State evolution prediction method and device and electronic equipment

Abstract

The invention provides a state evolution prediction method and apparatus, and an electronic device. The method comprises the steps of obtaining a to-be-detected image of a target object; inputting a to-be-detected image into the state stage prediction model, and determining a current state stage and at least one candidate evolution stage of the target object; wherein the state stage prediction model learns a state development path from a normal state to at least one abnormal stage state from a historical sample database; querying similar historical samples and corresponding actual state evolution records based on the image features of the to-be-detected image and the current state stage to generate a state evolution reference path; based on the current state stage, the at least one candidate evolution stage and the state evolution reference path, the probability distribution of the target object entering different state stages is determined, the crossing from static diagnosis to time sequence risk early warning is realized through the dynamic evolution prediction model, and a quantifiable clinical decision basis is provided for early lesion management.

Description

TECHNICAL FIELD

[0001] Embodiments of the present application relate to the field of information technology, and in particular to a state evolution prediction method and device and electronic equipment. BACKGROUND

[0002] Currently, medical image processing and artificial intelligence technology have been widely applied in endoscopic assisted diagnosis field, and a classification model based on deep learning can effectively identify typical lesion structures and anatomical sites. However, in clinical practice, there are a large number of "weak positive" images with atypical signs, fuzzy boundaries or weak features. Such images may correspond to the early stage or atypical manifestation of the disease, and accurate identification and dynamic evaluation of such images are crucial for early intervention of the disease. How to intelligently analyze and predict the evolution trend of such images has become a key research direction to improve the efficiency of clinical diagnosis.

[0003] At present, existing endoscopic assisted diagnosis systems mainly rely on static deep learning models, and make instant judgments on single examination images through end-to-end classification or segmentation networks. Although such technology performs well in identifying typical lesions, its function is limited to the detection of signs at the current time, and lacks the ability to quantitatively analyze the dynamic development of the disease.

[0004] The existing technology in processing endoscopic images is essentially still limited to the classification or segmentation level at a single time point, and the output result can only reflect the lesion state at the examination time. This static characteristic causes the system to completely ignore the essential laws of disease as a dynamic evolution process: neither can it construct the transfer relationship between the stages of disease development based on historical data, nor can it predict the potential development direction of the current signs to recovery, stability, deterioration or cross-level deterioration. Ultimately, the assisted diagnosis system lacks predictability of the course of evolution, and it is difficult to meet the actual needs of dynamic monitoring and trend prediction of early lesions in clinical practice. SUMMARY

[0005] The present application provides a state evolution prediction method, device and electronic equipment to realize the leap from static diagnosis to time series risk warning through a dynamic evolution prediction model, and provides a quantifiable basis for clinical decision-making for early lesion management.

[0006] In a first aspect, the embodiments of the present application provide a state evolution prediction method, which comprises:

[0007] obtaining a to-be-detected image of a target object; wherein the to-be-detected image is a human tissue surface image collected by an endoscope;

[0008] input the to-be-detected image into a state stage prediction model to determine a current state stage and at least one candidate evolution stage of the target object; wherein the state stage prediction model learns a state development path from a historical sample database through pre-training, and the state development path includes a state development relationship from a normal state to at least one abnormal stage state;

[0009] query, based on image features of the to-be-detected image and the current state stage, similar historical samples and corresponding actual state evolution records from the historical sample database to generate a state evolution reference path corresponding to the target object;

[0010] determine, based on the current state stage, the at least one candidate evolution stage and the state evolution reference path, a probability distribution of the target object entering different state stages in the future.

[0011] In a second aspect, an embodiment of the present application further provides a state evolution prediction device, and the device comprises:

[0012] a data acquisition module configured to acquire a to-be-detected image of a target object; wherein the to-be-detected image is a human tissue surface image collected through an endoscope;

[0013] a state stage determination module configured to input the to-be-detected image into a state stage prediction model to determine a current state stage and at least one candidate evolution stage of the target object; wherein the state stage prediction model learns a state development path from a historical sample database through pre-training, and the state development path includes a state development relationship from a normal state to at least one abnormal stage state;

[0014] an evolution path reference module configured to query, based on image features of the to-be-detected image and the current state stage, similar historical samples and corresponding actual state evolution records from the historical sample database to generate a state evolution reference path corresponding to the target object;

[0015] a state evolution prediction module configured to determine, based on the current state stage, the at least one candidate evolution stage and the state evolution reference path, a probability distribution of the target object entering different state stages in the future.

[0016] In a third aspect, an embodiment of the present application further provides an electronic device, and the electronic device comprises:

[0017] one or more processors;

[0018] a storage device configured to store one or more programs,

[0019] When one or more programs are executed by one or more processors, the one or more processors realize the state evolution prediction method according to any of the embodiments of the present application.

[0020] In a fourth aspect, the embodiments of the present application further provide a storage medium containing computer executable instructions, which, when executed by a computer processor, are used to perform the state evolution prediction method according to any of the embodiments of the present application.

[0021] The embodiments of the present application provide a state evolution prediction method and device and electronic equipment, and the method comprises: acquiring a to-be-detected image of a target object, wherein the to-be-detected image is a human tissue surface layer image collected by an endoscope, and then inputting the to-be-detected image into a state stage prediction model to determine a current state stage and at least one candidate evolution stage of the target object, wherein the state stage prediction model learns a state development path from a historical sample database through pre-training, the state development path comprises a state development relationship from a normal state to at least one abnormal stage state, further, based on image features of the to-be-detected image and the current state stage, querying similar historical samples and corresponding actual state evolution records from the historical sample database to generate a state evolution reference path corresponding to the target object, and finally, based on the current state stage, the at least one candidate evolution stage and the state evolution reference path, determining a probability distribution of the target object entering different state stages in the future. The technical scheme of the present application realizes the transition from static judgment to dynamic deduction by constructing a state stage prediction model with time series prediction capability and fusing actual evolution records in the historical sample database, enhances the forward-looking analysis capability of the disease development trend, provides quantitative decision support with clinical predictability for early lesion monitoring and intervention, and thus improves the initiative and accuracy of the diagnosis and treatment process. BRIEF DESCRIPTION OF DRAWINGS

[0022] In order to more clearly illustrate the technical solutions of the example embodiments of the present application, the drawings needed in the description of the embodiments are briefly introduced as follows. Obviously, the drawings introduced are only a part of the drawings of the embodiments to be described by the present application, and not all the drawings. Those skilled in the art can also obtain other drawings according to these drawings without creating creative labor.

