A valve disease postoperative recovery evaluation system based on image recognition

Abstract

The application relates to the field of medical image processing and image recognition, in particular to a valve disease postoperative recovery evaluation system based on image recognition. The method comprises the following steps: connecting a hospital PACS and an EMR system, and constructing a standardized multi-time sequence postoperative image data set; after standardization processing, the system fuses the heart anatomical structure features, spatial position information and image gray features, adopts a multi-feature fusion algorithm to extract a heart region of interest and focuses on a valve related myocardial region; stable structure features are extracted in the myocardial middle layer, high-precision space-time alignment of the multi-time sequence images is realized, and structure change quantitative analysis is realized; the application combines the image analysis result and clinical information, comprehensively evaluates the valve disease postoperative recovery state, and outputs the result, so that the accuracy of postoperative image analysis, the time sequence consistency, the quantification and the objectivity of the recovery evaluation are improved, and reliable technical support is provided for clinical postoperative diagnosis and treatment.

Description

Technical Field

[0001] This invention relates to the field of medical image processing and image recognition technology, specifically to an image recognition-based postoperative recovery assessment system for valvular heart disease. Background Technology

[0002] Valvular heart disease is a common cardiovascular disease in clinical practice. Accurate assessment of postoperative recovery is crucial for optimizing treatment plans and improving patient prognosis. Currently, clinical postoperative assessment of valvular heart disease mainly relies on doctors manually analyzing postoperative images such as echocardiography and CT scans, using augmented reality imaging, and then making subjective judgments based on clinical medical records. This method is significantly influenced by the doctor's experience, and the assessment results lack objectivity and standardization. Furthermore, it is impossible to accurately compare images from multiple time points, making it difficult to quantify and capture subtle structural changes in the valve-related myocardial region, and it is easy to miss abnormal recovery signals.

[0003] With the development of medical imaging and image recognition technologies, related systems have been gradually applied in clinical practice, but there are still obvious shortcomings. Existing systems mostly rely on grayscale features or anatomical structural features to extract regions, without achieving effective fusion of multiple features, resulting in low accuracy in extracting regions of interest in the heart and an inability to accurately focus on valve-related myocardial regions. At the same time, they lack the ability to perform high-precision alignment and quantitative analysis of multi-temporal images, making it difficult to explore the correlation between temporal changes in images and postoperative recovery, and failing to meet the core clinical needs for standardized and quantitative assessment.

[0004] To address this, an image recognition-based postoperative recovery assessment system for valvular heart disease is proposed. Summary of the Invention

[0005] This invention provides an image recognition-based postoperative recovery assessment system for valvular heart disease, comprising five modules: image data construction, processing, feature recognition, multi-temporal image analysis, and assessment output. The system interfaces with hospital PACS and EMR systems to construct a standardized multi-temporal postoperative image dataset. After preprocessing, it integrates cardiac anatomical features, valve spatial location, and grayscale distribution characteristics. A multi-feature fusion algorithm extracts regions of interest in the heart and focuses on valve-related myocardial regions. It extracts stable structural features of the myocardial midlayer to achieve high-precision alignment and quantitative analysis of structural changes across multiple temporal images. Finally, by combining the quantitative image results with clinical information, it completes a comprehensive assessment of the postoperative recovery status of valvular heart disease and outputs the results.

[0006] To achieve the above objectives, the present invention provides the following technical solution: Postoperative imaging data construction module: used to acquire multimodal medical images and corresponding clinical information of patients with valvular heart disease at different time points after surgery, and to establish a unified format and time stamp to form a multi-time series of postoperative imaging data; Image processing module: Performs image processing on the multi-time series postoperative image data set to focus on valve-related anatomical structures, so that the images are focused on the valve-related myocardial region, forming structurally focused image data; Stable structure feature recognition module: Based on the image recognition model, the module extracts features from the focused image data of the structure, and uses the myocardial mid-layer region as the stable feature extraction region. Stable structure features are obtained by comparing features from multiple time-series images. Multi-temporal image alignment and change analysis module: Based on the stable structural features, establish the structural correspondence between postoperative images at different time points and generate postoperative structural change indicators for valvular heart disease. Recovery assessment and result output module: This module is used to assess the postoperative recovery status of patients with valvular heart disease by combining the structural change indicators and corresponding clinical information, and output the assessment results.

