Intestinal necrosis feature recognition system based on image analysis and storage medium

Abstract

The present application relates to the field of image processing, and specifically to an intestinal necrosis feature recognition system based on image analysis and a storage medium, wherein a computer program in the storage medium performs the following steps: obtaining medical images of a patient and dividing them into an expiration phase subset and an inspiration phase subset; performing spatial registration and motion artifact correction on the expiration phase subset and the inspiration phase subset to obtain multi-modal images; extracting a blood vessel obstruction feature, a perfusion injury feature and a cell necrosis feature in the multi-modal images; constructing a causal hierarchical relationship among the blood vessel obstruction feature, the perfusion injury feature and the cell necrosis feature based on an intestinal necrosis pathological mechanism; and determining a risk level of the patient having intestinal necrosis and a probability corresponding to the risk level based on the causal hierarchical relationship, the blood vessel obstruction feature, the perfusion injury feature and the cell necrosis feature using a random forest algorithm. The present application solves the problem of insufficient accuracy in recognizing intestinal necrosis features in the prior art.

Description

Technical Field

[0001] This invention relates to the field of image processing, specifically to an intestinal necrosis feature recognition system and storage medium based on image analysis. Background Technology

[0002] In the medical field, intestinal necrosis is a serious and rapidly developing intestinal disease. If it is not diagnosed and treated in time, it will greatly threaten the patient's life and health. However, the diagnosis of intestinal necrosis still faces many challenges with current technology.

[0003] From a detection technology perspective, a single imaging method is insufficient to comprehensively and accurately present intestinal lesions. While CT angiography can observe vascular morphology, it is inadequate in displaying changes at the intestinal cellular level; contrast-enhanced ultrasound can reflect blood perfusion but cannot clearly show the fine vascular structures; confocal laser microscopy fluorescence imaging, although capable of observation at the cellular level, struggles to obtain overall vascular and perfusion information. Furthermore, respiratory motion interference can cause image displacement and distortion, degrading image quality and severely impacting the physician's accurate judgment of lesion characteristics, increasing the risk of misdiagnosis and missed diagnosis.

[0004] In terms of data analysis, traditional diagnosis of intestinal necrosis relies heavily on physicians' subjective experience and simple imaging feature analysis, lacking in-depth exploration of the disease's underlying pathological mechanisms. Judging solely by observing obvious lesions in imaging ignores the causal relationship between vascular obstruction, perfusion injury, and cell necrosis. Furthermore, current diagnostic methods cannot quantify the risk of intestinal necrosis in patients, making it difficult for physicians to accurately assess the severity of the patient's condition and consequently hindering the timely development of scientific and effective personalized treatment plans. This not only delays the optimal treatment window but may also lead to overtreatment or undertreatment. Therefore, a more scientific and accurate technological solution for identifying intestinal necrosis is urgently needed, capable of deeply analyzing the pathological mechanisms of intestinal necrosis and achieving quantitative risk assessment. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides an image analysis-based intestinal necrosis feature recognition system and storage medium, which solves the problem of insufficient accuracy in recognizing intestinal necrosis features in existing technologies.

[0006] To achieve the above objectives, the present invention provides a computer-readable storage medium storing a computer program, the computer program including program instructions, which, when executed by a processor, cause the processor to perform the following steps: acquiring expiratory and inspiratory subsets divided from the patient's CT angiography images, ultrasound angiography images, and confocal laser microscopy endoscopic fluorescence images; performing spatial registration and motion artifact correction on the expiratory and inspiratory subsets to obtain a multimodal image; extracting vascular occlusion features, perfusion injury features, and cell necrosis features from the multimodal image; constructing a causal hierarchical relationship between the vascular occlusion features, the perfusion injury features, and the cell necrosis features based on the pathological mechanism of intestinal necrosis; and introducing a random forest algorithm to determine the risk level of the patient having intestinal necrosis and the probability corresponding to the risk level based on the causal hierarchical relationship, the vascular occlusion features, the perfusion injury features, and the cell necrosis features.

[0007] This invention obtains multimodal images by dividing the expiratory and inspiratory phases into subsets and combining spatial registration and motion artifact correction. This effectively overcomes image interference caused by respiratory motion, ensuring image accuracy and stability. Based on the pathological mechanism of intestinal necrosis, it constructs a causal hierarchical relationship between features, enabling in-depth analysis of the internal logic of disease development and understanding the correlation between features from the pathological root. Finally, it introduces a random forest algorithm to determine the risk level and corresponding probability of intestinal necrosis in patients, providing intuitive and quantitative evidence for clinical diagnosis and improving the overall accuracy of intestinal necrosis diagnosis and clinical treatment level.

[0008] Optionally, acquiring the expiratory and inspiratory subsets of the patient's CT angiography images, ultrasound contrast images, and confocal laser microscopy fluorescence images includes: acquiring the patient's body surface movement trajectory and the corresponding time information using an infrared body surface marker sensor placed on the patient's abdomen; and using the body surface movement trajectory and the time information to divide the patient's CT angiography images, ultrasound contrast images, and confocal laser microscopy fluorescence images into expiratory and inspiratory subsets.

[0009] This invention can capture changes in abdominal movement during a patient's breathing process in real time and accurately record the movement trajectory. Further segmentation of the image effectively solves the problem of interference caused by respiratory movement to multimodal image acquisition. Because the movement of abdominal organs during breathing can cause displacement and distortion in images of different modalities, segmentation based on movement trajectory and time information can integrate images of the same breathing phase, improving the accuracy of expiratory and inspiratory subsets.

[0010] Optionally, the step of spatial registration and motion artifact correction of the expiratory and inspiratory subsets to obtain a multimodal image includes: performing motion artifact correction on the first expiratory subset and the first inspiratory subset respectively to obtain a third expiratory subset and a third inspiratory subset; performing cross-modal coarse registration on the third expiratory subset and the second expiratory subset, and performing cross-modal coarse registration on the third inspiratory subset and the second inspiratory subset respectively to obtain a fourth expiratory subset and a fourth inspiratory subset; and performing elastic registration based on grayscale information on the fourth expiratory subset and the fourth inspiratory subset respectively to obtain a multimodal image.

[0011] This invention effectively improves the quality of multimodal images by performing spatial registration and motion artifact correction on subsets of the expiratory and inspiratory phases. Motion artifact correction eliminates image displacement and deformation caused by respiratory motion, resulting in clear details and structural stability in the third expiratory and inspiratory phase subsets. Cross-modal coarse registration unifies the spatial coordinates of images from different modalities, solving the problem of large-scale spatial positional deviations and enhancing the comparability between images. Elastic registration based on grayscale information further optimizes the registration effect, compensating for subtle tissue deformations and grayscale differences between modalities, achieving pixel-level precise alignment. The resulting multimodal images are highly consistent in space, integrating complementary information from CT angiography, ultrasound contrast imaging, and confocal laser microscopy fluorescence images, greatly improving the quality and diagnostic value of multimodal images.

