An automatic lesion recognition ultrasound system for real-time monitoring

Abstract

The present application relates to the technical field of lesion recognition, in particular to an automatic lesion recognition ultrasonic system for real-time monitoring, which comprises a delay path discrimination module, a gray level trend detection module, a texture feature screening module, a feature fusion sorting module and a region highlight labeling module. Based on continuous ultrasonic frames, the collected deep tissue echo path is analyzed, and the echo arrival time of each pixel point in the continuous frame is detected. The present application supports multi-type data fusion judgment through a comprehensive judgment process supported by multi-dimensional parameter collaborative screening, penetration behavior, gray level trend and texture aggregation. The regional abnormal priority sorting mode improves the hierarchy of lesion feature discrimination, provides partition directional recognition for local structural abnormalities and early micro-variation, and converts the feature judgment result into a high confidence region label by a weight aggregation method. The image output process automatically completes the real-time visual presentation of the high-risk area, improving the clarity and pertinence of the lesion region presentation.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of lesion recognition, and particularly relates to an automatic lesion recognition ultrasonic system for real-time monitoring. BACKGROUND

[0002] Lesion recognition relates to the technology of detecting and classifying abnormal changes that may exist in human tissues through medical imaging means, including the recognition, positioning and preliminary identification of abnormal structures in ultrasonic images, CT images or MRI images, aiming to provide auxiliary diagnosis basis for clinics to improve diagnosis efficiency and accuracy. Among them, the traditional automatic lesion recognition ultrasonic system for real-time monitoring refers to a kind of system that automatically identifies potential lesions by means of image feature changes when continuously observing specific tissue regions based on ultrasonic images, and is aimed at how to automatically identify possible lesion regions in real-time ultrasonic images to assist physicians in judging lesion states.

[0003] The prior art relies on single parameter static recognition or simple interframe gray contrast, and deep tissue penetration details are easily disturbed by signal attenuation. Gray dynamic fluctuation and complex texture changes cannot be effectively linked to determine. In actual operation, the sensitivity to small early lesion regions is low. The lack of systematic weight discrimination leads to the fact that the lesion priority region is easily ignored. The lesion boundary is easy to produce blur or recognition error under complex tissue conditions. High-risk signals cannot be efficiently distinguished under real-time monitoring, resulting in incomplete and intuitive clinical diagnosis auxiliary information. SUMMARY

[0004] The purpose of the present application is to solve the shortcomings in the prior art and to provide an automatic lesion recognition ultrasonic system for real-time monitoring.

[0005] In order to achieve the above purpose, the present application adopts the following technical scheme: an automatic lesion recognition ultrasonic system for real-time monitoring, the system comprises:

[0006] The delay path discrimination module detects the echo arrival time of the pixel points in multiple frames based on continuous ultrasonic frames, judges the delay change trend of the pixels, identifies the sound density abnormal signal, and obtains the penetration dynamic discrimination parameter;

[0007] The gray trend detection module compares the pixel gray intensity of the continuous frames of the target region based on the penetration dynamic discrimination parameter, identifies the gray change rate difference, filters out inconsistent signals, and obtains the trend offset feature set;

[0008] The texture feature screening module calculates the texture intensity difference of the high-frequency reflection points of the target region based on the trend offset feature set, analyzes the reflection feature trend of the center pixel, compares the texture change consistency of the target region and the labeled coordinates, and obtains the high-frequency fusion feature quantity;

[0009] The feature fusion sequencing module judges the target region multi-feature combination based on the high-frequency fusion feature quantity, analyzes the weight distribution of the trend deviation feature set and the penetration dynamic discrimination parameter, optimizes the multi-feature intersection point priority, and obtains a lesion discrimination priority index;

[0010] The region highlight labeling module identifies a lesion feature priority region in the image frame based on the lesion discrimination priority index, analyzes the region and the standard template matching degree, performs highlight labeling on the target pixel point, and obtains lesion region labeling data.

[0011] The present application improves that the penetration dynamic discrimination parameter comprises delay distribution data, penetration level information and frequency adjustment instruction, the trend deviation feature set comprises gray scale change distribution, trend type label and abnormal response number, the high-frequency fusion feature quantity comprises texture difference parameter, fusion channel identification and spatial consistency parameter, the lesion discrimination priority index comprises discrimination grading category, confidence parameter and preferred region index, and the lesion region labeling data comprises labeled pixel index, highlight level label and display category type.

[0012] The present application improves that the delay path discrimination module comprises:

[0013] The data frame receiving submodule analyzes the deep tissue echo signal of each pixel point based on the continuous ultrasonic frame, point-by-point compares the pixel positions of each frame, judges the echo arrival time change of the same pixel point between adjacent frames, calculates the time sequence difference of each pixel point under the frame sequence, and obtains pixel delay data;

[0014] The path delay extraction submodule judges the pixel point delay change trend based on the pixel delay data, analyzes the change direction and change continuity, identifies the pixel points with stable trend, optimizes the corresponding relationship between the spatial position and the time sequence, and obtains delay trend data;

[0015] The frequency adjustment identification submodule compares the delay characteristics with the echo response of the current acoustic density detection region based on the delay trend data, analyzes the association between the deep tissue echo path and the response mode, adjusts the ultrasonic emission frequency, identifies the pixel region with inconsistent structure characteristics, and obtains the penetration dynamic discrimination parameter.

[0016] The present application improves that the gray scale trend detection module comprises:

[0017] The gray scale contrast analysis submodule detects the pixel gray scale intensity sequence in the target region based on the penetration dynamic discrimination parameter, calculates the change rate of the corresponding pixel gray scale in adjacent frames, identifies the pixel points with different gray scale change amplitudes, and performs gray scale time sequence aggregation according to the gray scale change direction of the frame sequence, and obtains gray scale change feature data.

[0018] The structural relationship identification submodule judges the spatial distribution of pixel gray difference in the target region based on the gray change characteristic data, analyzes the structural relationship between the pixel gray difference and adjacent pixels, compares the gray change trend and abnormal behavior characteristics, identifies the pixels with consistent spatial characteristics, and obtains a gray structure compliance index;

[0019] The trend deviation screening submodule analyzes the gray change mode of each pixel point in the time sequence based on the gray structure compliance index, judges the relevance of the change trend and abnormal characteristics, identifies the pixels with inconsistent change amplitude and abnormal characteristics, and obtains a trend deviation feature set.

[0020] The texture feature screening module comprises:

[0021] The high-frequency reflection point identification submodule analyzes the gray sequence of the center pixel and the neighborhood of the target region based on the trend deviation feature set, judges whether the direction of the gray change between frames frequently switches, identifies the pixels with multiple direction reversals in multiple consecutive frames, and obtains a high-frequency reflection point mapping set.

[0022] The reflection trend matching submodule analyzes the gray sequence of each point in consecutive frames based on the high-frequency reflection point mapping set, judges whether the direction of the gray change between frames is consistent with the trend type, compares the direction distribution of the target region with the matching condition of the reference direction, identifies the reflection points meeting the direction requirement, and obtains a reflection trend direction distribution structure.