[0007] Preferably, the step of forming a multi-temporal postoperative image dataset includes: The postoperative image data construction module acquires multimodal medical images and corresponding clinical information of patients with valvular heart disease at different time points after surgery, and performs format unification and standardization processing on the multimodal medical images and clinical information. Postoperative time node identifiers are added to the standardized multimodal medical images and corresponding clinical information, and the data are integrated according to the order of the time node identifiers to form a multi-temporal postoperative image data set with a clear time order.

[0008] Preferably, the standardization process includes the following steps: The images in the multi-temporal postoperative image dataset are classified according to modality to form corresponding modal image subsets; For different modal subsets of images, corresponding interference suppression processing is performed to reduce the impact of noise, artifacts and modal differences on image analysis while preserving the original anatomical features of valve-related myocardial regions. The subsets of images from each modality, after interference suppression processing, are integrated to form standardized multi-temporal postoperative image data.

[0009] Preferably, the step of focusing on the valve-related myocardial region includes: After standardizing the images in the multi-temporal postoperative image dataset, the cardiac region of interest was extracted from the images, and interfering regions that were irrelevant to cardiac tissue or affected the analysis were removed. The image processing module limits the scope of image feature analysis to the myocardial tissue region around the valve, forming structurally focused image data for subsequent feature extraction.

[0010] Preferably, the step of extracting the region of interest in the heart includes: Based on standardized multi-temporal postoperative imaging data, the cardiac anatomical regions corresponding to the valve structures in the images were located. Based on the spatial relationships and structural boundary features of the aforementioned cardiac anatomical regions, the initial extent of the valve-related myocardial region is determined; During the determination and correction of the initial range, a multi-feature fusion algorithm is used to jointly constrain anatomical structural features and imaging features, and then perform regional constraints and continuity correction on the constrained region to form a cardiac region of interest for subsequent analysis.

[0011] Preferably, the steps of the multi-feature fusion algorithm include: Obtain anatomical structural features and corresponding imaging features within the region of interest of the heart; The anatomical structural features and image features are aligned and jointly modeled to form a fused feature representation. Based on the fused feature representation, feature constraint optimization is performed on the cardiac region of interest to improve the continuity and accuracy of the cardiac region of interest extraction results in multiple temporal images.

[0012] Preferably, the step of obtaining stable structural features includes: In structural focusing imaging data, the myocardial mid-layer region was identified as the main region for extracting stable features; Based on the image recognition model, anatomical structural morphological features, tissue texture features, and temporal motion features of the myocardial mid-layer region are extracted. Through multi-temporal feature comparison and screening, stable structural features of the myocardial mid-layer with temporal consistency and structural stability are obtained.

[0013] Preferably, the step of generating postoperative structural change indicators for valvular heart disease includes: Based on the stable structural features, multi-temporal alignment is performed on the structural focusing images at different time points to establish corresponding structural matching relationships. The changes in stable structural features in aligned multi-temporal images are quantitatively analyzed to generate structural change indicators that reflect the postoperative recovery of valvular heart disease. These structural change indicators include morphological change indicators and motion characteristic change indicators of the myocardial midlayer.

[0014] The beneficial effects of this invention are: By integrating medical image processing and image recognition technologies into the postoperative recovery assessment of valvular heart disease, this system effectively addresses the technical pain points of existing assessments, such as high subjectivity, low accuracy, insufficient quantification, and the lack of multi-feature fusion and weak temporal analysis capabilities. The system integrates cardiac anatomical features, valve spatial location, and image grayscale distribution characteristics, combined with a multi-feature fusion algorithm, to accurately extract regions of interest and focus on valve-related myocardial regions. This avoids the accuracy limitations of single-feature extraction, improving the accuracy and specificity of image region extraction. Utilizing the stable structural features of the myocardial midlayer, it achieves high-precision spatiotemporal alignment of postoperative images across multiple time sequences, accurately quantifying and capturing subtle structural changes in valve-related myocardial regions. This avoids the problems of inaccurate temporal comparisons and the tendency to miss early recovery abnormal signals in traditional manual assessments. Simultaneously, by combining the results of quantitative image analysis with clinical information for comprehensive evaluation, it significantly reduces reliance on physician experience, significantly improving the objectivity, standardization, and quantification of postoperative recovery assessment. This provides reliable technical support for clinical development of optimized treatment plans, contributing to improved overall postoperative treatment outcomes for valvular heart disease. Attached Figure Description

[0015] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is a structural diagram of a postoperative recovery assessment system for valvular heart disease based on image recognition, provided by the present invention. Detailed Implementation

[0016] The preferred embodiments of the present invention will be described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are for illustration and explanation only and are not intended to limit the present invention.

