etc.; for obstetric examinations, standard ultrasound images include: sections measuring the biparietal diameter / head circumference / abdominal circumference / femur length / humerus length of the fetus, sections measuring the amniotic fluid index / maximum amniotic fluid depth, sections measuring the nuchal translucency (NT), etc.; for vascular examinations, standard ultrasound images include: sections measuring the intima-media thickness (IMT) of the carotid artery, sections measuring the internal diameters and blood flow spectrum of various vessels, etc.; for musculoskeletal examinations, standard ultrasound images include: sections measuring the thickness and cross-sectional area of ​​tendons / ligaments, sections measuring the depth of joint effusion, etc.

[0046] Furthermore, the neural network accelerator selects the highest confidence score and compares it with a preset first classification threshold. If the highest confidence score is greater than the preset threshold, the neural network accelerator confirms the standard section type corresponding to the current candidate measurement frame and feeds back the classification result and confidence score to the central processing unit.

[0047] Finally, to completely eliminate misjudgments caused by occasional single-frame visual interference, the CPU combines the classification and recognition results of the current frame and its preceding and following frames to perform temporal smoothing. The CPU maintains a time sliding window of a preset length in system memory, counting the proportion of frames consecutively identified as having the same standard cross-section type within this window. Only when the number of frames with the same cross-section type exceeds a preset proportion does the CPU finally determine the current frame as the target key measurement image through temporal smoothing. This multi-level state confirmation and temporal filtering mechanism not only aligns with the manual scanning and freeze-up habits of clinicians but also greatly ensures the high accuracy and stability of subsequent automatic positioning and measurement calculations.

[0048] It should be noted that, in other embodiments of the present invention, the step of using a pre-trained deep learning model to perform key measurement image recognition on the digital video frame sequence and determine whether the current frame is a target key measurement image may also include only the single detection step of inputting the current frame marked as a candidate measurement frame into the pre-trained deep learning model, calculating the confidence score of the current frame belonging to various standard sections, and confirming the standard section type corresponding to the current frame when the highest confidence score is greater than the preset first classification threshold; or the single detection step may be combined with any one or two of the other two steps (i.e., determining whether the current frame contains the original frozen state identifier of the ultrasound imaging device and / or whether it contains the preset original measurement interface elements of the device, and determining the number of frames continuously confirmed as the same standard section type within a preset time window and the preset ratio of the recognition results of the current frame and its adjacent frames before and after it for time-series smoothing processing). The specific settings are determined according to actual usage needs (e.g., the computing power and performance impact of the device), and are not specifically limited here.

[0049] Therefore, the above embodiments achieve accurate capture of the doctor's clinical scanning intent and effective locking of the measurement timing by performing device-native freeze identification and measurement interface element detection on the digital video frame sequence, and combining deep learning model section classification confidence and multi-frame temporal smoothing decision, effectively filtering transient jitter and single-frame misjudgment during the scanning process, and ensuring high accuracy and high stability of key section recognition.

[0050] Step S30: If the current frame is determined to be the key measurement image of the target, the measurement area of ​​the target anatomical structure is located in the current frame, and automatic measurement calculation is performed based on the visual features in the current frame to obtain the target measurement value and the corresponding measurement mark. In one embodiment of the present invention, the steps of locating the measurement region of the target anatomical structure in the current frame and performing automatic measurement calculations based on visual features within the current frame to obtain the target measurement value and the corresponding measurement marker include: A pre-trained image segmentation model is used to extract the scale pattern at the edge of the ultrasound image in the current frame; By calculating the number of pixels between adjacent scale marks in the scale graphic, the mapping scale from the pixels of the current frame to the actual physical size can be calculated in reverse. Use a pre-trained target detection model or semantic segmentation model to identify and locate the boundary extreme points of the target anatomical structure in the current frame; Based on the measurement type formula corresponding to the target anatomical structure, the pixel distance, pixel area, or pixel contour geometric parameters between each boundary extreme point are calculated, and combined with the mapping scale, the target measurement values ​​corresponding to the target anatomical structure are calculated and converted.

[0051] Specifically, after determining that the current frame is the key measurement image for the target, the central processing unit (CPU) schedules the image data of that frame to the neural network accelerator. Since the pure video stream itself does not contain low-level data labels reflecting physical dimensions, the neural network accelerator first calls a pre-trained image segmentation model to specifically scan the outer edges of the non-ultrasound scan areas of the image, accurately identifying and extracting the scale graphic mask natively displayed by the ultrasound imaging device. Subsequently, the CPU performs morphological analysis on this scale graphic mask, automatically identifying the center position of two adjacent scale marks (usually representing a fixed physical distance of one centimeter or ten millimeters) and accurately calculating the number of pixels in a straight line between these two scale marks. The CPU uses the fixed physical distance divided by this number of pixels to inversely calculate the pixel-to-millimeter mapping scale of the current frame and temporarily stores it in the system's memory, thus establishing a benchmark for subsequent physical dimension conversions.

[0052] Next, the central processing unit (CPU) retrieves the corresponding measurement item instructions from the knowledge graph based on the standard section type of the previously determined target key measurement image, and controls the neural network accelerator to load the corresponding high-precision target detection model or semantic segmentation model. The neural network accelerator performs forward inference calculations on the clean ultrasound scan image region in the current frame, outputting the region of interest bounding box or pixel-level segmentation mask of the target anatomical structure (e.g., fetal skull, liver boundary, or specific cyst). The CPU further performs contour extraction and geometric extremum search on these bounding boxes or segmentation masks to accurately locate the coordinates of multiple boundary extremum points that conform to clinical measurement standards (e.g., the farthest pixels at both ends of the major axis or the inner and outer edges of specific tissues). Specifically, for example, for the measurement of the maximum oblique diameter of the right lobe of the liver, the upper and lower boundary points of the right lobe of the liver are located; for the measurement of the biparietal diameter of the fetus, the near and far inner edges of the fetal skull are located.

[0053] Furthermore, after successfully locating the boundary extrema, the central processing unit (CPU) performs calculations according to the standard measurement type formula corresponding to the current measurement item. For example, for diameter measurements, the CPU calculates the Euclidean pixel distance between the two boundary extrema; for area measurements, the CPU calculates the total number of pixels enclosed by the polygonal contour or the pixel area of ​​an ellipse fitted based on the extrema; for perimeter measurements, it calculates geometric parameters such as the pixel perimeter of the line connecting the extrema or extracting the contour. It should be noted that measurement types include, but are not limited to: linear distance measurement (distance between two points), area measurement (area of ​​an ellipse or hand-drawn contour), perimeter measurement, angle measurement, and spectral parameter measurements such as blood flow velocity / resistance index. Based on this, the CPU reads the previously calculated mapping scale from memory, multiplies the obtained pixel-level parameters with the mapping scale, or performs corresponding dimensional conversion operations, accurately converting the virtual pixel space data into actual physical space data, ultimately obtaining the true and accurate target measurement value.

