Eye socket image diagnosis report generation model construction method and report generation method

Abstract

The invention relates to the technical field of medical data, and discloses an orbit image diagnosis report generation model construction method and a report generation method, and the orbit image diagnosis report generation model construction method comprises the steps: obtaining a knowledge graph of orbit image diagnosis; obtaining each target edge in the knowledge graph, two target entities corresponding to the target edge and a corresponding triple; the target edge indicates a target relationship between two entities corresponding to the two target edges; covering one of the two target entities and the target relationship corresponding to the triple; obtaining a target training sample based on the covered triple; correspondingly, obtaining a target training label based on the uncovered triple; constructing an orbit image diagnosis report generation model; and training the constructed orbit image diagnosis report generation model by using the target training sample and the corresponding target training label. According to the invention, the accuracy and normalization of the final orbit image diagnosis report can be improved.

Description

Technical Field

[0001] The present invention relates to the technical field of medical data processing, and in particular to a method for constructing a model for generating an orbital image diagnosis report and a method for generating a report. Background Art

[0002] Currently, imaging reports for orbital lesions are primarily text-based. The writing habits of different reporting physicians can lead to discrepancies in wording, consuming significant time and effort to assess report quality and review content. This hinders case analysis and statistical analysis, as well as multicenter data aggregation and evaluation. Furthermore, the orbital anatomy is delicate and complex, adjacent to structures such as the sinuses, sella turcica, and skull base. Some orbital lesions can extensively involve adjacent foramina and spaces, requiring extensive diagnostic descriptions and assessments. Junior physicians are prone to omissions and misreporting, increasing the workload and review difficulty for senior physicians. Summary of the Invention

[0003] In view of this, the present invention provides a method for constructing a model for generating an orbital imaging diagnosis report and a method for generating a report, so as to solve the problem of poor quality of orbital imaging diagnosis reports.

[0004] In a first aspect, the present invention provides a method for constructing a model for generating an orbital imaging diagnostic report, the method comprising:

[0005] Obtaining a knowledge graph for orbital imaging diagnosis, wherein entities corresponding to nodes in the knowledge graph include: one or more of anatomical site, pathological features, disease considerations, and recommended examination methods;

[0006] Obtaining each target edge in the knowledge graph and the triples corresponding to the two target entities corresponding to the target edge; the target edge indicates a target relationship between the two entities corresponding to the two target edges;

[0007] Covering one of the two target entities and the target relationship corresponding to the triple;

[0008] Based on the masked triples, a target training sample is obtained; correspondingly, based on the uncovered triples, a target training label is obtained;

[0009] Construct a model for generating orbital imaging diagnostic reports;

[0010] The constructed orbital image diagnosis report generation model is trained using the target training samples and the corresponding target training labels.

[0011] In an optional embodiment, after training the constructed orbital imaging diagnosis report generation model using the target training samples and the corresponding target training labels, the method further includes:

[0012] For the network layer in the orbital imaging diagnostic report generation model, create a low-order matrix, and add the low-order matrix to the corresponding network layer weight;

[0013] Obtain the first fine-tuning training sample;

[0014] The first fine-tuning training sample is used to fine-tune the orbital imaging diagnosis report generation model, and only the parameters of the low-order matrix are updated during the fine-tuning training process.

[0015] In an optional embodiment, after training the constructed orbital imaging diagnosis report generation model using the training samples and the corresponding training labels, the method further includes:

[0016] Obtaining a first orbital imaging diagnostic report, where the first orbital imaging diagnostic report is an orbital imaging diagnostic report that is screened according to a preset rule and is determined to need to be rewritten;

[0017] Rewriting the first orbital image diagnosis report using the orbital image diagnosis report generation model to obtain a second orbital image diagnosis report;

[0018] Obtaining a quality assessment result of the second orbital imaging diagnostic report;

[0019] Training a reward model based on the second orbital imaging diagnosis report and the corresponding quality assessment result;

[0020] Using the reward model, a quality assessment is performed on the orbital imaging diagnosis report output by the orbital imaging diagnosis report generation model based on the second fine-tuning training sample, and a corresponding reward is output;

[0021] Based on the rewards output by the reward model, the proximal strategy optimization method is used to fine-tune and optimize the orbital imaging diagnosis report generation model.