[0017] A postoperative recovery assessment system for valvular heart disease based on image recognition, such as Figure 1 As shown, it includes: Postoperative imaging data construction module: used to acquire multimodal medical images and corresponding clinical information of patients with valvular heart disease at different time points after surgery, and to establish a unified format and time stamp to form a multi-time series of postoperative imaging data; Furthermore, the step of forming a multi-temporal postoperative image data set includes: The postoperative image data construction module acquires multimodal medical images and corresponding clinical information of patients with valvular heart disease at different time points after surgery, and performs format unification and standardization processing on the multimodal medical images and clinical information. Postoperative time node identifiers are added to the standardized multimodal medical images and corresponding clinical information, and the data are integrated according to the order of the time node identifiers to form a multi-temporal postoperative image data set with a clear time order.

[0018] Furthermore, the standardization process includes the following steps: The images in the multi-temporal postoperative image dataset are classified according to modality to form corresponding modal image subsets; For different modal subsets of images, corresponding interference suppression processing is performed to reduce the impact of noise, artifacts and modal differences on image analysis while preserving the original anatomical features of valve-related myocardial regions. The subsets of images from each modality, after interference suppression processing, are integrated to form standardized multi-temporal postoperative image data.

[0019] Specifically, by connecting to the hospital's PACS and EMR systems, the system captures multimodal medical images of valvular heart disease patients at different time points after surgery (such as 1 day, 1 week, 1 month, 3 months after surgery), including core image types such as cardiac ultrasound, CT, and MRI images; at the same time, it acquires corresponding clinical information, covering clinical data related to postoperative recovery assessment, such as patient basic information, surgical parameters, postoperative signs, and laboratory test results.

[0020] The acquired imaging data and clinical information are screened to remove blurry, distorted, or incomplete invalid images, as well as redundant clinical data that is irrelevant to postoperative recovery of valvular heart disease. The screened valid data is cleaned to correct data entry errors and supplement missing key identification information to ensure the integrity and accuracy of the data.

[0021] Medical images of different modalities and formats are uniformly converted into a preset standard format, and parameters such as image resolution and pixel depth are standardized to eliminate image differences caused by different devices and formats; clinical information is structured and entered according to preset field specifications to form a unified format clinical dataset.

[0022] Each postoperative imaging data is assigned a unique time identifier to clearly mark the postoperative time node corresponding to the image, ensuring the temporal correlation of multiple time-series images; multiple time-series imaging data and corresponding clinical information of the same patient are linked and bound through the patient's unique identifier to achieve a one-to-one correspondence between images and clinical data.

[0023] By integrating standardized, time-stamped, and correlated multi-temporal imaging data and clinical data, a multi-temporal postoperative imaging dataset is constructed. The constructed dataset is classified and stored, and an indexing mechanism is established to facilitate quick access, querying, and management of data by subsequent modules, providing standardized and regulated data support for subsequent image processing and feature recognition.

[0024] Image processing module: Performs image processing on the multi-time series postoperative image data set to focus on valve-related anatomical structures, so that the images are focused on the valve-related myocardial region, forming structurally focused image data; Furthermore, the step of focusing on the valve-related myocardial region includes: After standardizing the images in the multi-temporal postoperative image dataset, the cardiac region of interest was extracted from the images, and interfering regions that were irrelevant to cardiac tissue or affected the analysis were removed. The image processing module limits the scope of image feature analysis to the myocardial tissue region around the valve, forming structurally focused image data for subsequent feature extraction.

[0025] Furthermore, the step of extracting the region of interest for the heart includes: Based on standardized multi-temporal postoperative imaging data, the cardiac anatomical regions corresponding to the valve structures in the images were located. Based on the spatial relationships and structural boundary features of the aforementioned cardiac anatomical regions, the initial extent of the valve-related myocardial region is determined; During the determination and correction of the initial range, a multi-feature fusion algorithm is used to jointly constrain anatomical structural features and imaging features, and then perform regional constraints and continuity correction on the constrained region to form a cardiac region of interest for subsequent analysis.

[0026] Specifically, the multi-temporal images and related information output by the postoperative image data construction module are briefly imported, and basic parameters such as image pixel coordinate system and slice thickness are unified to eliminate residual differences in format standardization, laying the foundation for subsequent processing.

[0027] Preprocessing is performed on images of different modalities. Ultrasound images are filtered using Gaussian filtering to remove noise artifacts, while CT and MRI images are enhanced by equalizing grayscale histograms to improve the grayscale contrast of valve and myocardial regions, highlighting anatomical contours and facilitating subsequent localization.