[0023] Figure 1 A flowchart of a state evolution prediction method provided by the embodiments of the present application is shown in the figure;

[0024] Figure 2 A representation diagram of a state development path related to the embodiments of the present application is shown in the figure;

[0025] Figure 3A flowchart illustrating yet another state evolution prediction method provided in an embodiment of this application;

[0026] Figure 4 A flowchart illustrating yet another state evolution prediction method provided in an embodiment of this application;

[0027] Figure 5 This is a schematic diagram illustrating the implementation process of the state evolution prediction method involved in the embodiments of this application;

[0028] Figure 6 This is a schematic diagram of a state evolution prediction device provided in an embodiment of this application;

[0029] Figure 7 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0030] The present application will now be described in further detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the application and not intended to limit it. Furthermore, it should be noted that, for ease of description, the accompanying drawings show only the parts relevant to the present application, not the entire structure.

[0031] Before introducing the technical solution provided in this application, the application scenarios of the solution can be explained first. This embodiment is applicable to various scenarios that require dynamic prediction of early lesions in endoscopic images. Currently, although static deep learning models are widely used in lesion identification, the traditional method's design of single-point-of-time classification has obvious limitations. In practical applications, early lesions often present with atypical signs and weak features, and the disease development has a multi-directional evolutionary pattern. Static judgment is difficult to capture the dynamic evolutionary characteristics of the disease course, which easily leads to a lack of foresight in risk assessment and follow-up decisions. Therefore, there is an urgent need for a method that can perform multi-path inference based on current signs and historical evolutionary patterns to improve the predictive ability of disease development trends and clinical decision support value. This embodiment constructs a state development path map, combines image feature matching and similar case evolution records, to achieve probabilistic prediction of candidate evolutionary stages, thereby ensuring accurate and reliable disease risk assessment and effectively improving the dynamic prediction performance of the auxiliary diagnostic system.

[0032] Example 1

[0033] Figure 1 This is a flowchart illustrating a state evolution prediction method provided in an embodiment of this application. This embodiment can be applied to any situation where early dynamic prediction of lesions in endoscopic images is required. The method can be executed by a state evolution prediction device, which can be implemented in the form of software and / or hardware. The hardware can be a controller, such as a mobile terminal, a PC, or a server.

[0034] like Figure 1 As shown, the method includes:

[0035] S110. Obtain the image to be detected of the target object; wherein, the image to be detected is a surface image of human tissue acquired through an endoscope.

[0036] Here, the target object refers to the specific individual whose corresponding endoscopic image to be analyzed and processed; this individual is the object for which state evolution prediction is currently required. The image to be detected specifically refers to the surface image of human tissue of the target object acquired through endoscopy, which requires state analysis and evolution prediction. In this technical solution, the image to be detected can be a real-time dynamic or static image acquired during clinical examinations using endoscopic equipment such as gastroscopy, colonoscopy, or bronchoscopy, targeting the mucosal surface of the target object's digestive tract, respiratory tract, or other internal hollow organs. For example, a digital image captured during gastroscopy showing slightly irregular mucosal texture in the gastric angle region serves as direct input to the state stage prediction model, carrying key visual information reflecting the current health status of the tissue.

[0037] In this embodiment, when it is necessary to predict the state evolution of a target object, the image to be detected can be obtained by calling the endoscopic image archiving system or by directly receiving a real-time video stream from the endoscopic device.

[0038] S120. Input the image to be detected into the state stage prediction model to determine the current state stage of the target object and at least one candidate evolution stage.

[0039] Among them, the state stage prediction model refers to a computational model based on a time-series modeling neural network structure that learns the state development path from a historical sample database through pre-training. Its function is to receive the image to be detected as input, determine the current state stage of the target object by analyzing the visual features of the image, and deduce at least one candidate evolution stage based on the state development relationship learned internally.

[0040] More specifically, the state stage prediction model learns state evolution paths from a historical sample database through pre-training. These paths include the state evolution relationships from a normal state to at least one abnormal stage. The historical sample database is a pre-built structured collection storing multiple historical sample data sets. Each historical sample data set contains a sequence of endoscopic images from an initial state to a final state, along with a clinically confirmed state stage label for each image. These accurately labeled time-series data sets collectively constitute the fundamental data source for training the state stage prediction model and generating the state evolution reference path.

[0041] The state development path refers to the logical framework learned by the state stage prediction model from a historical sample database through pre-training, used to describe the evolutionary patterns between state stages. This path defines the state development relationship from a normal state to at least one abnormal stage, constituting the core basis for inferring candidate evolution stages and calculating probability distributions. For an example, see the schematic diagram of the state development path. Figure 2 ,like Figure 2 As shown, the specific evolutionary path is as follows: Stage 0 (normal tissue) has a low probability of evolving into Stage 1 (mild abnormality); Stage 1 cases exhibit a dynamic characteristic of 10% probability of recovering to Stage 0, 60% probability of progressing to Stage 2 (moderate abnormality), and 30% probability of maintaining the current stage; Stage 2 cases show a three-way differentiation characteristic of 70% probability of progressing to Stage 3 (highly suspicious lesion), 20% probability of maintaining the current state, and 10% probability of improving to Stage 1; while Stage 3 points to the final evolutionary direction of developing into a solid lesion. This path, by quantitatively characterizing the transition probability between each stage, fully depicts the gradual evolutionary atlas from a normal state to a highly suspicious lesion.

[0042] This can be understood as follows: the state stage prediction model, through a pre-training process based on a historical sample database, has extracted and internalized the regular knowledge of state evolution from time-series data containing endoscopic image sequences and corresponding state stage labels, thereby establishing a probabilistic transition relationship network that can characterize the gradual transition from a normal state to different abnormal stages.

[0043] More specifically, the specific implementation steps involved in determining the state-stage prediction model may include:

[0044] (1) Construct a historical sample database, which includes multiple historical sample data. Each historical sample data includes an endoscopic image sequence from the initial state to the final state and a clinically confirmed state stage label corresponding to each image.

[0045] Among them, the endoscopic image sequence from the initial state to the final state refers to a series of endoscopic images arranged in chronological order in historical sample data, which completely record the entire process of a specific sample evolving from the earliest observable initial state to the final confirmed state.

