Pseotoxic cardiomyopathy auxiliary detection system based on image recognition

Abstract

The invention relates to the technical field of image recognition, in particular to an image-recognition-based septicotoxic cardiomyopathy auxiliary detection system, which comprises an image acquisition module for acquiring a left ventricular ultrasound image initial sequence based on a preset frame rate; receiving a frame rate correction instruction to recollect the ultrasonic diagnosis image sequence; the image checking module is used for performing periodic verification and neighborhood consistency checking on the basis of the ultrasonic image initial sequence; the optimization correction module is used for calculating the existence probability of a peak frame based on the ultrasonic image initial sequence so as to judge whether the sequence needs to be corrected or not; calculating a correction coefficient based on the historical standard sequence to determine a correction frame rate and send a correction frame rate instruction; the probability generation module is used for generating a myocardial probability graph sequence based on the left ventricular ultrasound image sequence; and the ROI segmentation module is used for generating a myocardial ROI mask image sequence based on the myocardial probability graph sequence. According to the method, the precision of the pyotoxic cardiomyopathy in the ultrasonic image analysis and detection process is improved.

Description

Technical Field

[0001] This invention relates to the field of image recognition technology, and in particular to an image recognition-based auxiliary detection system for septic cardiomyopathy. Background Technology

[0002] With the widespread adoption of bedside ultrasound technology in critical care medicine, it plays an increasingly important role in the rapid diagnosis and dynamic monitoring of septic cardiomyopathy. Two-dimensional speckle tracking imaging technology can quantitatively assess cardiac function by analyzing myocardial motion. Its core parameter, global longitudinal strain, has been proven to be of great value in the early detection of myocardial injury and prognosis prediction, providing crucial evidence for clinical practice.

[0003] However, current technologies heavily rely on operators manually selecting standard sections and key frames of the cardiac cycle, a process that is both time-consuming and introduces significant subjective variability. Furthermore, conventional ultrasound image analysis procedures lack automated verification and calibration of the raw sequence quality; a fixed acquisition frame rate may fail to accurately capture critical physiological moments; and traditional segmentation methods are sensitive to image quality fluctuations. These limitations result in high uncertainty and poor repeatability in the final measurement of myocardial function parameters, hindering their widespread application in precision medicine.

[0004] Chinese Patent Publication No. CN111093519A discloses an ultrasound image processing apparatus including an image processor device. The ultrasound image processing apparatus is adapted to: receive multiple ultrasound images, each ultrasound image imaging an invasive medical device relative to an anatomical feature of interest during a specific phase of a cardiac cycle, the anatomical feature of interest having different shapes at different stages of the cardiac cycle, the multiple ultrasound images covering at least two cardiac cycles, during which the invasive medical device is displaced relative to the anatomical feature of interest; compile multiple sets of the ultrasound images, wherein the ultrasound images in each set belong to the same phase of the cardiac cycle; and for each set: determine the displacement of the invasive medical device relative to the anatomical feature of interest between at least two ultrasound images in the set; and based on the determined displacement of the invasive medical device, generate an enhanced ultrasound image from one of the at least two ultrasound images by removing a shadowed area on the anatomical feature of interest caused by the invasive medical device from the ultrasound images. An ultrasound imaging system including such an ultrasound image processing device is also disclosed, as well as a method for implementing such an ultrasound image processing device and a computer program product that facilitates the implementation of such a method.

[0005] Therefore, the following problems exist in the existing technology:

[0006] The problem of inaccurate identification of myocardial images during the cardiac cycle is caused by a fixed acquisition frame rate and a lack of automated quality control. Summary of the Invention

[0007] Therefore, the present invention provides an image recognition-based auxiliary detection system for septic cardiomyopathy to overcome the problem in the prior art that the fixed acquisition frame rate and lack of automated quality control make it impossible to accurately identify myocardial images during the cardiac cycle.

[0008] To achieve the above objectives, the present invention provides an image recognition-based auxiliary detection system for septic cardiomyopathy, comprising: an image acquisition module for acquiring left ventricular ultrasound images, acquiring left ventricular ultrasound images of several cardiac cycles based on a preset frame rate to generate an initial sequence of ultrasound images, and acquiring left ventricular ultrasound images of several cardiac cycles at a specified frame rate according to a received corrected frame rate instruction to generate a sequence of ultrasound diagnostic images, wherein the image start frame and end frame are determined based on an electrocardiogram (ECG) signal.

[0009] The image inspection module is used to perform periodic verification based on the ECG gate signal corresponding to the initial sequence of the ultrasound image to determine whether the period of the initial sequence of the ultrasound image meets the standard. Based on the initial sequence of the ultrasound image that meets the standard, a neighborhood consistency check is performed according to the left ventricular area in the image to determine whether the neighborhood consistency of the initial sequence of the ultrasound image meets the expectation.

