Ultrasonic Image Retention Method and System

Abstract

The present invention is applicable to the field of medical imaging technology, and provides an ultrasonic image retention method and system, which are applied to an ultrasonic image retention system including an ultrasonic device, an artificial intelligence system and an ultrasonic workstation. The method includes that the artificial intelligence system obtains ultrasonic images from the ultrasonic device through a first video transmission line, generates image video signals, and outputs them to the ultrasonic workstation through a second video transmission line at the same time; when the artificial intelligence system analyzes the ultrasonic images in real time and identifies key image frames, it controls the output image video signals to pause updating within a preset time and freeze on the key image frames, and sends a trigger signal to the ultrasonic workstation through a control transmission line; the ultrasonic workstation receives the image video signals from the artificial intelligence system, and when receiving the trigger signal, retains the image data corresponding to the key image frames. The present invention solves the problems of insufficient real-time performance, accuracy and imperceptibility in the existing ultrasonic image retention.

Description

Technical Field

[0001] The present invention relates to the field of medical imaging technology, and particularly relates to a method and system for retaining ultrasonic images. Background Art

[0002] Due to its advantages of being non-invasive, real-time, portable, and relatively low-cost, ultrasound examination has become an indispensable imaging examination method in clinical medicine. In the conventional ultrasound examination process, an operating physician (doctor or technician) uses an ultrasound diagnostic device (hereinafter referred to as "ultrasound device" or "ultrasound host") for real-time scanning to observe the structure of tissues and organs and blood flow information. When a standard section, typical structure, or suspicious lesion with diagnostic value is found, the physician usually needs to manually operate (such as by freezing the image, pressing the save button on the device, using a handle or foot switch) to capture and save the selected static image or dynamic video segment. These saved images are then transmitted to or directly saved on an ultrasound imaging workstation (hereinafter referred to as "ultrasound workstation"). The ultrasound workstation is a key platform for physicians to perform image post-processing, measurement analysis, diagnosis writing, and finally generate an ultrasound report, and is often integrated with the hospital's Picture Archiving and Communication System (PACS).

[0003] In recent years, the application of artificial intelligence technology in the field of medical imaging has developed rapidly, especially showing great potential in ultrasound image analysis. AI algorithms can assist physicians in identifying standard anatomical sections, automatically measuring biological parameters, detecting and prompting suspicious lesions (such as nodules, masses, etc.), and evaluating image quality. In theory, the intervention of AI aims to improve the accuracy and consistency of diagnosis, reduce the workload of physicians, and improve the examination efficiency. However, in the practice of existing technologies, the application effect of AI technology is often limited by its integration with the existing clinical workflow. Currently, the common mode is that the artificial intelligence system may run as an independent software on a separate computer, or although it is connected to the ultrasound device / workstation, there is often a "gap" between its analysis results (such as the key frames or lesion positions prompted) and the image data that ultimately needs to be incorporated into the diagnostic report. Specifically:

[0004] Images with important clinical significance identified by an artificial intelligence system (such as a perfect standard section or a clear image of a positive lesion) often only stay on the interface of the artificial intelligence system or in its internal database. If a physician wants to use the key images identified by this artificial intelligence system for writing a report or archiving, they usually still need to return to the ultrasound device or the workstation interface and manually locate and perform the save operation again. This usually means that when the doctor observes the prompt of the artificial intelligence system or judges on their own that a certain frame of image needs to be saved, they must first perform a "freeze" operation on the ultrasound device, and then manually trigger the ultrasound workstation to capture the currently displayed frozen image through its connected image acquisition card by operating the handle, foot switch or keyboard shortcut keys connected to the ultrasound workstation. This not only fails to fully utilize the advantage of the artificial intelligence system in reducing the burden, but may also increase the workload of the physician due to the need to switch between different systems for observation and operation, and the manual re-capture may result in the failure to accurately save the best frame initially identified by the artificial intelligence system due to delay or operation errors. This operation process leads to an obvious barrier between the artificial intelligence system and the ultrasound workstation system. The recognition and diagnosis results of the artificial intelligence system cannot be directly and seamlessly integrated into the actual work process of ultrasound examiners. The artificial intelligence system only plays an auxiliary role in observation or prompt to a large extent, fails to deeply participate in and optimize the examination business itself, and cannot effectively reduce the operation burden of doctors and improve work efficiency.

[0005] In order to try to solve the problem of system isolation, the prior art has proposed a solution to specifically interface the artificial intelligence system with different ultrasound workstation systems, referring to Figure 3As shown in the figure, the ultrasound device is connected to both the artificial intelligence system and the ultrasound workstation system through video transmission lines. At the same time, the artificial intelligence system is docked with the ultrasound workstation system through an information interface. However, the architectures of ultrasound workstation systems from different manufacturers vary, and the interface standards are inconsistent, resulting in high development and maintenance costs, complex implementation, and poor versatility for this docking method. In addition, even if the interface docking is achieved, the data transmission is often not real-time and synchronous, and there may be problems such as "poor timeliness" (i.e., there is a delay from the completion of the artificial intelligence system's recognition to the data transmission to the workstation), "poor integrity" (images may be lost during data transmission due to network or other factors), and "backhaul delay". These risks prevent doctors from fully trusting the data transmitted back through the interface for directly issuing diagnostic reports. At the same time, after the artificial intelligence system is docked with the ultrasound workstation, the images collected by the artificial intelligence system are not transmitted to the ultrasound workstation in real time. A button needs to be clicked in the artificial intelligence system or the ultrasound workstation to trigger the transmission action to transmit the images collected by the artificial intelligence system. Specifically, during the ultrasound scan, the artificial intelligence system will perform real-time processing on the images. After the examination is completed or the key frame is obtained, click the image transmission button, and directly transmit the processing result (image) to the ultrasound workstation through the interface. At the same time, even if the artificial intelligence system can provide rich analysis results (such as measurement values, lesion locations), these information is often separated from the finally saved pure images, and doctors need to manually record or re-reference them in the report.

[0006] In summary, the existing technology has obvious deficiencies in automatically, accurately, and imperceptibly integrating the key ultrasound images identified by AI into the business process of the ultrasound workstation, lacking an effective solution that can overcome system barriers, ensure the real-time and accuracy of image retention, and not increase the additional operation burden on doctors. Summary of the Invention

[0007] Based on this, the purpose of the present invention is to provide a method and system for ultrasound image retention to fundamentally solve the problems of insufficient real-time, accuracy, and imperceptibility of existing ultrasound image retention.

[0008] A method for ultrasound image retention according to an embodiment of the present invention is applied to an ultrasound image retention system including an ultrasound device, an artificial intelligence system, and an ultrasound workstation. The artificial intelligence system is connected to the ultrasound device through a first video transmission line and is connected to the ultrasound workstation through a second video transmission line and a control transmission line. The method includes:

[0009] The artificial intelligence system obtains ultrasound images from the ultrasound device in real time through the first video transmission line and generates an image video signal including the ultrasound images, and at the same time continuously outputs the image video signal to the ultrasound workstation through the second video transmission line;

[0010] The artificial intelligence system analyzes the real-time ultrasound images obtained in real time, and when a key image frame is identified from the ultrasound images according to a preset rule, it controls the output image video signal to pause updating within a preset time and freeze on the key image frame, and sends a trigger signal to the ultrasound workstation through the control transmission line;

[0011] The ultrasound workstation receives the image video signal from the artificial intelligence system in real time through the second video transmission line, and when receiving the trigger signal through the control transmission line, it executes the built-in preset image saving function according to the trigger signal, and automatically saves the currently received image data corresponding to the key image frame that is frozen.