[0046] In this context, the state stage label refers to the standardized annotation of the clinically diagnosed pathological state level corresponding to each frame of endoscopic image in the historical sample database. These labels serve as ground truth data for supervised learning, providing crucial classification criteria for constructing a temporal state development path sample set and training the state stage prediction model. In this embodiment, the state stage label includes at least one stage from the following: normal state, low-grade lesion, high-grade lesion, early cancer, and invasive cancer. The standardized classification system used when annotating endoscopic images in the historical sample database fully covers the progressive pathological spectrum from healthy tissue to malignant lesions. Each label represents a clinically diagnosed independent lesion level or nature category, collectively constituting the basic state set describing the state development path.

[0047] In this embodiment, a structured data set containing time-series endoscopic images and corresponding pathological state annotations can be formed by collecting and organizing complete medical imaging data and authoritative diagnostic results of multiple independent cases. This database serves as the basic data source for training state stage prediction models and generating state evolution reference paths.

[0048] (2) Based on the endoscopic image sequences and their corresponding state stage labels in each historical sample data, construct a time-series state development path sample set.

[0049] In this technical solution, a structured dataset for training a temporal modeling neural network can be constructed by extracting the state evolution patterns from the endoscopic image sequences and their corresponding state stage labels of various historical sample data in a historical sample database. This dataset is the temporal state development path sample set. Each sample contains complete temporal information reflecting the transition relationships and evolution patterns between state stages.

[0050] (3) The state stage prediction model of the time-series modeling neural network structure is trained based on the sample set of the time-series state development path to obtain the state stage prediction model.

[0051] Among them, the training state stage prediction model of the time-series modeling neural network structure refers to the initial computational model that adopts a neural network architecture with time-series data processing capabilities and has not yet completed parameter training. After being trained with a sample set of time-series state development paths, this model has the ability to learn the state evolution law from the endoscope image sequence and predict the state stage.

[0052] In this embodiment, a sample set of temporal state development paths containing complete state evolution information is used as training data and input into the training model of the temporal modeling neural network structure. The model parameters are adjusted by the optimization algorithm so that the model can learn and master the dynamic evolution law between state stages. Finally, a mature state stage prediction model that can accurately predict the current state stage and candidate evolution stages based on the input image is obtained.

[0053] In this context, the current state stage refers to the specific health status level or pathological development position of the target object at the time of image acquisition. At least one candidate evolution stage refers to one or more potential developmental states that the target object may enter in the future, relative to the current state stage. These states constitute the set of candidate targets for subsequent probability calculations and risk assessments. More specifically, at least one candidate evolution stage includes the maintenance state, the improving state, the adjacent deterioration state, and the leapfrog deterioration state relative to the current state stage. This can be understood as the potential developmental states deduced by the state stage prediction model fully covering four basic evolutionary types: the maintenance state (maintaining the existing health status), the improving state (transforming towards a more benign direction), the adjacent deterioration state (developing towards an adjacent, more severe stage), and the leapfrog deterioration state (skipping intermediate stages and directly developing into a more severe abnormality).

[0054] In this technical solution, the image of the target object to be detected is passed as input data to the pre-trained state stage prediction model. After the model analyzes and processes the image features, it synchronously outputs the exact state stage classification result of the target object at the time of image acquisition, and infers several candidate evolution stages that it may enter in the future based on the pre-learned state development path.

[0055] For example, when an image to be detected is input showing slightly irregular mucosal texture in the gastric angle region, the state stage prediction model will output "low-grade lesion" as the current state stage, and simultaneously generate a set of candidate evolution stages including "maintaining state" (stable lesion), "improving state" (inflammation subsides), "adjacent deterioration state" (develops into high-grade lesion), and "trans-grade deterioration state" (directly develops into early cancer).

[0056] S130. Based on the image features and current state stage of the image to be detected, query similar historical samples and corresponding actual state evolution records from the historical sample database to generate a state evolution reference path corresponding to the target object.

[0057] Among them, similar historical samples refer to historical case data records that have a high degree of matching with the current target object's image and current state stage in feature space or clinical representation. Actual state evolution records refer to objective medical factual data recorded in the historical sample database, which are confirmed through subsequent clinical follow-up, and describe the complete evolution process of a specific historical sample from its initial state to its final state, including the sequence of state stages and corresponding timestamps.

[0058] Among them, the state evolution reference path refers to an empirical reference model used to describe the actual state stage evolution trajectory and pattern of cases with similar characteristics to the current state of the target object in history.

[0059] Specifically, the similarity between the feature vector of the image to be detected and the features of cases in the historical sample database can be calculated. Combined with the matching degree of the current state stage, a set of comparable historical cases can be retrieved. Then, based on the actual state evolution process of these cases, an empirical reference trajectory reflecting the potential development law can be constructed, which is the state evolution reference path corresponding to the target object.

[0060] For example, when the image to be detected of the target object shows that there are subtle erosion features in the gastric angle area and the current state stage is determined to be a low-grade lesion, 20 historical samples with similar erosion features and the same initial state stage are retrieved from the historical sample database by calculating the image feature similarity. It is then statistically analyzed that during the 3-year follow-up period, 12 of these samples remained unchanged, 5 reversed to a normal state, and 3 progressed to a high-grade lesion. Based on this, a state evolution reference path characterized by "60% maintenance state, 25% improvement state, and 15% adjacent deterioration state" is generated.

[0061] S140. Based on the current state stage, at least one candidate evolution stage, and the state evolution reference path, determine the probability distribution of the target object entering different state stages in the future.

[0062] Among them, the probability distribution of entering different state stages refers to the set of likelihood values ​​obtained through quantitative calculation to describe the target object entering each possible state stage within a specific time range in the future. This distribution represents the relative probability of all candidate evolution stages occurring in the form of probability values, constituting the quantitative basis for the final risk assessment.

[0063] In this embodiment, the current state stage and candidate evolution stage given by the state stage prediction model can be used as a benchmark. The image features of the image to be detected are matched with the corresponding features of the state evolution reference path in the historical sample database. The evolution trend weights output by the model are combined with the frequency or confidence of each state transition in the reference path. Through normalization calculation, the probability distribution of the target object entering the maintenance state, the improved state, the adjacent deterioration state or the cross-level deterioration state in a specific period of time in the future is formed and presented in the form of percentage confidence.