[0010] The optimization and correction module is used to calculate the probability of the presence of the peak frame of the left ventricular cavity area based on the changing trend of the start frame image, the end frame image and the corresponding next frame image in the initial sequence of ultrasound images that meet the expectations. Based on the probability of presence, it determines whether the initial sequence of ultrasound images needs to be corrected. Based on the change of the peak frame area of ​​the left ventricular cavity area in historical standard sequences, it calculates the correction coefficient of the current initial sequence of ultrasound images to determine the corrected acquisition frame rate. Based on the corrected acquisition frame rate, it sends a correction frame rate command to the image acquisition module.

[0011] The probability generation module is used to calculate the probability value of each pixel belonging to the myocardium based on each frame of the ultrasound diagnostic image sequence, so as to generate a myocardial probability map sequence.

[0012] The ROI segmentation module is used to extract key segmentation parameters based on the myocardial probability map sequence to determine the optimal segmentation pixel label configuration, so as to generate a myocardial ROI mask image sequence for auxiliary detection.

[0013] The key segmentation parameters include the probability value of each pixel, spatial neighborhood consistency, and temporal neighborhood smoothness.

[0014] Furthermore, the optimization and correction module determines the absolute value of the slope of the left ventricular cavity area change trend based on the absolute value of the rate of change of the left ventricular cavity area between the start frame image and the end frame image in the expected initial sequence of ultrasound images and the corresponding next frame image, so as to determine the probability of the existence of the peak frame of the left ventricular cavity area based on the absolute value of the slope and a preset absolute value of the slope threshold.

[0015] Furthermore, the optimization and correction module determines that the initial sequence of the ultrasound image needs to be corrected based on the presence probability being greater than or equal to a preset presence probability threshold.

[0016] Furthermore, the optimization and correction module determines the correction coefficient of the current ultrasound image initial sequence based on the ratio of the historical baseline value to the difference in left ventricular cavity area between the starting frame image or the ending frame image of the initial ultrasound image sequence and the corresponding next frame image.

[0017] The historical baseline value is determined based on the average difference in left ventricular cavity area between the peak frame image of the historical standard sequence and the next frame image.

[0018] Furthermore, the optimization and correction module determines the corrected acquisition frame rate based on the product of the correction coefficient and the preset frame rate, and sends a corrected frame rate command to the image acquisition module to trigger the image acquisition module to re-acquire left ventricular ultrasound images for several cardiac cycles.

[0019] Furthermore, the image inspection module calculates the current cycle length based on the initial sequence of the ultrasound image and the corresponding ECG gate signal, and compares the current cycle length with the preset standard cycle length range to determine whether the cycle of the initial sequence of the ultrasound image meets the standard.

[0020] Furthermore, the image inspection module generates an inter-frame left ventricular cavity area difference sequence based on the difference in left ventricular cavity area between each frame in the initial sequence of ultrasound images conforming to the standard and the adjacent frames. Based on the sign change pattern of the difference sequence, it determines whether the sign change conforms to the corresponding expectation of end-diastole and end-systole.

[0021] Furthermore, the probability generation module uses a Bayesian neural network to perform inference based on the left ventricular ultrasound image to calculate the probability value of each pixel in each frame of the left ventricular ultrasound image belonging to the myocardium, so as to generate a myocardial probability map sequence corresponding to each frame of the image.

[0022] Furthermore, it also includes a multi-source data module for acquiring multi-source clinical data and for performing auxiliary analysis on myocardial ROI mask image sequences based on the clinical data; wherein the clinical data includes myocardial injury biomarkers, inflammatory indicators, and hemodynamic parameters.

[0023] Furthermore, the acquired ultrasound image sequences also include images of the entire heart, including the right ventricle; a corresponding myocardial ROI mask image sequence is generated based on the acquired image sequences of one or more heart chambers.

[0024] Compared with existing technologies, the advantages of this invention lie in its ability to achieve intelligent processing of the entire process from ultrasound image acquisition to myocardial region segmentation by constructing a fully automated image recognition system. This system ensures the controllability of quality and the reliability of results at each processing stage through the collaborative work and quality control of multiple modules. This invention achieves intelligent screening of raw sequences through an automatic image quality inspection mechanism, periodic verification, and neighborhood consistency checks, ensuring data quality from the source. Through a dynamic frame rate optimization and correction mechanism, the system can adaptively adjust acquisition parameters based on sequence features, significantly improving the accuracy of capturing key physiological moments. Furthermore, through an optimized segmentation algorithm that integrates spatiotemporal constraints, it can accurately delineate myocardial boundaries, providing high-quality input for clinical strain analysis. Therefore, it comprehensively improves the accuracy, efficiency, and standardization of ultrasound image analysis and detection in septic cardiomyopathy.

[0025] Furthermore, this invention transforms the traditionally experience-dependent peak frame localization problem into a computable probabilistic problem by quantitatively analyzing the area change trend between keyframes and adjacent frames, thus achieving objectivity and standardization in the identification of key physiological phases. By introducing a precise calculation of the absolute value of the slope and a threshold comparison mechanism, it can sensitively capture subtle deviations between the currently set keyframe and the actual peak value, providing a reliable quantitative basis for subsequent correction decisions. This mathematical model-based evaluation method significantly reduces reliance on the operator's personal experience and enhances adaptability to different image qualities and patient physiological conditions, thereby fundamentally improving the accuracy and robustness of the entire system in identifying key physiological moments and laying a solid foundation for obtaining high-quality cardiac cycle sequences.

