Method and system for evaluating hepatic encephalopathy based on brain structural image

Abstract

The application discloses a kind of based on brain structure image's hepatic encephalopathy evaluation method and system, the method includes: obtaining the brain structure image data of subject brain;Based on brain structure image data, the feature parameter of characterizing brain tissue morphology is extracted;The characteristic parameter is handled by pre-training disease progression evaluation model, and the space-time progression trajectory of brain structure abnormality is inferred;Based on space-time progression trajectory, the disease subtype and disease progression stage to which the subject belongs are determined;Based on disease subtype and disease progression stage, evaluation information for characterizing the disease state of hepatic encephalopathy is generated.The space-time progression trajectory of brain structure abnormality is inferred based on cross-sectional image data, which overcomes the dependence on massive longitudinal tracking data, realizes the objective typing and staging of hepatic encephalopathy based on neurobiology, thereby effectively solving the technical problems of difficult to track disease evolution and individualized evaluation.

Description

Technical Field

[0001] This invention relates to the field of medical image processing technology, and in particular to a method and system for assessing hepatic encephalopathy based on brain structural images. Background Technology

[0002] Hepatic encephalopathy (HEE) is a major complication and cause of death in patients with cirrhosis, exhibiting significant heterogeneity in disease progression and treatment prognosis. Current classification methods for HEE primarily rely on clinical manifestations, such as the West-Haven classification. This syndrome-based grading approach fails to reflect the pathobiological mechanisms of the disease and neglects the crucial role of gray matter atrophy heterogeneity in disease progression. While neuroimaging can provide quantitative measurements of brain morphology, existing biomarker screening methods typically only compare brain regions with differences between positive and negative groups, exhibiting a lag and failing to reflect the causes and processes of adverse events, while also neglecting individual differences. More critically, identifying disease progression subtypes usually relies on massive longitudinal tracking data to depict the spatiotemporal trajectory of disease development, which is difficult to obtain in clinical practice. This prevents current technologies from effectively inferring spatiotemporal progression patterns of brain structural abnormalities based solely on cross-sectional data, hindering objective neurobiological-based classification and staging. Consequently, this limits effective tracking of HEE disease progression, risk stratification, and individualized prognostic assessment. Summary of the Invention

[0003] In view of this, the present invention proposes a method and system for assessing hepatic encephalopathy based on brain structural imaging, which can achieve objective classification and staging of hepatic encephalopathy based on neurobiology. The present invention provides the following technical solution: A method for assessing hepatic encephalopathy based on brain structural imaging, the method comprising: Obtain imaging data of the subject's brain structure; Based on the brain structure imaging data, feature parameters characterizing brain tissue morphology are extracted; The feature parameters are processed by a pre-trained disease progression assessment model to infer the spatiotemporal progression trajectory of brain structural abnormalities. The disease subtype and disease progression stage of the subject are determined based on the spatiotemporal progression trajectory. Based on the disease subtype and disease progression stage, assessment information is generated to characterize the disease status of hepatic encephalopathy.

[0004] Optionally, the step of extracting characteristic parameters representing brain tissue morphology based on the brain structure imaging data includes: Brain region segmentation and reconstruction were performed on the brain structure image data using brain imaging processing software. At least one of the following initial features is extracted from the whole brain region of interest: cortical thickness, subcortical gray matter volume, surface area, and curvature; The initial features are dimensionality reduced based on a multi-granularity brain map to obtain the feature parameters.

[0005] Optionally, the method further includes standardizing the feature parameters: Obtain brain structural imaging data and corresponding morphological features from healthy control groups; A linear regression model was constructed based on the morphological characteristics of the healthy control group, with age, sex, and intracranial volume as covariates. The standardized Z-score of the subject's characteristic parameters relative to the healthy control group is calculated using the linear regression model, and the standardized Z-score is used as the characteristic parameter input to the disease progression assessment model.

[0006] Optionally, the disease progression assessment model is configured to infer longitudinal disease evolution patterns based on cross-sectional image data; The inferred spatiotemporal progression trajectory of brain structural abnormalities includes: Brain structural abnormalities are modeled as an ordered accumulation of discrete events, where each event corresponds to a morphological feature of a specific brain region that reaches a preset abnormality threshold. The most likely sequence of discrete events under different disease subtypes is inferred by the disease progression assessment model to form the spatiotemporal progression trajectory.

[0007] Optionally, the disease progression assessment model employs a subtype stage inference algorithm; The preset anomaly thresholds include multiple levels of Z-score thresholds; The inference of the most likely sequence of discrete events under different disease subtypes using the disease progression assessment model includes: Under the assumption of a predetermined number of subtypes, the event sequence of each subtype at the population level is estimated using the Markov chain Monte Carlo sampling method; The optimal number of subtypes is determined based on cross-validation information criteria.

[0008] Optionally, determining the disease subtype and disease progression stage of the subject based on the spatiotemporal progression trajectory includes: Calculate the degree of matching between the subject's characteristic parameters and each of the spatiotemporal progression trajectories; The probability distribution of the subjects belonging to each disease subtype is assigned based on the Bayesian posterior probability. The disease subtype corresponding to the maximum posterior probability is determined as the disease subtype to which the subject belongs, and the disease progression stage is determined based on the cumulative number of events under that disease subtype.