[0028] Based on a pre-defined valvular anatomical template (covering common valve types such as the mitral and aortic valves, constructed using prior clinical anatomical knowledge, and including key parameters such as typical valve contours and size ranges), and combined with contour detection algorithms and normalized cross-correlation similarity measurement methods from image recognition, coarse localization is performed on the preprocessed image. First, a template matching function is called to match the pre-defined valve template with the image region by region, calculating the similarity coefficient for each region. A similarity threshold is set (ranging from 0.7 to 0.8), and regions with similarity coefficients higher than the threshold are selected, quickly pinpointing the approximate area in the image related to the valve structure. Simultaneously, irrelevant tissues far from the valve (such as the lungs, pleural fat, and blood vessels) are excluded, narrowing the processing range for subsequent precise segmentation. This improves processing efficiency and reduces interference from irrelevant regions, ensuring the targeted nature of the coarse localization area.

[0029] Furthermore, the steps of the multi-feature fusion algorithm include: Obtain anatomical structural features and corresponding imaging features within the region of interest of the heart; The anatomical structural features and image features are aligned and jointly modeled to form a fused feature representation. Based on the fused feature representation, feature constraint optimization is performed on the cardiac region of interest to improve the continuity and accuracy of the cardiac region of interest extraction results in multiple temporal images.

[0030] Specifically, a multi-feature fusion segmentation algorithm is invoked to fuse three core features: cardiac anatomical feature points, valve spatial location, and image grayscale distribution. This allows for precise segmentation of the coarsely localized region, enabling accurate extraction of valve-related myocardial regions. First, key anatomical features such as the apex of the valve annulus, the inflection point of the myocardial wall boundary, and the connection point between the valve and myocardium are extracted. Let the set of feature points be denoted as . : The formula for calculating the distance from each feature point to the valve center is as follows: ; in, Let be the distance from the i-th key anatomical feature point to the center of the valve; Let x be the x-coordinate of the i-th key anatomical feature point in the image coordinate system; Let be the ordinate of the i-th key anatomical feature point in the image coordinate system; The x-coordinate of the valve center in the image coordinate system; The vertical coordinate of the valve center in the image coordinate system is given.

[0031] Anatomical feature vectors are constructed based on the distances between each feature point. The vectors are then normalized to ensure consistent feature dimensions. Secondly, based on prior knowledge of cardiac anatomy, the spatial region containing the valves is defined, and the minimum bounding rectangle of the valve region is determined. Calculate any pixel in the image To rectangle Spatial membership The formula for quantifying the spatial correlation between pixels and the valve region is: ; in, Spatial attenuation coefficient (range of values) Based on experimental verification, the optimal value was determined to be 0.05. pixel to rectangle The shortest Euclidean distance is used, and the spatial membership value ranges from [0,1]. The closer the value is to 1, the closer the pixel is to the core region of the valve. Based on this, a spatial location feature vector is constructed. (The core element is) ); Finally, calculate any pixel in the image. The original grayscale value To eliminate the influence of differences in grayscale value ranges, normalization is performed using a formula: ; in, These are the maximum and minimum gray values ​​of the coarsely located region image, respectively. The normalized gray value range is [0,1]. Constructing grayscale distribution feature vectors (The core element is) ), quantify the grayscale features of pixels to highlight the grayscale difference between the valve-related myocardial region and the background region.

[0032] Based on clinical evidence and validation from multiple sets of experimental data, and considering the importance of three types of features in valve-related myocardial region segmentation, reasonable fusion weights were assigned to the three types of features: anatomical feature weights. (Anatomical features directly determine the positional relationship between the valve and the myocardium), spatial position feature weights (Locking the myocardial region surrounding the valve), grayscale distribution feature weights (Distinguish the grayscale differences between myocardial tissue and surrounding unrelated tissues); use a weighted summation model to calculate the fusion feature value of each pixel. To achieve deep fusion of the three types of features, the formula is: ; in, For pixels The corresponding anatomical feature normalization value, the fusion feature value combines the advantages of the three types of features, effectively improves the accuracy of pixel classification, and provides a reliable basis for subsequent threshold segmentation.