[0026] Furthermore, by setting a reasonable probability threshold, this invention effectively avoids overcorrection caused by image noise or normal physiological variations while ensuring sensitive detection of abnormal sequences. This ensures efficient use of system resources and achieves an optimal balance between identifying sequences that truly require correction and tolerating normal fluctuations. When the probability exceeds this threshold, the system automatically triggers the correction process. This mechanism ensures that only sequences with significant quality problems enter the correction stage, guaranteeing both the quality of the output sequences and maintaining the overall operational efficiency of the system. This intelligent decision-making mechanism significantly improves the system's practicality and clinical applicability, enabling it to adapt to the actual needs of different medical environments.

[0027] Furthermore, this invention provides a clear reference benchmark for calculating the correction coefficient by systematically and quantitatively comparing the performance of the current sequence with historical standard sequences, ensuring the scientific rigor and reliability of the correction process. By calculating the ratio of the mean standard deviation to the actual measured value, the degree of deviation in keyframe localization of the current sequence is intuitively reflected, providing a reliable mathematical basis for precise frame rate adjustment. This correction model, built upon a large amount of standard data, can effectively adapt to different population conditions, significantly improving the system's adaptability and accuracy in various clinical scenarios. Through this standardized correction method, the system can learn from historical experience and continuously optimize its performance, providing strong technical support for obtaining high-quality cardiac ultrasound sequences.

[0028] Furthermore, this invention, through an innovative mechanism of dynamically adjusting the acquisition frame rate, enables the system to adapt to different image quality conditions, effectively solving the problem of insufficient temporal resolution that may result from fixed frame rate acquisition. This closed-loop feedback control mechanism, which adjusts acquisition parameters based on real-time calculated correction coefficients, can improve the temporal resolution quality of the original image sequence from the source, ensuring more accurate capture of key physiological moments during new acquisition processes. This adaptive acquisition strategy lays a solid foundation for the accuracy of all subsequent analysis steps.

[0029] Furthermore, this invention, through an automated periodicity verification mechanism, can effectively filter out invalid data caused by acquisition errors or equipment malfunctions at the source, ensuring the basic quality of the input sequence. This verification method based on a reasonable range can reliably identify abnormal sequences that do not conform to objective laws, providing a qualified time series basis for subsequent analysis and processing.

[0030] Furthermore, this invention improves the objectivity and accuracy of detection by analyzing the differential characteristics of left ventricular cavity area changes between consecutive frames. This identification method based on differential sequence sign change patterns, through this quantitative analysis, enables the system to provide consistent and repeatable phase identification results, providing a reliable time reference for subsequent strain analysis and functional assessment.

[0031] Furthermore, through the multiple inference mechanism of Bayesian neural networks, this invention can not only provide a probability estimate of whether each pixel belongs to the myocardium, but also quantify the uncertainty of the prediction results. This provides richer information for subsequent processing steps, enabling the system to clearly distinguish between regions with strong determinism and regions with high uncertainty, and providing an important basis for subsequent quality control and result interpretation.

[0032] Furthermore, this invention constructs a multi-dimensional parameter system that comprehensively describes the morphological features of myocardium by comprehensively utilizing the probability values ​​of each pixel, spatial neighborhood consistency, and temporal neighborhood smoothness, providing a rich and reliable data foundation for accurate segmentation. Specifically, the calculation of spatial neighborhood consistency ensures the smoothness and rationality of the segmentation boundary in the spatial dimension; the constraint of temporal neighborhood smoothness guarantees the continuity of the segmentation results in the temporal dimension. This synergistic effect of multi-dimensional features enables the system to generate segmentation results that maintain a high degree of consistency in both the spatiotemporal dimensions.

[0033] Furthermore, this invention constructs a unified energy optimization framework using a temporal conditional random field model, integrating pixel-level probabilistic information, spatial constraints, and temporal smoothness requirements into a single mathematical model, achieving a globally optimal balance of segmentation results in both spatial and temporal dimensions. This energy-minimization-based optimization method simultaneously considers local features and global consistency, yielding precise and spatiotemporally continuous myocardial contours. These high-quality segmentation results provide high-quality input for clinical strain analysis, thereby improving the overall accuracy, efficiency, and standardization of ultrasound image analysis and detection of septic cardiomyopathy. Attached Figure Description

[0034] Figure 1 This is a module connection diagram of the image recognition-based auxiliary detection system for septic cardiomyopathy according to an embodiment of the present invention;

[0035] Figure 2 A flowchart illustrating the workflow for calculating the probability of the presence of peak frames of left ventricular cavity area in an embodiment of the present invention;

[0036] Figure 3 This is a logic diagram for determining whether the initial sequence of the ultrasound image needs to be corrected according to an embodiment of the present invention;

[0037] Figure 4 This is a flowchart illustrating the workflow for determining the correction coefficients of the initial sequence of the current ultrasound image in an embodiment of the present invention. Detailed Implementation

[0038] To make the objectives and advantages of the present invention clearer, the present invention will be further described below with reference to embodiments; it should be understood that the specific embodiments described herein are merely for explaining the present invention and are not intended to limit the present invention.