[0022] In an optional embodiment, obtaining the quality assessment result of the second orbital imaging diagnostic report includes:

[0023] Obtaining a quality assessment model for orbital imaging diagnostic reports, where positive and negative samples used to train the quality assessment model are manually screened orbital imaging diagnostic reports or orbital imaging diagnostic reports classified according to the seniority of the doctors who wrote the reports;

[0024] The quality assessment model is used to perform quality assessment on the second orbital imaging diagnosis report.

[0025] In an optional embodiment, obtaining a knowledge graph for orbital imaging diagnosis includes:

[0026] Acquiring medical knowledge data related to orbital imaging diagnosis, wherein the medical knowledge data includes data in an expert knowledge base of orbital imaging reports;

[0027] Extracting entities and relationships between entities from the medical knowledge data using a general large model;

[0028] The knowledge graph of orbital imaging diagnosis is constructed based on the entities extracted from the general large model and the relationships between the entities.

[0029] In an optional embodiment, the method for constructing a model for generating an orbital imaging diagnostic report further includes:

[0030] Obtain third orbital imaging diagnostic report;

[0031] Rewriting the third orbital image diagnosis report using an orbital image diagnosis report generation model to obtain a fourth orbital image diagnosis report;

[0032] Obtaining a fifth orbital imaging diagnostic report obtained by manually modifying the fourth orbital imaging diagnostic report;

[0033] The third orbital image diagnosis report is used as an optimization training sample, and the fifth orbital image diagnosis report is used as an optimization training label to perform optimization training on the orbital image diagnosis report generation model.

[0034] In a second aspect, the present invention provides a method for generating an orbital imaging diagnostic report, comprising:

[0035] Obtain the sixth orbital imaging diagnostic report;

[0036] The sixth orbital image diagnostic report is rewritten using the orbital image diagnostic report generation model to obtain a seventh orbital image diagnostic report; wherein, the orbital image diagnostic report generation model is constructed using the orbital image diagnostic report generation model construction method described in the first aspect or any corresponding embodiment thereof.

[0037] In an optional embodiment, the method for generating an orbital imaging diagnosis report further includes:

[0038] Comparing the sixth orbital imaging diagnostic report and the seventh orbital imaging diagnostic report to generate modification suggestions; and / or,

[0039] Output the review result of the sixth orbital imaging diagnostic report.

[0040] In a third aspect, the present invention provides a device for constructing a model for generating an orbital imaging diagnostic report, the device comprising:

[0041] A knowledge graph acquisition module is used to obtain a knowledge graph for orbital imaging diagnosis, wherein the entities corresponding to the nodes in the knowledge graph include: one or more of anatomical parts, pathological characteristics, disease considerations, and recommended examination methods;

[0042] A triple acquisition module is configured to acquire a triple corresponding to each target edge in the knowledge graph and two target entities corresponding to the target edge; the target edge indicates a target relationship between the two entities corresponding to the two target edges;

[0043] A covering module, configured to cover one of the two target entities and the target relationship corresponding to the triple;

[0044] A target sample construction module is used to obtain a target training sample based on the masked triples; correspondingly, to obtain a target training label based on the uncovered triples;

[0045] A model building module is used to build a model for generating orbital imaging diagnostic reports;

[0046] The model training module is used to train the constructed orbital image diagnosis report generation model using the target training samples and the corresponding target training labels.

[0047] In a fourth aspect, the present invention provides a device for generating an orbital imaging diagnostic report, the device comprising:

[0048] A report acquisition module is used to obtain the sixth orbital imaging diagnosis report;

[0049] A rewriting module is used to rewrite the sixth orbital image diagnosis report using the orbital image diagnosis report generation model to obtain a seventh orbital image diagnosis report; wherein, the orbital image diagnosis report generation model is constructed using the orbital image diagnosis report generation model construction method described in the first aspect or any corresponding embodiment thereof.