[0064] For example, when the state stage prediction model determines that the target object is currently in a "low-grade lesion" and lists four candidate evolution stages: maintenance, improvement, adjacent deterioration, and cross-grade deterioration, the image features of the image to be detected can be compared with the subsequent actual trends of similar low-grade cases in the state evolution reference path. It is found that 40% of the cases in the path remain in a low-grade lesion after six months, 30% improve back to normal, 20% progress to an adjacent high-grade lesion, and 10% directly cross-grade deteriorate to early cancer. Therefore, combining the model confidence and the path statistical frequency, the probability distribution of the target object entering each state stage in the next six months is output as follows: maintenance state 40%, improvement state 30%, adjacent deterioration state 20%, and cross-grade deterioration state 10%.

[0065] This application provides a state evolution prediction method. The method involves acquiring a target object's image to be detected, wherein the image is a surface image of human tissue acquired via an endoscope. The image is then input into a state stage prediction model to determine the target object's current state stage and at least one candidate evolution stage. The state stage prediction model learns state development paths from a historical sample database through pre-training. These paths include state development relationships from a normal state to at least one abnormal state. Furthermore, based on the image features of the target object and its current state stage, similar historical samples and corresponding actual state evolution records are queried from the historical sample database to generate a state evolution reference path corresponding to the target object. Finally, based on the current state stage, at least one candidate evolution stage, and the state evolution reference path, the probability distribution of the target object entering different state stages in the future is determined. The technical solution of this application, by constructing a state stage prediction model with time-series prediction capabilities and integrating actual evolution records from a historical sample database, realizes the transformation from static judgment to dynamic deduction, enhances the prospective analysis capability of disease development trends, and provides clinically predictive quantitative decision support for the monitoring and intervention of early lesions, thereby improving the initiative and accuracy of the diagnosis and treatment process.

[0066] Example 2

[0067] Figure 3This is a schematic diagram of a state evolution prediction method provided in an embodiment of this application. Based on the foregoing embodiments, this embodiment provides a more detailed description of steps S120 and S130. For specific implementation details, please refer to the technical solution of this embodiment. Technical terms that are the same as or corresponding to those in the above embodiments will not be repeated here.

[0068] like Figure 3 As shown, the method specifically includes the following steps:

[0069] S210. Obtain the image to be detected of the target object; wherein, the image to be detected is a surface image of human tissue acquired through an endoscope.

[0070] S220. Input the image to be detected corresponding to the target object into the state stage prediction model. Based on the feature extraction module in the state stage prediction model, perform anomaly detection and lesion region segmentation processing on the image to be detected, and extract the feature vector representing the nature of the lesion in the image to be detected.

[0071] The feature extraction module refers to the computational component in the state-stage prediction model that automatically extracts feature vectors that characterize the key properties of lesions by performing anomaly detection and lesion region segmentation on the input image to be detected. A feature vector is a high-dimensional data structure that quantitatively represents the key properties and visual patterns of lesions in the image in a multi-dimensional numerical form.

[0072] In this embodiment, after the image of the target object to be detected is sent into the state stage prediction model, the feature extraction module built into the model first performs anomaly detection on the image to locate suspicious areas, and further completes the lesion area segmentation to accurately separate the suspicious areas from normal tissue. Then, multi-dimensional attributes such as color, texture, shape and deep semantics are calculated in this area to form a feature vector that can quantitatively describe the nature of the lesion.

[0073] S230. Input the feature vector into the stage classification module in the state stage prediction model to predict the current state stage of the target object; and determine at least one candidate evolution stage corresponding to the current state stage based on the state development path learned by the state stage prediction model.

[0074] Among them, the stage classification module is a sub-network in the state stage prediction model that is specifically used to determine the state of feature vectors.

[0075] In this embodiment, after receiving the feature vector representing the nature of the lesion, the stage classification module, based on its learned state development path knowledge, first classifies the feature vector into the corresponding state stage label space through mapping and discrimination logic, thereby outputting the current state stage of the target object; then, under the same state development path framework, it retrieves subsequent nodes that are directly connected or logically reachable from the current state stage, and selects at least one candidate evolution stage including the maintenance state, the improved state, the adjacent deterioration state, and the cross-level deterioration state, providing a set of evolution directions for subsequent probability distribution calculation.

[0076] S240. Determine the similarity between the first image feature of the image to be detected and the second image feature of each historical sample in the historical sample database, and obtain the image similarity result.

[0077] Here, the first image feature refers to the feature vector representing the nature of lesions obtained from the image of the target object after processing by the feature extraction module, used for subsequent similarity comparison. The second image feature refers to the feature vector stored in the historical sample database after each historical sample has been pre-processed by the feature extraction module, used for similarity comparison with the first image feature of the image to be detected. The image similarity result refers to the numerical measure used to quantify the degree of closeness between the two in the feature space.

[0078] In this embodiment, the first image feature output by the feature extraction module of the image to be detected can be compared with the second image feature corresponding to each historical sample in the historical sample database by point-to-point or embedding space distance measurement, and the comprehensive similarity between the two in terms of color, texture, shape and deep semantic level can be calculated to obtain the image similarity result in a quantitative form.

[0079] S250. Compare the current state stage with the state stage labels marked on each historical sample in the historical sample database to obtain the state stage comparison results.

[0080] Among them, the state stage comparison result refers to the matching degree assessment result obtained by matching and comparing the current state stage of the target object with the state stage labels marked by each historical sample in the historical sample database, which is used to quantify the consistency between the two in the stage of disease development.

[0081] In this embodiment, the current state stage predicted by the stage classification module can be compared spatially with the state stage labels contained in each historical sample in the historical sample database. By determining whether the two belong to the same state category or are at the same evolutionary level, a state stage comparison result reflecting the consistency of the correspondence is generated, providing a stage-level matching basis for subsequent joint image similarity result screening of target historical samples.

[0082] S260. Based on the image similarity results and the state stage comparison results, retrieve at least one target historical sample that is similar to the target medical record from the historical sample database.

[0083] Among them, the target historical sample refers to one or more historical case data records that are significantly similar to the target medical record in terms of image features and disease development stage.

[0084] In this embodiment, several historical samples that are closest to the features of the image to be detected can be retained by sorting the image similarity results. Then, samples that are consistent with the current state stage label or are in the same evolution path can be filtered out by the state stage comparison results. Finally, at least one target historical sample that simultaneously satisfies visual feature similarity and state stage matching can be extracted to construct a state evolution reference path.