[0012] In addition, for an ultrasound image saving method according to the above embodiments of the present invention, the following additional technical features may also be included:

[0013] Further, the step of controlling the output image video signal to pause and freeze the key image frame within a preset time includes:

[0014] The artificial intelligence system processes the ultrasound images obtained through the first video transmission line in real time, continuously stores the processed image frames in the frame buffer, and attaches a corresponding preset identifier to the data identified as the key image frame;

[0015] The artificial intelligence system locks the data of the key image frame specified by the preset identifier obtained from the frame buffer within a preset time, encodes it into an image video signal, and sends it to the ultrasound workstation through the second video transmission line.

[0016] Further, before or during the step of the artificial intelligence system locking the data of the key image frame specified by the preset identifier obtained from the frame buffer within a preset time, the method further includes:

[0017] The artificial intelligence system obtains the original pixel data of the identified key image frame;

[0018] The artificial intelligence system generates the required marked watermark content to be superimposed, and determines the preset rendering position and style of the marked watermark content on the key image frame;

[0019] The artificial intelligence system draws or fuses the generated marked watermark content onto the original pixel data of the obtained key image frame according to the determined preset rendering position and style, and generates the modified key image frame data with the marked watermark content.

[0020] Further, the step of the artificial intelligence system generating the required marked watermark content to be superimposed includes:

[0021] The artificial intelligence system generates marked watermark content including fixed text or graphic identifiers according to preset rules; and / or

[0022] The artificial intelligence system generates marked watermark content including diagnostic auxiliary metadata based on the diagnostic auxiliary metadata extracted or calculated during the real-time analysis of the identified key video frames.

[0023] Further, the diagnostic auxiliary metadata includes at least one of the identified positioning information, calculated biometric parameters, and analyzed confidence scores;

[0024] The step of determining the preset rendering position and style of the marked watermark content on the key video frame screen includes:

[0025] The artificial intelligence system determines the preset rendering position of the associated marked watermark content on the key video frame screen according to the positioning information in the diagnostic auxiliary metadata;

[0026] The artificial intelligence system determines the style of the marked watermark content on the key video frame screen according to the positioning information, biometric parameters, and confidence scores in the diagnostic auxiliary metadata.

[0027] Further, the step of identifying key video frames from the ultrasound images according to preset rules includes:

[0028] The artificial intelligence system uses a deep learning model to perform real-time analysis on the ultrasound images, and identifies the video frames including predefined standard sections in the analyzed images as key video frames;

[0029] The artificial intelligence system uses a lesion detection algorithm to detect and locate the suspicious lesion areas in the ultrasound images, and identifies the video frames including the suspicious lesions that meet the preset lesion screening conditions in the detected images as key video frames.

[0030] Further, the step of the artificial intelligence system using a deep learning model to perform real-time analysis on the ultrasound images and identifying the video frames including predefined standard sections in the analyzed images as key video frames includes:

[0031] The artificial intelligence system preprocesses each video frame in the ultrasound images in real time, and inputs each preprocessed video frame into a pre-trained deep learning model for forward propagation calculation;

[0032] The artificial intelligence system obtains the classification result and the corresponding confidence score of each video frame output by the deep learning model;

[0033] When the classification result of the target image frame is a predefined standard section and the confidence score is higher than the recognition threshold of the predefined standard section, the artificial intelligence system identifies the analyzed target image frame containing the predefined standard section as a key image frame.

[0034] Further, the steps for the artificial intelligence system to use a lesion detection algorithm to detect and locate suspicious lesion areas in the ultrasound image and identify the image frame containing the suspicious lesion that meets the preset lesion screening conditions as a key image frame include:

[0035] The artificial intelligence system preprocesses each image frame in the ultrasound image in real time and inputs each preprocessed image frame into a pre-trained lesion detection or segmentation model for processing;

[0036] The artificial intelligence system obtains the bounding box or segmentation mask of the candidate lesion area in each image frame output by the lesion detection or segmentation model, as well as the category and confidence of each candidate lesion area;

[0037] The artificial intelligence system post-processes each candidate lesion area detected in each image frame and calculates the data features of each candidate lesion area, where the data features include size, aspect ratio, and internal echo pattern;

[0038] The artificial intelligence system compares the data features calculated in each image frame with the preset lesion screening conditions, and the preset lesion screening conditions include a minimum size and a specific echo pattern;

[0039] When the suspicious lesion in the candidate lesion area of the target image frame meets the preset lesion screening conditions and the confidence is higher than the preset lesion detection threshold, the artificial intelligence system identifies the target image frame containing the suspicious lesion that meets the preset lesion screening conditions as a key image frame.

[0040] Further, the steps for the artificial intelligence system to control the output video signal of the image to pause updating and freeze on the key image frame within a preset time include:

[0041] The artificial intelligence system performs image quality evaluation on the identified key image frame to obtain a quality score or grade;

[0042] The artificial intelligence system determines whether the quality score or grade of the key image frame meets the preset quality threshold;

[0043] If so, the artificial intelligence system controls the output video signal of the image to pause updating and freeze on the key image frame.

[0044] Another object of another embodiment of the present invention is to provide an ultrasonic image retention system, the system includes an ultrasonic device, an artificial intelligence system and an ultrasonic workstation, the artificial intelligence system is connected to the ultrasonic device through a first video transmission line, and is connected to the ultrasonic workstation through a second video transmission line and a control transmission line;

[0045] The artificial intelligence system is configured to real-time obtain ultrasonic images from the ultrasonic device through the first video transmission line and generate an image video signal including the ultrasonic images, and at the same time continuously output the image video signal to the ultrasonic workstation through the second video transmission line;

[0046] The artificial intelligence system is further configured to real-time analyze the real-time obtained ultrasonic images, and when a key image frame is identified from the ultrasonic images according to a preset rule, control the output image video signal to pause updating within a preset time and freeze at the key image frame, and send a trigger signal to the ultrasonic workstation through the control transmission line;

[0047] The ultrasonic workstation is configured to real-time receive the image video signal from the artificial intelligence system through the second video transmission line, and when receiving the trigger signal through the control transmission line, execute its built-in preset image saving function according to the trigger signal, and automatically retain the currently received image data corresponding to the key image frame that is frozen.

[0048] The ultrasonic image retention method provided by the embodiments of the present invention connects an artificial intelligence system and an ultrasonic device through a first video transmission line, ensuring that the artificial intelligence system can directly and real-time obtain the image data stream from the ultrasonic device. By connecting the artificial intelligence system and an ultrasonic workstation through a second video transmission line and a control transmission line, independent channels for the artificial intelligence system to transmit image content and send control instructions to the ultrasonic workstation are established respectively. By continuously outputting image signals through the second video transmission line, the ultrasonic workstation can display in real-time a picture synchronized with the processing of the artificial intelligence system. When the artificial intelligence system recognizes a key image frame, it actively controls the video output to pause and freeze on this key image frame, and sends a trigger signal to the ultrasonic workstation through the control transmission line, triggering the ultrasonic workstation to execute its built-in native image saving function, bypassing the existing complex software interface dependencies. Moreover, when receiving the trigger signal, the ultrasonic workstation can accurately capture the frozen key image frame. At the same time, by superimposing diagnostic auxiliary metadata (such as measurement values, positioning information, confidence levels, etc.) analyzed by the artificial intelligence system on the automatically saved key image frames, the retained images themselves carry preliminary intelligent analysis results, providing richer and more direct information support for subsequent diagnostic decision-making and report writing, and solving the problems of insufficient real-time performance, accuracy, and imperceptibility in the existing ultrasonic image retention. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figure 1 is a flowchart of the ultrasonic image retention method in the first embodiment of the present invention;

[0050] Figure 2 is a structural diagram of the ultrasonic image retention system in the second embodiment of the present invention;

[0051] Figure 3 is a structural diagram during ultrasonic image retention in the prior art;

[0052] The following specific embodiments will further illustrate the present invention in conjunction with the above-mentioned drawings. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0053] To facilitate the understanding of the present invention, the present invention will be described more comprehensively below with reference to the relevant drawings. Several embodiments of the present invention are shown in the drawings. However, the present invention can be implemented in many different forms and is not limited to the embodiments described herein. On the contrary, these embodiments are provided to make the disclosure of the present invention more thorough and comprehensive.