[0050] In a fifth aspect, the present invention provides a computer device comprising: a memory and a processor, the memory and the processor being communicatively connected to each other, computer instructions stored in the memory, and the processor executing the computer instructions to thereby execute the orbital imaging diagnostic report generation model construction method of the above-mentioned first aspect or any corresponding embodiment thereof, or execute the orbital imaging diagnostic report generation method of the above-mentioned second aspect or any corresponding embodiment thereof.

[0051] In a sixth aspect, the present invention provides a computer-readable storage medium having computer instructions stored thereon, the computer instructions being used to enable a computer to execute the orbital imaging diagnostic report generation model construction method of the above-mentioned first aspect or any corresponding embodiment thereof, or to execute the orbital imaging diagnostic report generation method of the above-mentioned second aspect or any corresponding embodiment thereof.

[0052] In the seventh aspect, the present invention provides a computer program product comprising computer instructions, the computer instructions being used to enable a computer to execute the orbital imaging diagnostic report generation model construction method of the above-mentioned first aspect or any corresponding embodiment thereof, or to execute the orbital imaging diagnostic report generation method of the above-mentioned second aspect or any corresponding embodiment thereof.

[0053] This embodiment provides a method for constructing an orbital imaging diagnosis report generation model that collaborates with a large language model and a knowledge graph, as well as a corresponding report generation method. The constructed orbital imaging diagnosis report generation model can rewrite low-quality orbital imaging diagnosis reports, such as orbital imaging diagnosis reports written by junior doctors, to improve the accuracy and standardization of the final orbital imaging diagnosis report. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] In order to more clearly illustrate the specific embodiments of the present invention or the technical solutions in related technologies, the following briefly introduces the drawings required for use in the specific embodiments or related technical descriptions. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0055] Figure 1 1 is a flow chart of a method for constructing a model for generating an orbital imaging diagnosis report according to an embodiment of the present invention;

[0056] Figure 2 is a flow chart of a method for generating an orbital imaging diagnosis report according to an embodiment of the present invention;

[0057] Figure 3 is a structural block diagram of a device for constructing a model for generating an orbital imaging diagnosis report according to an embodiment of the present invention;

[0058] Figure 4 is a structural block diagram of a device for generating an orbital image diagnosis report according to an embodiment of the present invention;

[0059] Figure 5 Schematic diagram of the hardware structure of a computer device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0060] To make the purpose, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without making creative efforts shall fall within the scope of protection of the present invention.

[0061] According to an embodiment of the present invention, an embodiment of a method for constructing a model for generating an orbital imaging diagnostic report is provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of executable computer instructions, and although a logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0062] In this embodiment, a method for constructing a model for generating an orbital imaging diagnosis report is provided, which can be used in various computer devices, such as medical equipment, Figure 1 FIG. 1 is a flow chart of a method for constructing a model for generating an orbital imaging diagnosis report according to an embodiment of the present invention. Figure 1 As shown, the process includes the following steps:

[0063] Step S101: Obtain a knowledge graph of orbital imaging diagnosis, where the entities corresponding to the nodes in the knowledge graph include: one or more of anatomical parts, pathological characteristics, disease considerations, and recommended examination methods.

[0064] Specifically, each edge of the knowledge graph represents the relationship between different entities, including: the relationship between anatomical sites and pathological features, the involvement relationship (indicating the anatomical sites affected by the pathological features), the relationship between pathological features and disease considerations, and the relationship between examination recommendations and pathological features / disease considerations. Orbital images include magnetic resonance imaging (MRI) and computed tomography (CT) images.

[0065] In some optional specific implementations, step S101, i.e., obtaining a knowledge graph for orbital imaging diagnosis, includes:

[0066] Step S1011 , obtaining medical knowledge data related to orbital imaging diagnosis, wherein the medical knowledge data includes data in an expert knowledge base of orbital imaging reports.

[0067] If the medical knowledge data related to orbital imaging diagnosis is unstructured, it can be structured first. In addition, the professional terminology of orbital imaging diagnosis can be standardized through consensus on imaging diagnosis of various ophthalmic diseases, anatomical atlases, and textbooks.