[0009] Optionally, the assessment information includes early warning information on the risk of developing obvious hepatic encephalopathy; The generation of assessment information for characterizing the state of hepatic encephalopathy includes: The disease subtype and disease progression stage are input into a pre-trained risk prediction model, which outputs the probability that the subject will develop significant hepatic encephalopathy within a preset time window.

[0010] Optionally, the assessment information includes cognitive recovery assessment information after liver transplantation; The generation of assessment information for characterizing the state of hepatic encephalopathy includes: The disease subtype and disease progression stage are input into a pre-trained outcome prediction model, which outputs a predicted value of the degree of improvement in cognitive function of the subject after liver transplantation.

[0011] This invention further discloses a system for assessing hepatic encephalopathy based on brain structural imaging, comprising: The data acquisition module is used to acquire brain structure imaging data of the subjects; The feature extraction module is used to extract feature parameters characterizing brain tissue morphology based on the brain structure image data. The trajectory inference module is used to process the feature parameters through a pre-trained disease progression assessment model to infer the spatiotemporal progression trajectory of brain structural abnormalities. The subtyping and staging module is used to determine the disease subtype and disease progression stage of the subject based on the spatiotemporal progression trajectory. The information generation module is used to generate assessment information to characterize the disease status of hepatic encephalopathy based on the disease subtype and disease progression stage.

[0012] The present invention further discloses a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described method.

[0013] According to the technical solution of the present invention, by acquiring brain structure imaging data of subjects and extracting characteristic parameters characterizing brain tissue morphology, and processing the characteristic parameters using a pre-trained disease progression assessment model, the spatiotemporal progression trajectory of brain structural abnormalities can be inferred based on cross-sectional imaging data, thereby overcoming the technical bottleneck of traditional disease progression subtype identification relying on massive longitudinal tracking data. Furthermore, based on the spatiotemporal progression trajectory, the disease subtype and disease progression stage of the subject are determined, realizing the transformation of hepatic encephalopathy from syndrome-based clinical manifestation grading to objective neurobiological classification, solving the problems of existing classification methods ignoring gray matter atrophy heterogeneity and failing to reflect the pathobiological mechanisms of the disease. Finally, based on the disease subtype and disease progression stage, assessment information for characterizing the disease state of hepatic encephalopathy is generated, which can quantify individualized disease progression locations and predict the risk of significant hepatic encephalopathy onset and the degree of cognitive recovery after liver transplantation. This overcomes the shortcomings of traditional biomarker screening methods, which only compare differences between groups and have a lag effect and do not consider individual differences, providing objective and interpretable technical basis for the stratified management, primary prevention, and resource allocation optimization of hepatic encephalopathy patients. Attached Figure Description

[0014] For illustrative and not limiting purposes, the present invention will now be described in conjunction with embodiments and accompanying drawings, wherein: Figure 1 This is a flowchart illustrating the assessment method for hepatic encephalopathy based on brain structural imaging according to an embodiment of the present invention. Figure 2 This is a schematic diagram of the components of the hepatic encephalopathy assessment system based on brain structural imaging in an embodiment of the present invention. Figure 3 This is a schematic diagram of the structure of the electronic device in an embodiment of the present invention; Figure 4 This is a schematic diagram of the subtype construction framework based on the SuStaIn algorithm in an embodiment of the present invention; Figure 5 This is a schematic diagram of the hierarchical prediction model framework in an embodiment of the present invention. Detailed Implementation

[0015] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present application, and not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present application.

[0016] It should be noted that, where there is no conflict, the embodiments and features of the embodiments in this application can be combined with each other. The embodiments of this application will be described in detail below with reference to the accompanying drawings.

[0017] refer to Figure 1 This embodiment discloses a method for assessing hepatic encephalopathy based on brain structural imaging, the method comprising the following steps: S100: Acquire brain structure imaging data of the subject.

[0018] In this step, acquiring brain structural imaging data of the subjects is fundamental to the subsequent evaluation process. The brain structural imaging data is preferably magnetic resonance imaging (MRI) data, specifically 3D-T1WI sequence images. The specific implementation process of this step includes subject recruitment, ethical review, data acquisition, and equipment parameter settings.

[0019] First, this embodiment uses a multi-center, large-sample cohort of data for model building and validation. Specifically, data acquisition involves two centers: a first center serves as the discovery cohort; and a second center serves as an independent validation cohort.

[0020] The subjects included a case group and a control group: Case group: Included patients with clinically or radiologically confirmed cirrhosis (CT, MRI, or ultrasound). Inclusion criteria included: age ≤ 60 years; normal vision, able to complete neuropsychological tests; right-handed. Exclusion criteria included: contraindications to MRI or claustrophobia; other organic brain lesions found on routine MRI (e.g., cerebral infarction, brain tumor, etc.); history of alcoholic cirrhosis or drug abuse, or currently taking medications that affect cognition; excessive head movement during examination, affecting data analysis; if the patient had undergone liver transplantation (LT), exclusion of severe postoperative complications (e.g., liver failure, or central nervous system complications, etc.).