[0054] Finally, based on the calculated pixel coordinates of the boundary extreme points and the obtained target measurement values, the central processing unit (CPU) generates corresponding virtual measurement marker data in the system memory (e.g., a crosshair indicating the start and end points of the measurement, a caliper line connecting two extreme points, or a dashed outline surrounding the lesion). The CPU packages these target measurement values ​​and measurement marker data and subsequently transfers them to the image overlay rendering module. Thus, without requiring underlying data interaction with the ultrasound imaging equipment, a complex automated measurement process from target localization to physical value calculation is completed through a closed loop of pure visual analysis.

[0055] Therefore, the above embodiments accurately extract the scale graphic of the edge of the ultrasound image by using an image segmentation model, and reverse calculate the mapping scale from the pixel to the physical size based on the pixel spacing of adjacent scale marks. This achieves the reverse calculation of physical size under the condition of no underlying digital communication metadata in pure video stream, overcomes the technical problem that conventional video analysis cannot directly obtain the real spatial dimensions, and ensures the absolute geometric accuracy and clinical usability of automatic measurement.

[0056] Furthermore, in one embodiment of the present invention, the steps of locating the measurement region of the target anatomical structure in the current frame and performing automatic measurement calculations based on visual features within the current frame to obtain the target measurement value and the corresponding measurement marker include: Query the preset knowledge graph or mapping table to obtain multiple measurement indicators associated with the standard cross-section type of the target key measurement image; The pre-trained target detection model and semantic segmentation model are invoked to synchronously and respectively locate multiple anatomical sub-regions or multiple key morphological landmarks corresponding to multiple measurement indicators in the current frame. Measurement calculations are performed in parallel for each anatomical sub-region or key morphological landmark to obtain a target set of measurement values ​​containing multiple measurement indicators.

[0057] Specifically, after determining the specific standard section type of the current target key measurement image, the central processing unit (CPU) actively accesses a pre-built knowledge graph or mapping table in the system memory. This knowledge graph embeds deep medical ultrasound examination standard logic. By comparing the current section type, the CPU quickly retrieves and obtains multiple measurement index tasks strongly related to that section. For example, if the current section is confirmed to be a standard section specific to obstetrics, the CPU will simultaneously extract multiple parallel measurement index requirements such as biparietal diameter and head circumference required for that section from the mapping table, thereby setting a specific calculation list for the subsequent multi-target comprehensive analysis.

[0058] Next, the central processing unit (CPU) issues multi-task parallel scheduling instructions to the neural network accelerator processor based on the extracted list of multiple measurement indicators. Upon responding to the instructions, the neural network accelerator processor simultaneously loads and invokes the pre-trained object detection model and semantic segmentation model in memory to perform synchronous inference on the current frame. Within the same forward propagation computation cycle, the object detection model quickly outlines multiple anatomical sub-regions distributed in the image, while the semantic segmentation model performs pixel-level fine segmentation of these sub-regions, accurately identifying and extracting multiple key morphological landmarks (such as the junction of tissue inner and outer edges or the endpoints of specific bones) corresponding to each measurement indicator. This multi-model fusion synchronous processing method ensures that even when multiple complex measurement targets exist on the same screen, the system can complete the accurate spatial localization of all targets in a very short time.

[0059] Furthermore, after the neural network accelerator outputs the masks of all sub-regions and the coordinates of key morphological landmarks, the central processing unit (CPU) initiates a multi-threaded concurrent computation mechanism. For each anatomical sub-region or each set of key morphological landmarks, the CPU automatically retrieves the corresponding specific spatial geometry algorithm formula. In multiple computation threads, the CPU executes in parallel computational tasks such as Euclidean distance calculations between extrema, perimeter integration of closed contours, and pixel-to-surface accumulation of connected regions. Combining this with the previously obtained pixel-to-actual-physical-size mapping scale, the CPU synchronously converts all pixel-level results into actual physical-size values.

[0060] Finally, the CPU summarizes and formats the multiple physical dimension values ​​obtained from these multi-threaded parallel calculations. The CPU packages the results of all independent measurements extracted from the same frame into a comprehensive target measurement value set containing multiple measurement indicators. Simultaneously, the CPU matches corresponding visual measurement marker coordinate parameters to each value in this set and stores them together in the system memory. This efficiently completes the automated calculation of multiple indicators for complex cross-sections on the same screen and prepares the data for subsequent image fusion and overlay.

[0061] Therefore, the above embodiments obtain related measurement indicators by querying a preset knowledge graph and call the model to simultaneously locate multiple anatomical sub-regions or landmarks to perform calculations in parallel. This achieves high-concurrency calculation of multiple measurement indicators on the same screen under complex cross-sections, effectively avoids repeated measurement operations on the same structure, and significantly shortens the time consumption of joint measurement of multiple indicators.

[0062] Step S40: Overlay the target measurement value and measurement mark onto the current frame to generate a processed image frame carrying measurement information; In one embodiment of the present invention, the step of overlaying the target measurement value and measurement marker onto the current frame to generate a processed image frame carrying measurement information includes: At the corresponding position of the anatomical structure in the current frame, draw caliper line segments and outlines as measurement marks according to preset visual rendering parameters; Identify the edge blank areas of non-ultrasound scan image regions in the current frame and obtain the coordinate range of the edge blank areas; The target measurement values ​​are plotted as structured text within the coordinate range of the blank area at the edge to generate a processed image frame, so as to avoid the target measurement values ​​obscuring the original lesion image of the anatomical structure.

[0063] Specifically, the central processing unit (CPU) transmits the previously calculated pixel coordinates of the extreme points of the target anatomical structure boundary, the geometric coordinates of the segmented contour, and the target measurement values ​​to the image overlay rendering module. The image overlay rendering module reads preset visual rendering parameters from the system memory, including the color, transparency, line width, and endpoint cursor style of the measurement lines. Subsequently, using the spatial coordinates of the anatomical structure in the current frame as a reference, the image overlay rendering module precisely draws high-contrast caliper lines between the corresponding extreme points and draws closed or semi-closed contour lines along the edges of the anatomical tissue's shape, serving as graphical measurement markers to visually indicate the measurement basis.

[0064] Next, to ensure that the added text information does not interfere with the doctor's observation of tissue structures and lesion details in the scan image, the image overlay rendering module and the central processing unit work together to perform regional topology analysis on the current frame. Based on the previously configured boundary parameters of the effective ultrasound scanning area, or through scanning calculations of the pixel distribution in each region of the image, the system accurately identifies the non-ultrasound scanning image areas outside the effective ultrasound fan-shaped or rectangular scanning field of view in the current frame, i.e., the edge blank areas that are pure black or have no effective echo signal. The image overlay rendering module automatically calculates and locks the safe rectangular boundary coordinate range of this edge blank area in the two-dimensional pixel coordinate system, using it as a safe rendering position specifically for arranging text information.