[0085] S270. The actual state evolution data recorded by each target historical sample is determined as a state evolution reference path.

[0086] Specifically, the complete actual change process carried by each selected target historical sample, which consists of the initial state, a time-series image sequence, and corresponding state stage labels, can be independently extracted as a comparable evolutionary trajectory, which can be used as a reference path for state evolution for comparison and weighting in subsequent probability distribution calculations.

[0087] S280. Based on the current state stage, at least one candidate evolution stage, and state evolution reference path, determine the probability distribution of the target object entering different state stages in the future.

[0088] The technical solution of this application, when determining the current state stage and at least one candidate evolution stage of a target object, inputs the image to be detected corresponding to the target object into a state stage prediction model. Based on the feature extraction module in the state stage prediction model, anomaly detection and lesion region segmentation are performed on the image to be detected, extracting feature vectors representing the nature of lesions in the image to be detected. Then, the feature vectors are input into the stage classification module in the state stage prediction model to predict the current state stage of the target object. And, based on the state development path learned by the state stage prediction model, at least one candidate evolution stage corresponding to the current state stage is determined. The technical solution of this application, through anomaly detection and lesion region segmentation processing of the feature extraction module, achieves accurate quantitative representation of the nature of lesions. Then, through the stage classification module combined with the pre-learned state development path, it can not only accurately determine the current state stage, but also deduce a complete set of candidate evolution stages based on the pathological evolution law, thereby elevating a single image classification into a dynamic disease assessment process with temporal prediction capabilities.

[0089] The technical solution of this application, when determining the state evolution reference path corresponding to the target object, determines the similarity between the first image feature of the image to be detected and the second image feature of each historical sample in the historical sample database to obtain an image similarity result. Subsequently, it compares the current state stage with the state stage labels marked on each historical sample in the historical sample database to obtain a state stage comparison result. Thus, based on the image similarity result and the state stage comparison result, at least one target historical sample similar to the target medical record is retrieved from the historical sample database, and the actual state evolution data recorded by each target historical sample is determined as a state evolution reference path. The technical solution of this application, by comprehensively calculating the image similarity result and the state stage comparison result, accurately retrieves the target historical sample highly similar to the target medical record from the historical sample database, and transforms its actual state evolution data into a state evolution reference path. This integrates the prediction results of a single model with the historical evolution laws of the real world, significantly improving the clinical credibility and empirical support of state evolution prediction.

[0090] Example 3

[0091] Figure 4 This is a schematic diagram of a state evolution prediction method provided in an embodiment of this application. Based on the foregoing embodiments, this embodiment provides a more detailed description of step S140. For specific implementation details, please refer to the technical solution of this embodiment. Technical terms that are the same as or corresponding to those in the above embodiments will not be repeated here.

[0092] like Figure 4 As shown, the method specifically includes the following steps:

[0093] S310. Obtain the image to be detected of the target object; wherein, the image to be detected is a surface image of human tissue acquired through an endoscope.

[0094] S320. Input the image to be detected into the state stage prediction model to determine the current state stage of the target object and at least one candidate evolution stage.

[0095] Among them, the state stage prediction model learns the state development path from the historical sample database through pre-training. The state development path includes the state development relationship from the normal state to at least one abnormal stage state.

[0096] S330. Based on the image features and current state stage of the image to be detected, query similar historical samples and corresponding actual state evolution records from the historical sample database to generate a state evolution reference path corresponding to the target object.

[0097] S340. Compare at least one candidate evolution stage with the actual state evolution data recorded in the historical samples of each target, and count the frequency of occurrence of each candidate evolution stage.

[0098] The frequency of occurrence of each candidate evolution stage refers to the statistical value of the number of times the evolution results consistent with the candidate evolution stages determined by the state stage prediction model occur in the actual state evolution data recorded by each target historical sample retrieved from the historical sample database.

[0099] In this embodiment, in the target historical sample set corresponding to the state evolution reference path, the matching relationship between the subsequent actual state changes and the candidate evolution stage list is checked one by one. The number of times each candidate evolution stage appears in the subsequent trajectory of all target historical samples is accumulated, thereby quantifying the frequency of occurrence of each evolution direction in historical experience and providing a numerical basis for subsequent weight assignment.

[0100] S350. Initial weights are assigned based on the frequency of occurrence, and the confidence scores of each candidate evolution stage are weighted and corrected in conjunction with the state stage prediction model to obtain the corrected weights.

[0101] The initial weight refers to the initial importance value directly assigned based on the frequency of each candidate evolution stage in the actual state evolution data of the target historical sample, used to characterize the historical probability of the evolution path occurring. The adjusted weight refers to the optimized weight value obtained by weighting and fusing the initial weight assigned based on the frequency of occurrence with the confidence scores of each candidate evolution stage by the state stage prediction model, which can simultaneously reflect historical statistical patterns and model prediction confidence.

[0102] In this embodiment, the frequency of each candidate evolution stage obtained from the statistical analysis of the target historical samples can be converted into initial weights. Then, the confidence score output by the state stage prediction model for the same candidate evolution stage is used as the model side confidence score. The initial weights are weighted and fused so that the final corrected weights reflect both the frequency of historical experience and the confidence of model prediction, thereby improving the accuracy and stability of the subsequent probability distribution.

[0103] S360. Normalize the corrected weights to obtain the probability distribution of the target object entering each state stage at different future time points.

[0104] Specifically, the weights obtained after weighted correction of all candidate evolution stages can be uniformly scaled by a total of 100%, so that the weight value of each stage can be directly converted into the probability of occurrence, thereby outputting the probability distribution of the target object entering the maintenance state, the improved state, the adjacent deterioration state, or the cross-level deterioration state at subsequent time points.

[0105] The technical solution of this application, when determining the probability distribution of a target object entering different state stages in the future, specifically includes: comparing at least one candidate evolution stage with the actual state evolution data recorded in the historical samples of each target, statistically analyzing the occurrence frequency of each candidate evolution stage, assigning initial weights based on the occurrence frequency, and then weighting and correcting the confidence scores of each candidate evolution stage in conjunction with the state stage prediction model to obtain corrected weights. Finally, the corrected weights are normalized to obtain the probability distribution of the target object entering each state stage at different future time points. This technical solution establishes initial weights by statistically analyzing the occurrence frequency of candidate evolution stages in historical actual data, then weights and corrects them in conjunction with the confidence scores of the state stage prediction model itself, and finally transforms the corrected weights into a standardized probability distribution through normalization. This effectively integrates historical experience data with model prediction confidence, significantly improving the accuracy and clinical interpretability of probability distribution calculation.