[0068] In addition, medical knowledge data can also include clinical orbital imaging reports. Specifically, a large number of clinical orbital imaging reports can be collected first, and then a classifier can be used to perform quality screening on the collected clinical orbital imaging reports (mainly to evaluate the quality of the report from the perspective of standardization and accuracy). The classifier can be trained in advance. During the training process, the orbital imaging reports that have been manually reviewed by senior doctors are used as positive samples, and the reports submitted for the first time by junior doctors and that are significantly different from the final review results are used as negative samples. The classifier can also be used for quality screening of various sample data used in this application for training, fine-tuning, and optimizing the orbital imaging diagnostic report generation model.

[0069] Step S1012: Using a general large model, entities and relationships in the medical knowledge data are extracted, and relationships between entities are extracted.

[0070] For example, you can input a prompt word into the general large model: Please extract the entities in the following description: "Bone destruction of the right frontal bone and upper outer wall of the orbit accompanied by a huge soft tissue mass, involving the right orbit, right anterior ethmoid sinus, and right frontal intracranial region, diffuse multiple bone destruction of the skull, considering malignant lesions, PET-CT and systemic examination recommended", and the general large model will return the extracted entities and types:

[0071] Anatomical location: right frontal bone;

[0072] Anatomical location: upper outer wall of the orbit;

[0073] Pathological features: bone destruction;

[0074] Pathological features: soft tissue mass;

[0075] Anatomical location: right orbit;

[0076] Anatomical location: right anterior ethmoid sinus;

[0077] Anatomical location: right frontal intracranial;

[0078] Pathological features: diffuse multiple bone destruction of the skull;

[0079] Disease considerations: malignant lesions;

[0080] Recommended examination: PET-CT;

[0081] Examination recommendations: full body system examination.

[0082] You can also enter the prompt word "Extract the relationship between the above entities" into the general model. The general model will return the relationship between the extracted entities:

[0083] (1) Relationship between anatomical site and pathological characteristics:

[0084] Right frontal bone -> bone destruction;

[0085] Upper outer wall of orbit -> bone destruction;

[0086] Upper outer orbital wall -> soft tissue mass;

[0087] (2) Involvement relationship (indicating the anatomical sites affected by the pathological characteristics):

[0088] Soft tissue mass -> involving -> right orbit;

[0089] Soft tissue mass -> involvement -> right anterior ethmoid sinus;

[0090] Soft tissue mass -> involvement -> right frontal intracranial area;

[0091] (3) Relationship between pathological characteristics and disease considerations:

[0092] Bone destruction -> Consider -> Malignant lesions;

[0093] Soft tissue mass -> Consider -> malignant lesion;

[0094] (4) Relationship between examination recommendations and pathological features / disease considerations:

[0095] Malignant lesions -> Recommendations -> PET-CT;

[0096] Malignant lesions -> Recommendations -> Full body system examination.

[0097] These relationships reveal the connections between anatomical sites, pathologic features, disease considerations, and recommended investigations.

[0098] Step S1013: constructing a knowledge graph for orbital imaging diagnosis based on the entities and relationships between the entities extracted from the general large model. Constructing a knowledge graph means representing entities and relationships in the form of a graph.

[0099] In the embodiment of the present invention, the knowledge graph for orbital imaging diagnosis is constructed by representing the entities (such as disease names, symptoms, treatment plans, etc.) and relationships (such as the association between diseases and symptoms, the correspondence between diseases and treatment plans, etc.) in the structured knowledge base in the form of a graph. The specific knowledge graph construction process can be, for example:

[0100] First, the specialized terminology for orbital imaging diagnosis can be standardized using consensus on imaging diagnosis of various ophthalmic diseases, anatomical atlases, and textbooks. Entity normalization is the process of standardizing the entities extracted from medical knowledge text using a large language model (LLM) into a unified form. First, based on a framework for entity attribute similarity, a computer-generated entity alignment is performed by defining similarity metrics and combinations. Finally, through expert calibration and review, standardized, unified nomenclature is achieved.