[0065] Furthermore, the image overlay rendering module extracts the target measurement values ​​and their corresponding anatomical measurement item names, arranging them into multi-line aligned structured text content according to standardized typesetting logic. Based on pre-set font size, line spacing, and text background transparency parameters, the image overlay rendering module precisely renders this structured text within the coordinate range of the locked edge blank area. Because this text area is completely isolated from the core ultrasound scan image area in space, regardless of the number of measurement items, the generated text information will not intrude into or obscure any anatomical structures or suspicious lesions within the scan field of view, ensuring the integrity and purity of the original image information.

[0066] Finally, the image overlay rendering module utilizes image layer fusion processing to merge the marker layer containing caliper lines and contours, the data layer containing structured text, and the original ultrasound image layer of the current frame pixel-by-pixel according to a set transparency ratio. This results in a standard processed image frame carrying complete, clear, and unobstructed measurement information. The central processing unit then writes this processed image frame into the output buffer for subsequent scheduling to the analog data output module for video transmission and hardware synchronous acquisition.

[0067] Therefore, the above embodiments automatically identify the edge blank areas of non-ultrasound scan image areas and draw the target measurement values ​​in the form of structured text within the coordinate range of the blank area, thereby achieving spatial physical isolation between the data text and the core ultrasound echo field of view. This ensures that the measurement results are clear and intuitive while avoiding visual obstruction of key diagnostic information such as the internal anatomical structures and lesion areas.

[0068] Furthermore, in one embodiment of the present invention, the step of overlaying and drawing the target measurement value and measurement mark onto the current frame further includes: The received video signal is analyzed to obtain the original video frame resolution and image aspect ratio parameters output by the ultrasound imaging device; Based on the difference between the original video frame resolution and the resolution calculated by the benchmark model, the corresponding overlay rendering scaling factor and coordinate translation vector are calculated. Based on the overlay rendering scaling factor and coordinate translation vector, the absolute pixel coordinate mapping relationship of the measurement mark on the current frame, the drawing line width of the caliper line segment, and the font size of the structured text of the target measurement value are dynamically adjusted. The rendering was performed using the adjusted mapping relationship and style parameters to ensure that the superimposed measurement marks under the output resolution of different models of ultrasound imaging equipment could be aligned with the edges of anatomical structures without causing visual distortion.

[0069] Specifically, after locking onto the incoming video signal, the video acquisition and processing module parses the video stream. The central processing unit (CPU) reads the parsing results from the video acquisition and processing module to obtain the horizontal pixel width and vertical pixel height of the original video frame output by the currently connected ultrasound imaging device, and then calculates the aspect ratio parameters of the current original image (e.g., a standard 4:3 or 16:9 aspect ratio). The CPU temporarily stores the parsed original resolution and aspect ratio parameters in the system memory as a reference for subsequent geometric transformations.

[0070] Next, since the deep learning models running in the neural network accelerator processor typically perform feature extraction and boundary prediction based on a preset fixed-size reference resolution, this reference resolution often differs from the actual output resolution of various external ultrasound imaging devices. The central processing unit (CPU) compares the resolution of the original video frames it reads with the model's reference calculation resolution, calculates the size ratio difference along the horizontal and vertical axes, and thus accurately calculates the corresponding overlay rendering scaling factor. Simultaneously, if the image was centered or had borders filled during the preprocessing stage, the CPU also calculates the corresponding coordinate translation compensation amount, generating the coordinate translation vector required for geometric alignment.

[0071] Furthermore, the central processing unit synchronously sends the calculated overlay rendering scaling factor and coordinate translation vector to the image overlay rendering module. Based on this scaling factor and translation vector, the image overlay rendering module establishes a reverse mapping channel from the model's computational coordinate system to the original video frame's pixel coordinate system. Specifically, the image overlay rendering module dynamically remaps the boundary extreme points and contour point coordinates output by the neural network accelerator, accurately restoring them to absolute pixel coordinates at the current original resolution. On this basis, the image overlay rendering module also performs proportional adaptive scaling of the rendering style according to the scaling factor, including dynamically adjusting the pixel line width of the caliper line segments according to the resolution, and dynamically scaling the font size and line height of the structured measurement text, ensuring that the overlay elements present a visually appropriate proportion for human observation on displays with different resolutions.

[0072] Finally, the image overlay rendering module utilizes the adjusted absolute pixel coordinate mapping relationship and adaptive style parameters to perform high-precision layer rendering and drawing on the high-resolution original pixel layer of the current frame. Through this real-time geometric inverse transformation and style adaptive mechanism, visual distortion problems such as coordinate drift, stretching deformation, or text blurring caused by the inconsistency between the model calculation scale and the device output scale are completely eliminated. This ensures that regardless of the model or era of the ultrasound imaging equipment connected to the front end, the generated caliper endpoints and measurement contour lines can be effectively attached to and aligned with the actual anatomical tissue boundaries.

[0073] Therefore, the above embodiments, by analyzing the physical resolution and aspect ratio of the original video signal, dynamically calculating the rendering scaling factor and coordinate translation vector to adaptively adjust the marker mapping relationship, line width and font size, achieve compatibility and distortion-free presentation for ultrasound imaging equipment of multiple brands, eras and resolutions, ensuring that the measurement markers can fit the anatomical contours on display terminals of different specifications.

[0074] Step S50: Transmit the processed image frame to the ultrasound reporting workstation transparently through the analog video acquisition card interface, and simultaneously send a hardware acquisition trigger signal to the ultrasound reporting workstation through the analog acquisition control interface to control the ultrasound reporting workstation to automatically acquire and store the currently transmitted processed image frame. In one embodiment of the present invention, the steps of transparently transmitting the processed image frame to the ultrasound reporting workstation through the analog video acquisition card interface and simultaneously sending a hardware acquisition trigger signal to the ultrasound reporting workstation through the analog acquisition control interface include: The processed image frame carrying measurement information is written into the output frame buffer of the analog video acquisition card module so that the video signal conforming to the standard video protocol is continuously output to the ultrasound reporting workstation. Simultaneously with writing the processed image frame carrying measurement information into the output frame buffer, or within a preset delay, a hardware acquisition trigger signal is sent to the ultrasound reporting workstation via the analog acquisition control interface. The hardware acquisition trigger signals include one or more of the following: simulated USB HID keyboard shortcut commands, TTL level change signals output through specific pins of the VGA interface, serial port COM commands, or Joystick protocol signals.

[0075] Specifically, after the image overlay rendering module generates a processed image frame containing anatomical measurement markers and structured numerical text, the central processing unit directly writes the data array of this processed image frame into the output frame buffer of the analog video capture card interface integrated within the analog data output module. The analog data output module, based on a pre-programmed standard Universal Serial Bus video device protocol or high-speed extended bus video protocol, virtualizes and disguises itself as a standard physical external video capture card hardware to the ultrasound reporting workstation's operating system. Therefore, the ultrasound reporting workstation does not need to install any dedicated third-party drivers or custom software; it can continuously read the video data stream from the output frame buffer through the system's native standard video capture channel and display the image with measurement results in real time in the workstation's image preview window, achieving completely transparent transmission and plug-and-play functionality at the video level.