[0033] The optimal segmentation threshold is calculated using the maximum inter-class variance method. This allows for the determination of the optimal threshold, maximizing the distinction between valve-related myocardial regions and background regions. The formula is as follows: ; in, Thresholds The pixel percentage of the two segmented regions (target region and background region). These are the grayscale mean values ​​for the two types of regions, The mean gray level of the entire image for the coarsely located region is taken as the inter-class variance. The largest As the optimal segmentation threshold Segmentation is determined based on the optimal threshold: if If the target region is identified as the valve-related myocardial region, then the target region is identified as the background region, thus completing the initial segmentation. To eliminate isolated noise points, regional holes, and jagged edges in the initial segmentation results, morphological opening and closing operations were used to optimize the segmentation results. The images of valve-related myocardial regions after precise segmentation were calibrated to correct structural shifts caused by differences in shooting angle and patient position, ensuring the consistency of the target region in images of different time series and modalities. At the same time, the image edges were smoothed and the image details were optimized. The calibrated and optimized images of valve-related myocardial regions are categorized and organized according to time markers to generate structural focusing image data. Processing markers are added and patient information is associated with the data, which is then output to the stable structural feature recognition module to provide accurate data support for subsequent feature extraction.

[0034] Stable structure feature recognition module: Based on the image recognition model, the module extracts features from the focused image data of the structure, and uses the myocardial mid-layer region as the stable feature extraction region. Stable structure features are obtained by comparing features from multiple time-series images. Furthermore, the step of obtaining stable structural features includes: In structural focusing imaging data, the myocardial mid-layer region was identified as the main region for extracting stable features; Based on the image recognition model, anatomical structural morphological features, tissue texture features, and temporal motion features of the myocardial mid-layer region are extracted. Through multi-temporal feature comparison and screening, stable structural features of the myocardial mid-layer with temporal consistency and structural stability are obtained.

[0035] Specifically, the structural focused image and associated parameters output by the image processing module are imported, and the focused image is fine-tuned to unify the pixel resolution and grayscale threshold, remove minor noise at the edges, and ensure that the myocardial region has a clear outline, thus laying the foundation for subsequent feature extraction.

[0036] Based on prior knowledge of cardiac anatomy and image segmentation algorithms, the myocardial region in the structural focusing image is segmented into layers. By comparing the grayscale and texture differences of different layers of the myocardium, the middle layer of the myocardium (which is least affected by postoperative edema and inflammation and is the core area for stable feature extraction) is accurately located. The outer and inner layers of the myocardium, which are easily disturbed, are removed, thus narrowing the feature extraction range.

[0037] The preset image recognition model is invoked to perform preliminary multi-dimensional feature extraction on the localized myocardial mid-layer region. The core extraction consists of two main categories: morphological features and texture features. Texture features are accurately extracted using the gray-level co-occurrence matrix algorithm, while morphological features are used as supplementary features to ensure that the extracted features are comprehensive, accurate, and can effectively reflect the stable structural characteristics of the myocardial mid-layer.

[0038] The focused image of the myocardial mid-layer region is subjected to gray-level compression processing, compressing the original 256 gray levels to L levels (L is 16~32, preferably 24 levels). This can effectively reduce the computational complexity of subsequent algorithms and reduce the computation time, while preserving the core texture and gray-level features of the myocardial mid-layer region, avoiding feature redundancy caused by too many gray levels, and ensuring that the compressed image can still clearly reflect the microstructural differences of myocardial tissue.

[0039] For the middle layer of the myocardium, core morphological features reflecting its overall structural morphology are extracted, mainly including four categories: contour perimeter, region area, myocardial thickness, and morphological complexity.

[0040] The four categories of morphological features together constitute a subset of morphological features, providing a foundation for subsequent stable feature selection.

[0041] The gray-level co-occurrence matrix algorithm is used to accurately extract texture features. This algorithm can effectively capture the gray-level distribution patterns and micro-texture changes within myocardial tissue, and is suitable for extracting stable texture features from the middle layer of myocardium. The formula for constructing the gray-level co-occurrence matrix is ​​as follows: ; in, As the core element of the gray-level co-occurrence matrix, it represents the probability that a pixel with gray level i and a pixel with gray level j will appear simultaneously in an image of the middle layer of myocardium under specific distance and orientation conditions. This represents the gray level of a single pixel in a mid-myocardial image after gray-level compression. The gray level of another pixel that satisfies specific distance and orientation conditions with respect to a pixel with gray level i. The distance (step) between two pixels, which takes a value of 1 to 3 pixels (preferably 2), is used to define the pixel spacing for texture feature extraction. The relative direction between two pixels; The total number of all pixel pairs in the myocardial mesothelial region that satisfy the following conditions: pixel spacing d, relative direction θ, and gray levels i and j. This represents the total number of all possible pixel pairs within the myocardial midlayer region that satisfy a pixel spacing of d and a relative direction of θ.