[0106] Based on the above embodiments, the method may optionally further include: generating personalized follow-up and treatment recommendations based on the current state stage, at least one candidate evolution stage, state evolution reference path, and probability distribution; and generating a state reasoning report file containing text explanations and visual charts based on the current state stage, at least one candidate evolution stage, state evolution reference path, and follow-up and treatment recommendations.

[0107] Among them, follow-up and treatment recommendations refer to personalized guidance plans generated for target subjects based on a comprehensive analysis of the current state stage, candidate evolution stage, state evolution reference path and probability distribution, regarding the frequency of subsequent medical observation, the schedule of follow-up examinations and targeted treatment plans.

[0108] Among them, the state reasoning report refers to a comprehensive disease assessment and prediction document that is generated after systematic integration and analysis based on the current state stage, candidate evolution stages, state evolution reference path, and follow-up and treatment recommendations. It includes both text descriptions and graphical displays.

[0109] In this embodiment, based on the current state stage, and combined with at least one candidate evolution stage, state evolution reference path, and its corresponding probability distribution, the follow-up interval, examination items, and intervention measures are automatically matched according to the risk level and time urgency to form personalized follow-up and treatment recommendations. Subsequently, the current state stage, candidate evolution stages, state evolution reference path, and follow-up and treatment recommendations are input into the report generation engine. Through natural language templates and visualization components, a state reasoning report file containing text explanations and visualization charts is output, realizing the integrated presentation of diagnosis, prediction, decision-making basis, and action guidance. This achieves seamless connection from state recognition and evolution prediction to clinical guidance, reduces the information integration burden on doctors, lowers the risk of misdiagnosis and missed diagnosis, and supports patient education and consultation communication, thereby improving the overall efficiency and quality of diagnosis and treatment.

[0110] For example, Figure 5 As shown in Figure 5, the implementation process of the state evolution prediction method involved in the embodiments of the invention is illustrated. This technical solution uses an image sign recognition and feature extraction module to segment lesion regions and extract feature vectors from the input endoscopic image; a disease stage modeling module determines the current state stage and candidate evolution stages based on the pre-learned state development path; a case similarity reasoning module retrieves similar cases from the historical database and generates a state evolution reference path; a multi-path risk prediction module integrates model confidence and historical statistical patterns to calculate the temporal probability distribution of each future state; and finally, a structured report generation module integrates all analysis results and outputs a structured diagnostic report containing visualization charts, text explanations, and personalized follow-up suggestions.

[0111] Example 4

[0112] Figure 6 This is a schematic diagram of a state evolution prediction device provided in an embodiment of this application. The device includes:

[0113] The data acquisition module 410 is used to acquire the image to be detected of the target object; wherein the image to be detected is a surface image of human tissue acquired by an endoscope.

[0114] The state stage determination module 420 is used to input the image to be detected into the state stage prediction model to determine the current state stage of the target object and at least one candidate evolution stage; wherein, the state stage prediction model learns the state development path from the historical sample database through pre-training, and the state development path includes the state development relationship from the normal state to at least one abnormal stage state.

[0115] The evolution path reference module 430 is used to query similar historical samples and corresponding actual state evolution records from the historical sample database based on the image features of the image to be detected and the current state stage, so as to generate a state evolution reference path corresponding to the target object.

[0116] The state evolution prediction module 440 is used to determine the probability distribution of the target object entering different state stages in the future based on the current state stage, the at least one candidate evolution stage, and the state evolution reference path.

[0117] This application provides a state evolution prediction device. When applied, the device acquires a target object image, which is a surface image of human tissue obtained through an endoscope. The image is then input into a state stage prediction model to determine the target object's current state stage and at least one candidate evolution stage. The state stage prediction model learns state development paths from a historical sample database through pre-training. These paths include state development relationships from a normal state to at least one abnormal state. Furthermore, based on the image features of the target image and the current state stage, similar historical samples and corresponding actual state evolution records are queried from the historical sample database to generate a state evolution reference path corresponding to the target object. Finally, based on the current state stage, at least one candidate evolution stage, and the state evolution reference path, the probability distribution of the target object entering different state stages in the future is determined. The technical solution of this application, by constructing a state stage prediction model with time-series prediction capabilities and integrating actual evolution records from a historical sample database, realizes the transformation from static judgment to dynamic deduction, enhances the prospective analysis capability of disease development trends, and provides clinically predictive quantitative decision support for the monitoring and intervention of early lesions, thereby improving the initiative and accuracy of the diagnosis and treatment process.

[0118] Based on the above-described apparatus, optionally, the at least one candidate evolution stage includes a maintenance state, an improved state, an adjacent deterioration state, and a cross-level deterioration state relative to the current state stage.

[0119] Optionally, based on the above-mentioned device, the state evolution prediction device further includes: a state prediction model training module, used to construct a historical sample database, the historical sample database including multiple historical sample data, each historical sample data including an endoscopic image sequence from an initial state to a final state, and a clinically diagnosed state stage label corresponding to each image; based on the endoscopic image sequence and its corresponding state stage label in each of the historical sample data, a temporal state development path sample set is constructed; and the state stage prediction model to be trained is trained on the temporal modeling neural network structure based on the temporal state development path sample set to obtain the state stage prediction model.

[0120] Based on the above-mentioned device, optionally, the status stage label includes at least one of the following stages: normal state, low-grade lesion, high-grade lesion, early cancer, and invasive cancer.

[0121] Based on the above-mentioned device, optionally, the state stage determination module 420 includes:

[0122] The feature vector determination unit is used to input the image to be detected corresponding to the target object into the state stage prediction model, and perform anomaly detection and lesion region segmentation processing on the image to be detected based on the feature extraction module in the state stage prediction model, and extract the feature vectors representing the nature of the lesions in the image to be detected.

[0123] The state stage determination unit is used to input the feature vector into the stage classification module in the state stage prediction model to predict the current state stage of the target object; and to determine at least one candidate evolution stage corresponding to the current state stage based on the state development path learned by the state stage prediction model.