[0076] Next, in order to ensure that the image captured by the ultrasound reporting workstation is strictly aligned with the current image frame with the measurement results on the timeline, the central processing unit monitors the output queue status of the analog video acquisition card interface in real time. When it is confirmed that the processed image frame carrying the measurement information has completely entered the effective output frame buffer area pushed to the ultrasound reporting workstation, and after a preset hardware transmission delay (to ensure that the ultrasound reporting workstation has completed the reception and rendering of the image frame on its screen preview layer), the central processing unit immediately sends a trigger signal generation command to the analog acquisition control interface.

[0077] Furthermore, after receiving instructions from the central processing unit, the analog acquisition control interface sends hardware-level physical electrical signals to the ultrasound reporting workstation via physical control cables, according to the trigger protocol mode pre-set for the ultrasound reporting workstation model. Specifically, the analog acquisition control interface can support various physical simulation methods: including simulating key press events of universal serial bus human-machine interaction devices (such as keyboards or mice) and sending preset single-key or combination shortcut key scan codes to the workstation; or outputting specific TTL logic high-low level transition pulses through specific idle pins in the video graphics array interface to trigger external interrupts of the acquisition card with hardware levels; or sending control codes in a specific format through the serial communication interface, or outputting key status packets conforming to the standard Joystick protocol.

[0078] Finally, when the ultrasound reporting workstation's operating system and application software capture the aforementioned physical signals at the underlying hardware port, they interpret it as the operating physician manually pressing a physical foot pedal switch, handle acquisition button, or keyboard acquisition shortcut key on-site. The ultrasound reporting workstation then automatically invokes its internally preset native image saving program, instantly capturing and saving the single frame currently being received and displayed through the analog video capture card interface (i.e., the processed image frame containing precise caliper measurements and values), and automatically archives it into the current patient's local image database and structured report template. At this point, the entire transmission and acquisition loop is completely driven by hardware-level simulation, completely eliminating dependence on the underlying proprietary software interfaces of different brands of workstations, achieving a highly efficient workflow where doctors are completely unaware and data is automatically retained.

[0079] Therefore, the above embodiments, by writing the processed image frames into the output buffer and simultaneously sending hardware acquisition trigger signals via universal serial bus human-machine interaction commands, specific interface pin level changes, serial port commands, or handle protocols, realize hardware-level physical simulation and access for various existing ultrasound reporting workstations, breaking down the barrier of high-cost customized development of proprietary software interfaces for workstations from different manufacturers.

[0080] Furthermore, in one embodiment of the present invention, while performing the step of transparently transmitting the processed image frame to the ultrasound reporting workstation through the analog video capture card interface, the method further includes: The extracted target measurement values ​​and corresponding examination item identifiers are encapsulated and converted into standard medical structured data format files, including JSON, XML, or DICOM SR formats; The medical structured data format file is sent to the target network address of the ultrasound reporting workstation via a preset network interface. After being parsed by the ultrasound reporting workstation, the target measurement value is directly filled into the corresponding field of the patient report template that the ultrasound reporting workstation is currently in an active receiving state.

[0081] Specifically, while the central processing unit (CPU) controls the analog video capture card interface to push video streams to the ultrasound reporting workstation, the CPU simultaneously extracts all target measurement values ​​derived from the analysis of the current frame from the system's memory, along with the corresponding standard anatomical examination item identification codes. Following a pre-configured system interface protocol, the CPU encapsulates and converts the extracted measurement item names, numerical values, units of measurement, and confidence levels into a cross-platform standard medical structured data format file. This file can be in JSON, XML, or DICOM SR format, and does not require or include the patient's personally identifiable information during file construction.

[0082] Next, the central processing unit (CPU) invokes the network communication interface module built into the auxiliary device to establish a connection channel with the external network via local area network communication protocols (such as HTTP / REST API, HL7, DICOM Secondary Capture, etc.). Based on pre-stored communication configuration parameters, the CPU packages the encapsulated medical structured data format file as the data payload into standard network data packets, and pushes them via wired network interface or wireless communication channel to the target network communication address bound to the ultrasound reporting workstation and the preset application receiving port, achieving real-time distribution of measurement data over the network channel.

[0083] Furthermore, after the data receiving service running in the background of the ultrasound reporting workstation listens for and captures the network data packet, it automatically calls a syntax parser that matches the file format to quickly unpack and decompose the incoming standard medical structured data format file, accurately extracting the name codes and specific numerical content of each measurement indicator. During the parsing process, the ultrasound reporting workstation does not forcibly compare the patient's identity information in the data packet, but directly obtains the current front-end business status of the workstation.

[0084] Finally, when the ultrasound reporting workstation detects that it is currently in the active reception and examination entry interface of a specific patient, the form mapping logic inside the workstation will automatically perform key-value pairing of the parsed anatomical examination item identifiers with the preset input box fields in the structured report template of the currently active patient. After successful pairing, the ultrasound reporting workstation directly and automatically fills in, refreshes, and saves the corresponding target measurement values ​​in the corresponding data unit of the patient's report template. Therefore, this embodiment of the invention, while automatically capturing and storing marked images through the video channel, can also automatically and accurately fill in structured values ​​in parallel through the network channel, constructing a dual-channel automated workflow for collaborative delivery of images and data, completely eliminating the tedious operation and error risk of manual data entry by clerks copying values ​​from images and manually typing them into the report.

[0085] Therefore, the above embodiments, by encapsulating the extracted measurement values ​​and project identifiers into standard medical structured data files and pushing them to the ultrasound reporting workstation in real time via a network interface, realize the parallel delivery of image capture and retention and automatic filling of report values ​​through two channels. This eliminates the tedious process of manual data entry by clerks copying values ​​from images and manually entering them into reports, forming a complete closed loop of integrated automated delivery of images and data.

[0086] Furthermore, in one embodiment of the present invention, after the steps of locating the measurement region of the target anatomical structure in the current frame and performing automatic measurement calculation based on the visual features within the current frame to obtain the target measurement value and the corresponding measurement mark, the method further includes: The overall confidence score is calculated by weighting the boundary extreme point confidence scores output by the object detection model or semantic segmentation model and the image quality score of the current frame. Determine whether the overall confidence score is greater than the preset security assessment threshold; If the overall confidence score is less than or equal to the preset safety assessment threshold, a processed image frame containing early warning information is generated, and the instruction to generate a hardware acquisition trigger signal is intercepted to block the automatic acquisition and storage action of the ultrasound report workstation. If the overall confidence score is greater than the preset security assessment threshold, then the step of overlaying the target measurement value and measurement mark onto the current frame is executed.