[0021] Control Group: A normal control group was recruited, for example, in the Tianjin area where Center 1 is located. Inclusion criteria included: no history of acute or chronic liver disease; normal liver function results from a physical examination within the past six months; no history of neuropsychiatric disorders; normal vision; age, sex, and education level matched to the patient; right-handed. Exclusion criteria included: contraindications to MRI or claustrophobia; organic or functional lesions of the nervous system; unhealthy lifestyle habits (such as alcoholism) and a history of drug abuse; excessive head movement during the examination, which would affect data analysis.

[0022] All research protocols were reviewed and approved by the hospital's ethics committee, and all participants signed informed consent forms.

[0023] Data acquisition is then performed using a superconducting magnetic resonance scanner. In a preferred example of this embodiment, a Siemens Magnetom Prisma 3.0T magnetic resonance scanner (Siemens Medical Systems, Allegra, Germany) is used. Of course, those skilled in the art will understand that other 3.0T magnetic resonance scanners with equivalent resolution and signal-to-noise ratio can also be used in this embodiment.

[0024] The acquired sequence is a 3D-T1 WIMPRAGE image sequence. Specific image acquisition parameters are as follows: Scanning direction: Sagittal scan; Number of floors: 176; Repeat time / echo time / reversal time (TR / TE / TI): 1900ms / 3ms / 900ms; Layer thickness: 1mm; Field of view (FOV): 250mm × 250mm; Resolution: 256×256; Flip angle: 9°.

[0025] After acquiring brain structural imaging data, a data quality control step is also included. Data with excessive head movement during the examination, artifacts in the image, or that does not meet the above acquisition parameters are discarded to ensure that the quality of the brain structural imaging data input into the subsequent evaluation model meets the analysis requirements. In addition, for liver transplant patients in the case group, longitudinal follow-up data must be collected at preset time points (such as 1 month, 3-6 months, and 1-2 years after surgery) for subsequent model training and outcome prediction validation.

[0026] Through the above steps, high-quality brain structure imaging data of the subjects are obtained, providing a reliable data foundation for the subsequent extraction of characteristic parameters that characterize brain tissue morphology.

[0027] S200: Based on the brain structure imaging data, extract characteristic parameters that characterize the morphology of brain tissue.

[0028] The main objective of this step is to quantitatively analyze the brain structure imaging data acquired in step S100 to obtain characteristic parameters that objectively and quantitatively reflect morphological changes in brain tissue. These characteristic parameters will serve as input data for subsequent disease progression assessment models. The specific implementation process of this step includes brain region segmentation and reconstruction, initial feature extraction, region of interest construction, feature dimensionality reduction, and standardization.

[0029] Specifically, brain imaging software is preferably used to process the brain structure image data. In a preferred example of this embodiment, FreeSurfer software (e.g., version 6.0 or 7.2) is used for automated processing based on 3D-T1WI images to achieve fine segmentation and reconstruction of the whole brain structure, providing an anatomical basis for subsequent feature extraction.

[0030] Subsequently, based on the reconstructed brain structure model, multi-dimensional morphological features across the entire brain were extracted as initial features. Specific extracted feature parameters include, but are not limited to: Cortical thickness: Characterizes the variation in thickness of the gray matter in the cortex; Subcortical gray matter volume: characterizes the size of subcortical nuclei; Surface area: Characterizes the unfolded area of ​​the cortical surface; Curvature: Characterizes the degree of bending of cortical folds; Total intracranial volume (ICV): used as a covariate in subsequent standardization.

[0031] In this embodiment, the above-mentioned multi-dimensional morphological features of the entire brain of all subjects are calculated to form a high-dimensional initial feature set.

[0032] To systematically analyze brain structural features, this embodiment constructs multiple regions of interest (ROIs) covering the entire brain based on a multi-granularity brain atlas. The multi-granularity brain atlas can be an existing standard brain atlas or a custom atlas tailored to research needs. By mapping the extracted initial features onto these ROIs, morphological measurements within each ROI are obtained. For example, the entire brain can be divided into N ROIs, each corresponding to a set of morphological indicators.

[0033] Since the initial features have a high dimensionality and may contain redundant information, this embodiment uses a spectral decomposition algorithm to reduce the dimensionality of the initial features. Specifically, a feature covariance matrix is ​​constructed, and principal components are extracted through eigenvalue decomposition to extract the maximum representation vector of morphological feature heterogeneity in the low-dimensional space. This step aims to retain key information reflecting the heterogeneity of brain structural atrophy while reducing computational complexity, resulting in the dimensionality-reduced feature parameters.

[0034] To eliminate the influence of individual differences (such as age, sex, and head size) on morphological features and to make the features comparable among different subjects, this embodiment also includes standardization processing of the feature parameters. The specific steps are as follows: A linear regression model was constructed by acquiring brain structural imaging data and corresponding morphological features from a healthy control group. Age, sex, and total intracranial volume were used as covariates, and the morphological measurement value of each region of interest (ROI) was used as the dependent variable. Based on the morphological features of the healthy control group, a linear regression model was constructed.

[0035] Calculate the standardized Z-score: Expected values ​​for the morphological features of each ROI for the subjects were calculated using the linear regression model. The standardized Z-score for each ROI was calculated by subtracting the expected values ​​from the observed values ​​and dividing by the standard deviation of the control group residuals.