[0101] After entities are extracted and normalized from the medical knowledge data, large-scale models and network topology methods can be used to extract entity relationships (e.g., "bone" vs. "destruction") and attributes (e.g., "CT" vs. "hyperdensity," "enhancement" vs. "marked enhancement," and "progressive enhancement" vs. "cavernous hemangioma"). The specific process for extracting entity relationships and attributes using large-scale models and network topology methods is as follows: 1. Initial network construction: Entities in the medical knowledge data text are used as nodes, and relationships between entities as edges to construct an initial network topology. Edge weights can be set based on factors such as relationship strength and frequency of occurrence. For example, if the entities "diffuse multiple bone destruction of the skull" and "malignant lesion" appear multiple times in the text and have a clear relationship description, the edge weight between them can be set higher. 2. Network feature extraction: Various features are extracted from the constructed network, such as node degree, betweenness centrality, and clustering coefficient, as well as edge features such as weight and number of shared neighbors. These features can help better understand the network structure and the relationships between entities. 3. Relationship Prediction and Optimization: Utilize extracted network features, combined with machine learning algorithms such as support vector machines or random forests, to predict and optimize unknown relationships. Graph neural networks can also be used to learn and reason directly on the network structure to more accurately extract entity relationships.

[0102] Finally, an "expert knowledge base for orbital imaging reports" is established, and this knowledge base is fused with the knowledge extracted from the cleaned clinical orbital imaging report data (specific methods include synonym fusion, dictionary matching, and entity modification). Specifically, the process of result fusion and verification can be as follows: the relationship extraction results obtained based on the large language model and network topology method are fused, and duplicate and conflicting relationships are removed to obtain the final entity relationship extraction results. The results are then verified and evaluated through manual annotation, and the knowledge graph structure and methods are continuously optimized. This improves the accuracy of the entities and relationships extracted through the LLM, reduces errors and conflicts, and enhances the integrity and richness of the relationships. The knowledge graph constructed by experts is simple and concise, but difficult to generalize and expand.

[0103] The embodiment of the present invention uses a general large model to assist in the construction of the knowledge graph, which can improve the efficiency of knowledge graph construction and reduce costs. In addition, in other embodiments, the orbital imaging diagnosis report generation model obtained after knowledge graph enhanced model training and fine-tuning can be used to extract entities of medical knowledge data and the relationships between entities. For example, for subsequent newly added medical knowledge data such as guideline consensus and imaging reports, the orbital imaging diagnosis report generation model obtained after a round of knowledge graph enhanced model training and fine-tuning can be used to extract entities and the relationships between entities, thereby updating the knowledge graph.

[0104] Step S102, obtaining each target edge in the knowledge graph, and the triples corresponding to the two target entities corresponding to the target edge; the target edge indicates the target relationship between the two entities corresponding to the two target edges.

[0105] Specifically, after the knowledge graph for orbital imaging diagnosis is constructed, a specialized graph database (such as Neo4j, ArangoDB, JanusGraph, etc.) can be used to store the knowledge graph. These databases optimize the query and traversal of graph structures, making them very suitable for processing complex relational data. In the knowledge graph, triples (subject-predicate-object) are a common data representation method, which can be directly mapped to nodes and edges in the graph database. Specifically, each triple can be regarded as an edge in the graph, where the subject and object are nodes, and the predicate is the type or attribute of the edge. For example, in a graph database such as Neo4j, you can use the CREATE statement to create nodes and edges to represent a triple. The following is an example:

[0106] ① / / Create two nodes

[0107] ②CREATE(a:Entity{name:malignant lesion})

[0108] ③CREATE(b:Entity{name:'PET-CT'})

[0109] ④ / / Create a relationship (edge) from a to b, type 'suggestion'

[0110] ⑤CREATE(a)-[:suggestion]->(b)

[0111] Step S103: Cover one of the two target entities and the target relationship corresponding to the triple.

[0112] Specifically, the triples in the knowledge graph, such as malignant lesion -> suggestion -> PET-CT, can be converted into a piece of token text and the entities or relationships can be masked.