[0087] Specifically, after the neural network accelerator completes the location and measurement calculation of the boundary extreme points of the target anatomical structure, the central processing unit (CPU) extracts the classification probability value and location confidence of each boundary extreme point from the neural network accelerator. Simultaneously, the CPU calls its internally integrated image quality assessment algorithm to perform multi-dimensional quantitative analysis of the current frame's sharpness, contrast, and tissue echo uniformity, calculating the image quality score for the current frame. Subsequently, the CPU retrieves preset weight allocation parameters from the system memory and performs a weighted summation calculation on the extracted boundary extreme point confidence and the image quality score, thereby generating a comprehensive confidence score that reflects both the current image's sharpness and the reliability of the measurement and location.

[0088] Next, the central processing unit (CPU) compares the calculated overall confidence score with a pre-defined safety assessment threshold embedded in the system in real time. This safety assessment threshold, calibrated based on extensive clinical expert evaluation data, serves as a safety threshold for determining whether the automated measurement results possess medical diagnostic usability. Based on the comparison results, the CPU sets internal logic execution branch status flags to determine the rendering mode of the subsequent image overlay rendering module and whether to issue a hardware trigger command to the analog data output module.

[0089] Furthermore, if the central processing unit (CPU) determines that the current overall confidence score is less than or equal to the preset safety assessment threshold, it indicates that the current image may have interference factors such as excessive probe pressure, blurred image edges, or incomplete visualization of anatomical structures. At this time, the CPU controls the image overlay rendering module to enter the safety warning mode, overlaying and drawing only conspicuous warning text in yellow or other eye-catching styles (such as displaying a low confidence warning or prompting manual review) in the preset edge area of ​​the current frame, generating a processed image frame containing warning information, and forcibly intercepting the control process of generating hardware acquisition trigger signals at the system's underlying level, completely prohibiting the analog data output module from sending any analog button or level trigger commands to the ultrasound reporting workstation, thereby completely blocking the automatic screenshot and retention actions of the ultrasound reporting workstation, and completely returning the diagnostic decision-making power to the doctor for manual retesting and confirmation.

[0090] Finally, if the central processing unit (CPU) determines that the current overall confidence score is greater than the preset security assessment threshold, it confirms that the automatic measurement result is highly accurate and reliable. The CPU then releases the security interception state and sends a normal rendering command to the image overlay rendering module, controlling it to execute the standard overlay drawing process. This process accurately draws and merges the structured text containing high-precision caliper lines, measurement contours, and target measurement values ​​onto the current frame, generating a standard processed image frame carrying complete measurement information. This drives the subsequent analog video output and hardware synchronously triggers the retention process, thereby achieving a highly automated workflow while ensuring the absolute rigor and security of medical data.

[0091] Therefore, the above embodiments construct a comprehensive confidence assessment system by integrating the confidence of boundary extreme points and image quality scores, and actively generate early warnings and intercept hardware acquisition trigger commands in the low confidence state. This realizes the linkage between intelligent assistance and clinical medical safety risk control, effectively preventing erroneous measurement data under blurred or non-standard sections from being automatically stored in the database, and ensuring the rigor and safety of ultrasound reports.

[0092] In summary, the video analysis-based automatic measurement assistance method for ultrasound examination described in the above embodiments of the present invention receives video signals output by ultrasound imaging equipment in real time and converts them into digital video frame sequences. It then uses a pre-trained deep learning model to identify key measurement images within the digital video frame sequences, achieving fully automatic, continuous, and real-time intelligent perception of key measurement images during ultrasound scanning, eliminating the need for manual pausing or marking of sections by the physician. Furthermore, by automatically locating the measurement area of ​​the target anatomical structure in the current frame identified as the target key measurement image, and automatically performing measurement calculations based on visual features to obtain the target measurement value and corresponding measurement markers, it achieves fully automatic extraction of morphological and dimensional parameters of the anatomical structure. This completely liberates ultrasound physicians from tedious manual caliper work and the mechanical operation of selecting measurement items, effectively reducing the physician's workload and significantly improving the overall efficiency of ultrasound scanning. Finally, by directly overlaying the target measurement value and measurement markers onto the current frame to generate a processed image frame carrying measurement information, it achieves intuitive fusion and visualization of the measurement basis (graphical markers) and measurement results (numerical text) on the original image, ensuring automatic... The system enhances the intuitiveness, traceability, and transparency of measurement results. By transparently transmitting processed image frames to the ultrasound reporting workstation via an analog video capture card interface, it achieves plug-and-play access for various brands and versions of ultrasound reporting workstations without the need for driver installation or software modification. This effectively avoids the high costs and compatibility issues associated with customizing dedicated software interfaces for different workstation systems. Furthermore, by simultaneously sending hardware acquisition trigger signals to the ultrasound reporting workstation via the analog acquisition control interface to control its automatic acquisition and retention while transmitting processed image frames, it achieves time-series synchronization and automatic archiving of key image frames containing measurement information without human intervention. This eliminates human errors such as mishearing, misrecording, and omissions caused by doctors' verbal instructions and manual transcription by clerks in the traditional process. It forms a seamless, automated closed loop from video stream input to intelligent recognition, automatic measurement, layer overlay, and finally hardware-triggered saving. This solves the problems of heavy manual measurement burden for doctors, high error rates in verbal communication and manual transcription during existing ultrasound examinations, and high costs and poor system compatibility of existing automation solutions when integrating with ultrasound reporting workstations.

[0093] Example 2 Please see Figure 2 This is a schematic diagram of a video analysis-based automatic measurement auxiliary system for ultrasound examination provided in the second embodiment of the present invention. For ease of explanation, only the parts related to the embodiment of the present invention are shown. The system is applied to an automatic measurement auxiliary device for ultrasound examination connected to an ultrasound imaging device and an ultrasound reporting workstation, respectively. The system includes: The video acquisition module 11 is used to receive video signals output from the ultrasound imaging device in real time and convert the video signals into a digital video frame sequence; Image recognition module 12 is used to perform key measurement image recognition on the digital video frame sequence using a pre-trained deep learning model, and to determine whether the current frame is a target key measurement image; The automatic measurement module 13 is used to locate the measurement area of ​​the target anatomical structure in the current frame if the image recognition module determines that the current frame is a key measurement image of the target, and to perform automatic measurement calculation based on the visual features in the current frame to obtain the target measurement value and the corresponding measurement mark. The overlay drawing module 14 is used to overlay the target measurement value and measurement mark onto the current frame to generate a processed image frame carrying measurement information. The acquisition trigger module 15 is used to transparently transmit the processed image frame to the ultrasound reporting workstation through the analog video acquisition card interface, and simultaneously send a hardware acquisition trigger signal to the ultrasound reporting workstation through the analog acquisition control interface to control the ultrasound reporting workstation to automatically acquire and retain the currently transmitted processed image frame.