[0054] It should be noted that when an element is referred to as "fixedly provided on" another element, it can be directly on the other element or there may also be an intermediate element. When an element is considered to be "connected to" another element, it can be directly connected to the other element or there may be an intermediate element at the same time. The terms "vertical", "horizontal", "left", "right" and similar expressions used herein are for illustrative purposes only.

[0055] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by those of ordinary skill in the technical field to which this invention belongs. The terms used herein in the description of the present invention are for the purpose of describing specific embodiments only and are not intended to limit the present invention. The term "and / or" used herein includes any and all combinations of one or more of the related listed items.

[0056] Embodiment 1

[0057] Please refer to Figure 1 , which shows the method for retaining ultrasonic images in the first embodiment of the present invention. For the convenience of description, only the parts related to the embodiments of the present invention are shown. The method for retaining ultrasonic images provided by the embodiments of the present invention includes:

[0058] Step S10, the artificial intelligence system obtains ultrasonic images from the ultrasonic device in real time through the first video transmission line and generates an image video signal containing the ultrasonic images, and at the same time continuously outputs the image video signal to the ultrasonic workstation through the second video transmission line;

[0059] In one embodiment of the present invention, the method is applied to an ultrasound image storage system including an ultrasound device, an artificial intelligence system and an ultrasound workstation, wherein the artificial intelligence system is connected to the ultrasound device via a first video transmission line, and is connected to the ultrasound workstation via a second video transmission line and a control transmission line. Specifically, the ultrasound device is the source of real-time ultrasound images and has a standard video output interface (such as HDMI, DisplayPort, DVI or analog signal interface). The artificial intelligence system is usually deployed on a computer with strong computing power, which is equipped with a first image acquisition card / interface, a high-performance processor (CPU / NPU / TPU), a video output interface, and a first control transmission line interface (such as a USB port), wherein the first image acquisition card is connected to the video output interface of the ultrasound device through a first video transmission line (such as an HDMI line) for real-time reception of the original video signal from the ultrasound device; wherein the video output interface is connected to the second image acquisition card / interface of the ultrasound workstation through a second video transmission line (such as an HDMI line) for outputting the image video signal output by the artificial intelligence system to the ultrasound workstation; wherein the first control transmission line interface is connected to the second control transmission line interface (such as a USB port) of the ultrasound workstation through a control transmission line (such as a USB copy line with a HID analog chip), for outputting the trigger signal output by the artificial intelligence system to the ultrasound workstation. The ultrasound workstation (USW) is usually a console or an independent image processing workstation that is matched with the ultrasound device, which is used for image post-processing and report writing. The ultrasound workstation includes a second image acquisition card / interface, a second control transmission line interface, a display, a user input device, and a workstation software, etc., wherein the second image acquisition card / interface is connected to the video output interface of the artificial intelligence system through a second video transmission line, which is used to receive the image video signal from the artificial intelligence system and display it in the display; wherein the second control transmission line interface is a USB port connected to the artificial intelligence system through a control transmission line, which is used to receive the trigger signal output by the artificial intelligence system; wherein the user input device generally includes a keyboard, a handle and a foot switch, which can trigger the preset image saving function built into the ultrasound workstation according to the user input; wherein the workstation software has image display, processing, and reporting functions, and can trigger its built-in image saving program through specific keyboard shortcuts (such as F1, which can be configured in the software). The control transmission line is a USB connection line with a keyboard simulation function (such as a USB KVM data transmission line or a USB copy line). Specifically, the control transmission line is a USB male-to-male physical connection line with a built-in chip, and its main purpose is to quickly transfer files between two computers through built-in software. At the same time, it also has a keyboard sharing auxiliary function, that is, the keyboard of one computer is synchronously operated through the keyboard of another computer.

[0060] Further, in an embodiment of the present invention, during the startup phase of the artificial intelligence system, it is necessary to load and initialize the driver program of the first image acquisition card. According to the performance of the first image acquisition card and the output specifications of the ultrasonic device, the acquisition parameters are configured. For example, its resolution is set to a resolution matching the output of the ultrasonic device; its frame rate is set to the frame rate of the ultrasonic device output; a suitable color space / format input format is selected and internal conversion is performed as appropriate for subsequent processing; an attempt is made to synchronize frames with the input signal to ensure the integrity of capture. At this time, the first image acquisition card driver program starts to work, continuously captures the incoming video signal from the video transmission line, decodes each frame of image data and transmits it to the memory of the artificial intelligence system host (usually the kernel buffer). The main application program of the artificial intelligence system reads the captured original video frame data from the kernel buffer at high speed by calling the API functions provided by the first image acquisition card SDK, and quickly sends it into a multi-thread safe input circular buffer (for example, a fixed-size first-in-first-out queue). Using a circular buffer can smooth the possible fluctuations between the acquisition frame rate and the processing frame rate and prevent data loss. At the same time, the buffer size needs to be adjusted according to the processing capacity and latency requirements (for example, buffer 10 - 30 frames). Further, one or more independent preprocessing threads / processes take out the original frame data from the input circular buffer, and then perform necessary preprocessing operations according to the configuration, such as cropping the ROI, denoising, and color / brightness adjustment. Among them, cropping the ROI is to crop out the effective ultrasonic image area according to the preset coordinate rectangle if the captured full-screen image contains a menu bar, etc. Denoising is to apply a denoising algorithm with low computational cost, such as fast median filtering or mean filtering, to preliminarily improve the image quality and prepare for subsequent analysis. Color / brightness adjustment may be to perform histogram equalization or gamma correction to make the image performance more consistent under different devices or lighting conditions.

[0061] Further, through data routing / passthrough, the preprocessed (or even raw, unprocessed) video frame data is directly and quickly routed to the data output, so that the image seen by the ultrasound workstation is as close as possible to the real-time image of the ultrasound device. At the same time, a copy of the preprocessed frame data is also made or passed to the parallel analysis engine through the zero-copy mechanism, and the analysis engine starts performing complex operations such as recognition and detection on it. Importantly, this analysis process does not block the path of the main data flow to the output end. At this time, the analysis results (if there are preliminary results at this time) are usually not superimposed on the output image at this stage to maintain the "original" nature of the output. The above data output mainly writes the selected frame data into an output frame buffer, then obtains the latest frame data from the output frame buffer, and encodes it according to preset parameters. The preset parameters can be the resolution compatible with the acquisition card of the ultrasound workstation, the target frame rate, and the encoding format selected for low latency. Then, the encoded video data is packaged into an image video signal that conforms to standards (such as HDMI, DisplayPort), and through the video output port of the artificial intelligence system, the generated image video signal is continuously sent to the second image acquisition card of the connected ultrasound workstation through the second video transmission line.

[0062] Step S20, the artificial intelligence system analyzes the real-time ultrasound image it obtains in real time, and when it identifies a key image frame from the ultrasound image according to a preset rule, it controls the output image video signal to pause updating within a preset time and freeze on the key image frame, and sends a trigger signal to the ultrasound workstation through the control transmission line;

[0063] Among them, in an embodiment of the present invention, while the artificial intelligence system continuously sends the generated image video signal to the second image acquisition card of the connected ultrasound workstation through the second video transmission line, its parallel analysis engine is also processing the real-time obtained image frames in parallel and determining whether they meet the preset rules. If so, they are identified as key image frames. The above steps of identifying key image frames from the ultrasound image according to the preset rules include:

[0064] The artificial intelligence system uses a deep learning model to perform real-time analysis on the ultrasound image, and identifies the analyzed image frames containing predefined standard sections as key image frames;

[0065] The artificial intelligence system uses a lesion detection algorithm to detect and locate suspicious lesion areas in the ultrasound image, and identifies the detected image frames containing suspicious lesions that meet the preset lesion screening conditions as key image frames.