[0036] Model input preparation: The standardized Z-score is used as the final input to the feature parameters of the disease progression assessment model.

[0037] Through step S200, standardized and dimensionality-reduced characteristic parameters representing the morphology of brain tissue were obtained. These parameters not only eliminated the interference of physiological individual differences, but also focused on reflecting the pathological brain structural changes associated with hepatic encephalopathy, providing high-quality data input for inferring the spatiotemporal progression trajectory through the disease progression assessment model in the subsequent step S300.

[0038] S300: The feature parameters are processed by a pre-trained disease progression assessment model to infer the spatiotemporal progression trajectory of brain structural abnormalities.

[0039] In this step, the standardized feature parameters (i.e., standardized Z-scores) obtained in step S200 are processed using a pre-trained disease progression assessment model. The disease progression assessment model aims to reconstruct the trajectory of gray matter atrophy progression in an individual from cross-sectional image data, thereby inferring the spatiotemporal progression pattern of brain structural abnormalities. In this embodiment, the disease progression assessment model is constructed using the Subtype and Stage Inference (SuStaIn) algorithm, referencing... Figure 4 This is a schematic diagram of the subtype construction framework based on the SuStaIn algorithm in this embodiment.

[0040] "Pre-training" refers to the model having completed parameter learning and optimization based on the discovery cohort (e.g., data from 300 cirrhosis patients and 100 healthy controls collected at Center 1). During the pre-training phase, the model uses unsupervised machine learning to extract population-level disease progression patterns from cross-sectional data without relying on longitudinal follow-up data. Specifically, the model models brain structural abnormalities as an ordered accumulation of discrete events and estimates the most probable event sequence for each subtype at the population level using the Markov Chain Monte Carlo (MCMC) sampling method.

[0041] To quantify the progression of brain structural abnormalities, this embodiment pre-sets multiple abnormality thresholds for each region of interest (ROI) to characterize different degrees of pathological abnormalities. Specifically, each ROI is assigned three abnormality thresholds: Z=1, Z=2, and Z=3. Therefore, each ROI corresponds to three potential abnormal events, each event representing the degree of atrophy of that ROI (i.e., the absolute value of the standardized Z-score) reaching the corresponding threshold.

[0042] If the system contains N ROIs, the total number of potential events is 3N. The disease progression assessment model models the disease progression process as an ordered accumulation of the aforementioned discrete events. Different disease subtypes correspond to different event sequences, i.e., different spatiotemporal progression trajectories. For example, one subtype may show the frontal lobe region reaching the Z=1 threshold first, followed by the hippocampus region; while another subtype may show abnormalities occurring first in the thalamus region.

[0043] When applying a pre-trained model to process the feature parameters of the current subjects, the specific steps include the following: Subtype Assumptions and Optimization: The model is run under the assumption of k=2 to 6 subtypes. For each k value, 25 random starting points are used for optimization to avoid getting trapped in local optima.

[0044] MCMC Sampling: After each optimization, 1,000,000 Markov Chain Monte Carlo (MCMC) iterations are run. The sampling method infers the most likely order of discrete events across different disease subtypes. This process, based on continuous Z-score data, calculates the probability of observing current subject characteristic parameters under a given subtype using maximum likelihood estimation.

[0045] Optimal model determination: The optimal number of subtypes is determined based on the cross-validation information criterion.

[0046] Trajectory generation: Based on the output of the optimal model, the spatiotemporal progression trajectory is generated. Each trajectory represents a specific spatiotemporal progression pattern of gray matter atrophy, containing a chronological list of all discrete events under that subtype.

[0047] The core advantage of this step lies in the configuration of the disease progression assessment model, which infers longitudinal disease evolution patterns based on cross-sectional imaging data. Traditional methods require massive amounts of longitudinal tracking data to depict disease progression, while this embodiment utilizes the characteristics of the SuStaIn algorithm to simulate a virtual longitudinal trajectory of disease development by integrating data from a large number of cross-sectional subjects. This means that for the current subject, only a single collection of brain structure imaging data (cross-sectional data) is needed to match their relative position on a certain spatiotemporal progression trajectory through the model, thereby inferring the evolutionary path of their brain structural abnormalities.

[0048] Aside from Z-score standardization, this embodiment does not perform additional sorting, discretization, or clustering preprocessing on the feature parameters input to the disease progression assessment model to maintain the original distribution characteristics of the data. The model directly calculates the probability distribution of events based on continuous Z-score data. Through the above processing, the disease progression assessment model outputs several statistically significant spatiotemporal progression trajectories. Each trajectory reflects a specific spatiotemporal evolutionary pattern of brain structural atrophy during the development of hepatic encephalopathy, providing a model basis for the subsequent step S400 to determine the disease subtype and disease progression stage of the subject.

[0049] S400: Determine the disease subtype and disease progression stage of the subject based on the spatiotemporal progression trajectory.

[0050] Specifically, in this step, based on the spatiotemporal progression trajectory at the group level inferred in step S300, individualized subtype classification and disease severity staging are performed for the current subject. The disease progression assessment model can evaluate the correlation between an individual's MRI data and the pathological progression stage of each trajectory, thereby achieving individualized subtype staging. The specific implementation process of this step includes matching degree calculation, subtype probability distribution allocation, optimal subtype determination, and progression stage localization.