[0094] Furthermore, in one embodiment of the present invention, the image recognition module 12 includes: The state feature detection unit is used to perform state feature detection on the digital video frame sequence to determine whether there is a native frozen state identifier of the ultrasound imaging device and / or whether it contains preset native measurement interface elements of the device in the current frame. The candidate frame marking unit is used to mark the current frame as a candidate measurement frame if it is determined that the current frame contains the original frozen state identifier of the ultrasound imaging device or contains preset original measurement interface elements of the device. The section classification unit is used to input the current frame, which is labeled as a candidate measurement frame, into the pre-trained deep learning model and calculate the confidence score of the current frame belonging to each standard section. The slice type confirmation unit is used to select the slice category corresponding to the highest confidence score. If the highest confidence score is greater than a preset first classification threshold, the standard slice type corresponding to the current frame is confirmed. The temporal smoothing decision unit is used to perform temporal smoothing processing by combining the recognition results of the current frame and its adjacent frames. If the number of frames that are continuously identified as having the same standard cross-section type within a preset time window exceeds a preset proportion, the current frame is determined to be the target key measurement image.

[0095] Furthermore, in one embodiment of the present invention, the automatic measurement module 13 includes: The scale graphic extraction unit is used to extract the scale graphic of the edge of the ultrasound image in the current frame using a pre-trained image segmentation model; The mapping scale calculation unit is used to calculate the mapping scale from the pixels of the current frame to the actual physical size by calculating the number of pixels between adjacent scale marks in the scale graphic. The boundary extreme point localization unit is used to identify and locate the boundary extreme points of the target anatomical structure in the current frame using a pre-trained target detection model or semantic segmentation model. The physical value conversion unit is used to calculate the pixel distance, pixel area or pixel contour geometric parameters between each boundary extreme point according to the measurement type formula corresponding to the target anatomical structure, and calculate and convert the target measurement value corresponding to the target anatomical structure in combination with the mapping scale.

[0096] Furthermore, in one embodiment of the present invention, the automatic measurement module 13 further includes a parallel measurement unit, the parallel measurement unit comprising: The indicator dictionary query subunit is used to query a preset knowledge graph or mapping table to obtain multiple measurement indicators associated with the standard cross-section type of the target key measurement image; The multi-target synchronous localization subunit is used to call the pre-trained target detection model and semantic segmentation model to synchronously and respectively locate multiple anatomical sub-regions or multiple key morphological landmarks corresponding to the multiple measurement indicators in the current frame. The parallel measurement subunit is used to perform measurement calculations in parallel for each anatomical subregion or key morphological landmark to obtain a set of target measurement values ​​containing multiple measurement indicators.

[0097] Furthermore, in one embodiment of the present invention, the automatic measurement module 13 further includes a security interception unit, the security interception unit comprising: The comprehensive confidence score calculation subunit is used to calculate the current automatically measured comprehensive confidence score by weighting the confidence scores of the boundary extreme points output by the object detection model or semantic segmentation model and the image quality score of the current frame. A security threshold assessment subunit is used to determine whether the overall confidence score is greater than a preset security assessment threshold. The acquisition blocking subunit is used to generate a processed image frame containing early warning information if the comprehensive confidence score is less than or equal to a preset safety assessment threshold, and to intercept the instruction to generate the hardware acquisition trigger signal, so as to block the automatic acquisition and storage action of the ultrasound report workstation. A drawing enable subunit is used to trigger the overlay drawing module 14 to perform the operation of overlaying the target measurement value and measurement mark onto the current frame if the overall confidence score is greater than a preset security assessment threshold.

[0098] Furthermore, in one embodiment of the present invention, the overlay rendering module 14 includes: The measurement mark drawing unit is used to draw caliper line segments and contour lines as measurement marks at the corresponding positions of the anatomical structures in the current frame according to preset visual rendering parameters. An edge blank area identification unit is used to identify edge blank areas in the non-ultrasound scan image region of the current frame and obtain the coordinate range of the edge blank area. The structured text drawing unit is used to draw the target measurement values ​​in the form of structured text within the coordinate range of the edge blank area, and generate the processed image frame to avoid the target measurement values ​​from obscuring the original lesion image of the anatomical structure.

[0099] Furthermore, in one embodiment of the present invention, the overlay rendering module 14 further includes an adaptive rendering unit, the adaptive rendering unit comprising: The video signal parameter parsing subunit is used to parse the received video signal and obtain the original video frame resolution and image aspect ratio parameters output by the ultrasound imaging device. The geometric transformation parameter calculation subunit is used to calculate the corresponding overlay rendering scaling factor and coordinate translation vector based on the difference between the original video frame resolution and the baseline model calculation resolution. The style parameter dynamic adjustment subunit is used to dynamically adjust the absolute pixel coordinate mapping relationship of the measurement mark in the current frame, the drawing line width of the caliper line segment, and the font size of the structured text of the target measurement value based on the overlay rendering scaling factor and coordinate translation vector. The adaptive layer rendering subunit is used to render and draw using the adjusted mapping relationship and style parameters to ensure that the measurement marks superimposed on the output resolution of different models of ultrasound imaging equipment can be aligned with the edges of anatomical structures without producing visual distortion.

[0100] Furthermore, in one embodiment of the present invention, the acquisition triggering module 15 includes: The frame buffer output unit is used to write the processed image frame carrying measurement information into the output frame buffer area of ​​the analog video acquisition card module so that the video signal conforming to the standard video protocol is continuously output to the ultrasound reporting workstation. The timing synchronization triggering unit is used to send the hardware acquisition trigger signal to the ultrasound reporting workstation through the analog acquisition control interface at the same time as the processed image frame carrying measurement information is written into the output frame buffer or within a preset delay. The hardware acquisition trigger signal includes one or more of the following: simulated USB HID keyboard shortcut key commands, TTL level change signals output through specific pins of the VGA interface, serial port COM commands, or Joystick protocol signals.

[0101] Furthermore, in one embodiment of the present invention, the video acquisition module 11 includes an image preprocessing unit, the image preprocessing unit comprising: The coordinate acquisition and reading subunit is used to read the pre-configured coordinate parameters of the acquisition area; The fixed-region cropping subunit is used to crop a fixed region of each frame in the digital video frame sequence according to the coordinate parameters of the acquisition area, remove the original non-scanning area interface of the ultrasound imaging device, and extract the clean ultrasound scan image area as the input image for deep learning model processing.

[0102] Furthermore, in one embodiment of the present invention, the system further includes a motion distortion suppression module, the motion distortion suppression module comprising: Motion vector calculation unit, used to calculate inter-frame motion vectors or optical flow velocities between consecutive digital video frame sequences; The fast motion determination unit is used to determine that the current image is in a probe fast motion distortion state when the inter-frame motion vector or optical flow velocity is greater than a preset motion distortion threshold, and to pause the image recognition module 12 from performing the step of key measurement image recognition of the digital video frame sequence using a pre-trained deep learning model until the inter-frame motion vector or optical flow velocity is less than the preset motion distortion threshold.