[0039] Preferred embodiments of the present invention will now be described with reference to the accompanying drawings. Those skilled in the art should understand that these embodiments are merely illustrative of the technical principles of the present invention and are not intended to limit the scope of protection of the present invention.

[0040] It should be noted that in the description of this invention, the terms "upper", "lower", "left", "right", "inner", "outer", etc., which indicate directions or positional relationships, are based on the directions or positional relationships shown in the accompanying drawings. This is only for the convenience of description and is not intended to indicate or imply that the device or element must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, it should not be construed as a limitation of this invention.

[0041] Furthermore, it should be noted that, in the description of this invention, unless otherwise explicitly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0042] The following are relevant explanations in this application to facilitate a better understanding of the present invention:

[0043] Septic cardiomyopathy is usually diagnosed using echocardiography as an important tool.

[0044] Echocardiography is the most important imaging tool for the clinical diagnosis of septic cardiomyopathy. It is a key technology that uses ultrasound to non-invasively visualize the heart structure and assess cardiac function. This examination uses an ultrasound probe placed on the chest wall to emit high-frequency sound waves and receive echo signals, generating dynamic two-dimensional images of the heart in real time, providing objective evidence for the early identification of myocardial dysfunction.

[0045] The cardiac cycle refers to the complete mechanical activity of the heart from the start of one contraction to the start of the next, usually beginning with the appearance of the R wave on an electrocardiogram (ECG). A standard cardiac cycle includes systole and diastole. At end-diastole, ventricular filling is complete, and the ventricle is at its largest volume; at end-systole, ventricular ejection is complete, and the ventricle is at its smallest volume. End-diastole and end-systole are two key physiological phases. Accurate identification of these key phases is crucial in the echocardiographic assessment of septic cardiomyopathy because they are the benchmarks for calculating cardiac function parameters such as ejection fraction and strain. This invention provides a reliable basis for the objective diagnosis of septic cardiomyopathy by improving the accuracy of identifying these key phases.

[0046] Furthermore, the ultrasound images of the left ventricle acquired in this invention can also be replaced by ultrasound images of the right ventricle. The processing method is the same as that of the ultrasound images of the left ventricle, and will not be described in detail here.

[0047] Please see Figure 1The diagram shown is a module connection diagram of an image recognition-based auxiliary detection system for septic cardiomyopathy according to an embodiment of the present invention. The embodiment of the present invention provides an image recognition-based auxiliary detection system for septic cardiomyopathy, comprising:

[0048] The image acquisition module is used to acquire left ventricular ultrasound images, acquire left ventricular ultrasound images of several cardiac cycles based on a preset frame rate to generate an initial sequence of ultrasound images, and acquire left ventricular ultrasound images of several cardiac cycles at a specified frame rate according to a received corrected frame rate instruction to generate a sequence of ultrasound diagnostic images, wherein the start frame and end frame of the image are determined based on the electrocardiogram gate signal.

[0049] The image inspection module is used to perform periodic verification based on the ECG gate signal corresponding to the initial sequence of the ultrasound image to determine whether the period of the initial sequence of the ultrasound image meets the standard. Based on the initial sequence of the ultrasound image that meets the standard, a neighborhood consistency check is performed according to the left ventricular area in the image to determine whether the neighborhood consistency of the initial sequence of the ultrasound image meets the expectation.

[0050] The optimization and correction module is used to calculate the probability of the presence of the peak frame of the left ventricular cavity area based on the changing trend of the start frame image, the end frame image and the corresponding next frame image in the initial sequence of ultrasound images that meet the expectations. Based on the probability of presence, it determines whether the initial sequence of ultrasound images needs to be corrected. Based on the change of the peak frame area of ​​the left ventricular cavity area in historical standard sequences, it calculates the correction coefficient of the current initial sequence of ultrasound images to determine the corrected acquisition frame rate. Based on the corrected acquisition frame rate, it sends a correction frame rate command to the image acquisition module.

[0051] The probability generation module is used to calculate the probability value of each pixel belonging to the myocardium based on each frame of the ultrasound diagnostic image sequence, so as to generate a myocardial probability map sequence.

[0052] The ROI segmentation module is used to extract key segmentation parameters based on the myocardial probability map sequence to determine the optimal segmentation pixel label configuration, so as to generate a myocardial ROI mask image sequence for auxiliary detection.

[0053] The key segmentation parameters include the probability value of each pixel, spatial neighborhood consistency, and temporal neighborhood smoothness.