[0051] First, the subject characteristic parameters (i.e., the standardized Z-scores of each ROI) obtained in step S2 are input into the disease progression assessment model. The model calculates the degree of matching between the subject's characteristic parameters and each of the spatiotemporal progression trajectories. Specifically, based on the maximum likelihood estimation principle, the model calculates the probability that the subject belongs to each spatiotemporal progression trajectory (i.e., each disease subtype) given the current image feature observations.

[0052] Subsequently, a probability distribution for each disease subtype is assigned to the subject based on the Bayesian posterior probability. The probability distribution ranges from 0 to 1. For example, if the model identifies the optimal number of subtypes as 3, the model will output three posterior probability values ​​for the subject belonging to subtype 1, subtype 2, and subtype 3, and the sum of these three probability values ​​is 1.

[0053] Furthermore, based on the calculated probability distribution, the disease subtype corresponding to the maximum posterior probability is determined as the disease subtype to which the subject belongs.

[0054] In practice, the posterior probability values ​​of the subjects belonging to each disease subtype are compared, and the subtype label corresponding to the highest probability value is selected. For example, if the probability of a subject belonging to subtype 1 is 0.7, the probability of belonging to subtype 2 is 0.2, and the probability of belonging to subtype 3 is 0.1, then the subject is determined to belong to subtype 1. This subtype represents which evolutionary path in the population best matches the spatiotemporal progression pattern of gray matter atrophy in the subject's brain.

[0055] After determining the disease subtype to which the subject belongs, the disease progression stage is determined based on the cumulative number of events under that disease subtype.

[0056] In the disease progression assessment model, disease progression is modeled as an ordered accumulation of discrete events. As described in step S300, each ROI has multiple anomaly thresholds set (e.g., Z=1, Z=2, Z=3), and each threshold being triggered is considered an event occurring. The disease progression stage corresponds to the number of events that have occurred in that subtype trajectory.

[0057] In practice, the model combines the subject's imaging characteristics to determine how many discrete events have been triggered in the event sequence for that specific subtype (i.e., the absolute value of the Z-score for the corresponding brain region has reached or exceeded the threshold defined for that event). The number of triggered events is mapped to a disease progression stage value. For example, if a subtype trajectory contains 30 event sequences, and the subject has met the conditions for the occurrence of the first 15 events, then the subject's disease progression stage is determined to be stage 15. The higher the stage value, the more severe the accumulation of brain structural abnormalities, and the later the disease progression.

[0058] Finally, through the above processing, the model outputs individualized staging results for each subject, including: Subtype label: Indicates the neural structural subtype to which the subject belongs (e.g., subtype 1, subtype 2, etc.); Stage value: Indicates the specific progress position of the subject on the subtype trajectory (e.g., stage 0 to stage N).

[0059] This result enables the identification of subtypes and stages of individual data, overcoming the limitation of traditional methods that rely on longitudinal follow-up data for staging. Only a single cross-sectional image is needed to infer the relative position of the subject in the longitudinal evolution of the disease.

[0060] Through step S400, this embodiment not only clarifies which gray matter atrophy pattern (subtype) the subject belongs to, but also quantifies the severity (stage) of their disease. This fine-grained classification and staging result can more accurately reflect the pathobiological heterogeneity of hepatic encephalopathy, providing a core quantitative basis for generating assessment information to characterize the disease state of hepatic encephalopathy in the subsequent step S500.

[0061] S500: Based on the disease subtype and disease progression stage, generate assessment information to characterize the disease status of hepatic encephalopathy.

[0062] In this step, based on the disease subtype and disease progression stage of the subject determined in step S400, and combined with the subject's clinical baseline information, assessment information is generated to characterize the disease status of hepatic encephalopathy. This assessment information aims to provide clinicians with quantitative references regarding the patient's disease progression trend, complication risk, and treatment prognosis, thereby enabling stratified management and precise intervention for patients with hepatic encephalopathy.

[0063] This embodiment calls a pre-trained disease outcome prediction model. The prediction model preferably employs a multi-task hierarchical prediction architecture based on a graph neural network (GNN), referencing... Figure 5 This is a schematic diagram of the hierarchical prediction model framework in this embodiment. In the model input construction stage, not only are the disease subtype labels and disease progression stage values ​​output in step S400 used as global feature inputs, but the original brain tissue morphological feature parameters extracted in step S200 are also fused.

[0064] Specifically, a covariant network of individual brain region structures is constructed, using regions of interest (ROIs) across the entire brain as nodes and morphological feature correlations between brain regions as edges. Simultaneously, multimodal structural parameters such as gray matter volume, cortical thickness, and morphological features are extracted to form a feature pool, which is then reconstructed as a graph structure input. This input construction method can simultaneously capture the degree of atrophy in local brain regions (reflected by stage values) and the topological connectivity anomalies of the global brain network (reflected by the graph structure), thereby improving prediction accuracy.

[0065] Firstly, a warning sign of the risk of developing obvious hepatic encephalopathy (OHE): In this embodiment, the assessment information includes early warning information on the risk of developing obvious hepatic encephalopathy (OHE).