[0012] Optionally, the step of constructing a causal hierarchy relationship between the vascular occlusion feature, the perfusion injury feature, and the cell necrosis feature based on the intestinal necrosis pathological mechanism includes: constructing a directed acyclic graph (DAG) that clearly defines the causal relationship between the vascular occlusion feature, the perfusion injury feature, and the cell necrosis feature based on the intestinal necrosis pathological mechanism; calculating a first quantified causal effect value of the vascular occlusion feature on the perfusion injury feature using propensity score matching based on the DAG; calculating a second quantified causal effect value of the perfusion injury feature on the cell necrosis feature using instrumental variable method based on the DAG; constructing a causal hierarchy relationship matrix, and filling the first quantified causal effect value and the second quantified causal effect value into the causal hierarchy relationship matrix according to the edge relationships of the DAG, thereby obtaining the causal hierarchy relationship.

[0013] This invention constructs a directed acyclic graph (DAG) with clearly defined causal relationships, which intuitively presents the causal logic between vascular occlusion, perfusion injury, and cell necrosis. It employs propensity score matching to calculate the first quantitative causal effect value of vascular occlusion on perfusion injury, effectively controlling for confounding factors and accurately measuring the strength of the causal relationship between the two, providing precise data for assessing the extent of vascular occlusion's impact on perfusion injury. The instrumental variable method is used to calculate the second quantitative causal effect value of perfusion injury on cell necrosis, resolving potential endogeneity issues and ensuring the accuracy of causal effect calculations. Adding these quantitative causal effect values ​​to a causal hierarchy matrix results in a hierarchy that not only reflects the causal order between features but also uses specific numerical values ​​to reflect the degree of causal influence, improving the scientific rigor and accuracy of the causal hierarchy.

[0014] Optionally, the causal hierarchy satisfies the following formula:

[0015]

[0016]

[0017]

[0018] in, The causal hierarchy matrix is... This is the first quantified causal effect value. This is the second quantified causal effect value. The total number of samples used for intestinal necrosis feature recognition in the multimodal images. This represents the number of samples containing characteristics of vascular occlusion. For the first One sample showed characteristics of vascular occlusion. Indicator functions, For the first A perfusion injury characteristic value, In the propensity score matching method, the first is the Perfusion injury characteristics of a sample with vascular occlusion matched with an unoccluded control sample. For the first Predicted values ​​of perfusion injury characteristics for each sample. The average of the predicted values ​​of perfusion injury characteristics for all samples For the first Cell necrosis characteristic values ​​of each sample This represents the average of the cell necrosis characteristic values ​​for all samples.

[0019] The causal hierarchy formula of this invention accurately characterizes the causal relationship between vascular occlusion, perfusion injury and cell necrosis by constructing a causal hierarchy matrix and related quantitative calculations, thereby improving the accuracy and scientific nature of the causal hierarchy construction.

[0020] Optionally, determining the risk level of the patient's intestinal necrosis and the probability corresponding to the risk level using the random forest algorithm based on the causal hierarchy, the vascular occlusion features, the perfusion injury features, and the cell necrosis features includes: constructing an intestinal necrosis feature recognition model based on the random forest algorithm; making predictions using the intestinal necrosis feature recognition model based on the causal hierarchy, the vascular occlusion features, the perfusion injury features, and the cell necrosis features; and obtaining the risk level of the patient's intestinal necrosis and the probability corresponding to the risk level based on the prediction results.

[0021] This invention utilizes the random forest algorithm to effectively handle multi-source heterogeneous data when constructing a feature recognition model for intestinal necrosis. Specifically, it integrates causal hierarchical relationships, vascular occlusion, perfusion injury, and cell necrosis features, fully exploring the nonlinear relationships between these complex features. This allows for efficient analysis of input patient-related features, capturing subtle but crucial pathological information, and ultimately obtaining risk levels and corresponding probability results. Compared to traditional single-feature analysis methods, the comprehensive multi-feature prediction is more comprehensive and accurate, reducing the risk of missed diagnoses and misdiagnoses.

[0022] Optionally, the step of constructing the intestinal necrosis feature recognition model based on the random forest algorithm includes: obtaining sample data including historical causal hierarchy, historical vascular occlusion features, historical perfusion injury features, historical cell necrosis features, and historical intestinal necrosis risk level based on historical intestinal necrosis image data of patients with intestinal necrosis; and training the random forest algorithm using the sample data to obtain the intestinal necrosis feature recognition model.

[0023] This invention collects historical image data from patients with intestinal necrosis, encompassing multi-dimensional information such as causal hierarchy, various pathological features, and risk levels, providing comprehensive and rich data support for model training. When training with this sample data, the random forest algorithm can effectively handle complex relationships and high-dimensional features in the data, reducing the risk of overfitting in the intestinal necrosis feature recognition model and improving its generalization ability.

[0024] Optionally, training the random forest algorithm using the sample data to obtain the intestinal necrosis feature recognition model includes: assigning an initial weight metric value to each sample data using the historical intestinal necrosis image data based on the pathological mechanism of intestinal necrosis; setting the sample sampling probability of the random forest algorithm according to the initial weight metric value; and training the random forest algorithm using the sample data to obtain the intestinal necrosis feature recognition model based on the sample sampling probability.

[0025] Assigning initial weights to sample data based on the pathological mechanism of intestinal necrosis highlights the importance of samples related to key pathological features, allowing model training to focus more on information reflecting the essence of the disease. Setting sampling probabilities according to weights enables the algorithm to sample important samples more frequently during training, avoiding interference from a large number of non-critical samples, improving training efficiency and quality, and further enhancing the performance of the intestinal necrosis feature recognition model.

[0026] Optionally, the sample sampling probability satisfies the following formula:

[0027]

[0028] in, For the first The probability of a sample being selected. For the first The initial weight metric for each sample data. The total number of samples used for intestinal necrosis feature recognition in the multimodal images. For the first The initial weight metric for each sample data.

[0029] The sampling probability formula of this invention determines the sampling probability by measuring the initial weight of the samples. This enables the random forest algorithm to prioritize the extraction of key samples with high weights, avoiding model bias caused by sample imbalance. It allows the model training to focus more on important pathological information, improves the training relevance and efficiency, and thus enhances the accuracy and reliability of the intestinal necrosis feature recognition model.

[0030] Another aspect of the present invention provides an image analysis-based intestinal necrosis feature recognition system, comprising an input device, a processor, an output device, and a memory, wherein the input device, processor, output device, and memory are interconnected, the memory includes the computer-readable storage medium described in the preceding aspect of the present invention, the memory is used to store a computer program, the computer program includes program instructions, and the processor is configured to invoke the program instructions.