[0023] The spatial consistency extraction submodule calculates the gray change parameter difference of each direction of the target region and the labeled coordinates based on the reflection trend direction distribution structure, compares the change characteristics of the horizontal and vertical directions, optimizes the difference expression of the directions of the two regions, analyzes the difference degree between the regions, obtains a spatial consistency offset, and obtains a high-frequency fusion feature quantity.

[0024] The feature fusion sorting module comprises:

[0025] The multi-feature combination analysis submodule analyzes the spatial coordinates based on the high-frequency fusion feature quantity, judges the corresponding regions of the coordinate points in the trend deviation feature set and the penetration dynamic discrimination parameter, and compares the joint distribution of various features at the same spatial position, to obtain feature combination correlation data.

[0026] The feature weight calculation submodule analyzes the expression of each feature in the spatial distribution in the feature overlapping region based on the feature combination correlation data, judges the distribution density of each feature in the target region, optimizes the combination structure of various features in the region, and obtains a feature distribution weight parameter.

[0027] The priority level division submodule compares the weight ranking of multiple feature intersection areas based on the feature distribution weight parameters, judges the feature combination performance of the center point of the feature overlap area, identifies the spatial location with multiple feature aggregation advantages, divides the feature performance priority level, and obtains the lesion judgment priority index.

[0028] The present invention is improved in that the region highlighting module includes:

[0029] The pixel feature rendering submodule optimizes the grayscale distribution, boundary features and texture structure of the pixel set based on the lesion determination priority index, compares the rendering levels required for the classification, determines the brightness adjustment configuration affected by the confidence parameter, and obtains a multi-parameter pixel rendering sequence.

[0030] The region matching submodule determines the texture structure, grayscale contour and lateral symmetry of the target region based on the multi-parameter pixel rendering sequence, calculates the matching with the standard template reference vector and texture reference, identifies the performance that conforms to the structural consistency characteristics, and obtains the structural consistency matching index.

[0031] The highlight data generation submodule, based on the aforementioned structural consistency matching index, compares the degree of abrupt changes in the central texture, the grayscale deviation ratio, and the connectivity of the boundary texture, using the formula: Calculate the highlighting index, combine it with the region category and structural distribution number, and perform highlighting marking on the target pixels to obtain the lesion region annotation data. Indicates the first Highlighting index of each pixel Indicates the first The grayscale deviation of each pixel This represents the average grayscale deviation ratio of all pixels within the current analysis area. Indicates the first Adjustment parameters for each spatial structure reference item, Indicates the first The pixel in the first Texture connectivity amplitude under each spatial structure reference item Indicates the total number of spatial structure reference items. Indicates the first The normalized texture gradient term for each pixel. This represents the average texture gradient at the corresponding position in the standard template.

[0032] The present invention is improved in that the continuous ultrasound frame refers to multiple ultrasound image frames acquired in real time and arranged sequentially at a short time interval, the high-frequency reflection point refers to the pixel point in the ultrasound image that exhibits high-frequency gray-scale changes, and the texture intensity difference refers to the gray-scale intensity difference between any two high-frequency reflection points in the target area.

[0033] Compared with the prior art, the application has the advantages and positive effects that:

[0034] In the application, the deep echo information and the pixel gray dynamic change realize time sequence linkage processing, the spatial texture aggregation feature is screened through multi-dimensional parameter cooperation, the comprehensive judgment process of penetration behavior, gray trend and texture aggregation supports multi-type data fusion judgment, the regional anomaly priority sorting mode improves the hierarchy of lesion feature judgment, the partition directional recognition is provided for local structure anomaly and early micro variation, the weight aggregation method converts the feature judgment result into high confidence region labeling, the image output process automatically completes the real-time visual presentation of high-risk areas, and the clarity and pertinence of the lesion area presentation are improved. BRIEF DESCRIPTION OF DRAWINGS

[0035] Figure 1 The system flowchart of the application is shown in the figure;

[0036] Figure 2 The flowchart of the delay path judgment module in the application is shown in the figure;

[0037] Figure 3 The flowchart of the gray trend detection module in the application is shown in the figure;

[0038] Figure 4 The flowchart of the texture feature screening module in the application is shown in the figure;

[0039] Figure 5 The flowchart of the feature fusion sorting module in the application is shown in the figure;

[0040] Figure 6 The flowchart of the regional highlight labeling module in the application is shown in the figure. DETAILED DESCRIPTION

[0041] In order to make the purpose, technical scheme and advantages of the application clearer, the application will be further described in detail below in combination with the drawings and examples. It should be understood that the specific examples described herein are only used to explain the application and not to limit the application.

[0042] In the description of the application, it should be understood that the orientations or positional relationships indicated by the terms "length", "width", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer" and the like are based on the orientations or positional relationships shown in the drawings, and are only for the convenience of describing the application and simplifying the description, and therefore cannot be understood as indicating or implying that the devices or elements referred to must have a particular orientation, be constructed and operated in a particular orientation, and therefore cannot be understood as limiting the application. In addition, in the description of the application, the meaning of "multiple" is two or more, unless otherwise specifically limited. EMBODIMENT

[0043] Please refer toFigure 1 The application provides a technical solution: an automatic lesion recognition ultrasonic system for real-time monitoring, comprising:

[0044] The delay path discrimination module analyzes the collected deep tissue echo path based on consecutive ultrasonic frames, judges the delay change trend of the same position pixel by detecting the echo arrival time of each pixel point in the consecutive frames, adjusts the working frequency of the ultrasonic probe, identifies the signals that do not conform to the tissue characteristics in the sound density detection area, and obtains a penetration dynamic discrimination parameter;

[0045] The gray trend detection module compares the pixel gray intensity of the target area under consecutive frames based on the penetration dynamic discrimination parameter, identifies the pixel points with different gray change rates, analyzes the gray difference and structure distribution relationship in the target area, judges the compliance degree of the gray change trend and abnormal characteristics, and eliminates the signals that do not meet the requirements of the trend change amplitude, and obtains a trend offset feature set;

[0046] The texture feature screening module calculates the texture intensity difference between the high-frequency reflection points in the target area based on the trend offset feature set, analyzes the reflection feature change trend of the center pixel of the consecutive frames, adjusts the spatial directionality feature, and compares whether the texture change of the target area and the labeled coordinates is consistent, and obtains a high-frequency fusion feature quantity;

[0047] The feature fusion sorting module judges the combination relationship of multiple features in the target area based on the high-frequency fusion feature quantity, analyzes the weight situation of the trend offset feature set and the penetration dynamic discrimination parameter in the spatial distribution, optimizes the priority division of the intersection point of multiple features, and obtains a lesion determination priority index;

[0048] The region highlight labeling module identifies the region with lesion feature priority in the image frame based on the lesion determination priority index, adjusts the visualization rendering parameter of the pixel, analyzes the matching condition of the determination region and the standard template, and performs highlight labeling on the target pixel point, and obtains lesion region labeling data.