[0066] In practice, the disease subtype and disease progression stage are input into a pre-trained risk prediction model. This risk prediction model, trained with historical cohort data (such as longitudinal follow-up data from 300 cirrhosis patients in the aforementioned cohort), is able to learn the nonlinear mapping relationship between different subtypes and stages and the onset time of OHE.

[0067] The model outputs the probability that the subject will develop significant hepatic encephalopathy within a preset time window. This preset time window can be set according to clinical needs, such as 6 months, 1 year, or 2 years. For example, the model outputs an 85% probability that a subject will develop OHE within the next year.

[0068] In addition, the model can set risk thresholds based on the stage of disease progression. For example, when a subject is identified as belonging to a specific high-risk subtype (such as subtype 1) and the disease progression stage exceeds stage 15, the system automatically generates a "high-risk" warning signal, prompting clinicians to provide primary preventive treatment (such as using drugs like lactulose or rifaximin) to reduce the incidence of OHE.

[0069] Secondly, assessment of cognitive recovery after liver transplantation (LT): In this embodiment, the assessment information includes cognitive recovery assessment information after liver transplantation (LT).

[0070] In practice, for cirrhotic patients scheduled for liver transplantation, the disease subtype and disease progression stage are input into a pre-trained outcome prediction model. This outcome prediction model is trained based on a longitudinal follow-up cohort after liver transplantation (such as postoperative follow-up data from 100 patients who underwent LT surgery in Center 1 mentioned above), and can reflect the correlation between preoperative neural structural damage patterns and the degree of postoperative cognitive function improvement.

[0071] The model outputs a predicted value for the degree of improvement in cognitive function of the subjects after liver transplantation. The degree of improvement in cognitive function can be characterized by the change in scores on neuropsychological scales, such as the improvement score on the Hepatic Encephalopathy Psychometric Scale (PHES).

[0072] For example, the model predicts that a subject's PHES score will improve by 5 points at 6 months post-surgery, indicating good cognitive recovery; while another subject is predicted to improve by only 1 point or not at all, suggesting possible persistent post-operative cognitive impairment. This assessment information can be used for pre-operative doctor-patient communication, donor-recipient matching optimization, and post-operative rehabilitation planning. In situations of donor shortage, incorporating cognitive benefits into donor allocation considerations helps to optimize resource allocation.

[0073] Within the multi-task hierarchical prediction model, a mechanism combining a general feature learning stage and a multi-task prediction stage is adopted.

[0074] In the general feature learning phase, the pseudo-labels output by the subtyping model (i.e., the subtypes determined in step S400) serve as supervision, employing graph neural networks to deeply fuse topological connections and multimorphological features of a covariant network. An attention mechanism is used to adaptively learn the contribution weights of each feature to disease progression and subtyping. For example, the model may automatically assign higher weights to morphological features in key brain regions such as the frontal lobe and thalamus, as these areas are more sensitive to the progression of hepatic encephalopathy.

[0075] In the multi-task prediction phase, the parameters of the general feature extractor are frozen, and only the multi-task prediction head is fine-tuned to achieve accurate prediction of post-LT cognitive scores and OHE onset time, respectively. Through transfer learning of general features under specific clinical tasks, the model achieves end-to-end mapping from brain imaging features to disease outcomes.

[0076] Finally, the system summarizes the above prediction results and generates assessment information to characterize the disease state of hepatic encephalopathy. This assessment information can be a structured data report or a visualized risk assessment map.

[0077] The report includes: Neural structural subtype classification results: indicate the type of atrophy pattern to which the patient belongs; Disease progression stage score: quantifies the current severity of the pathology; OHE seizure risk probability: the likelihood of an attack within a specific future time period; Predictive value of cognitive benefit after LT surgery: the expected degree of cognitive function recovery.

[0078] Through step S500, this embodiment transforms the abstract neuroimaging subtypes and staging results into specific clinical assessment information. This not only achieves a comprehensive characterization of the disease status of HE patients, but also provides objective and quantitative technical tools for primary prevention of OHE patients and personalized assessment of cognitive recovery after LT surgery, completing a closed loop from imaging data analysis to clinical auxiliary decision-making.

[0079] refer to Figure 2 This embodiment further discloses a hepatic encephalopathy assessment system based on brain structural imaging. This system can be deployed on a hospital image post-processing workstation, a cloud server, or a terminal device with computing capabilities. The system includes a data acquisition module 21, a feature extraction module 22, a trajectory inference module 23, a classification and staging module 24, and an information generation module 25. The modules communicate with each other via a data bus or network interface to collaboratively complete the identification, staging, and disease status assessment of the neural structural subtypes of patients with hepatic encephalopathy. The specific functions and implementation methods of each module are described in detail below.