[0113] Step S104: obtaining a target training sample based on the masked triples; correspondingly, obtaining a target training label based on the uncovered triples.

[0114] In an embodiment of the present invention, a training sample for a model for generating orbital imaging diagnosis reports is constructed based on a knowledge graph of orbital imaging diagnosis, so that the model can directly learn the knowledge contained in the knowledge graph during the training phase, thereby enhancing the training of the model.

[0115] In addition to constructing training samples based on triplets from the knowledge graph, it is also necessary to construct training samples based on clinical orbital imaging reports. For example, a large number of orbital imaging diagnostic reports submitted by junior physicians and those manually reviewed by senior physicians are collected. The orbital imaging diagnostic reports submitted by junior physicians are used as training samples, while the orbital imaging diagnostic reports manually reviewed by senior physicians are used as training labels.

[0116] For the collected orbital imaging diagnostic reports, pre-processing operations such as quality filtering, deduplication, and privacy removal can be performed. Specifically, quality filtering can be based on a classifier, and the classifier used for quality filtering can be the classifier mentioned above. Specifically, a classifier is used to evaluate the quality of the collected orbital imaging diagnostic reports, and then low-quality orbital imaging diagnostic reports are deleted based on the evaluation results. Heuristic quality filtering can also be used, for example, by designing a series of rules and strategies to identify and delete low-quality data, such as: keyword filtering, indicator filtering, statistical filtering, etc.

[0117] Step S105: constructing an orbital image diagnosis report generation model.

[0118] Step S106: using the target training samples and the corresponding target training labels, the constructed orbital image diagnosis report generation model is trained.

[0119] Specifically, during training, batch size directly affects the model's training speed, convergence, generalization ability, and resource efficiency. During training, based on reports and the increase in high-quality data, the batch size needs to be dynamically increased to improve training stability. In the initial 0.1% to 0.5% of training steps, a linear warm-up strategy is used to gradually increase the learning rate until it reaches a maximum value of approximately 5×10 -5 to 1×10 -4 After that, a cosine decay strategy is adopted in subsequent steps to gradually reduce the learning rate to about 10% of the maximum value until the training loss converges. The Adam optimizer combines the advantages of momentum and RMSprop, automatically adjusting the learning rate of each parameter by calculating the first-order moment estimate and the second-order moment estimate of the gradient.

[0120] This embodiment provides a method for constructing an orbital imaging diagnosis report generation model that collaborates with a large language model and a knowledge graph. The constructed orbital imaging diagnosis report generation model can rewrite low-quality orbital imaging diagnosis reports, such as orbital imaging diagnosis reports written by junior doctors, to improve the accuracy and standardization of the final orbital imaging diagnosis report.

[0121] In some optional specific implementations, step S106, i.e., after training the constructed orbital image diagnosis report generation model using the target training samples and the corresponding target training labels, further includes:

[0122] Step S107, creating a low-order matrix for the network layer in the orbital imaging diagnosis report generation model, and adding the low-order matrix to the corresponding network layer weight;

[0123] Step S108, obtaining a first fine-tuning training sample;

[0124] Step S109: fine-tune the orbital imaging diagnosis report generation model using the first fine-tune training sample, and only update the parameters of the low-order matrix during the fine-tune training process.

[0125] In an embodiment of the present invention, a low-rank adaptation (LoRA) model fine-tuning method is used to fine-tune the orbital imaging diagnostic report generation model. When using LoRA (Low-Rank Adaptation) for model fine-tuning, the scaling factor (α / r) and Dropout are two hyperparameters, and the update of the model parameters needs to be multiplied by the scaling factor α / r. It is generally set to 1. The larger the scaling factor, the greater the weight of the fine-tuning part. Dropout represents the probability of dropout of the LoRA layer, the purpose of which is to avoid overfitting. For small models such as 13B, it can be set to 10%, for example, and for larger models such as 65B, it can be set to 5%, for example.

[0126] The embodiment of the present invention realizes efficient fine-tuning of parameters of an orbital imaging diagnosis report generation model.