[0049] The penetration dynamic discrimination parameter includes delay distribution data, penetration level information and frequency adjustment instruction, the trend offset feature set includes gray change distribution, trend type label and abnormal response number, the high-frequency fusion feature quantity includes texture difference parameter, fusion channel identifier and spatial consistency parameter, the lesion determination priority index includes determination classification category, confidence parameter and preferred region index, and the lesion region labeling data includes labeled pixel index, highlight level label and display category type.

[0050] In the delay path discrimination module, the continuous ultrasound frames refer to multiple ultrasound image frames collected in real time and arranged in sequence in a short time interval, reflecting the acoustic response changes of the detected tissue in a dynamic state; the deep tissue echo path refers to the propagation path of the ultrasound wave from the emission point to the reflection point and back to the probe when passing through the deep tissue of the body, which is commonly used to distinguish the echo response characteristics of deep structures; the echo arrival time refers to the time taken by the ultrasound wave to return to the probe after emission, reflecting the speed difference between different tissues or interfaces and the thickness information of the tissue; the delay change trend refers to the change direction or rate of the echo arrival time of the same pixel point in consecutive image frames, which is used to capture the dynamic characteristics of deep structures; the acoustic density detection area refers to the analysis target block determined by the system on the ultrasound image according to the acoustic signal density (such as echo intensity distribution), excluding background noise and non-structure components; the signal not meeting the tissue characteristics refers to the ultrasound echo signal determined as non-physiological and non-target tissue characteristics after screening by the echo path, arrival time, acoustic density, etc., which is usually clutter, artifact, etc.

[0051] In the gray level trend detection module, the target area refers to the key analysis area selected by the delay path discrimination module, which usually has a risk of lesions or abnormal performance; the pixel gray level intensity refers to the brightness value of each pixel in the ultrasound image, reflecting the local echo energy size, which is a key parameter for describing tissue density and acoustic characteristics; the gray level change rate refers to the change speed of the gray level intensity of the same pixel position in consecutive frames, which is used to monitor the dynamic changes of the tissue and the progression of the lesion; the gray level difference refers to the difference in gray level intensity between different pixels or pixel blocks in the target area, which is used to describe the heterogeneity of the tissue structure; the structure distribution relationship refers to the geometric and topological relationship between the gray level of each pixel point in the target area and its spatial neighborhood, revealing the clustering or dispersion characteristics of the abnormal area in structure; the gray level change trend refers to the change direction and pattern of the gray level intensity over time or frame sequence, which is used to capture the rules of lesion development or regression; the abnormal feature refers to the preset gray level change rule that is significantly different from the normal tissue dynamic behavior, which usually reflects the potential pathological process; the signal with an inconsistent trend change amplitude refers to the signal whose change amount does not meet the preset screening standard after dynamic trend analysis, which is excluded from the subsequent recognition steps.

[0052] In the texture feature screening module, the high-frequency reflection point refers to a pixel point in the ultrasound image that presents high-frequency gray scale changes, which often represents the microstructure of the tissue or the micro-abnormalities of early lesions; the texture intensity difference refers to the gray scale intensity difference between any two high-frequency reflection points in the target region, which is used to measure the texture uniformity or abnormal spots inside the tissue; the reflection feature change trend refers to the change law of the reflection intensity of the center pixel and the neighborhood over time in multiple ultrasound images, which is used to determine the dynamic evolution of the microstructure; the spatial directionality feature refers to the distribution difference of the texture changes in different spatial directions (such as radial and tangential directions), which is helpful for positioning the growth or diffusion direction of the abnormal texture; the labeling coordinates refer to the pixel coordinates set by the gray scale trend detection module for the suspicious regions, which are used for subsequent fine texture comparison and region positioning; and the consistency of texture changes refers to the synchronization or asynchronization of the texture intensity, structure and other features between the target region and the labeling coordinate region over time, which is used for lesion consistency discrimination.

[0053] In the feature fusion sorting module, the combination relationship of multiple features refers to the joint distribution characteristics of multi-dimensional parameters such as penetration, gray scale and texture in space or time, which is the basis for comprehensive lesion determination; the weight condition refers to the proportion of each feature in the comprehensive determination, reflecting the influence of different parameters on the priority determination of the lesion; and the intersection point of multiple features refers to the pixel point or region where different feature parameters overlap in space, which is usually the potential lesion position that the system pays attention to first.

[0054] In the region highlight labeling module, the region with priority lesion features refers to the pixel point or region with the most abnormalities or lesion risks after comprehensive determination and sorting, which is highlighted for subsequent visualization; the visualization rendering parameter refers to the specific settings for adjusting the display effect of the ultrasound image, such as brightness, contrast, color, etc., which is used to enhance the visual recognition of the target region; the standard template refers to the pre-set or trained lesion feature comparison atlas, which is used for spatial structure and shape matching with real-time detection results; and the highlight mark refers to the identification of the target region on the image in a special color or style through image processing, which facilitates quick identification by humans or systems.

[0055] Please refer to Figure 2 , the delay path discrimination module comprises:

[0056] The data frame receiving submodule analyzes the deep tissue echo signals of each pixel point based on consecutive ultrasound frames, compares each pixel position point by point, judges the change of the echo arrival time of the same pixel point between adjacent frames, calculates the time sequence difference of each pixel point in the frame sequence, and obtains the pixel delay data;

[0057] In the process of continuously acquiring ultrasound image frames, multiple frames of images are arranged in time sequence, the position of each pixel in the image and its corresponding gray value are read row by row and column by column in each frame of image, and the gray value reflects the echo intensity of the tissue to the ultrasound wave. For the same spatial position pixel point, the gray value change in different frames of images is read to judge whether it is in a dynamic change state. For pixels with a gray value difference greater than 10 in three consecutive frames, it is considered that there is an effective signal of tissue structure response. The frame sequence time stamp of each frame of image is read, the sampling time of the same pixel position in the current frame and the previous frame is compared, the echo arrival time difference of the pixel point is calculated, and the pixel point with a change greater than ±5 microseconds is marked as a significant delay change. This operation is performed on all pixels in each frame, and the pixel points in the entire image are traversed. The echo time change results of each pixel in the consecutive frames are recorded and classified in turn, and a pixel delay database containing the image spatial position and frame sequence time axis is constructed. In an example, for example, the echo times of a certain pixel point in three consecutive frames are 8.12 microseconds, 8.18 microseconds and 8.21 microseconds, the changes between adjacent frames are 0.06 microseconds and 0.03 microseconds respectively, and the average value is 0.045 microseconds. Compared with the set reference value 0.02 microseconds, if it exceeds, the point is added to the delay mark list. After continuously processing multiple pixel points in this way, a group of pixel data with delay time change record is finally formed in the image, which is used as the input data for subsequent delay path extraction.

[0058] The path delay extraction submodule judges the delay change trend of the pixel points based on the pixel delay data, analyzes the change direction and change continuity, identifies the pixel points with stable trend, optimizes the corresponding relationship between the spatial position and the time sequence, and obtains delay trend data.