[0080] The data acquisition module 21 is used to acquire brain structural imaging data of the subject. This module is equipped with a standard medical imaging communication interface (such as a DICOM interface) and can connect to a hospital's magnetic resonance imaging (MRI) equipment or picture archiving and communication system (PACS). In a preferred example of this embodiment, the data acquisition module is configured to receive 3D-T1 WIMPRAGE sequence images acquired from a Siemens Magnetom Prisma 3.0T superconducting MRI scanner. Specific parameters include: sagittal scan, 176 slices, TR / TE / TI = 1900ms / 3ms / 900ms, slice thickness 1mm, FOV = 250mm × 250mm, resolution 256 × 256, and flip angle 9°. Furthermore, this module also includes a data quality control unit for automatically detecting head movement and artifacts in the image data. Data with excessive head movement or that does not conform to the above acquisition parameters is discarded or marked to ensure that the quality of the brain structural imaging data input to subsequent modules meets the analysis requirements. For patients after liver transplantation, this module also supports obtaining longitudinal follow-up imaging data at preset time points (such as 1 month, 3-6 months, 1-2 years after surgery).

[0081] Feature extraction module 22 is used to extract morphological features representing brain tissue based on the brain structure imaging data. This module integrates or calls brain imaging processing software (such as FreeSurfer version 6.0 or 7.2) to automate the processing of the brain structure imaging data. Specifically, this module is configured to perform motion correction, craniotomy, tissue classification (gray matter, white matter, cerebrospinal fluid), cortical surface reconstruction, and subcortical structure segmentation. Based on this, at least one of the following initial features is extracted from the whole-brain region of interest: cortical thickness, subcortical gray matter volume, surface area, and curvature. This module also incorporates a multi-granularity brain atlas to construct multi-spatial regions of interest (ROIs) covering the entire brain and performs dimensionality reduction on the initial features using a spectral decomposition algorithm. Furthermore, this module stores morphological feature data and linear regression model coefficients of a healthy control group and is configured to calculate the standardized Z-score of the subject's feature parameters relative to the healthy control group, using age, sex, and intracranial volume as covariates. The standardized Z-score is then output as the final feature parameter to the trajectory inference module.

[0082] The trajectory inference module 23 is used to process the feature parameters through a pre-trained disease progression assessment model to infer the spatiotemporal progression trajectory of brain structural abnormalities. This module is loaded with a pre-trained subtype stage inference (SuStaIn) algorithm model. This model has completed parameter learning based on historical cohort data (such as a multicenter cohort of cirrhosis patients) and is configured to infer longitudinal disease evolution patterns based on cross-sectional image data. Specifically, this module models brain structural abnormalities as an ordered accumulation process of discrete events, where each event corresponds to a specific brain region whose morphological features reach a preset abnormality threshold (e.g., Z=1, Z=2, and Z=3). This module is configured to estimate the event sequence of each subtype at the population level using a Markov chain Monte Carlo (MCMC) sampling method under a preset number of subtype assumptions (e.g., k=2-6), and determine the optimal number of subtypes based on cross-validation information criteria, thereby forming several statistically significant spatiotemporal progression trajectories. Each trajectory reflects a specific spatiotemporal evolutionary pattern of brain structural atrophy during the development of hepatic encephalopathy.

[0083] The classification and staging module 24 is used to determine the disease subtype and disease progression stage of the subject based on the spatiotemporal progression trajectory. This module is configured to calculate the matching degree between the subject's characteristic parameters and each spatiotemporal progression trajectory. Specifically, this module assigns a probability distribution (range 0-1) to each disease subtype for the subject based on Bayesian posterior probabilities. The module compares the posterior probability values ​​of the subject belonging to each disease subtype and determines the subtype corresponding to the highest posterior probability as the subject's disease subtype. After determining the subtype, the module determines the disease progression stage based on the cumulative number of events under that subtype, i.e., determining how many discrete events have been triggered in the event sequence of that specific subtype (the absolute value of the Z-score of the corresponding brain region of the subject has reached or exceeded the threshold defined for that event), and maps the number of triggered events to a disease progression stage value. The higher the stage value, the more severe the cumulative degree of brain structural abnormalities.

[0084] Information generation module 25 is used to generate assessment information characterizing the state of hepatic encephalopathy based on the disease subtype and disease progression stage. This module is loaded with a pre-trained multi-task hierarchical prediction model (such as a graph neural network (GNN) based model). This module is configured to input the disease subtype and disease progression stage into the prediction model and, in conjunction with the subject's clinical baseline information, output specific assessment information. In one application scenario of this embodiment, the assessment information includes early warning information on the risk of developing overt hepatic encephalopathy (OHE). This module outputs the probability that the subject will develop OHE within a preset time window (e.g., 1 year) and generates a high-risk warning signal when the probability exceeds a threshold. In another application scenario, the assessment information includes cognitive recovery assessment information after liver transplantation (LT). This module outputs a predicted value (e.g., PHES score improvement score) of the subject's cognitive function improvement after liver transplantation. Finally, this module integrates the above subtype classification results, stage scores, risk probabilities, and predicted values ​​into a structured data report or a visualized risk assessment atlas for clinicians' reference.

[0085] Figure 3 A schematic diagram of the physical structure of an electronic device provided in an embodiment of the present invention, such as... Figure 3 As shown, the electronic device 50 includes: a processor 501, a memory 502, and a bus 503; The processor 501 and the memory 502 communicate with each other via the bus 503; the processor 501 is used to call the program instructions in the memory 502 to execute the methods provided in the above-described embodiments.

[0086] This embodiment provides a non-transitory computer-readable storage medium that stores computer instructions that cause a computer to execute the methods provided in the above-described embodiments.