[0124] Based on the above-mentioned device, optionally, the evolution path reference module 430 includes:

[0125] A similarity determination unit is used to determine the similarity between the first image feature of the image to be detected and the second image feature of each historical sample in the historical sample database, and to obtain an image similarity result;

[0126] The state stage comparison unit is used to compare the current state stage with the state stage labels marked on each of the historical samples in the historical sample database to obtain the state stage comparison result.

[0127] A similar historical sample determination unit is used to retrieve at least one target historical sample that is similar to the target medical record from the historical sample database based on the image similarity result and the state stage comparison result.

[0128] An evolution path reference unit is used to determine the actual state evolution data recorded by each of the target historical samples as a state evolution reference path.

[0129] Based on the above-mentioned device, optionally, the state evolution prediction module 440 includes:

[0130] The frequency determination unit is used to compare the at least one candidate evolution stage with the actual state evolution data recorded by each target historical sample, and to count the frequency of occurrence of each candidate evolution stage.

[0131] The corrected weight determination unit is used to assign initial weights based on the frequency of occurrence, and to perform weighted correction on the confidence scores of each candidate evolution stage in combination with the state stage prediction model to obtain corrected weights.

[0132] The probability distribution determination unit is used to normalize the modified weights to obtain the probability distribution of the target object entering each state stage at different future time points.

[0133] Optionally, based on the above-mentioned device, the state evolution prediction device further includes: a structured report production module, used to generate personalized follow-up and treatment recommendations based on the current state stage, the at least one candidate evolution stage, the state evolution reference path, and the probability distribution; and to generate a state reasoning report file containing text explanations and visual charts based on the current state stage, the at least one candidate evolution stage, the state evolution reference path, and the follow-up and treatment recommendations.

[0134] The state evolution prediction device provided in this application embodiment can execute the state evolution prediction method provided in any embodiment of this application, and has the corresponding functional modules and beneficial effects of the execution method.

[0135] It is worth noting that the various units and modules included in the above system are only divided according to functional logic, but are not limited to the above division, as long as the corresponding functions can be achieved; in addition, the specific names of each functional unit are only for easy differentiation and are not used to limit the protection scope of the embodiments of this application.

[0136] Example 5

[0137] Figure 7 This is a schematic diagram of the structure of a controller provided in an embodiment of this application. Figure 7 A block diagram is shown of an exemplary controller 50 suitable for implementing embodiments of the present application. Figure 7 The controller 50 shown is merely an example and should not impose any limitations on the functionality and scope of use of the embodiments of this application.

[0138] likeFigure 7 As shown, the controller 50 is presented in the form of a general-purpose computing device. The components of the controller 50 may include, but are not limited to: one or more processors or processing units 501, system memory 502, and bus 503 connecting different system components (including system memory 502 and processing unit 501).

[0139] Bus 503 represents one or more of several bus architectures, including a memory bus or memory controller, a peripheral bus, a graphics acceleration port, a processor, or a local bus using any of the various bus architectures. For example, these architectures include, but are not limited to, the Industry Standard Architecture (ISA) bus, the Micro Channel Architecture (MAC) bus, the Enhanced ISA bus, the Video Electronics Standards Association (VESA) local bus, and the Peripheral Component Interconnect (PCI) bus.

[0140] The controller 50 typically includes a variety of computer system readable media. These media can be any available media that can be accessed by the controller 50, including volatile and non-volatile media, and removable and non-removable media.

[0141] System memory 502 may include computer system readable media in the form of volatile memory, such as random access memory (RAM) 504 and / or cache memory 505. Controller 50 may further include other removable / non-removable, volatile / non-volatile computer system storage media. By way of example only, storage system 506 may be used to read and write non-removable, non-volatile magnetic media (…). Figure 7 Not shown; usually referred to as a "hard drive"). Although Figure 7 As not shown, a disk drive for reading and writing to a removable non-volatile disk (e.g., a "floppy disk") and an optical disk drive for reading and writing to a removable non-volatile optical disk (e.g., a CD-ROM, DVD-ROM, or other optical media) may be provided. In these cases, each drive may be connected to bus 503 via one or more data media interfaces. Memory 502 may include at least one program product having a set (e.g., at least one) of program modules configured to perform the functions of the embodiments of this application.

[0142] A program / utility 508 having a set (at least one) of program modules 507 may be stored, for example, in memory 502. Such program modules 507 include, but are not limited to, an operating system, one or more application programs, other program modules, and program data. Each or some combination of these examples may include an implementation of a network environment. Program modules 507 typically perform the functions and / or methods described in the embodiments of this application.

[0143] The controller 50 can also communicate with one or more external devices 509 (e.g., keyboard, pointing device, display 510, etc.), and with one or more devices that enable a user to interact with the controller 50, and / or with any device that enables the controller 50 to communicate with one or more other computing devices (e.g., network card, modem, etc.). This communication can be performed via input / output (I / O) interface 511. Furthermore, the controller 50 can also communicate with one or more networks (e.g., local area network (LAN), wide area network (WAN), and / or public networks, such as the Internet) via network adapter 512. As shown, network adapter 512 communicates with other modules of the controller 50 via bus 503. It should be understood that, although... Figure 7 As not shown, other hardware and / or software modules can be used in conjunction with controller 50, including but not limited to: microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives, and data backup storage systems.

[0144] The processing unit 501 executes various functional applications and page processing by running programs stored in the system memory 502, such as implementing the state evolution prediction method provided in the embodiments of this application.

[0145] Example 6

[0146] This application also provides a storage medium containing computer-executable instructions, which, when executed by a computer processor, are used to perform a state evolution prediction method, the method comprising:

[0147] Acquire an image of the target object to be detected; wherein, the image to be detected is a surface image of human tissue acquired through an endoscope;

[0148] The image to be detected is input into the state stage prediction model to determine the current state stage and at least one candidate evolution stage of the target object; wherein, the state stage prediction model learns the state development path from the historical sample database through pre-training, and the state development path includes the state development relationship from the normal state to at least one abnormal stage state.

[0149] Based on the image features of the image to be detected and the current state stage, similar historical samples and corresponding actual state evolution records are queried from the historical sample database to generate a state evolution reference path corresponding to the target object.

[0150] Based on the current state stage, the at least one candidate evolution stage, and the state evolution reference path, the probability distribution of the target object entering different state stages in the future is determined.