[0103] Furthermore, in one embodiment of the present invention, the system further includes a data network transmission module, the data network transmission module comprising: The data encapsulation and conversion unit is used to encapsulate and convert the extracted target measurement values ​​and corresponding examination item identifiers into standard medical structured data format files, including JSON, XML or DICOM SR formats. The network-targeted push unit is used to send the medical structured data format file to the target network address where the ultrasound reporting workstation is located through a preset network interface, so that the ultrasound reporting workstation can parse it and directly fill the target measurement value into the corresponding field of the patient report template that the ultrasound reporting workstation is currently in an active receiving state.

[0104] The automatic measurement auxiliary system for ultrasound examination based on video analysis provided in this embodiment of the invention has the same implementation principle and technical effect as the aforementioned method embodiment. For the sake of brevity, any parts not mentioned in the device embodiment can be referred to the corresponding content in the aforementioned method embodiment.

[0105] Example 3 In another aspect, the present invention also proposes an automatic measurement auxiliary device for ultrasound examination, please refer to [link to relevant documentation]. Figure 3 The image shows an automatic measurement auxiliary device for ultrasound examination in the third embodiment of the present invention. It is connected to an ultrasound imaging device and an ultrasound reporting workstation, and includes a memory 20, a processor 10, and a program 30 stored in the memory 20 and executable on the processor 10. When the processor 10 executes the program 30, it implements the automatic measurement auxiliary method for ultrasound examination based on video analysis as described in the above method embodiment.

[0106] In some embodiments, the processor 10 may be a central processing unit (CPU), controller, microcontroller, microprocessor or other data processing chip, used to run program code stored in memory 20 or process data, such as executing access restriction programs.

[0107] The memory 20 includes at least one type of readable storage medium, such as flash memory, hard disk, multimedia card, card-type memory (e.g., SD or DX memory), magnetic memory, magnetic disk, optical disk, etc. In some embodiments, the memory 20 can be an internal storage unit of the automated ultrasound measurement auxiliary device, such as the hard disk of the device. In other embodiments, the memory 20 can be an external storage device of the automated ultrasound measurement auxiliary device, such as a plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, etc., equipped on the device. Furthermore, the memory 20 can include both internal and external storage units. The memory 20 can be used not only to store application software and various data installed on the automated ultrasound measurement auxiliary device, but also to temporarily store data that has been output or will be output.

[0108] It should be pointed out that, Figure 3 The structure shown does not constitute a limitation on the automatic measurement auxiliary device for ultrasound examination. In other embodiments, the automatic measurement auxiliary device for ultrasound examination may include fewer or more components than shown, or combine certain components, or have different component arrangements.

[0109] This invention also proposes a computer-readable medium storing a program that, when executed by a processor, implements the video analysis-based automatic measurement auxiliary method for ultrasound examination as described in the foregoing method embodiments.

[0110] Those skilled in the art will understand that the logic and / or steps represented in the flowcharts or otherwise described herein, for example, can be considered as a ordered list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-included system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "medium" can mean any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device.

[0111] More specific examples of media (a non-exhaustive list) include: electrical connections (electronic devices) with one or more wires, portable computer disk drives (magnetic devices), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Furthermore, computer-readable media can even be paper or other suitable media on which the program can be printed, since the program can be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in computer memory.

[0112] It should be understood that various parts of the present invention can be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented in software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0113] In the description of this specification, references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0114] The embodiments described above are merely illustrative of several implementations of the present invention, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the appended claims.

Claims

1. A video-analysis-based automated measurement assistance method for ultrasound examination, characterized in that, An automated measurement auxiliary device for ultrasound examination, respectively connected to an ultrasound imaging device and an ultrasound reporting workstation, wherein the method comprises: It receives video signals from ultrasound imaging equipment in real time and converts the video signals into a digital video frame sequence; A pre-trained deep learning model is used to identify key measurement images in the digital video frame sequence to determine whether the current frame is a target key measurement image. If the current frame is determined to be a key measurement image of the target, the measurement area of ​​the target anatomical structure is located in the current frame, and automatic measurement calculation is performed based on the visual features in the current frame to obtain the target measurement value and the corresponding measurement mark. The target measurement value and measurement mark are superimposed and drawn onto the current frame to generate a processed image frame carrying measurement information; The processed image frame is transparently transmitted to the ultrasound reporting workstation through the analog video capture card interface, and a hardware acquisition trigger signal is simultaneously sent to the ultrasound reporting workstation through the analog acquisition control interface to control the ultrasound reporting workstation to automatically acquire and retain the currently transmitted processed image frame.

2. The automatic measurement assistance method for ultrasound examination based on video analysis according to claim 1, characterized in that, The step of using a pre-trained deep learning model to perform key measurement image recognition on the digital video frame sequence and determining whether the current frame is a target key measurement image includes: The digital video frame sequence is subjected to state feature detection to determine whether the current frame contains the original frozen state identifier of the ultrasound imaging device and / or whether it contains preset original measurement interface elements of the device. If it is determined that the current frame contains the original frozen state identifier of the ultrasound imaging device or contains preset original measurement interface elements of the device, then the current frame is marked as a candidate measurement frame. Input the current frame, which is marked as a candidate measurement frame, into the pre-trained deep learning model to calculate the confidence score of the current frame belonging to each standard section. Select the slice category corresponding to the highest confidence score. If the highest confidence score is greater than the preset first classification threshold, confirm the standard slice type corresponding to the current frame. The recognition results of the current frame and its adjacent frames are combined for temporal smoothing. If the number of frames that are continuously identified as having the same standard cross-section type within a preset time window exceeds a preset proportion, then the current frame is determined to be the target key measurement image.

3. The automatic measurement assistance method for ultrasound examination based on video analysis according to claim 1, characterized in that, The steps of locating the measurement region of the target anatomical structure in the current frame and automatically measuring and calculating based on the visual features within the current frame to obtain the target measurement value and the corresponding measurement marker include: The scale graphic of the edge of the ultrasound image in the current frame is extracted using a pre-trained image segmentation model; By calculating the number of pixels between adjacent scale marks in the scale graphic, the mapping scale from the pixels of the current frame to the actual physical size is calculated in reverse. The boundary extreme points of the target anatomical structure in the current frame are identified and located using a pre-trained target detection model or semantic segmentation model. Based on the measurement type formula corresponding to the target anatomical structure, the pixel distance, pixel area, or pixel contour geometric parameters between each boundary extreme point are calculated, and combined with the mapping scale, the target measurement values ​​corresponding to the target anatomical structure are calculated and converted.

4. The automatic measurement assistance method for ultrasound examination based on video analysis according to claim 1, characterized in that, The steps of locating the measurement region of the target anatomical structure in the current frame and automatically measuring and calculating based on the visual features within the current frame to obtain the target measurement value and the corresponding measurement marker include: Query a preset knowledge graph or mapping table to obtain multiple measurement indicators associated with the standard cross-section type of the target key measurement image; The pre-trained target detection model and semantic segmentation model are invoked to synchronously and respectively locate multiple anatomical sub-regions or multiple key morphological landmarks corresponding to the multiple measurement indicators in the current frame. Measurement calculations are performed in parallel for each anatomical sub-region or key morphological landmark to obtain a target set of measurement values ​​containing multiple measurement indicators.