[0059] The echo delay time of each pixel point in the continuous frames is checked one by one to see whether it presents a continuous rising or falling trend, and whether the change direction remains consistent. If the echo delay time of a certain pixel continuously increases or decreases in three frames of images, it is preliminarily judged that the delay trend is stable. For the pixel points with stable trends, it is further judged whether the fluctuation amplitude of the delay change is stable. The delay values in the time sequence are compared multiple times. If the change value fluctuates less in the set range, such as less than 0.01 microseconds, the pixel point is recorded as a stable trend pixel. The image coordinate position of the stable pixel point and its corresponding time sequence relationship are bound to establish the corresponding mapping of space and time, form a mapping table of image position to delay change sequence, and be used to describe the delay evolution process of the pixel in the time dimension. In the example, if the delay of a certain pixel in the continuous three frames is 0.04, 0.05 and 0.045 microseconds, the change fluctuation is small, the difference value is concentrated, it is judged as a stable trend point, and its position and the corresponding delay time of the frame sequence are arranged in sequence to form a record. In this way, a group of pixels with stable change characteristics is screened from the image, and the delay trend data is constituted together with the time delay sequence data of the pixels, for the next step.

[0060] The frequency adjustment identification sub-module compares the delay characteristics with the echo response of the current acoustic density detection area based on the delay trend data, analyzes the correlation between the echo path of the deep tissue and the response mode, adjusts the ultrasonic emission frequency, identifies the pixel area with inconsistent structure characteristics, and obtains the penetration dynamic discrimination parameter.

[0061] First, the delay mean value of each stable trend pixel is extracted, and the overall echo response characteristics of the region where the pixel is located are analyzed. The gray value of all effective pixels in the sound density detection area is extracted, the average level of the overall reflection intensity of the current region is judged, and the delay change of individual pixels is compared with the overall response characteristics of the region where they are located. If the delay value of a certain pixel is significantly different from the delay level of most pixels in the region, such as exceeding the set reference range of 0.015 microseconds, it is judged that the echo path of the pixel is abnormal. Further frequency adjustment processing is performed on such regions in combination with the gray value distribution. If the reflection intensity of the region is generally low, such as the average gray level being lower than 80 levels, it is judged that the current transmission frequency is insufficient for deep tissue penetration, and the transmission frequency is automatically reduced. The next frame of image is re-acquired, and the delay trend is analyzed again. If the gray value is found to be too high, such as exceeding 220 levels, and the delay is short, it is judged that there is reflection superposition or artifact interference, and the transmission frequency is increased. The new data is re-sampled and analyzed. After each frequency adjustment, the delay value and gray response of the pixels in the region are analyzed again. If the adjusted delay level enters the reference value interval and the gray distribution in the region tends to be concentrated, it is determined that the frequency adjustment is effective, and the frequency and the position of the corresponding pixel are recorded. If the frequency adjustment still appears the delay and gray deviation from the average level of the region, the frequency is continuously adjusted until the data converges. A group of regions with inconsistent delay trend and echo characteristics are extracted, and the image coordinates, average delay value and transmission frequency of the region at that time are combined to form a penetration dynamic discrimination parameter data set.

[0062] Please refer to Figure 3 , the gray trend detection module comprises:

[0063] The gray contrast analysis submodule detects the pixel gray intensity sequence in the target region based on the penetration dynamic discrimination parameter, calculates the change rate of the corresponding pixel gray in adjacent frames, identifies the pixels with different gray change amplitudes, and performs gray time sequence aggregation according to the gray change direction of the frame sequence to obtain gray change characteristic data;

[0064] The parameter marked as inconsistent structural features is extracted, and the pixel gray value sequence in the region is read frame by frame. The difference between the gray values of each coordinate position in adjacent frames is calculated, the difference between the current frame gray and the last frame gray is compared, and the change value is recorded for subsequent rate judgment. Repeat this operation on each target pixel to form a sequence composed of multiple gray difference values. When judging the gray change rate, if the absolute value of the gray difference between two consecutive frames exceeds 15, it is recorded as a fast-changing pixel. If the change in all consecutive frames does not exceed 5, it is recorded as a slow-changing pixel. The rest is medium rate change. According to the above classification results, further judge whether there is a pixel distribution pattern with uneven change rate distribution in the target area. For example, in a group of 5 consecutive images, the pixel gray sequence is 122, 139, 150, 152, 155, and the corresponding frame change is 17, 11, 2, 3. Since the first frame difference exceeds 15, it is a fast-changing pixel, and the point position is recorded as a high-change area. Then perform sequence aggregation operation on the gray change direction of each pixel in multiple frames. If the gray value continuously increases or decreases, it is marked as consistent direction. If the gray value changes alternately, it is marked as direction fluctuation. The judgment rule is that the gray change direction does not alternate within three and more frames, which is consistent, otherwise it is fluctuation. If the number of consistent frames accounts for more than 70% of the total number of frames, it is considered that the pixel has a significant gray time trend. Through the above steps, the gray change rate, change direction and change type of each pixel are aggregated to generate a gray change feature data containing change amplitude classification, direction mark and time sequence stability.

[0065] The structural relationship recognition submodule judges the spatial distribution of pixel gray difference in the target area based on the gray change feature data, analyzes the relationship between pixel gray difference and adjacent pixel structure, compares the gray change trend and abnormal behavior characteristics, and identifies the pixels with consistent spatial features to obtain the gray structure compliance index.

[0066] Read the gray level change type and gray level direction information of each pixel in the image frame sequence from the feature data, extract its position in the two-dimensional image coordinates, and select the other pixel points connected with the pixel in the adjacent 8 directions according to the pixel space proximity relationship. Compare the gray level change amplitude difference of the center pixel and its adjacent pixels one by one. If the gray level change type of the adjacent pixel is consistent with that of the center pixel and the difference is less than 20, mark it as a structure consistent relationship. Then judge the consistency level of each pixel point in the spatial distribution according to the number of structure consistent neighborhoods of each pixel point. If the number of structure consistent pixels exceeds 4, mark the point as a structure correlation point. At the same time, further compare the structure consistent relationship with the abnormal behavior characteristics. The abnormal behavior characteristics refer to the pixels marked as rapid change or obvious fluctuation in the gray level change. Further confirm whether the neighborhood around the pixel also has similar changes. If so, the point is marked as an abnormal structure aggregation point. If the gray level direction of a certain pixel is consistent upward, the change amplitude is 35, and its adjacent pixels are 32, 33, 36, 38, 31, 37, 39 and 35, it is judged that the gray level difference in its neighborhood is within 10 and the change direction is consistent, which is marked as a structure consistent area. By counting the intersection area of all structure consistent pixels and abnormal behavior points, the pixels meeting the double conditions of structure trend and gray level trend are extracted, and a gray structure consistent index data set is constructed.

[0067] The trend deviation screening submodule analyzes the gray level change pattern of each pixel point in the time sequence based on the gray structure consistent index, judges the relevance of the change trend and the abnormal characteristics, identifies the pixels with inconsistent change amplitude and abnormal characteristics, and obtains a trend deviation feature set.