[0054] In this embodiment, firstly, the image acquisition module acquires an initial sequence of left ventricular ultrasound images based on a preset frame rate of 55-65Hz. This sequence automatically identifies the R wave to determine the start and end frames through ECG gating signals. Subsequently, upon receiving a frame rate correction instruction from the optimization and correction module, the module immediately re-acquires the optimized image sequence at the new frame rate.

[0055] The image inspection module performs periodic verification on the initial sequence, calculates the total sequence duration and verifies whether the total sequence duration is within the standard range of 0.33 to 1.5 seconds; for the verified sequences, a neighborhood consistency check is performed, and by calculating the inter-frame left ventricular cavity area difference sequence and analyzing its sign change pattern, it is determined whether it conforms to the identification rules of end-diastolic and end-systolic phases.

[0056] The optimization and correction module then calculates the probability of peak frames in the start and end frames of the inspected sequence. When the probability exceeds the threshold, it calculates the correction coefficient by comparing with historical standard sequences, thereby determining the correction frame rate and triggering re-acquisition.

[0057] The probability generation module uses a Bayesian neural network to perform multiple inferences based on the re-acquired image sequence, calculates the probability value of each pixel belonging to the myocardium, and generates a myocardial probability map sequence containing uncertainty information.

[0058] The ROI segmentation module extracts key parameters such as pixel probability values, spatial neighborhood consistency, and temporal neighborhood smoothness from the probability map sequence. It uses a temporal conditional random field model to construct an energy function and solves for the optimal pixel label configuration by minimizing the function, ultimately generating a high-precision myocardial ROI mask image sequence.

[0059] Specifically, the image acquisition module is used to acquire left ventricular ultrasound images at a set acquisition frame rate, which includes a preset frame rate and a corrected acquisition frame rate.

[0060] Specifically, the image inspection module calculates the current cycle length based on the initial sequence of the ultrasound image and the corresponding ECG gate signal, and compares the current cycle length with a preset standard cycle length range to determine whether the cycle of the initial sequence of the ultrasound image meets the standard.

[0061] In this embodiment, the image inspection module accurately calculates the current cycle duration by jointly analyzing the ultrasound image sequence and the corresponding ECG gating signal. The system first automatically identifies two consecutive R waves from the ECG signal as the start and end markers of the cardiac cycle, and accurately measures the time interval between these two R waves, using this as the baseline cycle duration based on the physiological signal. Simultaneously, the system calculates the cycle duration corresponding to the image sequence based on the total number of frames in the ultrasound image sequence and a preset frame rate. Subsequently, the module cross-validates these two cycle durations: ensuring that the calculated cycle duration falls within a reasonable physiological range of 0.33 seconds to 1.5 seconds, and verifying that the difference between the image cycle duration and the ECG signal cycle duration does not exceed a preset tolerance, preferably set to 50 milliseconds. Only when both conditions are met is the sequence deemed compliant; if either condition is not met, the system determines that the sequence acquisition is asynchronous or abnormal, automatically excludes the sequence, and triggers a re-acquisition process. This dual-validation mechanism fully utilizes the physiological timing accuracy of the ECG signal and the continuity of the image sequence, significantly improving the reliability of cycle duration determination.

[0062] Specifically, the image inspection module generates an inter-frame left ventricular cavity area difference sequence based on the difference in left ventricular cavity area between each frame in the initial sequence of ultrasound images conforming to the standard and the adjacent frames. Based on the sign change pattern of the difference sequence, it determines whether the sign change conforms to the corresponding expectation of end-diastole and end-systole.

[0063] In this embodiment, the image inspection module calculates the difference in left ventricular cavity area between each frame and its preceding and following adjacent frames for the sequence that has passed periodic verification, thereby generating an inter-frame left ventricular cavity area difference sequence. Then, it analyzes the sign change pattern of the difference sequence to check whether there are clear zero-crossing points from positive to negative and from negative to positive. These points correspond to candidate images of end-diastole and end-systole, respectively. If the sign change pattern is chaotic or does not conform to the expected pattern, it is determined that the neighborhood consistency of the sequence does not meet expectations. At this time, the system immediately terminates the subsequent processing of the current sequence and sends a re-acquisition command to the image acquisition module.

[0064] It is understandable that the left ventricular cavity area can be determined using any method available in the prior art, such as identifying the cavity region through traditional image segmentation methods like region growing and level sets. Alternatively, it can be determined based on the myocardial ROI mask image sequence generated after the specific implementation of the probability generation module and the ROI segmentation module described below.

[0065] Please see Figure 2 As shown, it is a flowchart of the process for calculating the probability of the presence of the peak frame of the left ventricular cavity area in an embodiment of the present invention.

[0066] Specifically, the optimization and correction module determines the absolute value of the slope of the left ventricular cavity area change trend based on the absolute value of the rate of change of the left ventricular cavity area between the start frame image and the end frame image in the expected initial sequence of ultrasound images and the corresponding next frame image. Based on the absolute value of the slope and a preset absolute value threshold, the probability of the existence of the peak frame of the left ventricular cavity area is determined.