5. The automatic measurement auxiliary method for ultrasound examination based on video analysis according to claim 1, characterized in that, After the steps of locating the measurement region of the target anatomical structure in the current frame and performing automatic measurement calculations based on visual features within the current frame to obtain the target measurement value and the corresponding measurement marker, the method further includes: The overall confidence score is calculated by weighting the boundary extreme point confidence scores output by the object detection model or semantic segmentation model and the image quality score of the current frame. Determine whether the overall confidence score is greater than a preset security assessment threshold; If the overall confidence score is less than or equal to the preset safety assessment threshold, a processed image frame containing early warning information is generated, and the instruction to generate the hardware acquisition trigger signal is intercepted to block the automatic acquisition and retention action of the ultrasound report workstation. If the overall confidence score is greater than the preset security assessment threshold, then the step of overlaying the target measurement value and measurement mark onto the current frame is executed.

6. The automatic measurement assistance method for ultrasound examination based on video analysis according to claim 1, characterized in that, The step of overlaying the target measurement value and measurement marker onto the current frame to generate a processed image frame carrying measurement information includes: At the location corresponding to the anatomical structure in the current frame, caliper line segments and contour lines are drawn as measurement marks according to preset visual rendering parameters; Identify the edge blank areas of non-ultrasound scan image regions in the current frame, and obtain the coordinate range of the edge blank areas; The target measurement values ​​are plotted in the form of structured text within the coordinate range of the blank area at the edge, and the processed image frame is generated to avoid the target measurement values ​​obscuring the original lesion image of the anatomical structure.

7. The automatic measurement assistance method for ultrasound examination based on video analysis according to claim 1, characterized in that, The step of overlaying the target measurement value and measurement marker onto the current frame includes: The received video signal is analyzed to obtain the original video frame resolution and image aspect ratio parameters output by the ultrasound imaging device; Based on the difference between the original video frame resolution and the resolution calculated by the benchmark model, the corresponding overlay rendering scaling factor and coordinate translation vector are calculated. Based on the overlay rendering scaling factor and coordinate translation vector, the absolute pixel coordinate mapping relationship of the measurement mark in the current frame, the drawing line width of the caliper line segment, and the font size of the structured text of the target measurement value are dynamically adjusted. The rendering was performed using the adjusted mapping relationship and style parameters to ensure that the superimposed measurement marks under the output resolution of different models of ultrasound imaging equipment could be aligned with the edges of anatomical structures without causing visual distortion.

8. The automatic measurement assistance method for ultrasound examination based on video analysis according to claim 1, characterized in that, The steps of transparently transmitting the processed image frame to the ultrasound reporting workstation via the analog video acquisition card interface and simultaneously sending a hardware acquisition trigger signal to the ultrasound reporting workstation via the analog acquisition control interface include: The processed image frame carrying measurement information is written into the output frame buffer of the analog video acquisition card module so that a video signal conforming to the standard video protocol is continuously output to the ultrasound reporting workstation. Simultaneously with or within a preset delay, the processed image frame carrying measurement information is written into the output frame buffer, and the hardware acquisition trigger signal is sent to the ultrasound reporting workstation via the analog acquisition control interface. The hardware acquisition trigger signal includes one or more of the following: simulated USB HID keyboard shortcut key commands, TTL level change signals output through specific pins of the VGA interface, serial port COM commands, or Joystick protocol signals.

9. The automatic measurement auxiliary method for ultrasound examination based on video analysis according to claim 1, characterized in that, Before the step of performing key measurement image recognition on the digital video frame sequence using a pre-trained deep learning model, the method further includes: Read the pre-configured acquisition area coordinate parameters; Based on the coordinate parameters of the acquisition area, each frame in the digital video frame sequence is cropped in a fixed region to remove the original non-scanning area interface of the ultrasound imaging device, and the pure ultrasound scan image area is extracted as the input image for deep learning model processing.

10. The automatic measurement assistance method for ultrasound examination based on video analysis according to claim 1, characterized in that, Before the step of performing key measurement image recognition on the digital video frame sequence using a pre-trained deep learning model, the method further includes: Calculate the inter-frame motion vector or optical flow velocity between consecutive digital video frame sequences; When the inter-frame motion vector or optical flow velocity is greater than the preset motion distortion threshold, it is determined that the current image is in a state of rapid probe movement distortion, and the step of using a pre-trained deep learning model to perform key measurement image recognition on the digital video frame sequence is paused until the inter-frame motion vector or optical flow velocity is less than the preset motion distortion threshold.

11. The automatic measurement assistance method for ultrasound examination based on video analysis according to claim 1, characterized in that, While performing the step of transparently transmitting the processed image frame to the ultrasound reporting workstation through the analog video capture card interface, the method also includes: The extracted target measurement values ​​and corresponding examination item identifiers are encapsulated and converted into standard medical structured data format files, including JSON, XML, or DICOM SR formats. The medical structured data format file is sent to the target network address of the ultrasound reporting workstation via a preset network interface. After being parsed by the ultrasound reporting workstation, the target measurement value is directly filled into the corresponding field of the patient report template that the ultrasound reporting workstation is currently in an active receiving state.

12. An automated measurement auxiliary system for ultrasound examination based on video analysis, characterized in that, An automated measurement auxiliary device for ultrasound examination, which is connected to an ultrasound imaging device and an ultrasound reporting workstation respectively, the system comprising: The video acquisition module is used to receive video signals output from the ultrasound imaging device in real time and convert the video signals into a digital video frame sequence; The image recognition module is used to perform key measurement image recognition on the digital video frame sequence using a pre-trained deep learning model, and to determine whether the current frame is a target key measurement image. An automatic measurement module is used to locate the measurement area of ​​the target anatomical structure in the current frame if the image recognition module determines that the current frame is a key measurement image of the target, and to perform automatic measurement calculation based on the visual features in the current frame to obtain the target measurement value and the corresponding measurement mark. The overlay drawing module is used to overlay the target measurement values ​​and measurement marks onto the current frame to generate a processed image frame carrying measurement information. The acquisition trigger module is used to transparently transmit the processed image frame to the ultrasound reporting workstation through the analog video acquisition card interface, and simultaneously send a hardware acquisition trigger signal to the ultrasound reporting workstation through the analog acquisition control interface to control the ultrasound reporting workstation to automatically acquire and retain the currently transmitted processed image frame.

13. An automatic measurement auxiliary device for ultrasound examination, characterized in that, The device is connected to both an ultrasound imaging device and an ultrasound reporting workstation, and includes a memory, a processor, and a program stored in the memory and executable on the processor. When the processor executes the program, it implements the video analysis-based automatic measurement assistance method for ultrasound examination as described in any one of claims 1-11.

Images