[0068] The gray value change sequence of each pixel point meeting the index in all frames is extracted in the image frame sequence, and the change curve is classified one by one. The change direction is divided into five categories: continuous rise, continuous fall, fluctuation rise, fluctuation fall and no obvious trend. For each trend, the standard trend sequence corresponding to the preset abnormal feature is matched and judged. The abnormal feature trend is set according to the gray change process of the clinically common lesion. For example, if the gray value rises from 120 to more than 180 within five frames, it is classified as a typical hyperplasia abnormality. If the gray value does not fluctuate obviously between frames but remains high, such as continuously greater than 200, it is classified as a high-density abnormality. In the trend matching process, it is judged whether the actual pixel sequence meets the above conditions. If it does not meet the conditions, it is marked as a non-abnormal trend point. At the same time, the gray value change amplitude is used to judge again whether it reaches the preset abnormal response threshold. The threshold is set to be greater than 10. If it is less than this value, the pixel point is excluded as an effective response point. For example, the gray value of a certain pixel in five consecutive frames is 128, 130, 133, 134 and 136. The average change of each frame is 2, which is less than the set value 10, and the trend classification is slow rise, which does not belong to the abnormal feature trend. Therefore, it is excluded. After the trend judgment and feature matching of all analysis points are completed, the pixel points that do not meet the abnormal change conditions are screened out, and the remaining part is recorded as the trend deviation feature set.

[0069] Please refer to Figure 4 , the texture feature screening module comprises:

[0070] The high-frequency reflection point identification submodule analyzes the gray sequence of the center pixel and the neighborhood of the target region based on the trend deviation feature set, judges whether the direction of gray value change between frames frequently switches, identifies the pixels that appear multiple times in multiple consecutive frames, and obtains a high-frequency reflection point mapping set.

[0071] The gray scale values corresponding to each pixel point and its adjacent pixels in the feature set in the continuous image frames are extracted. For the center pixel point selected in each frame image in the target region, the gray scale values in the previous and subsequent frames are extracted, a gray scale sequence is formed according to the time sequence, any two continuous gray scale values in the sequence are compared, and the change direction is recorded as rising or falling. After repeating the process to traverse the entire sequence, the direction change state between all frames is counted. If the same pixel point appears more than 2 times of gray scale change direction reversal in 5 continuous frames, for example, the first frame gray scale is 112, the second frame is 130 (rising), the third frame is 125 (falling), the fourth frame is 132 (rising), and the fifth frame is 129 (falling), the change direction is recorded as alternating more than 2 times, and it is determined as a pixel point with frequent direction switching. The judgment is based on the set direction reversal threshold of 2 times. If the number is less than the number, it is not counted in the high-frequency identification. Further, the corresponding judgment result of the center pixel point in its eight neighborhood pixels is combined. If there are more than 5 pixels with direction reversal number greater than or equal to 2 in the neighborhood, the center point is marked as a high-frequency reflection point. After judging, comparing and counting all pixels in all target regions in this way, the pixel set meeting the high-frequency direction reversal condition is extracted, and the mapping index structure is generated combined with the image coordinates to form the high-frequency reflection point mapping set.

[0072] The reflection trend matching sub-module analyzes the gray scale sequence of each point in the continuous frames based on the high-frequency reflection point mapping set, judges whether the direction and trend type of the inter-frame gray scale change are consistent, compares the matching of the direction distribution of the target region and the reference direction, identifies the reflection points meeting the direction requirements, and obtains the reflection trend direction distribution structure.

[0073] The pixel coordinates corresponding to each identified high-frequency reflection point are extracted one by one. Then, the grayscale value of the pixel in each frame is extracted from the original image frame sequence to form a grayscale change sequence of a single pixel. The direction of grayscale change between two consecutive frames of this sequence is judged, and the upward or downward direction label is recorded. The dominant direction type in the entire sequence is counted. Then, a preset reference sample of the baseline direction is read. For example, the typical grayscale change direction in the lesion formation process is a continuous upward direction. If the grayscale continuously increases for more than three frames, or if there is only one downward trend in any frame, it can still be judged as conforming to the baseline direction. During the judgment, the main trend direction of grayscale change of the target reflection point is compared with the baseline direction frame by frame. If the matching degree is higher than 70%, that is, more than 70% of the frames have the same grayscale change direction as the baseline direction, then the point is recorded as the directional trend. Consistent reflection points are identified, and when judging matching accuracy, the distribution of pixels with consistent directional trends within the entire image area is aggregated and calculated. The distribution density of such pixels within the image area is extracted, and a directional trend distribution map is constructed. This distribution map is further analyzed on a pixel block basis. If the reflection points in a certain image area match more than 30 directions and are concentrated in a 16×16 pixel area, it is marked as a directional clustering area. This judgment is based on a region number threshold of 30 and a spatial size threshold of 16 pixels. In this way, the trend direction of the identified high-frequency reflection points in the entire target area is analyzed inter-frame, matched with the reference direction, and the directional aggregation distribution is judged. Then, all pixels that meet the conditions and their coordinates, trend direction status, and spatial clustering status are output to obtain the reflection trend direction distribution structure.

[0074] The spatial consistency extraction submodule, based on the reflection trend direction distribution structure, calculates the differences in grayscale change parameters between the target area and the labeled coordinates in each direction, compares the change characteristics in the horizontal and vertical directions, optimizes the expression of the differences between the two areas, and uses the following formula:

[0075] ;

[0076] Perform inter-regional dissimilarity analysis to obtain spatial consistency offset. We obtained high-frequency fusion characteristic quantities, among which, Indicates the target region number Gray-scale variation parameters in each direction, Indicates the labeled area number Gray-scale variation parameters in each direction, Indicates the target region number The absolute difference in grayscale changes in each direction Indicates the number of directions involved in the comparison;

[0077] The spatial consistency offset is used to quantify the overall difference between the gray scale change parameters of the target region and the labeled region in different directions. By comparing the gray scale change parameters (such as change amplitude, change trend, etc.) of the target region and the labeled region in each direction, the gray scale change differences in all directions are weighted and normalized, and an index reflecting the spatial structure consistency or offset degree between the two regions is output. The larger the index, the more obvious the texture or structure difference between the target region and the labeled region in spatial distribution. The smaller the index, the higher the consistency of the two regions in each direction gray scale distribution and change characteristics. It is used to assist the automatic lesion recognition system to distinguish the degree of coincidence of a region and a standard lesion template in spatial texture features, and is a key input of high-frequency fusion features. The high-frequency fusion feature is based on the spatial consistency offset, further combined with other key parameters such as the reflection trend direction distribution structure, and the aggregation state of the overall high-frequency reflection characteristics of the region is determined by comprehensively considering various spatial and texture indexes. It usually aggregates information such as spatial consistency offset, direction distribution, trend change, etc., and is used to reflect the aggregation and spatial consistency of the high-frequency texture features of the target region.