[0031] The present invention provides an image analysis-based intestinal necrosis feature recognition system, which is compact, stable, highly integrated, and simple in construction. It can stably execute the steps of the program instructions in the computer-readable storage medium provided in the preceding aspect of the present invention, further enhancing the overall applicability and practical application capability of the present invention. Attached Figure Description

[0032] Figure 1 This is a flowchart of program instructions in a computer-readable storage medium according to an embodiment of the present invention;

[0033] Figure 2 This is a schematic diagram of an image analysis-based intestinal necrosis feature recognition system according to an embodiment of the present invention. Detailed Implementation

[0034] Specific embodiments of the present invention will now be described in detail. It should be noted that the embodiments described herein are for illustrative purposes only and are not intended to limit the invention. In the following description, numerous specific details are set forth in order to provide a thorough understanding of the invention. However, it will be apparent to those skilled in the art that these specific details are not necessary to practice the invention. In other instances, well-known circuits, software, or methods have not been specifically described to avoid obscuring the invention.

[0035] Throughout this specification, references to "an embodiment," "an embodiment," "an example," or "an example" mean that a particular feature, structure, or characteristic described in connection with that embodiment or example is included in at least one embodiment of the invention. Therefore, the phrases "in an embodiment," "in an embodiment," "an example," or "an example" appearing in various places throughout the specification do not necessarily refer to the same embodiment or example. Furthermore, specific features, structures, or characteristics can be combined in one or more embodiments or examples in any suitable combination and / or sub-combination. Moreover, those skilled in the art will understand that the illustrations provided herein are for illustrative purposes and are not necessarily drawn to scale.

[0036] Please see Figure 1 In one embodiment of the present invention, a computer-readable storage medium is provided, the computer-readable storage medium storing a computer program, the computer program including program instructions, which, when executed by a processor, cause the processor to perform the following steps:

[0037] Step S1: Obtain the expiratory and inspiratory subsets from the patient's CT angiography images, ultrasound contrast images, and confocal laser microscopy fluorescence images.

[0038] Specifically, obtaining the expiratory and inspiratory subsets from the patient's CT angiography images, ultrasound contrast images, and confocal laser microscopic fluorescence images includes the following sub-steps:

[0039] Step S101: The patient's body surface movement trajectory and the corresponding time information are obtained by using an infrared body surface marker sensor placed on the patient's abdomen.

[0040] In this embodiment, the patient is first assisted to lie in a comfortable supine position. The markers of the infrared body surface marker sensor are cleaned and affixed to the patient's abdomen (such as the periumbilical area, rectus abdominis muscle, and other key areas that can effectively reflect abdominal movement). The markers are ensured to be firmly attached and not obstructed by clothing or other objects to ensure stable signal transmission.

[0041] These infrared markers utilize active infrared emission technology, continuously emitting infrared signals that are captured in real-time by multiple high-sensitivity infrared cameras positioned around the patient. Using the triangulation principle of a multi-camera vision system, the three-dimensional spatial coordinates of each marker are precisely calculated. As the patient breathes and moves, the marker positions dynamically change, and the system continuously tracks and records their continuous positions, forming a body surface movement trajectory. Simultaneously, the system's built-in high-precision clock module adds a timestamp accurate to milliseconds to each trajectory point in real time, ensuring a strict correspondence between the movement trajectory and time information. During the process, the system monitors data quality in real time, automatically marking abnormal data points caused by the patient's vigorous movement or external interference, and after initial screening, ensuring the accuracy and completeness of the acquired body surface movement trajectory and time information.

[0042] Step S102: Using the body surface movement trajectory and the time information, the patient's CT angiography images, ultrasound angiography images, and confocal laser microscopy fluorescence images are divided into expiratory phase subsets and inspiratory phase subsets.

[0043] In this embodiment, the motion trajectory data is first preprocessed by using low-pass filtering to remove high-frequency noise and retaining low-frequency signals (approximately 0.1-0.3 Hz) that reflect respiratory motion. Based on the preprocessed trajectory data, key features of respiratory motion are extracted: using the displacement curve of the marker point in the vertical direction (the direction of the greatest displacement of the anterior abdominal wall) as the main reference, combined with respiratory physiological characteristics (abdomen protrudes outward during inhalation and contracts inward during exhalation), the complete respiratory cycle is divided by detecting the peak value (end of inspiration) and trough value (end of exhalation) of the curve.

[0044] For each CT angiography image, ultrasound contrast imaging image, and confocal laser microscopy endoscopic fluorescence image, its acquisition timestamp (accurate to milliseconds) is extracted and synchronized with the time information of the body surface motion trajectory. Specifically, if the acquisition time of an image falls between two adjacent troughs (i.e., the displacement curve rises from a trough to a peak and then falls back to a trough), it is determined to be the inspiratory phase; if it falls from a peak to the next peak (the displacement curve falls from a peak to a trough and then rises back to a peak), it is determined to be the expiratory phase. For patients with irregular respiratory rhythms, a dynamic time warping algorithm is used to align the motion trajectory with the time series to ensure the robustness of respiratory phase division.

[0045] Finally, according to the above rules, the images of the three modalities are classified into the inspiratory phase subset and the expiratory phase subset, respectively, forming image groups that synchronize the respiratory cycle, namely the expiratory phase subset and the inspiratory phase subset.

[0046] Step S2: Spatial registration and motion artifact correction are performed on the expiratory phase subset and the inspiratory phase subset to obtain a multimodal image.

[0047] The expiratory phase subset includes the first expiratory phase subset of the ultrasound contrast-enhanced images and the confocal laser endoscopic fluorescence images, and the second expiratory phase subset of the CT angiography images; the inspiratory phase subset includes the first inspiratory phase subset of the ultrasound contrast-enhanced images and the confocal laser endoscopic fluorescence images, and the second inspiratory phase subset of the CT angiography images.

[0048] Specifically, the step of performing spatial registration and motion artifact correction on the expiratory and inspiratory subsets to obtain a multimodal image includes the following sub-steps:

[0049] Step S201: Perform motion artifact correction on the first expiratory phase subset and the first inspiratory phase subset respectively to obtain the third expiratory phase subset and the third inspiratory phase subset.

[0050] In this embodiment, the Lucas-Kanade optical flow method is used to correct motion artifacts in the first expiratory subset and the first inspiratory subset, respectively.

[0051] The image sequences within each subset are preprocessed by converting color images to grayscale images and removing high-frequency noise using Gaussian filtering while preserving low-frequency motion information. Subsequently, an image pyramid is constructed for each image, with the original image at the bottom and each subsequent layer downsampled by 2×2 to reduce resolution and handle motion displacement at different scales (the bottom layer handles detailed motion, while the top layers handle large-scale global motion).