[0042] Based on the normalized gray-level co-occurrence matrix, four core texture features that can reflect the stability of the texture of the myocardial mesolayer are calculated: contrast (CON), uniformity (ASM), entropy (ENT), and correlation (COR). These four features capture the texture information of the myocardial mesolayer from different dimensions, are highly complementary, and can comprehensively reflect the microstructural characteristics of myocardial tissue. The mean fusion algorithm is used to take the average of similar texture features in four directions as the final value, ensuring comprehensive and stable texture features. The fusion formula is as follows: ; in, Let be the texture feature value in the k-th direction. Each direction corresponds to four feature values: contrast, uniformity, entropy, and correlation. The mean of the same type of texture features in the four directions is calculated as the final value of that type of texture feature. The final texture features of the four categories together constitute a subset of texture features.

[0043] The extracted morphological feature subsets (contour perimeter, region area, myocardial thickness, morphological complexity) are integrated with the fused texture feature subsets (contrast, uniformity, entropy, correlation) to construct a multi-dimensional initial feature set. This feature set comprehensively covers the macroscopic morphological and microscopic texture features of the myocardial midlayer, providing a sufficient feature foundation for subsequent stable feature selection. This ensures that the selected stable features can accurately reflect the structural stability of the myocardial midlayer and meet the needs of subsequent multi-temporal image alignment and change analysis.

[0044] The initial multidimensional features of the myocardium midlayer extracted from multiple time-series images of the same patient were compared one by one. A reasonable stability threshold was set, and features with changes less than the threshold across time nodes (stable features) were selected. Unstable features that are prone to fluctuation or are affected by temporary changes after surgery were removed, and core stable features were retained.

[0045] The selected stable features are weighted and fused together to form a unified set of stable myocardial midlayer structure features by combining the stability weights of each feature. At the same time, the features in the feature set are standardized to eliminate the differences in feature dimensions caused by different time points and different modal images, so as to ensure that the stable features of each time series are comparable.

[0046] The fused and standardized stable structural feature set is validated to determine whether the stability and recognizability of the features meet the requirements of subsequent multi-temporal image alignment. After successful validation, a unique patient identifier and time identifier are added to the feature set, and the corresponding images and processing parameters are associated with it. The result is then output to the multi-temporal image alignment and change analysis module to provide core feature support for subsequent modules.

[0047] Multi-temporal image alignment and change analysis module: Based on the stable structural features, establish the structural correspondence between postoperative images at different time points and generate postoperative structural change indicators for valvular heart disease. Furthermore, the step of generating postoperative structural change indicators for valvular heart disease includes: Based on the stable structural features, multi-temporal alignment is performed on the structural focusing images at different time points to establish corresponding structural matching relationships. The changes in stable structural features in aligned multi-temporal images are quantitatively analyzed to generate structural change indicators that reflect the postoperative recovery of valvular heart disease. These structural change indicators include morphological change indicators and motion characteristic change indicators of the myocardial midlayer.

[0048] Specifically, the system receives the stable structural feature set of the myocardial midlayer, related focused image data of the structure, and corresponding time markers output by the stable structural feature recognition module. It completes data reception and preliminary verification, removing abnormal samples with missing data or incorrect labeling. The alignment benchmark prioritizes stable texture features (features extracted through a gray-level co-occurrence matrix algorithm, such as contrast and uniformity) and stable morphological features (middle myocardial thickness, area, etc.) of the myocardial midlayer region, constructing a unified feature reference set. This reference set possesses advantages such as strong stability across time nodes, high recognition accuracy, and minimal susceptibility to temporary interference from postoperative edema and inflammation. It can effectively avoid alignment deviations caused by patient position shifts, differences in shooting angles, and fluctuations in equipment parameters in images from different time sequences. During the alignment process, the stability features of the myocardium midlayer in the earliest postoperative time series image (1 day postoperatively) were used as the baseline feature set. Subsequent time series images (1 week, 1 month postoperatively, etc.) were all referenced to this baseline feature set. Through the feature point matching algorithm, the stability features of subsequent time series images were accurately matched with the baseline feature set to achieve pixel-level alignment of the images. This ensured that the spatial location and anatomical structure of the valve-related myocardial region remained consistent in different time series images, laying a precise spatial foundation for subsequent structural change analysis.