[0066] Among them, the above steps of the artificial intelligence system using a deep learning model to perform real-time analysis on the ultrasound image and identifying the analyzed image frames containing predefined standard sections as key image frames include:

[0067] The artificial intelligence system pre - processes each image frame in the ultrasound image in real - time and inputs each pre - processed image frame into a pre - trained deep - learning model for forward propagation calculation;

[0068] The artificial intelligence system obtains the classification result and the corresponding confidence score of each image frame output by the deep - learning model;

[0069] If the classification result of the target image frame is a predefined standard section and the confidence score is higher than the recognition threshold of the predefined standard section, the artificial intelligence system identifies the target image frame containing the predefined standard section to be analyzed as a key image frame.

[0070] Specifically, the artificial intelligence system pre - processes each image frame in the ultrasound image in real - time. For example, it performs necessary size adjustment on the image frame to match the input size requirements of the pre - trained standard section classification model, and normalizes the pixel values (for example, subtracts the mean and divides by the standard deviation) to make it conform to the input distribution during model training. Then the pre - processed image frame is input into a pre - trained deep - learning model. This deep - learning model is usually a convolutional neural network, and its last layer is designed to output the probabilities of different standard section categories. Then, forward propagation calculation of the deep - learning model is performed to obtain a vector, where each element of the vector represents the probability or score that the input image frame belongs to a certain predefined standard section category (such as "four - chamber view", "left ventricular long - axis view", "fetal biparietal diameter", "S7 segment of the liver"). Then, the probability vector output by the deep - learning model is obtained, and the category with the highest probability and its corresponding confidence score are found. This highest confidence score is compared with the predefined recognition threshold for this specific standard section category (for example, for the four - chamber view, the threshold may be set to 0.95; for more difficult - to - identify sections, the threshold may be slightly lower). If the category corresponding to the highest probability is one of the predefined standard sections and its confidence score is higher than or equal to the corresponding recognition threshold, then the current frame is determined to be a key image frame containing this standard section. And an identification result event containing a "standard section" type label, the specific section name, and the confidence score is generated.

[0071] Among them, the steps of the above - mentioned artificial intelligence system using a lesion detection algorithm to detect and locate suspicious lesion areas in the ultrasound image and identifying the image frame containing suspicious lesions that meet the preset lesion screening conditions as key image frames include:

[0072] The artificial intelligence system pre - processes each image frame in the ultrasound image in real - time and inputs each pre - processed image frame into a pre - trained lesion detection or segmentation model for processing;

[0073] The artificial intelligence system obtains the bounding boxes or segmentation masks of the candidate lesion regions in each frame of the image output by the lesion detection or segmentation model, as well as the category and confidence of each candidate lesion region.

[0074] The artificial intelligence system performs post-processing on each candidate lesion region detected in each frame of the image, and calculates the data features of each candidate lesion region. The data features include size, aspect ratio, and internal echo pattern.

[0075] The artificial intelligence system compares the data features calculated in each frame of the image with the preset lesion screening conditions, and the preset lesion screening conditions include the minimum size and a specific echo pattern.

[0076] When there is a suspicious lesion in the candidate lesion region of the target image frame that meets the preset lesion screening conditions and the confidence is higher than the preset lesion detection threshold, the artificial intelligence system identifies the target image frame containing the suspicious lesion that meets the preset lesion screening conditions as a key image frame.

[0077] Specifically, the artificial intelligence system also preprocesses each frame of the ultrasound images in real time and then inputs the preprocessed image frames into a pre-trained lesion detection or segmentation model. Commonly used models include region proposal-based detection models (such as Faster R-CNN), single-stage detection models (such as YOLO, SSD), or pixel-level segmentation models (such as U-Net and its variants). Then, the forward propagation of the model is executed. At this time, for the detection model, it outputs the bounding box coordinates of a series of candidate lesion regions, the lesion category corresponding to each box (such as cysts, solid nodules, calcified lesions, etc.), and its confidence score. For the segmentation model, it outputs a segmentation mask of the same size as the input image frame, where each pixel is labeled as belonging to the background or a certain type of lesion region, usually accompanied by region-level confidence or probability maps for each pixel. Then, candidate lesion regions are extracted from the model output. For the detection model, it directly obtains the list of bounding boxes. For the segmentation model, it extracts the contours or bounding boxes of each independent candidate lesion region from the segmentation mask through connected component analysis or contour finding. Then, data features are calculated for each candidate lesion region that passes the preliminary filtering (defined by the bounding box or segmentation mask). The data features include size, shape, and internal echo pattern. At this time, the size mainly calculates the maximum diameter, short diameter, area, or volume estimate of the region (based on the number of pixels and pixel size). The shape mainly calculates shape descriptors such as aspect ratio, circularity, and edge regularity. The internal echo pattern mainly analyzes the pixel gray-scale distribution statistics (such as mean, standard deviation, entropy), texture features (such as gray-level co-occurrence matrix features) within the region, or applies a pre-trained echo pattern classifier (for example, to distinguish anechoic, hypoechoic, isoechoic, hyperechoic, mixed echo). Then, the data features of each calculated candidate lesion region are compared with the preset lesion screening conditions. These conditions are to further filter out regions with little clinical significance or likely artifacts. The conditions are specifically, for example, a minimum size threshold, specific echo pattern requirements, and shape regularity requirements. The minimum size threshold requires that the maximum diameter of the lesion must be greater than the preset millimeter size; the specific echo pattern requirements may only focus on anechoic (possibly cysts) or hypoechoic (possibly solid nodules) regions; the shape regularity requirements may filter out regions with extremely irregular shapes (possibly noise or artifacts). At this time, if there is at least one candidate lesion region in the target image frame that satisfies the above preset screening conditions and its original model confidence is also higher than or equal to a final, relatively high lesion detection threshold (for example, 0.8), then the current frame is determined to be a key image frame containing a suspicious lesion that meets the conditions. And an identification result event is generated that includes a label of the type "suspicious lesion", the lesion location (bounding box / mask), the possible category, the confidence score, and the feature information that meets the conditions.

[0078] Among them, when a key image frame is recognized from the ultrasonic image according to a preset rule, the output image video signal is controlled to pause updating within a preset time and freeze on the key image frame. The specific steps include:

[0079] The artificial intelligence system retrieves whether there is a previously recognized and triggered key image frame stored within a preset retrospective time window;

[0080] If there is a previous key image frame, the artificial intelligence system calculates the image content similarity between the currently recognized key image frame and the previous key image frame and determines whether it is higher than a preset similarity threshold;

[0081] When there is no previous key image frame or the image content similarity is not higher than the preset similarity threshold, the artificial intelligence system controls the output image video signal to pause updating within a preset time and freeze on the key image frame.

[0082] Furthermore, the steps of controlling the output image video signal to pause updating within a preset time and freeze on the key image frame specifically further include:

[0083] The artificial intelligence system performs image quality assessment on the recognized key image frame to obtain a quality score or level;

[0084] The artificial intelligence system determines whether the quality score or level of the key image frame meets a preset quality threshold;

[0085] If so, the artificial intelligence system controls the output image video signal to pause updating within a preset time and freeze on the key image frame.