[0052] At each level of the pyramid, the Shi-Tomasi corner detection algorithm is used to extract interest points (corners) in the image as feature points for tracking, and non-maximum suppression is used to ensure uniform distribution of feature points. Starting from the top of the pyramid, the Lucas-Kanade optical flow algorithm is applied to the feature points of adjacent frames. Based on the assumptions of "small motion" and "constant brightness," the two-dimensional motion vector of the feature points is iteratively calculated by minimizing the sum of squares of the pixel gray-level differences within the window of adjacent frames. Since the pyramid structure can refine motion estimation layer by layer, the coarse-scale motion vectors of the higher levels are passed to the lower levels as initial values, effectively handling large displacements caused by respiratory movements (such as the up-and-down peristalsis of abdominal organs).

[0053] Outlier filtering is performed on the calculated optical flow field using median filtering or neighborhood consistency checks to remove noise vectors that deviate from the global motion trend, retaining the effective optical flow that truly reflects tissue motion. Subsequently, using a reference frame (the frame with the smallest mean displacement) within the subset as a baseline, inverse motion compensation is performed on each frame using the optical flow field: for each pixel in the current frame, the corrected pixel value is calculated using bilinear interpolation based on the position of the optical flow vector mapped to the reference frame, eliminating artifacts such as image blurring and misalignment caused by respiratory motion or slight patient movement.

[0054] After frame-by-frame processing, the quality of the corrected image sequence is verified. By calculating the structural similarity index of adjacent frames, it is ensured that motion artifacts are effectively suppressed and that key features such as tissue texture and boundaries are not significantly distorted. Finally, the subsets of the third expiratory phase and the third inspiratory phase after motion artifact correction are obtained.

[0055] Step S202: Perform cross-modal coarse registration on the third expiratory phase subset and the second expiratory phase subset, and perform cross-modal coarse registration on the third inspiratory phase subset and the second inspiratory phase subset to obtain the fourth expiratory phase subset and the fourth inspiratory phase subset, respectively.

[0056] In this embodiment, cross-modal coarse registration refers to the technical process of eliminating large-scale spatial positional deviations in multiple medical images and achieving preliminary alignment of images of different modalities in a spatial coordinate system. A feature point matching algorithm is used to perform cross-modal coarse registration on the third and second expiratory phase subsets. First, the two types of images are preprocessed to unify the grayscale range and enhance contrast, reducing the impact of modal differences. The ORB algorithm is used to extract feature points and generate descriptors containing position, scale, and orientation information to ensure feature consistency across the multimodal images. Initial matching is performed using Hamming distance, and mismatched point pairs are eliminated using ratio testing, retaining stable feature points.

[0057] The RANSAC algorithm is used to estimate similarity transformation models (translation, rotation, scaling). Using CT images as a reference coordinate system, transformation parameters are iteratively optimized, and inlier points are filtered through reprojection errors to improve registration robustness. Feature points of ultrasound / endoscopic images are geometrically transformed according to the estimated model, and bilinear interpolation is used to fill pixels to avoid distortion.

[0058] After registration, the alignment accuracy of vascular structures and anatomical features is verified using indicators such as mutual information to ensure consistency of spatial coordinates in multimodal images. Finally, the registered ultrasound and CT images are integrated to form a subset of the fourth expiratory phase, providing a preliminary image basis for subsequent elastic registration and achieving spatial unification of multimodal images of the same respiratory phase.

[0059] Similarly, the fourth inhalation phase subset is obtained.

[0060] Step S203: Perform elastic registration based on grayscale information on the fourth expiratory phase subset and the fourth inspiratory phase subset respectively to obtain a multimodal image.

[0061] In this embodiment, the Demons algorithm is used to perform elastic registration of the fourth expiratory phase subset and the fourth inspiratory phase subset based on grayscale information. First, the registration parameters are initialized, including the number of iterations and the smoothing factor. Registration operations are then performed separately for the fourth expiratory phase subset and the fourth inspiratory phase subset. CT angiography images within the subsets are used as fixed images, while ultrasound contrast images and confocal laser microscopy fluorescence images are used as floating images.

[0062] The Demons algorithm, based on the grayscale information of an image, calculates the grayscale difference between a fixed image and a floating image, transforming it into a deformation field. The deformation field is continuously updated iteratively. In each iteration, the grayscale gradient of each pixel in the floating image is calculated, and combined with the grayscale difference between the fixed and floating images, the displacement vector of each pixel is obtained. To ensure the smoothness of the deformation field, a Gaussian filter is used to smooth the displacement vector, avoiding drastic local deformations.

[0063] As the number of iterations increases, the deformation field is continuously optimized, causing the floating image to gradually move closer to the fixed image. The iteration stops after reaching a preset number of iterations or when the deformation field converges. At this point, the floating image is resampled using the final deformation field, that is, the pixels on the floating image are mapped to new positions according to the displacement vectors in the deformation field, resulting in an image that is more spatially aligned with the fixed image.

[0064] After processing by the Demons algorithm, the fourth expiratory phase subset and the fourth inspiratory phase subset respectively completed the second spatial registration, resulting in the final multimodal image composed of the adjusted expiratory phase subset and inspiratory phase subset. The multimodal image achieved high spatial consistency, providing an accurate data foundation for subsequent image analysis and feature extraction.

[0065] It should be noted that the second spatial registration solves the problem of subtle deformation and grayscale consistency. Multimodal images refer to integrated images of CT angiography, ultrasound angiography and confocal laser microscopy fluorescence images.

[0066] Step S3: Extract vascular occlusion features, perfusion injury features, and cell necrosis features from the multimodal image.

[0067] In this embodiment, for CT angiography images, a vascular segmentation algorithm is used to accurately identify vascular structures, and vascular occlusion features are extracted by calculating indicators such as the degree of vascular stenosis and the area of ​​the occlusion region. In ultrasound contrast imaging and CT perfusion imaging, the blood flow perfusion of the region of interest is analyzed, such as calculating parameters like blood flow velocity and perfusion volume changes, to obtain perfusion injury features. Confocal laser microscopy fluorescence images are used to observe changes at the cellular level. Based on fluorescence labeling, the percentage of necrotic cells and changes in cell morphology are statistically analyzed to extract cell necrosis features.

[0068] Step S4: Based on the pathological mechanism of intestinal necrosis, construct the causal hierarchical relationship between the vascular occlusion feature, the perfusion injury feature, and the cell necrosis feature.

[0069] Specifically, the step of constructing a causal hierarchy among the vascular occlusion features, the perfusion injury features, and the cell necrosis features based on the pathological mechanism of intestinal necrosis includes the following sub-steps:

[0070] Step S401: Based on the pathological mechanism of intestinal necrosis, construct a directed acyclic graph that clarifies the causal relationship between the vascular obstruction characteristics, the perfusion injury characteristics, and the cell necrosis characteristics.