[0078] The original values are mapped to the interval [0, 1] with 255 as the upper limit and 0 as the lower limit for normalization of each parameter. The normalization interval is set to , and the number of direction samples is set to . The original values and the normalized results are as follows:

[0079] , and the normalized results are as follows: ;

[0080] , and the normalized results are as follows: ;

[0081] , and the normalized results are as follows: ;

[0082] Substitute the normalized parameters into the operation as follows, numerator part:

[0083] ;

[0084] ;

[0085] Denominator part:

[0086] ;

[0087] Substitute the calculation formula:

[0088] ; according to the reference interval division standard of the spatial consistency offset, wherein:

[0089] When , it is defined as a high consistency interval, indicating that the target region and the labeled coordinates have minimal differences in multi-directional gray scale variation parameters, and have significant spatial structure consistency;

[0090] When , it is defined as a moderate deviation interval, indicating that there are partial directional gray scale variation differences between regions, but the overall texture similarity is maintained;

[0091] When , it is defined as a significant deviation interval, indicating that the directional gray scale variation difference between regions is significant, and does not have structural consistency characteristics;

[0092] When , it is defined as a high inconsistency interval, indicating that the texture gray scale features between the two regions have serious deviation, and the spatial structure is different.

[0093] Therefore, the current value is located in the high consistency interval, indicating that the target region is highly similar to the labeled region in spatial directional texture distribution. The result represents the comprehensive deviation degree of directional gray scale matching, and the result is directly used to construct a high-frequency fusion feature as a dominant weight factor. Through joint analysis with other features (such as reflection trend direction distribution), it further participates in the fusion and sorting of subsequent lesion judgment priority indicators, and promotes the lesion region to obtain a higher lesion feature priority in the entire image. The formula normalizes the gray scale variation difference item and participates in the ratio operation, so that different directional parameters are in the same calculation standard, ensuring that the numerical comparison of directional feature differences has a unified dimension, so that the spatial deviation can accurately reflect the consistency degree of the directional gray scale structure of the two regions.

[0094] Referring to Figure 5 , the feature fusion sorting module includes:

[0095] The multi-feature combination analysis submodule is based on high-frequency fusion features, analyzes spatial coordinates, judges the corresponding regions of the coordinate points in the trend deviation feature set and the penetration dynamic discrimination parameters, and compares the joint distribution of various features at the same spatial position to obtain feature combination correlation data;

[0096] Based on the high-frequency fusion characteristic quantity, the image space coordinates of each feature point are extracted, and the coordinates of the image region involved in the trend offset feature set and the penetration dynamic judgment parameter are matched. It is judged whether the coordinates exist in multiple feature sets at the same time. If there is an intersection, it is marked as a feature coincidence point. For each coordinate point, the corresponding texture difference parameter, trend type label and delay distribution data are extracted from three data sources. Then the data is classified and combined according to the coordinates. The numerical performance of the point in the three types of data sources is counted. For the pixel points marked as direction fluctuation type in the gray trend offset, if they have delay abnormality mark in the penetration dynamic parameter, and the texture intensity difference is greater than 50 gray level in the high-frequency fusion, it is determined that the point is abnormal in the three types of features, and is classified into the high abnormal feature combination. Repeat the above processing in all pixel points, count the number of feature coincidences, and classify and count according to each feature combination condition. For example, a pixel point coordinate is (180, 220). The texture difference is 62 in the high-frequency fusion feature, the trend offset is marked as "strong fluctuation", and the delay value exceeds the reference interval by more than 5 microseconds in the penetration parameter. The point is defined as "three high expression combination point". Then count the spatial distribution of all such points, judge whether there is a concentrated area, and establish the corresponding label according to the spatial distribution mode of feature superposition for each type of combination. The feature combination correlation data is generated.

[0097] The feature weight calculation sub-module analyzes the expression of each feature in the spatial distribution in the feature coincidence area according to the feature combination correlation data, judges the distribution density of each feature in the target area, optimizes the combination structure of each type of feature in the area, and obtains the feature distribution weight parameter.

[0098] In the identified feature coincidence area, the frequency of each feature item in the area is counted, and the distribution matrix of the feature items in space is established in units of areas. The pixel number of each type of feature in the area is normalized, and the distribution density of the feature in the area is determined by dividing the number of occurrences by the total number of pixels in the area. The reference value is set to 10% of the density threshold, that is, if the frequency of a certain type of feature in the area exceeds 10%, it is recorded as a high-density distribution feature. The total of the three types of features in the area is extracted to determine whether it meets the feature aggregation condition. The aggregation judgment standard is set to more than the respective reference threshold in the area. If it meets, it is set as a combined effective area. Then the expression intensity comparison between features in the combined effective area is performed. If the proportion of a certain feature in the area exceeds 50%, it is determined to have a dominant expression relationship in the area, and the feature is recorded as the dominant feature in the area. After processing each effective area, the dominant feature label, the corresponding area feature coverage rate and the area position relationship are packaged as feature distribution weight parameters, such as the total number of pixels in area A is 256, of which the trend offset feature pixel is 68, accounting for 26.5%, the texture difference pixel is 41, accounting for 16%, and the delay exceeds the threshold pixel is 33, accounting for 12.9%. All of them exceed 10%, so A area is an effective aggregation area, and the trend offset is the highest coverage item, which is recorded as the dominant feature. The information is generated into a weight parameter data structure.

[0099] The priority level division sub-module compares the weight order of the multi-feature intersection area based on the feature distribution weight parameter, judges the center point feature combination performance of the feature coincidence area, identifies the spatial position with multi-feature aggregation advantage, divides the feature performance priority level, and obtains the lesion determination priority index;

[0100] The dominant feature type of all recorded valid aggregation area position points is read in sequence, and the expression proportion of the remaining features in the area is taken as the basis for sorting to determine the priority of the multi-feature intersection area. Within the intersection area, the feature combination of the center pixel point is extracted in detail to determine whether the pixel point has significant performance in three types of features. If the number of three types of features is higher than the set threshold value, it is marked as a high priority point. The priority level is divided into three levels: the first level is that all three types of features are significant, the second level is that any two types of features are significant, and the third level is that only one type of feature is significant. For example, the texture difference value of a certain area center point is 74, the trend label is continuous fluctuation, and the delay value is 6.2 microseconds, all of which exceed the respective reference standard, so it is divided into a first-level priority point. If another area center point only meets the conditions of texture feature and trend feature, but the delay value does not exceed the set standard, it is divided into a second level. In this way, the center points of all intersection areas are judged according to the coincidence and performance intensity of the three types of features to generate a lesion priority grading index structure. At the same time, a spatial coordinate label, a dominant feature identifier, and a comprehensive score value are established for each level, and the output is a lesion determination priority index.

[0101] Please refer to Figure 6 The region highlight annotation module includes:

[0102] The pixel feature rendering submodule optimizes the gray distribution, boundary feature, and texture structure of the pixel set based on the lesion determination priority index, compares the rendering levels required by the classification categories, judges the brightness adjustment configuration affected by the confidence parameter, and obtains a multi-parameter pixel rendering sequence.