[0086] Specifically, the artificial intelligence system performs image quality assessment on the identified key image frames to obtain a quality score or grade. There are various implementation methods, mainly including methods based on traditional image features and methods based on deep learning. Among them, the methods based on traditional image features mainly calculate clarity / sharpness metrics, calculate contrast metrics, calculate the estimated value of signal-to-noise ratio (SNR), and detect specific artifacts. When calculating the clarity / sharpness metrics, the Laplacian operator, gradient energy, edge density, or frequency domain analysis (such as the proportion of high-frequency component energy) can be used to evaluate the clarity of the image. Generally, the higher the score, the clearer the image. When calculating the contrast metrics, the standard deviation of the image gray value, dynamic range, or local contrast can be calculated. Generally, when the contrast is moderate, the quality is better. When calculating the estimated value of signal-to-noise ratio (SNR), it is mainly estimated by analyzing the noise level in the image background area and the intensity of the signal area. The higher the signal-to-noise ratio, the better. When detecting specific artifacts, specific detection algorithms designed for common ultrasound artifacts (such as acoustic shadows, reverberations, and sidelobe artifacts) are mainly used to output the probability of the presence of artifacts or the severity score. The fewer the artifacts, the better. Among the methods based on deep learning, a pre-trained deep learning model is usually used. The model directly takes the image as the input and outputs an end-to-end quality score (for example, in the range of 0-1, the higher the score, the better the quality) or a quality grade (for example, excellent, good, medium, poor). When performing the above image quality assessment, the above-mentioned multiple single metrics can be calculated, and then a final comprehensive quality score or quality grade can be obtained through weighted average or rule combination. Then, the obtained quality score is compared with a preset quality threshold (for example, it is required that the score > 0.75). Or the obtained quality grade is compared with a preset minimum acceptable grade (for example, it is required that the grade is "good" or "excellent"). This threshold / grade can be dynamically adjusted according to the type of key frames identified (for example, the quality requirements for standard sections may be higher than those for general lesions). If the calculated quality score is higher than or equal to the threshold, or the quality grade reaches or exceeds the minimum acceptable grade, it is determined that the image quality of the key image frame is qualified. At this time, the artificial intelligence system controls the output video signal of the image to pause updating within a preset time and freeze on the key image frame.

[0087] Among them, in an embodiment of the present invention, the step of controlling the output video signal of the image to pause and freeze the key image frame within a preset time includes:

[0088] The artificial intelligence system processes the ultrasound images obtained through the first video transmission line in real time, continuously stores the processed image frames in the frame buffer, and attaches a corresponding preset identifier to the data identified as key image frames;

[0089] The artificial intelligence system locks the data of the key image frame specified by the preset identifier obtained from the frame buffer within a preset time, encodes it into an image video signal, and sends it to the ultrasound workstation through the second video transmission line.

[0090] Specifically, the artificial intelligence system obtains and processes ultrasound images in real time through the first video transmission line, such as preprocessing, AI analysis (plane recognition, lesion detection, etc.), and then stores all the processed image frames (whether they are key image frames or not) in an orderly manner in a frame buffer. The frame buffer can be a circular queue that stores the image data of the most recent several seconds or several frames. When the artificial intelligence system does not recognize a key image frame or is not in the freeze frame period, the artificial intelligence system obtains data from the latest position of the frame buffer (i.e., the most recently processed image frame), encodes it into an image video signal, and then outputs it to ensure that the ultrasound workstation receives real-time updated ultrasound images. When the artificial intelligence system recognizes a certain frame as a key image frame, the artificial intelligence system attaches a preset identifier (for example, a specific flag bit, metadata tag) to the data of this frame. The artificial intelligence system locks the data of the key image frame specified by this preset identifier obtained from the frame buffer within the next preset time (for example, 0.5 seconds to 2 seconds, and this time should be sufficient for the ultrasound workstation to complete image acquisition). At the same time, the artificial intelligence system encodes the locked key image frame data (if necessary, for example, encodes it into an H.264 video frame), and sends it to the ultrasound workstation as an image video signal through the second video transmission line. Within the preset time, the artificial intelligence system continuously sends the data of this key image frame, so that the video stream received by the ultrasound workstation "pauses updating" and "freezes and displays" at this key image frame. At this time, the picture seen by the ultrasound workstation is stably frozen on the key image frame. At the same time, the artificial intelligence system starts an internal timer and sets it to the preset time. When the internal timer reaches the preset time, the artificial intelligence system releases the lock on the key image frame and resumes obtaining data from the latest position of the frame buffer, encoding it into an image video signal, and then outputting it.

[0091] Further, in an embodiment of the present invention, before or during the step of the artificial intelligence system locking the data of the key image frame specified by the preset identifier obtained from the frame buffer within a preset time, the artificial intelligence system can perform enhancement processing on this key image frame, and its steps specifically include:

[0092] The artificial intelligence system obtains the original pixel data of the recognized key image frame;

[0093] The artificial intelligence system generates the marked watermark content to be superimposed, and determines the preset rendering position and style of the marked watermark content on the key image frame screen;

[0094] The artificial intelligence system draws or fuses the generated marked watermark content onto the original pixel data of the acquired key video frame according to the determined preset rendering position and style, generating the modified key video frame data with the marked watermark content.

[0095] Further, the steps for the artificial intelligence system to generate the marked watermark content to be superimposed include:

[0096] The artificial intelligence system generates marked watermark content including fixed text or graphic identifiers according to preset rules; and / or

[0097] The artificial intelligence system generates marked watermark content including diagnostic auxiliary metadata based on the diagnostic auxiliary metadata extracted or calculated during the real-time analysis of the identified key video frame.

[0098] Further, the diagnostic auxiliary metadata includes at least one of the identified positioning information, calculated biometric parameters, and analyzed confidence scores; the steps for determining the preset rendering position and style of the marked watermark content on the key video frame screen include:

[0099] The artificial intelligence system determines the preset rendering position of the associated marked watermark content on the key video frame screen according to the positioning information in the diagnostic auxiliary metadata;

[0100] The artificial intelligence system determines the style of the marked watermark content on the key video frame screen according to the positioning information, biometric parameters, and confidence score in the diagnostic auxiliary metadata.

[0101] Specifically, after identifying the key video frame, the artificial intelligence system obtains the original pixel data of the key video frame (usually stored in the internal frame buffer), and then the artificial intelligence system can generate the superimposed marked watermark content according to preset rules and / or the diagnostic auxiliary metadata extracted or calculated during the real-time analysis of the identified key video frame. Specifically, the artificial intelligence system generates marked watermark content with fixed content according to preset rules, where the fixed content can be fixed text or graphic identifiers such as hospital logos, department names, AI-assisted recognition identifiers (such as "AI identified standardplane", "AI Auto-Capture", etc.), timestamps, etc., used to indicate that the key video frame is automatically recognized by the artificial intelligence system. Or the artificial intelligence system generates dynamic marked watermark content based on the diagnostic auxiliary metadata extracted or calculated during the real-time analysis of the key video frame. These diagnostic auxiliary metadata can include at least one of the identified positioning information, calculated biometric parameters, and analyzed confidence scores.

[0102] Specifically, the identified location information usually directly comes from the output of the lesion detection or segmentation model. For the detection model, it directly outputs the bounding box coordinates of the candidate region (e.g., the pixel coordinates of the upper left corner and the lower right corner), as well as the corresponding class label and confidence. At this time, the artificial intelligence system directly extracts these outputs as location information metadata, and at the same time, it may need to perform inverse normalization or scaling of the coordinates to match the original image or display resolution. For the segmentation model, it outputs a pixel-level probability map or segmentation mask (a binary map representing the pixels within the lesion or structural area). At this time, the artificial intelligence system needs to perform post-processing to extract the location information. The post-processing specifically includes thresholding, connected component analysis, and calculating the bounding box / contour. Specifically, thresholding applies a threshold (e.g., 0.5) to the probability map to obtain a binary segmentation mask. Connected component analysis is to find the independent connected regions in the mask (representing different lesion or structural instances). Calculating the bounding box / contour is to calculate the minimum bounding rectangle (Bounding Box) or extract its exact contour (a series of continuous boundary pixel point coordinates) for each connected region. At this time, the calculated bounding box or contour coordinates are the location information metadata.