[0071] In this embodiment, the main pathological mechanism of intestinal necrosis is that the mesenteric vessels become blocked due to thrombosis, spasm, etc., leading to insufficient blood supply to the intestine. Intestinal ischemia and hypoxia cause perfusion injury, resulting in microcirculatory disturbances, metabolic disorders, and inflammatory responses. Continuous ischemia depletes cellular energy, causing cell membrane rupture and cell necrosis. Ultimately, the morphology of the intestinal wall tissue changes.

[0072] When constructing a directed acyclic graph based on the pathological mechanism of intestinal necrosis, the first step is to delve into the occurrence and development of intestinal necrosis. Intestinal necrosis typically originates from vascular obstruction, such as when mesenteric vessels are blocked by thrombi, leading to insufficient local blood supply to the intestine and subsequently perfusion injury. Perfusion injury prevents tissues from obtaining sufficient oxygen and nutrients, affecting normal cellular metabolism and ultimately resulting in cell necrosis.

[0073] When constructing the directed acyclic graph (DAG), three nodes are established to represent vascular occlusion, perfusion injury, and cell necrosis characteristics, respectively. A directed edge is drawn from the vascular occlusion node to the perfusion injury node, indicating that vascular occlusion is the cause of perfusion injury; another directed edge is drawn from the perfusion injury node to the cell necrosis node, indicating that perfusion injury leads to cell necrosis. In this way, the unidirectional causal relationship among the three is clearly defined, and there are no loop paths, thus constructing a DAG that clearly presents the logical sequence from vascular occlusion to perfusion injury, and finally to cell necrosis.

[0074] Step S402: Based on the directed acyclic graph, the first quantitative causal effect value of the vascular occlusion feature on the perfusion injury feature is calculated using the propensity score matching method.

[0075] In this embodiment, based on the directed acyclic graph, the causal relationship between vascular occlusion features and perfusion injury features, as well as potential confounding factors (such as the patient's underlying disease, vascular anatomy, etc.), are first clarified. Then, the vascular occlusion feature values, perfusion injury feature values, and confounding factor data of the samples are extracted from the multimodal images, and preprocessing such as missing value imputation and outlier handling is performed. Next, a logistic regression model is constructed with confounding factors as independent variables to calculate the propensity score for vascular occlusion in each sample (i.e., the probability of vascular occlusion given confounding factors). Then, a suitable matching algorithm (such as nearest neighbor matching) is used to match each sample with vascular occlusion with a control sample without vascular occlusion that has a similar propensity score. Finally, the mean value of the difference in perfusion injury features between the sample with vascular occlusion and the matched control sample is calculated, which is the first quantified causal effect value, accurately characterizing the degree of causal influence of vascular occlusion on perfusion injury.

[0076] The first quantified causal effect value satisfies the following formula:

[0077]

[0078] in, This is the first quantified causal effect value. The total number of samples used for intestinal necrosis feature recognition in the multimodal images. This represents the number of samples containing characteristics of vascular occlusion. For the first One sample showed characteristics of vascular occlusion. Indicator functions, For the first A perfusion injury characteristic value, In the propensity score matching method, the first is the Perfusion injury characteristics of unobstructed control samples matched with samples exhibiting vascular occlusion.

[0079] In the above formula, It originates from feature extraction of vascular occlusion samples, such as by analyzing multimodal images (e.g., CT angiography images, ultrasound contrast images, etc.) to quantify the feature values ​​of the sample in terms of perfusion injury, which are used to characterize the degree of perfusion injury of the sample. From a sample set without vascular occlusion, a propensity score matching strategy (based on covariate modeling, such as patient underlying diseases, vascular anatomy, etc.) is used to find a control sample that is similar to the i-th vascular occlusion sample in terms of covariate distribution. Perfusion injury features are then extracted from this control sample to obtain the corresponding feature values.

[0080] Step S403: Based on the directed acyclic graph, the instrumental variable method is used to calculate the second quantitative causal effect value of the perfusion injury characteristics on the cell necrosis characteristics.

[0081] In this embodiment, the causal relationship between perfusion injury and cell necrosis is first clarified through a graph structure (perfusion injury may induce cell necrosis), and potential endogeneity issues are identified—that is, there may be unobserved confounding factors (such as microcirculatory status, inflammatory factors, etc.) that simultaneously affect both, causing traditional regression to fail to accurately quantify the causal effect. Therefore, it is necessary to screen instrumental variables (such as mesenteric artery hemodynamic parameters) that meet the exogenous condition. These variables must be directly related to the characteristics of perfusion injury but independent of the error term of cell necrosis characteristics.

[0082] Subsequently, a two-stage analysis using instrumental variables was conducted, combining the perfusion injury characteristics and the cell necrosis characteristics: In the first stage, a regression model was constructed using instrumental variables and covariates as inputs to predict the perfusion injury characteristics, obtaining predicted values ​​of perfusion injury after eliminating endogeneity effects; in the second stage, the predicted values ​​were used as independent variables, combined with covariates, to perform regression analysis on the cell necrosis characteristics, ultimately obtaining the causal effect value of perfusion injury on cell necrosis. This value, by quantifying "the expected change in cell necrosis risk for every unit change in perfusion injury," clearly characterizes the causal strength between the two, providing key parameters for subsequently constructing a causal hierarchy matrix, ensuring the scientific rigor and accuracy of intestinal necrosis risk assessment.

[0083] The second quantified causal effect value satisfies the following formula:

[0084]

[0085] in, This is the second quantified causal effect value. For the first Predicted values ​​of perfusion injury characteristics for each sample. The average of the predicted values ​​of perfusion injury characteristics for all samples For the first Cell necrosis characteristic values ​​of each sample This represents the average of the cell necrosis characteristic values ​​for all samples.

[0086] The above formula is derived based on the two-stage least squares (2SLS) method using instrumental variables. The specific process is as follows:

[0087] First, we introduce instrumental variables. (and perfusion injury characteristics) (Related, but not related to the error term of cell necrosis characteristics) and covariates (Including information such as the patient's body surface movement trajectory), characteristics of perfusion injury The regression yields the following formula:

[0088]

[0089] in, This is the intercept term of the first-stage regression equation, and is a constant representing the sum of all independent variables. , When it is zero The baseline value, Instrumental variable The regression coefficients in the first-stage regression measure right The extent of the impact Let be the regression coefficient of the covariate in the first stage regression, representing the regression coefficient of the first stage regression. Each covariate pair Influence weight, For the first The first sample One covariate is used to control for other factors that affect perfusion injury. This is the error term for the first-stage regression, representing terms that cannot be regressed. and Explanation of random factors.

[0090] Through regression calculation, the predicted values ​​of perfusion injury characteristics satisfy the following formula:

[0091]

[0092] in, for Estimates, obtained from sample data The fitting results for The estimated value, using sample data to evaluate instrumental variables Fitting of regression coefficients, for The estimated value is obtained by analyzing the covariates using sample data. Fitting of regression coefficients.