[0049] The alignment of multi-temporal images is verified using a preset alignment error threshold (error less than 2 pixels). If the alignment error exceeds the threshold, the alignment is returned for re-alignment. If the verification is successful, the subsequent change analysis stage is entered to ensure that the alignment accuracy meets the requirements of change analysis.

[0050] Based on aligned multi-temporal images, this study focuses on valve-related myocardial regions, exploring two core dimensions: morphological and textural changes. The morphological change dimension primarily quantifies the thickness of the myocardial media, the area of ​​the region, and subtle morphological shifts in the valve-myocardial connection area, using a baseline temporal sequence as a reference to calculate the amount and rate of change of each indicator. The textural change dimension mainly analyzes the dynamic changes in the uniformity, contrast, and entropy of the myocardial media texture, capturing the recovery trend of myocardial microstructure. The overall principle is to quantify core changes and avoid irrelevant interference.

[0051] By integrating quantitative analysis results from two dimensions—morphology and texture—standardized postoperative structural change indicators for valvular heart disease are generated, clearly marking the corresponding time points, change trends, and reference ranges for each indicator. Subsequently, the rationality of the change indicators is verified, and abnormal data caused by image noise and calculation bias are eliminated to ensure the accuracy and reliability of the indicators.

[0052] The standardized structural change indicators after verification are associated and bound with the corresponding time-series imaging data and stable feature data. Patient unique identifiers, time identifiers and processing parameters are added, and the data is uniformly output to the recovery assessment and result output module to provide accurate and reliable quantitative data support for postoperative recovery status assessment.

[0053] Recovery assessment and result output module: This module is used to assess the postoperative recovery status of patients with valvular heart disease by combining the structural change indicators and corresponding clinical information, and output the assessment results.

[0054] Specifically, it receives standardized structural change indicators, associated image data, and stable feature data output by the multi-temporal image alignment and change analysis module, completes data reception and preliminary verification, eliminates abnormal samples, ensures the accuracy of input data, and lays the foundation for recovery assessment.

[0055] Based on the pre-set clinical assessment criteria and the quantitative data as the core basis, the postoperative recovery status of valvular heart disease was assessed in multiple dimensions, as detailed below: The core assessment criteria are myocardial morphological change indicators (change rate of myocardial media thickness, change in regional area, etc.) and myocardial texture change indicators (uniformity, contrast, entropy value change, etc.). The earliest time sequence after surgery (1 day after surgery) is used as the benchmark. Combined with the clinically pre-set normal reference range, recovery trend threshold and expected standards for different stages after surgery, the recovery status is comprehensively determined.

[0056] The two core assessment dimensions consider both macroscopic and microscopic recovery: First, myocardial structure recovery assessment, focusing on the trend of changes in the morphology of the myocardial midlayer. If relevant quantitative indicators are within normal thresholds and show a stable recovery trend, the myocardial structure is considered to be recovering well; if the indicators are abnormal or show deteriorating fluctuations, the recovery is considered poor, and abnormal indicators and potential risks (such as residual inflammation) are marked. Second, valve-myocardial connection stability assessment, focusing on morphological deviations in the connection area and changes in the texture of the surrounding myocardium. If the deviation is within a safe threshold and the texture indicators tend to be stable, the connection is considered stable; if the indicators are abnormal, the stability is considered insufficient, suggesting potential problems such as poor healing. Finally, the assessment results from both dimensions are combined to comprehensively classify the recovery into four levels: "good recovery," "normal recovery," "poor recovery," and "requiring clinical intervention," forming a multi-time-series recovery trend comparison to ensure that the assessment is objective, accurate, and meets clinical needs.

[0057] By integrating assessment conclusions, multi-time series recovery trends, and relevant quantitative indicators, the rationality of the assessment results is verified to ensure consistency with input data and clinical standards, and to avoid assessment bias.

[0058] The assessment report is output in a standardized clinical report format, linking the patient's basic information, processing parameters of each module, and imaging data, providing accurate reference for clinical diagnosis and follow-up.