[0103] The calculated biometric parameters are usually obtained through further image processing or geometric calculations after the above specific structures or regions are identified. The biometric parameters specifically include distance, area, volume, and blood flow parameters. The distance measurement is mainly the Euclidean distance between two feature points calculated by the artificial intelligence system after identifying the two feature points that need to measure the distance. The area measurement is mainly the product of the number of pixels within the mask and the area represented by each pixel calculated by the artificial intelligence system after obtaining the segmentation mask of a closed region through the segmentation model. The volume estimation is mainly that the artificial intelligence system uses the corresponding geometric formula for estimation after detecting the major axis and minor axis for some regular shapes (such as ellipsoids). For irregular shapes, the artificial intelligence system estimates the volume through voxel accumulation (calculating the number of voxels within the segmentation mask multiplied by the voxel volume) or a method based on two-dimensional area integration (such as Simpson's rule). The blood flow parameter calculation is that if the key image frame is a Doppler spectrogram, the artificial intelligence system first automatically depicts the spectral envelope (finding the maximum flow velocity at each time point), and then based on the depicted envelope, calculates the peak systolic velocity (PSV), end-diastolic velocity (EDV), resistance index (RI), pulsatility index (PI), or time-averaged maximum velocity (TAMAX), etc., and uses these values to calculate the resistance index and pulsatility index. Therefore, the artificial intelligence system can calculate the major diameter and minor diameter according to the bounding box; can calculate the area, perimeter, circularity, or volume according to the segmentation mask; if specific measurement line segments are identified (such as the two lines for measuring the intima-media thickness IMT of the carotid artery), the average or maximum distance between the line segments can be calculated.

[0104] The confidence scores analyzed therein are usually part of the output of the above model itself, reflecting the degree of confidence of the model in its prediction results. For example, when identifying standard sections, the deep learning model directly outputs the probabilities of each category. At this time, the confidence score is usually the probability value of the predicted category by the model. When detecting and locating suspicious lesion areas, the detection model usually outputs a confidence score for each detected bounding box, indicating the possibility that the box contains the target and the possibility of belonging to a specific category. The segmentation model usually outputs a probability map of each pixel belonging to the foreground, and the regional-level confidence score is obtained by calculating the average or maximum value of the pixel probabilities within the segmented area.

[0105] Furthermore, the artificial intelligence system determines the preset rendering position of the marked watermark content on the key image frame according to the preset rules and / or the positioning information in the diagnostic assistance metadata. For example, for the fixed content described above, the artificial intelligence system can determine that the preset rendering position is a preset fixed area on the screen, such as one of the four corners of the image, or the information bar area at the top or bottom. For the dynamic content related to the diagnostic assistance metadata described above, the artificial intelligence system can determine the preset rendering position of the associated marked watermark content on the key image frame according to the positioning information in the diagnostic assistance metadata. For example, based on the positioning information in the diagnostic assistance metadata (such as bounding box coordinates, measurement lines, segmentation mask areas, key point positions), the spatial rendering position of the associated marked watermark content (such as graphic elements or text labels) on the key image frame is determined to ensure that the mark is adjacent to or superimposed on the corresponding image feature area. For example, the bounding box is directly drawn around the lesion, and the measurement value is marked next to the measurement line.

[0106] Furthermore, the artificial intelligence system determines the style of the marked watermark content on the key image frame according to the preset rules or the positioning information, biometric parameters, and confidence scores in the diagnostic assistance metadata. For example, for the fixed content described above, the artificial intelligence system determines the style to be the preset style on the screen, where the style includes font, font size, color, line type, line width, background, transparency, etc. For the dynamic content related to the diagnostic assistance metadata described above, the artificial intelligence system can determine the style of the marked watermark content on the key image frame according to the positioning information, biometric parameters, and confidence scores in the diagnostic assistance metadata. An artificial intelligence system internally maintains a preset style rule library, which defines what visual rendering styles should correspond to different types of metadata or different numerical ranges / levels of the same type of metadata. For example, this rule library can be:

[0107] Type-based mapping: For example, it is stipulated that all bounding boxes use red thin lines, all measurement values use white small fonts, and all marks with a confidence level lower than 0.8 use semi-transparent gray, etc.

[0108] Or it is a numerical / level-based conditional rule: for example, if the lesion diameter > 2.0 cm, the bounding box uses bold red lines. If the malignant risk score > 0.7, the label text uses red font. If the standard section confidence < 0.85, the section name label uses gray or has a question mark. If the image quality level is "excellent", a small green tick icon can be added in the corner.

[0109] At this time, the artificial intelligence system traverses each item in the obtained set of diagnostic assistance metadata. For each item of metadata, it parses its type (whether it is location information, biometric parameter, or confidence score) and specific content / value. Using the parsed type and value / content, it queries or matches in the style rule library to find the most suitable style rule. Then, according to the matched style rule, it determines the final rendering style parameters for the current metadata.

[0110] Furthermore, after the artificial intelligence system determines the rendering position and the determined rendering style of each marked watermark content, it calls the underlying graphics drawing API (such as OpenGL, DirectX, or the drawing functions of a specific image processing library). Using the determined position, style (font, font size, color, line type, line width, background, transparency, etc.), it draws or blends the corresponding text, graphics (bounding box, line, filled area) onto the pixel data of the key image frame to generate the modified key image frame data with the marked watermark content. That is to say, the artificial intelligence system draws or blends the generated marked watermark content onto the pixel data of the key image frame according to the determined position and style to generate the modified key image frame data with the marked watermark content.

[0111] Furthermore, when the artificial intelligence system controls the output video signal of the image to pause updating and freeze on the key image frame within a preset time, it also sends a trigger signal to the ultrasound workstation through the control transmission line to notify the ultrasound workstation that the image displayed in the current video stream is the key image frame that needs to be saved. In the embodiment of the present invention, the trigger signal is set as a pre-configured keyboard shortcut instruction.

[0112] Step S30, the ultrasound workstation receives the image video signal from the artificial intelligence system in real time through the second video transmission line, and when it receives the trigger signal through the control transmission line, it executes its built-in preset image saving function according to the trigger signal to automatically retain the currently received image data corresponding to the frozen key image frame;

[0113] Among them, in an embodiment of the present invention, the second image acquisition card of the ultrasound workstation continuously receives the video signal output from the artificial intelligence system through the connected second video transmission line. The acquisition card driver of the ultrasound workstation decodes the received signal into digital video frame data and renders it in real time on the display. When the artificial intelligence system controls its output video signal to freeze at a key image frame, the signal transmitted through the second video transmission line also becomes a static picture accordingly. Therefore, the video area from the artificial intelligence system displayed on the ultrasound workstation is also frozen at this key image frame correspondingly.