[0093] Next, the predicted values ​​obtained in the first stage will be... As an independent variable, the characteristics of cell necrosis Regression yields the following formula:

[0094]

[0095] in, This is the intercept term of the second-stage regression equation. This is the second quantified causal effect value, which measures the predicted value of perfusion injury characteristics. Characteristics of cell necrosis The degree of causal influence covariates The regression coefficients in the second-stage regression, This is the error term for the second-stage regression.

[0096] According to the principle of least squares, to minimize the sum of squared errors, the parameters need to be adjusted. By taking the partial derivative and setting it to zero, in the deviation form (i.e., assuming the covariates have been effectively controlled or centered, eliminating the influence of the variable mean on the calculation), through differentiation and algebraic simplification, we can ultimately derive the following: The expression, this process uses mathematical optimization techniques to ensure that the regression model... It can best fit the data and accurately quantify the predicted values ​​of perfusion injury characteristics. Characteristics of cell necrosis The causal effect.

[0097] Minimize the sum of squared errors Satisfy the following formula:

[0098]

[0099] Step S404: Construct a causal hierarchy matrix, and fill the first quantified causal effect value and the second quantified causal effect value into the causal hierarchy matrix according to the edge relationship of the directed acyclic graph to obtain the causal hierarchy relationship.

[0100] In this embodiment, when constructing the causal hierarchy matrix, the dimensions and element layout of the matrix are first determined. Based on the three factors involved in the directed acyclic graph—vascular occlusion characteristics, perfusion injury characteristics, and cell necrosis characteristics—a 3×3 matrix is ​​constructed.

[0101] The elements on the main diagonal of the matrix are all set to 1 because each feature has an identity relationship with itself. Then, based on the edge relationships of the directed acyclic graph, the first quantified causal effect value is filled into the intersection of the vascular occlusion feature and the perfusion injury feature in the matrix. This value reflects the strength of the causal effect from vascular occlusion to perfusion injury. Similarly, the second quantified causal effect value is filled into the intersection of the perfusion injury feature and the cell necrosis feature, reflecting the degree of causal influence of perfusion injury on cell necrosis. Other intersections in the matrix without direct causal relationships are filled with 0, thus completing the construction of the causal hierarchy matrix. Through this matrix, the causal relationships between various features are clearly and systematically integrated, forming a complete causal hierarchy.

[0102] Causal hierarchy refers to the progressive causal chain of “vascular obstruction → perfusion injury → cell necrosis” based on the pathological mechanism of intestinal necrosis, and provides a structured association basis driven by pathological mechanism for intestinal necrosis risk assessment by quantifying the causal effects of each link.

[0103] The aforementioned edge relationship refers to the direction of direct causal action:

[0104] An edge connecting vascular occlusion features to perfusion injury features indicates that vascular occlusion is a direct cause of perfusion injury. In the causal hierarchy matrix, this edge corresponds to the first row and second column. Filling ATE into this position means that the matrix directly reflects the degree of causal influence of vascular occlusion on perfusion injury through numerical values.

[0105] The edge from the perfusion injury feature to the cell necrosis feature indicates that perfusion injury is the direct cause of cell necrosis. In the matrix, this edge corresponds to the second row and third column position, realizing the causal quantification mapping from the directed acyclic graph to the matrix.

[0106] For locations without a direct causal relationship (such as the transition from vascular occlusion features to cell necrosis features), since vascular occlusion does not directly lead to cell necrosis in the current pathological mechanism and directed acyclic graph construction (but rather indirectly affects it through perfusion injury), these intersection locations without direct edge connections are filled with 0, indicating that there is no direct causal effect.

[0107] The causal hierarchy relationship satisfies the following formula:

[0108]

[0109]

[0110]

[0111] in, The causal hierarchy matrix is... This is the first quantified causal effect value. This is the second quantified causal effect value. The total number of samples used for intestinal necrosis feature recognition in the multimodal images. This represents the number of samples containing characteristics of vascular occlusion. For the first One sample showed characteristics of vascular occlusion. Indicator functions, For the first A perfusion injury characteristic value, In the propensity score matching method, the first is the Perfusion injury characteristics of a sample with vascular occlusion matched with an unoccluded control sample. For the first Predicted values ​​of perfusion injury characteristics for each sample. The average of the predicted values ​​of perfusion injury characteristics for all samples For the first Cell necrosis characteristic values ​​of each sample This represents the average of the cell necrosis characteristic values ​​for all samples.

[0112] Step S5: Introduce the random forest algorithm. Based on the causal hierarchy, the vascular occlusion features, the perfusion injury features, and the cell necrosis features, use the random forest algorithm to determine the risk level of the patient having intestinal necrosis and the probability corresponding to the risk level.

[0113] Specifically, determining the risk level of intestinal necrosis in the patient and the probability corresponding to the risk level using the random forest algorithm based on the causal hierarchy, the vascular occlusion characteristics, the perfusion injury characteristics, and the cell necrosis characteristics includes the following sub-steps:

[0114] Step S501: Construct an intestinal necrosis feature recognition model based on the random forest algorithm.

[0115] The construction of the intestinal necrosis feature recognition model based on the random forest algorithm specifically includes the following sub-steps:

[0116] Step S50101: Based on the historical intestinal necrosis image data of patients with intestinal necrosis, sample data including historical causal hierarchy, historical vascular occlusion characteristics, historical perfusion injury characteristics, historical cell necrosis characteristics, and historical intestinal necrosis risk level are obtained.

[0117] In this embodiment, historical CT angiography, ultrasound angiography, and confocal laser microscopy fluorescence images of patients with intestinal necrosis are first extracted from the hospital's imaging database. After spatial registration and motion artifact correction, historical vascular occlusion features (such as stenosis rate and occlusion status), historical perfusion injury features (such as time to peak flow and blood flow), and historical cell necrosis features (such as necrotic cell density and fluorescence intensity) are extracted for each patient according to a unified standard. Simultaneously, based on a directed acyclic graph constructed from historical data, the first and second quantified causal effect values ​​for each sample are calculated, forming a historical causal hierarchy matrix. Combined with clinical diagnostic results, the historical intestinal necrosis risk level (such as low, medium, and high risk) is labeled for each sample. Finally, the data is integrated into structured sample data containing multimodal features, causal effect parameters, and risk levels, providing standardized input for training the random forest model.

[0118] Step S50102: Use the sample data to train the random forest algorithm to obtain the intestinal necrosis feature recognition model.

[0119] The step of training the random forest algorithm using the sample data to obtain the intestinal necrosis feature recognition model specifically includes the following sub-steps:

[0120] Step S5010201: Based on the pathological mechanism of intestinal necrosis, an initial weight metric value is assigned to each sample data using the historical intestinal necrosis image data.