[0059] Finally, it should be noted that the above descriptions are merely preferred embodiments of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A postoperative recovery assessment system for valvular heart disease based on image recognition, characterized in that, include: Postoperative imaging data construction module: used to acquire multimodal medical images and corresponding clinical information of patients with valvular heart disease at different time points after surgery, and to establish a unified format and time stamp to form a multi-time series of postoperative imaging data; Image processing module: Performs image processing on the multi-time series postoperative image data set to focus on valve-related anatomical structures, so that the images are focused on the valve-related myocardial region, forming structurally focused image data; Stable structure feature recognition module: Based on the image recognition model, the module extracts features from the focused image data of the structure, and uses the myocardial mid-layer region as the stable feature extraction region. Stable structure features are obtained by comparing features from multiple time-series images. Multi-temporal image alignment and change analysis module: Based on the stable structural features, establish the structural correspondence between postoperative images at different time points and generate postoperative structural change indicators for valvular heart disease. Recovery assessment and result output module: This module is used to assess the postoperative recovery status of patients with valvular heart disease by combining the structural change indicators and corresponding clinical information, and output the assessment results.

2. The image recognition-based postoperative recovery assessment system for valvular heart disease according to claim 1, characterized in that, In the postoperative image data construction module, the step of forming a multi-temporal postoperative image data set includes: The postoperative image data construction module acquires multimodal medical images and corresponding clinical information of patients with valvular heart disease at different time points after surgery, and performs format unification and standardization processing on the multimodal medical images and clinical information. Postoperative time node identifiers are added to the standardized multimodal medical images and corresponding clinical information, and the data are integrated according to the order of the time node identifiers to form a multi-temporal postoperative image data set with a clear time order.

3. The postoperative recovery assessment system for valvular heart disease based on image recognition according to claim 2, characterized in that, The standardization process includes the following steps: The images in the multi-temporal postoperative image dataset are classified according to modality to form corresponding modal image subsets; For different modal subsets of images, corresponding interference suppression processing is performed to reduce the impact of noise, artifacts and modal differences on image analysis while preserving the original anatomical features of valve-related myocardial regions. The subsets of images from each modality, after interference suppression processing, are integrated to form standardized multi-temporal postoperative image data.

4. The postoperative recovery assessment system for valvular heart disease based on image recognition according to claim 1, characterized in that, In the image processing module, the step of focusing on the valve-related myocardial region includes: After standardizing the images in the multi-temporal postoperative image dataset, the cardiac region of interest was extracted from the images, and interfering regions that were irrelevant to cardiac tissue or affected the analysis were removed. The image processing module limits the scope of image feature analysis to the myocardial tissue region around the valve, forming structurally focused image data for subsequent feature extraction.

5. The postoperative recovery assessment system for valvular heart disease based on image recognition according to claim 4, characterized in that, The steps for extracting the region of interest in the heart include: Based on standardized multi-temporal postoperative imaging data, the cardiac anatomical regions corresponding to the valve structures in the images were located. Based on the spatial relationships and structural boundary features of the aforementioned cardiac anatomical regions, the initial extent of the valve-related myocardial region is determined; During the determination and correction of the initial range, a multi-feature fusion algorithm is used to jointly constrain anatomical structural features and imaging features, and then perform regional constraints and continuity correction on the constrained region to form a cardiac region of interest for subsequent analysis.

6. The image recognition-based postoperative recovery assessment system for valvular heart disease according to claim 5, characterized in that, The steps of the multi-feature fusion algorithm include: Obtain anatomical structural features and corresponding imaging features within the region of interest of the heart; The anatomical structural features and image features are aligned and jointly modeled to form a fused feature representation. Based on the fused feature representation, feature constraint optimization is performed on the cardiac region of interest to improve the continuity and accuracy of the cardiac region of interest extraction results in multiple temporal images.

7. The postoperative recovery assessment system for valvular heart disease based on image recognition according to claim 1, characterized in that, In the stable structure feature recognition module, the step of obtaining stable structure features includes: In structural focusing imaging data, the myocardial mid-layer region was identified as the main region for extracting stable features; Based on the image recognition model, anatomical structural morphological features, tissue texture features, and temporal motion features of the myocardial mid-layer region are extracted. Through multi-temporal feature comparison and screening, stable structural features of the myocardial mid-layer with temporal consistency and structural stability are obtained.

8. The postoperative recovery assessment system for valvular heart disease based on image recognition according to claim 1, characterized in that, In the multi-temporal image alignment and change analysis module, the step of generating postoperative structural change indicators for valvular heart disease includes: Based on the stable structural features, multi-temporal alignment is performed on the structural focusing images at different time points to establish corresponding structural matching relationships. The changes in stable structural features in aligned multi-temporal images are quantitatively analyzed to generate structural change indicators that reflect the postoperative recovery of valvular heart disease. These structural change indicators include morphological change indicators and motion characteristic change indicators of the myocardial midlayer.

Images