[0103] The spatial pixel position corresponding to each group of indicators is extracted, and the gray value distribution, edge transition intensity and texture difference of the region where the pixel point is located are archived. The hierarchical category label of each pixel point is read, and the corresponding level is divided from 1 to 3. Level 1 indicates that the three types of features are significantly overlapped, and level 3 indicates that only one type of feature is significantly expressed. According to the level division, the rendering level is set. Level 1 corresponds to the highest rendering brightness and the clearest texture boundary, and level 3 corresponds to the most basic rendering configuration. For the pixel points under the same level, the corresponding confidence parameter is read again. The confidence is set in percentage form. The brightness adjustment reference is set as the brightness increase interval in the gray range, which is 20 to 80. The brightness of the pixel with a confidence less than 50% is increased by 20, the brightness of the pixel with a confidence between 50% and 80% is increased by 40, and the brightness of the pixel with a confidence greater than 80% is increased by 80. At the same time, in terms of texture structure, the boundary sharpening degree is set according to the gray gradient change between each pixel and the 8-neighborhood. If the gray gradient change is greater than 30, it is defined as a boundary pixel. The edge sharpening weight is added to the pixel position, the contrast is adjusted and the texture contrast is enlarged. For example, the pixel level of a certain region is 1, the confidence is 85%, the original brightness is 140, and it is adjusted to 220. According to the fact that there are 4 directions with gray difference greater than 30 in the neighborhood of the point, it is judged to be a boundary point and the boundary contour performance is enhanced. The brightness value, texture information and boundary intensity of all adjusted pixels are arranged in spatial coordinate order to form a complete rendering sequence structure, and the output is a multi-parameter pixel rendering sequence.

[0104] The region matching determination sub-module determines the texture structure, gray contour and transverse symmetry of the target region based on the multi-parameter pixel rendering sequence, calculates the matching with the standard template reference vector and the texture reference, identifies the performance of the structure consistency feature, and obtains the structure consistency matching index.

[0105] Based on a multi-parameter pixel rendering sequence, texture feature vectors, brightness contrast values, and boundary structure information of each pixel are extracted from the rendering sequence. Target regions are constructed using spatial blocks as units. Texture structure clustering analysis is performed on each target region, using texture direction, brightness variation boundaries, and geometric distribution information as structural description features. A pre-defined standard template is read; this template is a pre-annotated texture reference image with typical lesion manifestations, including texture feature directions, boundary sharpness standards, and brightness distribution reference values. The feature vector of the target region is compared one-to-one with the reference vector in the standard template in terms of direction difference. If the difference is within 15 degrees, the direction is considered consistent. Then, the brightness contour of the region is compared with the brightness reference structure in the template. If the grayscale layer... If the difference between the mean values ​​of the two distributions is within 20, the brightness is considered to be consistent. Finally, the symmetry of the region image is judged in the left and right directions. The grayscale difference of each pixel in the region is compared with the corresponding pixel on the other side of the vertical central axis. If the vast majority of the differences are within 10, the region is considered to have a horizontally symmetrical structure. For example, in a certain matching region, the angle between the texture direction and the main direction of the template is 10 degrees, the average brightness is 160, the template value is 145, the difference is 15, and the average difference in left and right symmetry is 8. It is determined that the region is highly consistent with the template in all three features and is marked as a structurally consistent region. The above judgment and comparison process is repeated for all target regions. The coordinates, matching degree and label information of all regions that meet the texture structure, brightness contour and symmetry features are used to generate a structural consistency matching index.

[0106] The highlight data generation submodule, based on the structural consistency matching index, compares the degree of abrupt changes in the center texture, the proportion of grayscale deviation, and the connectivity of the boundary texture, using the following formula:

[0107] ;

[0108] Calculate the highlighting index, combine it with the region category and structural distribution number, and perform highlighting marking on the target pixels to obtain the lesion region annotation data. Indicates the first Highlighting index of each pixel Indicates the first The grayscale deviation ratio of the nth pixel represents the average grayscale deviation ratio of all pixels in the current analysis area, and represents the grayscale deviation ratio of the th pixel. Adjustment parameters for each spatial structure reference item, Indicates the first Texture connectivity magnitude of each pixel under the first spatial structure reference term. Indicates the total number of spatial structure reference items. This represents the normalized texture gradient term for the i-th pixel. This represents the average texture gradient at the corresponding position in the standard template;

[0109] The highlight index refers to a comprehensive index representing the highlight degree of the lesion characteristics of each pixel in a target region in an automatic lesion recognition ultrasound system, which is obtained by quantitative calculation based on multiple features such as gray scale, texture, and spatial structure after the lesion determination priority of each pixel in the target region is ranked, and is used to measure the abnormality or priority of a single pixel in the current analysis frame and region. The calculation is based on features such as gray scale deviation, texture connectivity under structural reference, and the difference between the texture gradient of the corresponding region of the standard template. The index reflects the "highlight degree" of each pixel in the automatic labeling process of the lesion region, that is, the pixel is more likely to be determined as a lesion region due to abnormal structure, gray scale distribution, or texture connection characteristics, and is then given a higher rendering level in subsequent steps to achieve automatic highlight display.

[0110] When comparing the central texture mutation degree, the gray scale differences of the target pixel in the four neighborhood directions are 6, 7, 5, and 8, respectively, and the original texture mutation intensity is calculated as The average texture gradient corresponding to the standard template is The lower limit of the region template is 4 and the upper limit is 8, and a linear normalization method is used, and the normalized value is , The gray scale deviation ratio is The original ratio is obtained by the ratio of the pixel gray scale 143 and the average gray scale 126 of the current region, which is The normalized value of the maximum ratio 1.1 and the minimum ratio 1 of the corresponding region is The average gray scale deviation ratio of the same region pixel set is The spatial structure reference item takes three direction feature channels The texture connectivity amplitude under each item is

[0111] , The parameter value is a dimensionless ratio feature and does not need to be normalized. The corresponding adjustment parameters are set as , The step-by-step calculation is as follows:

[0112] ;

[0113] ;

[0114] ;

[0115] ;

[0116] ;

[0117] ​Substitute calculation:

[0118] ;

[0119] Divide into three hierarchical paragraphs for marking decision logic:

[0120] When , it means that the pixel does not show obvious abnormalities in the three dimensions of gray offset, texture mutation and structure connectivity, and is classified as a non-lesion area, without any labeling;

[0121] When , it means that the pixel features have a certain degree of deviation, but do not have strong consistency or significant lesion indications, and are classified as low-level highlight labeling area. The system will decide whether to attach a label according to the spatial continuity and neighborhood conditions;

[0122] When , it means that the features of the pixel are highly consistent with the characteristics of the lesion area recognized by the system, and are classified as high-priority highlight labeling area. The pixel is directly included in the lesion labeling set.

[0123] The index obtained by this calculation falls into the last interval , so it can be determined that the pixel belongs to the "high-priority highlight labeling area"; the result shows that the pixel has high matching accuracy in structure and texture, and outstanding local gray abnormality. Its comprehensive characteristics meet the spatial distribution characteristics of the lesion area and the set standard of the organization structure response model, so its pixel position index, hierarchical label and rendering priority will be registered in the lesion area labeling data for subsequent call processing. The formula can uniformly measure multi-dimensional features by introducing normalized gray and texture parameters, taking into account abnormal deviation and structural synergy, and improve the accuracy and response ability of local lesion pixel recognition.

[0124] The above is only a preferred embodiment of the present application, and does not limit the form of the present application. Any skilled person in the art can use the disclosed technical content to make changes or modifications to equivalent embodiments applied to other fields, but any simple modification, equivalent change and modification made according to the technical essence of the present application to the above embodiments without departing from the technical solution content of the present application still belongs to the protection scope of the present application.