[0121] Based on the pathological mechanism of intestinal necrosis, we first conducted an in-depth analysis of historical intestinal necrosis image data to accurately extract relevant feature information on vascular occlusion, perfusion injury, and cell necrosis. Through quantitative analysis of these features, we determined the first historical quantitative causal effect value of vascular occlusion on perfusion injury and the second historical quantitative causal effect value of perfusion injury on cell necrosis. Subsequently, based on the importance of the first and second historical quantitative causal effect values, we assigned corresponding weights to these two historical quantitative causal effect values. Finally, we combined the weights and the two historical quantitative causal effect values ​​to calculate an initial weight metric.

[0122] The initial weight metric is a value that quantifies the importance of the sample data.

[0123] The initial weight metric values ​​satisfy the following formula:

[0124]

[0125]

[0126] in, For the first The initial weight metric for each sample data. and All are weights. For the first The first historical quantitative causal effect value of each sample data point For the first The second historical quantitative causal effect value of each sample data.

[0127] Step S5010202: Set the sample sampling probability of the random forest algorithm according to the initial weight metric value.

[0128] In this embodiment, setting the sample sampling probability based on the initial weight metric allows the random forest algorithm to prioritize more important samples during training (samples with higher weights contain more critical pathological information). This helps the model capture the core pathological features of intestinal necrosis, improves the accuracy of risk level assessment, enhances the fitting ability to the pathological mechanism, and makes the assessment results more consistent with clinical practice.

[0129] The sampling probability of the sample satisfies the following formula:

[0130]

[0131] in, For the first The probability of a sample being selected. For the first The initial weight metric for each sample data. The total number of samples used for intestinal necrosis feature recognition in the multimodal images. For the first The initial weight metric for each sample data.

[0132] Step S5010203: Based on the sample sampling probability, the random forest algorithm is trained using the sample data to obtain an intestinal necrosis feature recognition model.

[0133] In this embodiment, a training set is constructed by drawing samples with replacement from the sample data based on the sampling probability. Samples with higher weights are more likely to be selected due to their higher sampling probability. For each training set, a decision tree is constructed using the sample's vascular occlusion, perfusion injury, cell necrosis features, and corresponding intestinal necrosis risk level labels. This process is repeated to generate multiple decision trees, forming a random forest. Because the sampling probability makes the model focus more on important samples, the trained intestinal necrosis feature recognition model can effectively capture the correlation between key pathological features and risk levels, improving the accuracy and reliability of intestinal necrosis feature recognition and risk assessment.

[0134] Step S502: Based on the causal hierarchy, the vascular occlusion characteristics, the perfusion injury characteristics, and the cell necrosis characteristics, the intestinal necrosis feature recognition model is used to make predictions.

[0135] In this embodiment, the causal hierarchy is organized into structured data to ensure that it can be read by the intestinal necrosis feature recognition model. This relationship reflects the causal logic among vascular occlusion, perfusion injury, and cell necrosis. Simultaneously, the vascular occlusion features, perfusion injury features, and cell necrosis features are standardized to unify the data format and measurement units.

[0136] The processed causal hierarchy, vascular occlusion features, perfusion injury features, and cell necrosis features are input into the trained intestinal necrosis feature recognition model. The random forest inside the intestinal necrosis feature recognition model analyzes the input data based on the patterns and rules learned during training, and finally the intestinal necrosis feature recognition model outputs the prediction results.

[0137] Step S503: Based on the prediction results, obtain the risk level of the patient having intestinal necrosis and the probability corresponding to the risk level.

[0138] In this embodiment, the predicted results include risk levels and corresponding probabilities. These probabilities reflect the quantitative assessment of the likelihood of a patient being at each risk level, based on the mapping relationship between features learned during training and risk levels. Next, the probability with the highest value is selected from all output probabilities; the risk level corresponding to this probability is the predicted risk level of bowel necrosis that the patient is currently most likely to be in.

[0139] Furthermore, if the high-risk probability is greater than 70%, combined with the causal hierarchy in the pathological mechanism (such as the progressive effect of vascular occlusion → perfusion injury → cell necrosis), the patient can be determined to be at high risk of intestinal necrosis. This suggests that clinicians should immediately initiate emergency assessment and intervention based on the degree of occlusion on CT angiography and the area of ​​cell necrosis on endoscopic fluorescence images. If the medium-risk probability is between 30% and 70%, it is marked as a suspicious state. It is recommended to dynamically monitor changes in perfusion injury through ultrasound contrast imaging or to perform tissue biopsy to rule out the possibility of early necrosis. If the low-risk probability is <30%, it is determined to be at low risk of intestinal necrosis, but a follow-up plan should be developed based on the patient's medical history (such as history of thrombosis, vasculitis) to continuously track the evolution of pathological features.

[0140] Furthermore, the dispersion of the probability distribution can help determine the reliability of the results: if the probabilities of each level are close (e.g., low 30%, medium 35%, high 35%), it indicates that the feature information is atypical, and we need to be wary of the risk of the model missing complex pathological states. It is recommended to further verify the results by combining the spatial registration results of multimodal images (e.g., structural details after correction of organ motion artifacts in the respiratory cycle). Finally, by combining the quantified probability with the clinically preset risk threshold (e.g., high risk corresponds to a probability of intestinal necrosis > 70%), a closed loop of "feature input → probability calculation → level determination → clinical decision-making" is formed.

[0141] like Figure 2 As shown, in another aspect, the present invention also provides an image analysis-based intestinal necrosis feature recognition system, including an input device, a processor, an output device, and a memory, wherein the input device, processor, output device, and memory are interconnected, the memory includes the aforementioned computer-readable storage medium, the memory is used to store a computer program, the computer program includes program instructions, and the processor is configured to call the program instructions.

[0142] In this embodiment, the input device is used to provide the system with relevant input data or instructions. In an image-based intestinal necrosis feature recognition system, the input device may include common human-computer interaction interface devices such as a keyboard, mouse, and touchscreen. Through the input device, doctors or researchers can input necessary parameters.

[0143] The processor is the core component of the system, responsible for executing computer program instructions and performing data processing and analysis. In the image analysis-based intestinal necrosis feature recognition system, the processor analyzes and interprets the input experimental data by running pre-programmed algorithms and models. The processor can be a central processing unit (CPU), a graphics processing unit (GPU), or other dedicated processing unit.

[0144] The memory is used to store computer programs, data, and parameters required by the system. It may include random access memory (RAM) for temporary data storage and processing, and persistent memory (such as hard disks or solid-state drives) for long-term data storage and preservation.

[0145] The output device is used to present the results of system processing and analysis to users or external devices. The output device can be a monitor, printer, charting device, etc. Through the output device, the system can display the prediction results, which can be used as a reference for doctors, researchers, or patients to assist in decision-making and communication.