[0114] Furthermore, the second control transmission line interface of the ultrasound workstation is connected to the first control transmission line interface of the artificial intelligence system through the control transmission line. The operating system or dedicated driver of the ultrasound workstation monitors this port and waits for signal input. When the artificial intelligence system sends a trigger signal through the control transmission line, the ultrasound workstation will receive this trigger signal, recognize it as a standard keyboard event, and pass the keyboard event to the workstation software that is currently active or has registered this global shortcut key. The workstation software has a preset internal setting to bind the shortcut key to its built-in "image saving" function module. When the workstation software receives the keyboard event, it automatically activates the image saving function and immediately sends an instruction to the second image acquisition card driver or API to capture the single-frame image data that is currently being received and displayed through the second video transmission line. Since the artificial intelligence system is in a state of pausing and freezing the output of the key image frame at this time, the key image frame recognized and locked by the artificial intelligence system is exactly what the ultrasound workstation captures. Furthermore, when the artificial intelligence system draws or fuses the required superimposed marked watermark content on the original pixel data of the key image frame, the key image frame captured by the ultrasound workstation is the one with the marked watermark content added, which is convenient for distinguishing whether it is manually collected by the doctor or automatically collected by the artificial intelligence system, improves work efficiency, reduces the operations of the doctor manually repeating the collection or searching for images, and at the same time, adding the marked watermark in the key image frame greatly facilitates the doctor's subsequent report writing and review.

[0115] In summary, in the above-described embodiments of the present invention, the ultrasonic image retention method ensures that the artificial intelligence system can directly and real-time obtain the image data stream from the ultrasonic device by connecting between the artificial intelligence system and the ultrasonic device through the first video transmission line. By connecting between the artificial intelligence system and the ultrasonic workstation through the second video transmission line and the control transmission line, independent channels for the artificial intelligence system to transmit image content and send control instructions to the ultrasonic workstation are respectively established. By continuously outputting the image signal through the second video transmission line, the ultrasonic workstation can display the picture synchronized with the processing of the artificial intelligence system in real time. When the artificial intelligence system recognizes a key image frame, it actively controls the video output to pause and freeze on the key image frame, and sends a trigger signal to the ultrasonic workstation through the control transmission line, so as to trigger the ultrasonic workstation to execute its built-in native image saving function, bypassing the existing complex software interface dependencies. Moreover, when receiving the trigger signal, the ultrasonic workstation can accurately capture the frozen key image frame. At the same time, by superimposing the diagnostic auxiliary metadata (such as measurement values, positioning information, confidence levels, etc.) analyzed by the artificial intelligence system on the automatically saved key image frame, the retained image itself carries preliminary intelligent analysis results, providing richer and more direct information support for subsequent diagnostic decision-making and report writing, and solving the problems of insufficient real-time performance, accuracy, and imperceptibility in the existing ultrasonic image retention.

[0116] Embodiment 2

[0117] Please refer to Figure 2 , which is a schematic structural diagram of an ultrasonic image retention system provided by the second embodiment of the present invention. For the sake of convenience of description, only the parts related to the embodiments of the present invention are shown. The system includes an ultrasonic device, an artificial intelligence system, and an ultrasonic workstation. The artificial intelligence system is connected to the ultrasonic device through a first video transmission line, and is connected to the ultrasonic workstation through a second video transmission line and a control transmission line;

[0118] The artificial intelligence system is used to real-time obtain the ultrasonic image from the ultrasonic device through the first video transmission line and generate an image video signal including the ultrasonic image, and at the same time continuously output the image video signal to the ultrasonic workstation through the second video transmission line;

[0119] The artificial intelligence system is further used to real-time analyze the obtained ultrasonic image, and when a key image frame is recognized from the ultrasonic image according to a preset rule, control the output image video signal to pause updating and freeze on the key image frame within a preset time, and send a trigger signal to the ultrasonic workstation through the control transmission line;

[0120] The ultrasound workstation is used to receive the image video signal from the artificial intelligence system in real time through the second video transmission line, and when the trigger signal is received through the control transmission line, it executes the built-in preset image saving function according to the trigger signal, and automatically saves the currently received image data corresponding to the key image frame that is frozen.

[0121] Further, in an embodiment of the present invention, the artificial intelligence system is also used to process the ultrasound images obtained through the first video transmission line in real time, continuously store the processed image frames in the frame buffer, and attach corresponding preset identifiers to the data identified as key image frames;

[0122] The artificial intelligence system is also used to lock the data of the key image frame specified by the preset identifier obtained from the frame buffer within a preset time, encode it into an image video signal, and send it to the ultrasound workstation through the second video transmission line.

[0123] Further, in an embodiment of the present invention, the artificial intelligence system is also used to obtain the original pixel data of the identified key image frame;

[0124] The artificial intelligence system is also used to generate the marked watermark content to be superimposed, and determine the preset rendering position and style of the marked watermark content on the key image frame screen;

[0125] The artificial intelligence system is also used to draw or fuse the generated marked watermark content onto the original pixel data of the obtained key image frame according to the determined preset rendering position and style, and generate the modified key image frame data with the marked watermark content.

[0126] Further, in an embodiment of the present invention, the artificial intelligence system is also used to generate the marked watermark content including fixed text or graphic identifiers according to preset rules;

[0127] The artificial intelligence system is also used to generate the marked watermark content including diagnostic auxiliary metadata according to the diagnostic auxiliary metadata extracted or calculated during the real-time analysis of the identified key image frame.

[0128] Further, in an embodiment of the present invention, the diagnostic auxiliary metadata includes at least one of the identified positioning information, calculated biometric parameters, and analyzed confidence scores;

[0129] The artificial intelligence system is also used to determine the preset rendering position of the associated marked watermark content on the key image frame screen according to the positioning information in the diagnostic auxiliary metadata;

[0130] The artificial intelligence system is also used to determine the style of the marked watermark content on the key image frame according to the positioning information, biometric parameters and confidence score in the diagnostic assistance metadata.

[0131] Further, in an embodiment of the present invention, the artificial intelligence system is also used to perform real-time analysis on the ultrasound image by using a deep learning model, and identify the image frame containing a predefined standard section analyzed as a key image frame;

[0132] The artificial intelligence system is also used to detect and locate the suspicious lesion area in the ultrasound image by using a lesion detection algorithm, and identify the image frame containing the suspicious lesion that meets the preset lesion screening conditions as a key image frame.

[0133] Further, in an embodiment of the present invention, the artificial intelligence system is also used to preprocess each image frame in the ultrasound image in real time, and input each preprocessed image frame into a pre-trained deep learning model for forward propagation calculation;

[0134] The artificial intelligence system is also used to obtain the classification result of each image frame output by the deep learning model and the corresponding confidence score;

[0135] When the classification result of the target image frame is a predefined standard section and the confidence score is higher than the recognition threshold of the predefined standard section, the artificial intelligence system is also used to identify the analyzed target image frame containing the predefined standard section as a key image frame.

[0136] Further, in an embodiment of the present invention, the artificial intelligence system is also used to preprocess each image frame in the ultrasound image in real time, and input each preprocessed image frame into a pre-trained lesion detection or segmentation model for processing;

[0137] The artificial intelligence system is also used to obtain the bounding box or segmentation mask of the candidate lesion area in each image frame output by the lesion detection or segmentation model, as well as the category and confidence of each candidate lesion area;

[0138] The artificial intelligence system is also used to post-process each candidate lesion area detected in each image frame, and calculate the data features of each candidate lesion area, where the data features include size, aspect ratio and internal echo pattern;

[0139] The artificial intelligence system is also used to compare the data features calculated in each image frame with the preset lesion screening conditions, and the preset lesion screening conditions include the minimum size and a specific echo pattern;

[0140] When a suspicious lesion in the candidate lesion area of the target image frame meets the preset lesion screening conditions and the confidence level is higher than the preset lesion detection threshold, the artificial intelligence system is further configured to identify the target image frame including the suspicious lesion that meets the preset lesion screening conditions as a key image frame.

[0141] Further, in an embodiment of the present invention, the artificial intelligence system is further configured to perform image quality assessment on the identified key image frame to obtain a quality score or grade;

[0142] The artificial intelligence system is further configured to determine whether the quality score or grade of the key image frame meets a preset quality threshold;

[0143] If so, the artificial intelligence system is further configured to control the output video signal of the image to pause updating within a preset time and freeze on the key image frame.