[0087] Those skilled in the art will understand that all or part of the steps of the above-described method implementation can be implemented by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When the program is executed, it performs the steps of the above-described method implementation. The aforementioned storage medium includes various storage media capable of storing program code, such as ROM, RAM, magnetic disk, or optical disk.

[0088] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0089] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., including several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods of each embodiment or some parts of the embodiments.

[0090] The specific embodiments described above do not constitute a limitation on the scope of protection of this invention. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can occur depending on design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this invention should be included within the scope of protection of this invention.

Claims

1. A method for assessing hepatic encephalopathy based on brain structural imaging, characterized in that, The method includes: Obtain imaging data of the subject's brain structure; Based on the brain structure imaging data, feature parameters characterizing brain tissue morphology are extracted; The feature parameters are processed by a pre-trained disease progression assessment model to infer the spatiotemporal progression trajectory of brain structural abnormalities. The disease subtype and disease progression stage of the subject are determined based on the spatiotemporal progression trajectory. Based on the disease subtype and disease progression stage, assessment information is generated to characterize the disease status of hepatic encephalopathy.

2. The method for assessing hepatic encephalopathy according to claim 1, characterized in that, The extraction of characteristic parameters representing brain tissue morphology based on the brain structure imaging data includes: Brain region segmentation and reconstruction were performed on the brain structure image data using brain imaging processing software. At least one of the following initial features is extracted from the whole brain region of interest: cortical thickness, subcortical gray matter volume, surface area, and curvature; The initial features are dimensionality reduced based on a multi-granularity brain map to obtain the feature parameters.

3. The method for assessing hepatic encephalopathy according to claim 2, characterized in that, The method further includes standardization processing of the feature parameters: Obtain brain structural imaging data and corresponding morphological features from healthy control groups; A linear regression model was constructed based on the morphological characteristics of the healthy control group, with age, sex, and intracranial volume as covariates. The standardized Z-score of the subject's characteristic parameters relative to the healthy control group is calculated using the linear regression model, and the standardized Z-score is used as the characteristic parameter input to the disease progression assessment model.

4. The method for assessing hepatic encephalopathy according to claim 1, characterized in that, The disease progression assessment model is configured to infer longitudinal disease evolution patterns based on cross-sectional image data; The inferred spatiotemporal progression trajectory of brain structural abnormalities includes: Brain structural abnormalities are modeled as an ordered accumulation of discrete events, where each event corresponds to a morphological feature of a specific brain region that reaches a preset abnormality threshold. The most likely sequence of discrete events under different disease subtypes is inferred by the disease progression assessment model to form the spatiotemporal progression trajectory.

5. The method for assessing hepatic encephalopathy according to claim 4, characterized in that, The disease progression assessment model employs a subtype-stage inference algorithm. The preset anomaly thresholds include multiple levels of Z-score thresholds; The inference of the most likely sequence of discrete events under different disease subtypes using the disease progression assessment model includes: Under the assumption of a predetermined number of subtypes, the event sequence of each subtype at the population level is estimated using the Markov chain Monte Carlo sampling method; The optimal number of subtypes is determined based on cross-validation information criteria.

6. The method for assessing hepatic encephalopathy according to claim 1, characterized in that, The determination of the disease subtype and disease progression stage of the subject based on the spatiotemporal progression trajectory includes: Calculate the degree of matching between the subject's characteristic parameters and each of the spatiotemporal progression trajectories; The probability distribution of the subjects belonging to each disease subtype is assigned based on the Bayesian posterior probability. The disease subtype corresponding to the maximum posterior probability is determined as the disease subtype to which the subject belongs, and the disease progression stage is determined based on the cumulative number of events under that disease subtype.

7. The method for assessing hepatic encephalopathy according to claim 1, characterized in that, The assessment information includes early warning information on the risk of developing obvious hepatic encephalopathy; The generation of assessment information for characterizing the state of hepatic encephalopathy includes: The disease subtype and disease progression stage are input into a pre-trained risk prediction model, which outputs the probability that the subject will develop significant hepatic encephalopathy within a preset time window.

8. The method for assessing hepatic encephalopathy according to claim 1, characterized in that, The assessment information includes cognitive recovery assessment information after liver transplantation; The generation of assessment information for characterizing the state of hepatic encephalopathy includes: The disease subtype and disease progression stage are input into a pre-trained outcome prediction model, which outputs a predicted value of the degree of improvement in cognitive function of the subject after liver transplantation.

9. A system for assessing hepatic encephalopathy based on brain structural imaging, characterized in that, include: The data acquisition module is used to acquire brain structure imaging data of the subjects; The feature extraction module is used to extract feature parameters characterizing brain tissue morphology based on the brain structure image data. The trajectory inference module is used to process the feature parameters through a pre-trained disease progression assessment model to infer the spatiotemporal progression trajectory of brain structural abnormalities. The subtyping and staging module is used to determine the disease subtype and disease progression stage of the subject based on the spatiotemporal progression trajectory. The information generation module is used to generate assessment information to characterize the disease status of hepatic encephalopathy based on the disease subtype and disease progression stage.

10. A computer-readable storage medium, characterized in that, The storage medium stores a computer program that, when executed by a processor, implements the method of any one of claims 1-8.

Images