[0146] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention, and they should all be covered within the scope of the claims and specification of the present invention.

Claims

1. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, the computer program including program instructions that, when executed by a processor, cause the processor to perform the following steps: Acquire expiratory and inspiratory subsets from the patient’s CT angiography images, ultrasound contrast images, and confocal laser microscopy fluorescence images. Spatial registration and motion artifact correction are performed on the expiratory and inspiratory subsets to obtain a multimodal image; Vascular occlusion features, perfusion injury features, and cell necrosis features are extracted from the multimodal images; Based on the pathological mechanism of intestinal necrosis, a causal hierarchical relationship is constructed among the vascular occlusion characteristics, the perfusion injury characteristics, and the cell necrosis characteristics. A random forest algorithm is introduced to determine the risk level of intestinal necrosis and the probability corresponding to the risk level of the patient based on the causal hierarchy, the vascular occlusion characteristics, the perfusion injury characteristics, and the cell necrosis characteristics.

2. The computer-readable storage medium according to claim 1, characterized in that, The acquisition of expiratory and inspiratory subsets, derived from the patient's CT angiography images, ultrasound contrast images, and confocal laser endoscopic fluorescence images, includes: By using an infrared body surface marker sensor placed on the patient's abdomen, the patient's body surface movement trajectory and the corresponding time information are obtained; Using the body surface movement trajectory and the time information, the patient's CT angiography images, ultrasound contrast images, and confocal laser microscopy fluorescence images are divided into expiratory and inspiratory subsets.

3. A computer-readable storage medium according to claim 1, characterized in that, The expiratory phase subset includes a first expiratory phase subset of the ultrasound contrast-enhanced images and the confocal laser microendoscopic fluorescence images, and a second expiratory phase subset of the CT angiography images. The inspiratory phase subset includes a first inspiratory phase subset of the ultrasound contrast-enhanced images and the confocal laser microendoscopic fluorescence images, and a second inspiratory phase subset of the CT angiography images. Spatial registration and motion artifact correction are performed on the expiratory and inspiratory phase subsets to obtain multimodal images, including: Motion artifact correction is performed on the first expiratory phase subset and the first inspiratory phase subset respectively to obtain the third expiratory phase subset and the third inspiratory phase subset; Cross-modal coarse registration is performed on the third expiratory phase subset and the second expiratory phase subset, and cross-modal coarse registration is performed on the third inspiratory phase subset and the second inspiratory phase subset, respectively, to obtain the fourth expiratory phase subset and the fourth inspiratory phase subset; Elastic registration based on grayscale information is performed on the fourth expiratory phase subset and the fourth inspiratory phase subset respectively to obtain a multimodal image.

4. A computer-readable storage medium according to claim 1, characterized in that, The causal hierarchy among the vascular occlusion features, perfusion injury features, and cell necrosis features, based on the pathological mechanism of intestinal necrosis, includes: Based on the pathological mechanism of intestinal necrosis, a directed acyclic graph was constructed to clarify the causal relationship between the vascular occlusion feature, the perfusion injury feature, and the cell necrosis feature; Based on the directed acyclic graph, the first quantitative causal effect value of the vascular occlusion feature on the perfusion injury feature is calculated using the propensity score matching method. Based on the directed acyclic graph, the instrumental variable method is used to calculate the second quantitative causal effect value of the perfusion injury characteristics on the cell necrosis characteristics; A causal hierarchy matrix is ​​constructed, and the first quantified causal effect value and the second quantified causal effect value are filled into the causal hierarchy matrix according to the edge relationship of the directed acyclic graph to obtain the causal hierarchy relationship.

5. A computer-readable storage medium according to claim 4, characterized in that, The causal hierarchy relationship satisfies the following formula: in, The causal hierarchy matrix is... This is the first quantified causal effect value. This is the second quantified causal effect value. The total number of samples used for intestinal necrosis feature recognition in the multimodal images. This represents the number of samples containing characteristics of vascular occlusion. For the first One sample showed characteristics of vascular occlusion. Indicator functions, For the first A perfusion injury characteristic value, In the propensity score matching method, the first is the Perfusion injury characteristics of a sample with vascular occlusion matched with an unoccluded control sample. For the first Predicted values ​​of perfusion injury characteristics for each sample. The average of the predicted values ​​of perfusion injury characteristics for all samples For the first Cell necrosis characteristic values ​​of each sample This represents the average of the cell necrosis characteristic values ​​for all samples.

6. A computer-readable storage medium according to claim 1, characterized in that, The determination of the risk level of intestinal necrosis and the probability corresponding to the risk level using the random forest algorithm based on the causal hierarchy, the vascular occlusion characteristics, the perfusion injury characteristics, and the cell necrosis characteristics includes: A feature recognition model for intestinal necrosis was constructed based on the aforementioned random forest algorithm; Based on the causal hierarchy, the vascular occlusion characteristics, the perfusion injury characteristics, and the cell necrosis characteristics, predictions are made using the intestinal necrosis feature recognition model. Based on the prediction results, the risk level of the patient having intestinal necrosis and the probability corresponding to the risk level are obtained.

7. A computer-readable storage medium according to claim 6, characterized in that, The intestinal necrosis feature recognition model constructed based on the random forest algorithm includes: Based on historical intestinal necrosis image data of patients with intestinal necrosis, sample data including historical causal hierarchy, historical vascular occlusion characteristics, historical perfusion injury characteristics, historical cell necrosis characteristics, and historical intestinal necrosis risk level were obtained. The random forest algorithm is trained using the sample data to obtain a feature recognition model for intestinal necrosis.

8. A computer-readable storage medium according to claim 7, characterized in that, The step of training the random forest algorithm using the sample data to obtain the intestinal necrosis feature recognition model includes: Based on the pathological mechanism of intestinal necrosis, an initial weight metric value is assigned to each sample data using the historical intestinal necrosis image data; The sample sampling probability is set according to the initial weight metric value in the random forest algorithm. Based on the sample sampling probability, the random forest algorithm is trained using the sample data to obtain an intestinal necrosis feature recognition model.

9. A computer-readable storage medium according to claim 8, characterized in that, The sampling probability of the sample satisfies the following formula: in, For the first The probability of a sample being selected. For the first The initial weight metric for each sample data. The total number of samples used for intestinal necrosis feature recognition in the multimodal images. For the first The initial weight metric for each sample data.

10. A system for identifying intestinal necrosis features based on image analysis, characterized in that, The device includes an input device, a processor, an output device, and a memory, wherein the input device, processor, output device, and memory are interconnected, the memory includes a computer-readable storage medium as described in any one of claims 1 to 9, the memory is used to store a computer program, the computer program includes program instructions, and the processor is configured to invoke the program instructions.

Images