[0151] The computer storage medium in this application embodiment can be any combination of one or more computer-readable media. A computer-readable medium can be a computer-readable signal medium or a computer-readable storage medium. A computer-readable storage medium can be, for example,—but not limited to—an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of computer-readable storage media (a non-exhaustive list) include: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In this document, a computer-readable storage medium can be any tangible medium that contains or stores a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.

[0152] Computer-readable signal media may include data signals propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals may take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. Computer-readable signal media may also be any computer-readable medium other than computer-readable storage media, capable of sending, propagating, or transmitting programs for use by or in connection with an instruction execution system, apparatus, or device.

[0153] Program code contained on a computer-readable medium may be transmitted using any suitable medium, including—but not limited to—wireless, wire, optical fiber, RF, etc., or any suitable combination thereof.

[0154] Computer program code for performing the operations of the embodiments of this application can be written in one or more programming languages ​​or a combination thereof. Programming languages ​​include object-oriented programming languages—such as Java, Smalltalk, and C++—and conventional procedural programming languages—such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0155] Note that the above description is merely a preferred embodiment and the technical principles employed in this application. Those skilled in the art will understand that this application is not limited to the specific embodiments described herein, and various obvious changes, readjustments, and substitutions can be made without departing from the scope of protection of this application. Therefore, although this application has been described in detail through the above embodiments, this application is not limited to the above embodiments. Many other equivalent embodiments may be included without departing from the concept of this application, and the scope of this application is determined by the scope of the appended claims.

Claims

1. A method for predicting state evolution, characterized in that, The method includes: Acquire an image of the target object to be detected; wherein, the image to be detected is a surface image of human tissue acquired through an endoscope; The image to be detected is input into a state stage prediction model to determine the current state stage and at least one candidate evolution stage of the target object; wherein, the state stage prediction model learns the state development path from a historical sample database through pre-training, and the state development path includes the state development relationship from a normal state to at least one abnormal stage state. Based on the image features of the image to be detected and the current state stage, similar historical samples and corresponding actual state evolution records are queried from the historical sample database to generate a state evolution reference path corresponding to the target object. Based on the current state stage, the at least one candidate evolution stage, and the state evolution reference path, the probability distribution of the target object entering different state stages in the future is determined.

2. The method according to claim 1, characterized in that, The at least one candidate evolutionary stage includes a maintenance state, an improved state, an adjacent deterioration state, and a cross-level deterioration state relative to the current state stage.

3. The method according to claim 1, characterized in that, The process of determining the state stage prediction model includes: Construct a historical sample database, which includes multiple historical sample data, each of which includes an endoscopic image sequence from the initial state to the final state and a clinically diagnosed state stage label corresponding to each image; Based on the endoscopic image sequences and their corresponding state stage labels in each of the historical sample data, a temporal state development path sample set is constructed. The state stage prediction model is trained based on the sample set of temporal state development paths to obtain the state stage prediction model of the temporal modeling neural network structure.

4. The method according to claim 3, characterized in that, The status stage label includes at least one of the following stages: normal state, low-grade lesion, high-grade lesion, early cancer, and invasive cancer.

5. The method according to claim 1, characterized in that, The step of inputting the image to be detected into a state stage prediction model to determine the current state stage of the target object and at least one candidate evolution stage includes: The image to be detected corresponding to the target object is input into the state stage prediction model. Based on the feature extraction module in the state stage prediction model, the image to be detected is subjected to anomaly detection and lesion region segmentation, and feature vectors representing the nature of lesions in the image to be detected are extracted. The feature vector is input into the stage classification module in the state stage prediction model to predict the current state stage of the target object; and, based on the state development path learned by the state stage prediction model, at least one candidate evolution stage corresponding to the current state stage is determined.

6. The method according to claim 1, characterized in that, The step of querying similar historical samples and corresponding actual state evolution records from the historical sample database based on the image features of the image to be detected and the current state stage to generate a state evolution reference path corresponding to the target object includes: The similarity between the first image feature of the image to be detected and the second image feature of each historical sample in the historical sample database is determined to obtain the image similarity result; By comparing the current state stage with the state stage labels marked on each historical sample in the historical sample database, the state stage comparison result is obtained; Based on the image similarity results and the state stage comparison results, at least one target historical sample similar to the target medical record is retrieved from the historical sample database; The actual state evolution data recorded by each of the target historical samples are respectively determined as a state evolution reference path.

7. The method according to claim 1, characterized in that, The step of determining the probability distribution of the target object entering different state stages in the future based on the current state stage, the at least one candidate evolution stage, and the state evolution reference path includes: The at least one candidate evolution stage is compared with the actual state evolution data recorded in each of the target historical samples, and the frequency of occurrence of each candidate evolution stage is counted. Initial weights are assigned based on the frequency of occurrence, and the confidence scores of each candidate evolution stage are weighted and corrected in conjunction with the state stage prediction model to obtain corrected weights. The corrected weights are normalized to obtain the probability distribution of the target object entering each state stage at different future time points.

8. The method according to claim 1, characterized in that, The method further includes: Based on the current state stage, the at least one candidate evolution stage, the state evolution reference path, and the probability distribution, personalized follow-up and treatment recommendations are generated. Based on the current state stage, the at least one candidate evolution stage, the state evolution reference path, and the follow-up and treatment recommendations, a state reasoning report file containing text explanations and visual charts is generated.

9. A state evolution prediction device, characterized in that, The device includes: The data acquisition module is used to acquire the image to be detected of the target object; wherein, the image to be detected is a surface image of human tissue acquired through an endoscope; The state stage determination module is used to input the image to be detected into the state stage prediction model to determine the current state stage of the target object and at least one candidate evolution stage; wherein, the state stage prediction model learns the state development path from the historical sample database through pre-training, and the state development path includes the state development relationship from the normal state to at least one abnormal stage state. An evolution path reference module is used to query similar historical samples and corresponding actual state evolution records from the historical sample database based on the image features of the image to be detected and the current state stage, so as to generate a state evolution reference path corresponding to the target object. The state evolution prediction module is used to determine the probability distribution of the target object entering different state stages in the future based on the current state stage, the at least one candidate evolution stage, and the state evolution reference path.

10. An electronic device, characterized in that, The electronic device includes: At least one processor; and A memory communicatively connected to the at least one processor; wherein, The memory stores a computer program that can be executed by the at least one processor, the computer program being executed by the at least one processor to enable the at least one processor to perform the state evolution prediction method according to any one of claims 1-8.

Images