[0067] In this embodiment, the optimization and correction module first extracts the start frame and end frame from the initial sequence of ultrasound images that meet the expectations, calculates the absolute value of the rate of change of the left ventricular cavity area between each frame and the corresponding next frame, and takes the minimum value among the absolute values ​​as the absolute value of the slope of the trend of change of the left ventricular cavity area; the ratio of the preset absolute value of the slope to the absolute value of the slope of the trend of change of the left ventricular cavity area is determined as the probability of the existence of the peak frame of the left ventricular cavity area.

[0068] Please see Figure 3 As shown, it is a logic diagram for determining whether the initial sequence of the ultrasound image needs to be corrected according to an embodiment of the present invention.

[0069] Specifically, the optimization and correction module determines that the initial sequence of the ultrasound image needs to be corrected based on the probability of existence being greater than or equal to a preset probability threshold.

[0070] In this embodiment, after calculating the probability of existence, the optimization and correction module compares it with a preset probability threshold. Preferably, the probability threshold can be set to any value between 1.2 and 1.3, and the preset absolute slope threshold is the average of the rate of change of left ventricular cavity area between frames. When the probability of existence is greater than or equal to the threshold, the system determines that the currently set start or end frame is very likely not a true peak or valley frame. In the ultrasound image sequence acquired at the current frame rate, there are large values ​​of left ventricular area between adjacent frames where images were not acquired. That is, it is determined that the initial ultrasound image sequence needs to start the correction process to ensure that only sequences with significant deviations will enter the subsequent correction stage. When the probability of existence is less than the preset probability threshold, it is determined that the initial ultrasound image sequence does not need to start the correction process, and the probability generation module is triggered for subsequent analysis.

[0071] Please see Figure 4 As shown, it is a flowchart of the process for determining the correction coefficient of the initial sequence of the current ultrasound image in an embodiment of the present invention.

[0072] Specifically, the optimization and correction module determines the correction coefficient of the current ultrasound image initial sequence based on the ratio of the historical baseline value to the difference in left ventricular cavity area between the starting or ending frame image and the corresponding next frame image of the initial ultrasound image sequence.

[0073] The historical baseline value is determined based on the average difference in left ventricular cavity area between the peak frame image of the historical standard sequence and the next frame image.

[0074] In this embodiment, after determining that correction is needed, the optimization and correction module selects the most standard sequence from the pre-stored historical standard database, calculates the absolute value of the area difference between the precise peak frame and the next frame of these sequences, and takes the average of these absolute values ​​as the average standard deviation; then calculates the absolute value of the area difference between the start frame or end frame of the current sequence and the next frame; finally, the average standard deviation is divided by the absolute value of the area difference of the current sequence, and the resulting ratio is the correction coefficient.

[0075] In this embodiment, the most standard ultrasound image sequence is determined by selecting frames whose absolute peak positions are determined by all available detection methods—including but not limited to zero-crossing analysis of the first derivative of the area change curve, verification of the precise correspondence with the R wave of the synchronized electrocardiogram signal, and selection of consecutive dense frames. The accuracy of the peak frames in these selected sequences has been verified.

[0076] Specifically, the optimization and correction module determines the corrected acquisition frame rate based on the product of the correction coefficient and the preset frame rate, and sends a corrected frame rate command to the image acquisition module to trigger the image acquisition module to reacquire left ventricular ultrasound images for several cardiac cycles.

[0077] In this embodiment, the optimization and correction module multiplies the calculated correction coefficient by the initial preset frame rate, and the product is the corrected acquisition frame rate. Then, it sends an instruction to the image acquisition module to re-acquire the left ventricular ultrasound image sequence during the cardiac cycle, which includes this corrected frame rate. After receiving the instruction, the image acquisition module immediately uses the optimized frame rate as a parameter to re-acquire the left ventricular ultrasound image sequence, thereby ensuring that key peak frames can be captured more accurately in the new acquisition process, and generating an ultrasound diagnostic image sequence.

[0078] Specifically, the probability generation module uses a Bayesian neural network to perform inference based on the left ventricular ultrasound image to calculate the probability value of each pixel in each frame of the left ventricular ultrasound image belonging to the myocardium, so as to generate a myocardial probability map sequence corresponding to each frame of the image.

[0079] In this embodiment, the probability generation module inputs the high-quality left ventricular ultrasound image sequence obtained by the image acquisition module, i.e., the aforementioned ultrasound diagnostic image sequence, into a pre-trained Bayesian neural network. Through multiple rounds of forward propagation inference, and by statistically analyzing the multiple prediction results for each pixel, the frequency of its classification as myocardium is calculated. Finally, this frequency value is used as the probability value of the pixel belonging to myocardium, and the myocardial probability map of the entire sequence is output. It is understood that the specific steps of the Bayesian neural network's image recognition inference are existing technology and will not be elaborated here. This invention can also identify myocardial tissue in images based on any existing technology.

[0080] Specifically, the ROI segmentation module extracts the key segmentation parameters based on the myocardial probability map sequence. The spatial neighborhood consistency is determined by calculating the average difference between the probability values ​​of each pixel and its neighboring pixels, and the temporal neighborhood smoothness is determined by calculating the average difference between the probability values ​​of each pixel in the current frame and the corresponding position probability values ​​in the previous and next frames.