Claims

1. An automatic lesion identification ultrasound system for real-time monitoring, characterized in that, The system includes: The delay path discrimination module is based on continuous ultrasound frames, detects the echo arrival time of pixels in multiple frames, judges the pixel delay change trend, identifies abnormal sound density signals, and obtains the penetration dynamic discrimination parameters. The delay path discrimination module includes: The data frame receiving submodule analyzes the deep tissue echo signal of each pixel based on continuous ultrasound frames, compares the pixel position of each frame point by point, judges the change in the echo arrival time of the same pixel between adjacent frames, calculates the temporal difference of each pixel in the frame sequence, and obtains pixel delay data. The path delay extraction submodule, based on the pixel delay data, determines the pixel delay change trend, analyzes the change direction and continuity, identifies pixels with stable trends, optimizes the correspondence between spatial location and time series, and obtains delay trend data. Based on the delay trend data, the frequency adjustment and recognition submodule compares the delay characteristics with the echo response of the current acoustic density detection area, analyzes the correlation between the echo path and response mode of deep tissue, adjusts the ultrasonic emission frequency, identifies pixel areas with inconsistent structural features, and obtains the penetration dynamic discrimination parameters. The grayscale trend detection module compares the pixel grayscale intensity of consecutive frames in the target area based on the penetration dynamic discrimination parameters, identifies the difference in grayscale change rate, filters out inconsistent signals, and obtains a trend offset feature set. The grayscale trend detection module includes: The grayscale comparison analysis submodule detects the pixel grayscale intensity sequence within the target area based on the penetration dynamic discrimination parameters, calculates the rate of change of corresponding pixel grayscale in adjacent frames, identifies pixels with different grayscale change amplitudes, and performs grayscale temporal aggregation according to the grayscale change direction of the frame sequence to obtain grayscale change feature data. Based on the grayscale change feature data, the structural relationship recognition submodule determines the spatial distribution of pixel grayscale differences within the target area, analyzes the structural relationship between pixel grayscale differences and adjacent pixels, compares grayscale change trends and abnormal behavior characteristics, identifies pixels with matching spatial features, and obtains grayscale structure conformity index. The trend offset screening submodule analyzes the gray-scale change pattern of each pixel in the time series based on the gray-scale structure conformity index, judges the correlation between its change trend and abnormal features, identifies pixels whose change magnitude is inconsistent with abnormal features, and obtains the trend offset feature set. Based on the trend offset feature set, the texture feature screening module calculates the texture intensity difference of high-frequency reflection points in the target area, analyzes the reflection feature trend of the center pixel, compares the consistency of texture changes between the target area and the labeled coordinates, and obtains the high-frequency fusion feature quantity. Based on the high-frequency fusion feature quantity, the feature fusion and sorting module judges the combination of multiple features in the target area, analyzes the weight distribution of the trend offset feature set and the penetration dynamic discrimination parameter, optimizes the priority of the intersection point of multiple features, and obtains the priority index for lesion judgment. Based on the lesion determination priority index, the region highlighting and annotation module identifies the priority region of lesion features in the image frame, analyzes the matching degree between the region and the standard template, performs highlighting and marking on the target pixels, and obtains lesion region annotation data. The region highlighting module includes: The pixel feature rendering submodule optimizes the grayscale distribution, boundary features and texture structure of the pixel set based on the lesion determination priority index, compares the rendering levels required for the classification, determines the brightness adjustment configuration affected by the confidence parameter, and obtains a multi-parameter pixel rendering sequence. The region matching submodule determines the texture structure, grayscale contour and lateral symmetry of the target region based on the multi-parameter pixel rendering sequence, calculates the matching with the standard template reference vector and texture reference, identifies the performance that conforms to the structural consistency characteristics, and obtains the structural consistency matching index. The highlight data generation submodule, based on the aforementioned structural consistency matching index, compares the degree of abrupt changes in the central texture, the grayscale deviation ratio, and the connectivity of the boundary texture, using the formula: ； Calculate the highlighting index, combine it with the region category and structural distribution number, and perform highlighting marking on the target pixels to obtain the lesion region annotation data. Indicates the first Highlighting index of each pixel Indicates the first The grayscale deviation of each pixel This represents the average grayscale deviation ratio of all pixels within the current analysis area. Indicates the first Adjustment parameters for each spatial structure reference item, Indicates the first The pixel in the first Texture connectivity amplitude under each spatial structure reference item Indicates the total number of spatial structure reference items. Indicates the first The normalized texture gradient term for each pixel. This represents the average texture gradient at the corresponding position in the standard template.

2. The automatic lesion identification ultrasound system for real-time monitoring according to claim 1, characterized in that, The texture feature screening module includes: The high-frequency reflection point identification submodule analyzes the grayscale sequence of the center pixel and its neighborhood in the target area based on the trend offset feature set, determines whether the direction of grayscale change between frames switches frequently, identifies pixels that have reversed direction multiple times in multiple consecutive frames, and obtains a high-frequency reflection point mapping set. The reflection trend matching submodule analyzes the grayscale sequence of each point in consecutive frames based on the high-frequency reflection point mapping set, determines whether the direction of grayscale change between frames is consistent with the trend type, compares the matching of the direction distribution of the target area with the reference direction, identifies reflection points that meet the direction requirements, and obtains the reflection trend direction distribution structure. The spatial consistency extraction submodule calculates the difference in grayscale change parameters between the target area and the labeled coordinates in each direction based on the reflection trend direction distribution structure, compares the change characteristics in the horizontal and vertical directions, optimizes the difference expression between the two regions, performs inter-regional difference analysis, obtains the spatial consistency offset, and obtains the high-frequency fusion feature quantity.

3. The automatic lesion identification ultrasound system for real-time monitoring according to claim 1, characterized in that, The feature fusion and sorting module includes: The multi-feature combination analysis submodule analyzes spatial coordinates based on the high-frequency fusion feature quantity, determines the corresponding region of the coordinate point in the trend offset feature set and the penetration dynamic discrimination parameter, and compares the joint distribution of various features at the same spatial location to obtain feature combination association data. The feature weight calculation submodule analyzes the spatial distribution of each feature within the feature overlap region based on the feature combination association data, determines the distribution density of each feature within the target region, optimizes the combination structure of various features within the region, and obtains the feature distribution weight parameters. The priority level division submodule compares the weight ranking of multiple feature intersection areas based on the feature distribution weight parameters, judges the feature combination performance of the center point of the feature overlap area, identifies the spatial location with multiple feature aggregation advantages, divides the feature performance priority level, and obtains the lesion judgment priority index.

4. The automatic lesion identification ultrasound system for real-time monitoring according to claim 1, characterized in that, The continuous ultrasound frame refers to multiple ultrasound image frames acquired in real time and arranged sequentially at short time intervals. The high-frequency reflection point refers to the pixel point in the ultrasound image that exhibits high-frequency gray-scale changes. The texture intensity difference refers to the gray-scale intensity difference between any two high-frequency reflection points in the target area.

Images