[0144] The ultrasonic image retention system provided by the embodiment of the present invention has the same implementation principle and the same technical effects as those in the foregoing method embodiment. For the sake of brief description, for the parts not mentioned in the device embodiment, reference may be made to the corresponding content in the foregoing method embodiment.

[0145] In the description of this specification, the descriptions with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples", etc. mean that the specific features, structures, materials, or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials, or characteristics described may be combined in any one or more embodiments or examples in a suitable manner.

[0146] The above embodiments only represent several implementation manners of the present invention. The descriptions thereof are relatively specific and detailed, but should not be construed as a limitation on the scope of the present invention. It should be noted that for those of ordinary skill in the art, without departing from the concept of the present invention, several modifications and improvements can still be made, and these all belong to the protection scope of the present invention. Therefore, the protection scope of the present invention should be subject to the appended claims.

Claims

1. An ultrasonic image retention method, characterized in that, An ultrasound image retention system applied to an ultrasound device, an artificial intelligence system, and an ultrasound workstation. The artificial intelligence system is connected to the ultrasound device through a first video transmission line and is connected to the ultrasound workstation through a second video transmission line and a control transmission line. The method includes: The artificial intelligence system obtains ultrasound images from the ultrasound device in real time through the first video transmission line and generates an image video signal containing the ultrasound images, and simultaneously continuously outputs the image video signal to the ultrasound workstation through the second video transmission line; The artificial intelligence system analyzes the ultrasound images obtained in real time, and when a key image frame is identified from the ultrasound images according to a preset rule, it controls the output image video signal to pause updating within a preset time and freeze on the key image frame, and sends a trigger signal to the ultrasound workstation through the control transmission line; The ultrasound workstation receives the image video signal from the artificial intelligence system in real time through the second video transmission line, and when receiving the trigger signal through the control transmission line, executes its built-in preset image saving function according to the trigger signal, and automatically retains the currently received image data corresponding to the key image frame that is frozen; The step of identifying a key image frame from the ultrasound images according to a preset rule includes: The artificial intelligence system uses a deep learning model to analyze the ultrasound images in real time, and identifies the image frames containing predefined standard sections as key image frames; The artificial intelligence system uses a lesion detection algorithm to detect and locate suspicious lesion areas in the ultrasound images, and identifies the image frames containing suspicious lesions that meet the preset lesion screening conditions as key image frames, including: The artificial intelligence system preprocesses each frame of the ultrasound images in real time and inputs each preprocessed frame of the ultrasound images into a pre-trained lesion detection or segmentation model for processing; The artificial intelligence system obtains the bounding boxes or segmentation masks of the candidate lesion areas in each frame of the ultrasound images output by the lesion detection or segmentation model, as well as the category and confidence level of each candidate lesion area; The artificial intelligence system post-processes each candidate lesion area detected in each frame of the ultrasound images, calculates the data features of each candidate lesion area, and the data features include size, aspect ratio, and internal echo pattern; The artificial intelligence system compares the data features calculated in each frame of the ultrasound images with the preset lesion screening conditions, and the preset lesion screening conditions include a minimum size and a specific echo pattern; When there is a suspicious lesion in the candidate lesion area of the target image frame that meets the preset lesion screening conditions and the confidence level is higher than the preset lesion detection threshold, the artificial intelligence system identifies the target image frame containing the suspicious lesion that meets the preset lesion screening conditions as a key image frame.

2. The ultrasonic image retention method according to claim 1, characterized in that The step of controlling the output image video signal to pause and freeze the key image frame within a preset time includes: The artificial intelligence system processes the ultrasonic images obtained through the first video transmission line in real time, continuously stores the processed image frames in the frame buffer, and attaches corresponding preset identifiers to the data identified as key image frames; The artificial intelligence system locks the data of the key image frames specified by the preset identifier obtained from the frame buffer within a preset time, encodes it into an image video signal, and sends it to the ultrasonic workstation through the second video transmission line.

3. The ultrasonic image retention method according to claim 2, wherein Before or during the step of the artificial intelligence system locking the data of the key image frames specified by the preset identifier obtained from the frame buffer within a preset time, the method further includes: The artificial intelligence system obtains the original pixel data of the identified key image frames; The artificial intelligence system generates the marker watermark content to be superimposed, and determines the preset rendering position and style of the marker watermark content on the key image frame screen; The artificial intelligence system draws or fuses the generated marker watermark content onto the original pixel data of the obtained key image frames according to the determined preset rendering position and style, generating modified key image frame data with the marker watermark content.

4. The ultrasonic image retention method according to claim 3, characterized in that, The step of the artificial intelligence system generating the marker watermark content to be superimposed includes: The artificial intelligence system generates marker watermark content including fixed text or graphic identifiers according to preset rules; and / or The artificial intelligence system generates marker watermark content including diagnostic auxiliary metadata according to the diagnostic auxiliary metadata extracted or calculated during the real-time analysis of the identified key image frames.

5. The ultrasonic imaging retention method according to claim 4, wherein The diagnostic auxiliary metadata includes at least one of the identified positioning information, calculated biometric parameters, and analyzed confidence scores; The step of determining the preset rendering position and style of the marker watermark content on the key image frame screen includes: The artificial intelligence system determines the preset rendering position of the associated marker watermark content on the key image frame screen according to the positioning information in the diagnostic auxiliary metadata; The artificial intelligence system determines the style of the marker watermark content on the key image frame screen according to the positioning information, biometric parameters, and confidence score in the diagnostic auxiliary metadata.

6. The ultrasonic image retention method according to claim 1, wherein, The step of the artificial intelligence system using a deep learning model to perform real-time analysis on the ultrasonic images and identifying the image frames containing predefined standard sections analyzed as key image frames includes: The artificial intelligence system preprocesses each image frame in the ultrasonic images in real time, and inputs each preprocessed image frame into a pre-trained deep learning model for forward propagation calculation; The artificial intelligence system obtains the classification result of each image frame output by the deep learning model and the corresponding confidence score; If the classification result of the target image frame is a predefined standard section and the confidence score is higher than the recognition threshold of the predefined standard section, the artificial intelligence system identifies the target image frame containing the predefined standard section analyzed as a key image frame.

7. The ultrasonic image retention method according to claim 1, wherein The step of controlling the output image video signal to pause updating and freeze on the key image frame within a preset time includes: The artificial intelligence system performs image quality assessment on the identified key image frames to obtain a quality score or grade; The artificial intelligence system determines whether the quality score or grade of the key image frames meets a preset quality threshold; If so, the artificial intelligence system controls the output video signal of the image to pause updating within a preset time and freeze on the key image frame.

8. An ultrasound image retention system, characterized in that, Implement the ultrasonic image retention method according to any one of claims 1-7. The system includes an ultrasonic device, an artificial intelligence system, and an ultrasonic workstation. The artificial intelligence system is connected to the ultrasonic device through a first video transmission line, and is connected to the ultrasonic workstation through a second video transmission line and a control transmission line; The artificial intelligence system is used to obtain ultrasonic images from the ultrasonic device in real time through the first video transmission line and generate a video signal of the image containing the ultrasonic images, and at the same time continuously output the video signal of the image to the ultrasonic workstation through the second video transmission line; The artificial intelligence system is also used to analyze the ultrasonic images obtained in real time, and when a key image frame is identified from the ultrasonic images according to a preset rule, control the output video signal of the image to pause updating within a preset time and freeze on the key image frame, and send a trigger signal to the ultrasonic workstation through the control transmission line; The ultrasonic workstation is used to receive the video signal of the image from the artificial intelligence system in real time through the second video transmission line, and when receiving the trigger signal through the control transmission line, execute the built-in preset image saving function according to the trigger signal, and automatically retain the currently received image data corresponding to the key image frame that is frozen.

Images