[0081] In this embodiment, the ROI segmentation module extracts three key parameters from the myocardial probability map sequence: the probability value of each pixel itself, reflecting the possibility that it independently belongs to the myocardium; spatial neighborhood consistency by calculating the average difference between the current pixel and the pixel probability values ​​in its eight-connected neighborhood, which is used to constrain the smoothness of the segmentation boundary; and temporal neighborhood smoothness by calculating the average difference between the pixel probability values ​​of the current frame and the corresponding pixel probability values ​​of the adjacent frames, which is used to ensure the stability of the segmentation result in the time dimension.

[0082] Specifically, the ROI segmentation module constructs an energy function using a temporal conditional random field model based on the key segmentation parameters. By minimizing the energy function, it determines the optimal pixel label configuration for segmentation, thereby generating the myocardial ROI mask image sequence.

[0083] In this embodiment, the ROI segmentation module constructs an energy function that integrates single-frame pixel probability, spatial neighborhood consistency, and temporal neighborhood smoothness based on the extracted key segmentation parameters using a temporal conditional random field model. Subsequently, optimization techniques such as graph cut algorithms are used to minimize this energy function to find the pixel label configuration that minimizes the total energy. This optimal label configuration is the final segmentation result, which generates a high-quality myocardial ROI mask image sequence.

[0084] Understandably, the energy function is a composite objective function that integrates image appearance features, spatial constraints, and temporal consistency. The independent variable of this function is the set of label configurations for all pixels in the entire ultrasound image sequence, i.e., the complete combination of binary labels assigning each pixel as either "myocardial" or "non-myocardial." The dependent variable is the energy evaluation value, which characterizes the overall quality of the current label configuration. Specifically, the energy function consists of three key parts: the first part is a data term, the value of which depends on the degree of matching between the initial predicted probability of each pixel and the actual assigned label; this term decreases when a pixel is correctly labeled. The second part is a spatial smoothing term, used to measure the consistency of labels between adjacent pixels; this term increases when there are unnecessary abrupt changes in labels within the spatial neighborhood. The third part is a temporal smoothing term, used to evaluate the label stability of the same anatomical location across different time frames; this term increases when there are unreasonable label jumps between adjacent frames. By combining these three parts into a unified energy function and employing a graph cut optimization algorithm to find the label configuration scheme that minimizes the total energy, the optimal segmentation result that conforms to image grayscale features while maintaining spatial continuity and temporal consistency is ultimately achieved.

[0085] Understandably, pixel label configuration is a complete answer table that assigns "myocardial (1)" or "non-myocardial (0)" to each pixel; temporal conditional random fields quantify single-frame probability, spatial continuity, and temporal smoothness into a unified energy function. The lowest total energy corresponds to the highest joint probability, meaning that this configuration respects appearance evidence, ensures boundary coherence, and prevents label flickering between adjacent frames. By solving for this minimum value using a graph cut algorithm, the most self-consistent label assignment for the entire video under the three criteria can be obtained in one go, thereby directly generating a high-quality myocardial ROI mask sequence with accurate boundaries and temporal coherence.

[0086] Specifically, it also includes a multi-source data module for acquiring multi-source clinical data and for performing auxiliary analysis on myocardial ROI mask image sequences based on the clinical data; wherein the clinical data includes myocardial injury biomarkers, inflammatory markers, and hemodynamic parameters.

[0087] Understandably, multi-source clinical data can be obtained through hospital information systems, laboratory information systems, or bedside monitoring equipment. The acquisition of multi-source clinical data can aid in the analysis of myocardial ROI mask image sequences, thereby improving the accuracy of auxiliary detection.

[0088] Specifically, the acquired ultrasound image sequence also includes images of the entire heart, including the right ventricle; a corresponding myocardial ROI mask image sequence is generated based on the acquired image sequence of one or more heart chambers.

[0089] By expanding the image acquisition range from the left ventricle to the right ventricle or the entire heart, a more comprehensive assessment of cardiac function can be achieved. By acquiring and analyzing ultrasound image sequences of the left and right ventricles or the entire heart, global patterns of myocardial motion abnormalities caused by sepsis can be captured more accurately, improving the comprehensiveness and reliability of auxiliary detection.

[0090] The technical solution of the present invention has been described above with reference to the preferred embodiments shown in the accompanying drawings. However, it will be readily understood by those skilled in the art that the scope of protection of the present invention is obviously not limited to these specific embodiments. Without departing from the principles of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will all fall within the scope of protection of the present invention.

Claims

1. An image recognition-based auxiliary detection system for septic cardiomyopathy, characterized in that, include: The image acquisition module is used to acquire left ventricular ultrasound images, acquire left ventricular ultrasound images of several cardiac cycles based on a preset frame rate to generate an initial sequence of ultrasound images, and acquire left ventricular ultrasound images of several cardiac cycles at a specified frame rate according to a received corrected frame rate instruction to generate a sequence of ultrasound diagnostic images, wherein the start frame and end frame of the image are determined based on the electrocardiogram gate signal. The image inspection module is used to perform periodic verification based on the ECG gate signal corresponding to the initial sequence of the ultrasound image to determine whether the period of the initial sequence of the ultrasound image meets the standard. Based on the initial sequence of the ultrasound image that meets the standard, a neighborhood consistency check is performed according to the left ventricular area in the image to determine whether the neighborhood consistency of the initial sequence of the ultrasound image meets the expectation. The optimization and correction module is used to calculate the probability of the presence of the peak frame of the left ventricular cavity area based on the changing trend of the start frame image, the end frame image and the corresponding next frame image in the initial sequence of ultrasound images that meet the expectations. Based on the probability of presence, it determines whether the initial sequence of ultrasound images needs to be corrected. Based on the change of the peak frame area of ​​the left ventricular cavity area in historical standard sequences, it calculates the correction coefficient of the current initial sequence of ultrasound images to determine the corrected acquisition frame rate. Based on the corrected acquisition frame rate, it sends a correction frame rate command to the image acquisition module. The probability generation module is used to calculate the probability value of each pixel belonging to the myocardium based on each frame of the ultrasound diagnostic image sequence, so as to generate a myocardial probability map sequence. The ROI segmentation module is used to extract key segmentation parameters based on the myocardial probability map sequence to determine the optimal segmentation pixel label configuration, so as to generate a myocardial ROI mask image sequence for auxiliary detection. The key segmentation parameters include the probability value of each pixel, spatial neighborhood consistency, and temporal neighborhood smoothness.

2. The image recognition-based auxiliary detection system for septic cardiomyopathy according to claim 1, characterized in that, The optimization and correction module determines the absolute value of the slope of the left ventricular cavity area change trend based on the absolute value of the rate of change of the left ventricular cavity area between the start and end frames in the initial sequence of ultrasound images that meet the expectations and the corresponding next frame image. Based on the absolute value of the slope and a preset absolute value threshold, the probability of the existence of the peak frame of the left ventricular cavity area is determined.

3. The image recognition-based auxiliary detection system for septic cardiomyopathy according to claim 2, characterized in that, The optimization and correction module determines that the initial sequence of the ultrasound image needs to be corrected based on the probability of existence being greater than or equal to a preset probability threshold.

4. The image recognition-based auxiliary detection system for septic cardiomyopathy according to claim 3, characterized in that, The optimization and correction module determines the correction coefficient of the current ultrasound image initial sequence based on the ratio of the historical baseline value to the difference in left ventricular cavity area between the starting or ending frame image and the corresponding next frame image of the initial ultrasound image sequence. The historical baseline value is determined based on the average difference in left ventricular cavity area between the peak frame image of the historical standard sequence and the next frame image.

5. The image recognition-based auxiliary detection system for septic cardiomyopathy according to claim 4, characterized in that, The optimization and correction module determines the corrected acquisition frame rate based on the product of the correction coefficient and the preset frame rate, and sends a corrected frame rate command to the image acquisition module to trigger the image acquisition module to reacquire left ventricular ultrasound images for several cardiac cycles.

6. The image recognition-based auxiliary detection system for septic cardiomyopathy according to claim 1, characterized in that, The image inspection module calculates the current cycle length based on the initial sequence of the ultrasound image and the corresponding ECG gate signal, and compares the current cycle length with the preset standard cycle length range to determine whether the cycle of the initial sequence of the ultrasound image meets the standard.

7. The image recognition-based auxiliary detection system for septic cardiomyopathy according to claim 6, characterized in that, The image inspection module generates an inter-frame left ventricular cavity area difference sequence based on the difference in left ventricular cavity area between each frame in the initial sequence of ultrasound images conforming to the standard and the adjacent frames. Based on the sign change pattern of the difference sequence, it determines whether the sign change conforms to the corresponding expectation of end-diastole and end-systole.

8. The image recognition-based auxiliary detection system for septic cardiomyopathy according to any one of claims 1-7, characterized in that, The probability generation module uses a Bayesian neural network to perform inference based on the left ventricular ultrasound image to calculate the probability value of each pixel in each frame of the left ventricular ultrasound image belonging to the myocardium, so as to generate a myocardial probability map sequence corresponding to each frame of the image.

9. The image recognition-based auxiliary detection system for septic cardiomyopathy according to claim 1, characterized in that, It also includes a multi-source data module for acquiring multi-source clinical data and for performing auxiliary analysis of myocardial ROI mask image sequences based on the clinical data; wherein the multi-source clinical data includes myocardial injury biomarkers, inflammatory indicators, and hemodynamic parameters.

10. The image recognition-based auxiliary detection system for septic cardiomyopathy according to claim 8, characterized in that, The acquired ultrasound image sequence also includes images of the entire heart, including the right ventricle; a corresponding myocardial ROI mask image sequence is generated based on the acquired image sequence of one or more heart chambers.

Images