[0127] In other optional specific implementations, efficient parameter fine-tuning algorithms such as prefix tuning, prompt tuning (P-Tuning) and adapter may also be used.

[0128] In some optional specific implementations, after training the constructed orbital imaging diagnostic report generation model using the training samples and the corresponding training labels, the method further includes:

[0129] Step S110 : obtaining a first orbital image diagnosis report, where the first orbital image diagnosis report is an orbital image diagnosis report that is screened according to a preset rule and is determined to need to be rewritten.

[0130] Step S111: rewrite the first orbital image diagnosis report using the orbital image diagnosis report generation model to obtain a second orbital image diagnosis report. The orbital image diagnosis report generation model may be a fine-tuned orbital image diagnosis report generation model.

[0131] Step S112: Obtain a quality assessment result of the second orbital imaging diagnosis report.

[0132] Step S113: training a reward model based on the second orbital image diagnosis report and the corresponding quality assessment result.

[0133] For example, first, the orbital imaging diagnostic report written by a junior doctor (i.e., the first orbital imaging diagnostic report) can be used as the input of the orbital imaging diagnostic report generation model to output a rewritten standardized report (i.e., the second orbital imaging diagnostic report). Then, the senior doctor sorts the rewritten report (i.e., the second orbital imaging diagnostic report) from good to bad (mainly considering the two aspects of standardization and accuracy). That is, the second orbital imaging diagnostic report rewritten by the orbital imaging diagnostic report generation model is scored and sorted according to the preferences of the senior doctors to obtain a quality assessment result. Multiple second orbital imaging diagnostic reports and the corresponding quality assessment results form a preference data set for training the reward model. Alternatively, the classifier mentioned above can be used to perform a rough classification and sorting. Finally, the sorted reports are used to train the reward model.

[0134] In step S114, the reward model is used to perform a quality assessment on the orbital imaging diagnostic report generated by the orbital imaging diagnostic report generation model based on the second fine-tuning training sample, and a corresponding reward is output. As described above, the quality assessment result here can also be a score or ranking. The second fine-tuning training sample can be completely different from the first fine-tuning training sample, or some of the second fine-tuning training sample can be the same as the first fine-tuning training sample.

[0135] Step S115: Based on the rewards output by the reward model, the proximal strategy optimization method is used to fine-tune and optimize the orbital imaging diagnostic report generation model. Reinforcement learning is used to optimize the reward model strategy, and the rewards are used to fine-tune and optimize the model using the proximal strategy optimization method.

[0136] The embodiment of the present invention implements alignment fine-tuning of an orbital image diagnosis report generation model.

[0137] The reinforcement learning in the embodiment of the present invention specifically refers to "Reinforcement Learning from Human Feedback". The core of the whole process is to use human feedback as a reward signal to guide the training process of the model. The objective function of the reward modeling uses a pairwise ranking loss function. The difference in rewards represents the logarithmic probability that the orbital imaging diagnostic report generation model outputs one rewritten report that is more favored than another rewritten report. The reinforcement learning algorithm used here is the proximal policy optimization method (PPO), an "on-policy" algorithm for training agents that directly learns and updates the current policy instead of learning from past experience. PPO was chosen mainly because it can effectively balance the stability and update speed of the model, enabling the model to continuously evolve towards a better strategy while maintaining performance. The proximal policy optimization method adjusts the parameters of the orbital imaging diagnostic report generation model based on the rewards output by the reward model. During the parameter adjustment process, the proximal policy optimization method calculates two key ratios:

[0138] Probability Ratio: The probability that the new policy will generate the current rewrite report / the probability that the old policy will generate the current rewrite report. This ratio measures how much the probability of generating the current rewrite report has changed after the policy update.

[0139] Advantage Function: The reward score given by the reward model minus the baseline value. The advantage function measures the improvement in the reward of the current rewrite report relative to the average reward. The baseline value is typically the average reward over a period of time, which is used to reduce variance and make training more stable.

[0140] The PPO algorithm combines probability ratios with an advantage function and introduces a clipping term. This clipping term limits the range of probability ratios, preventing excessive policy updates. The reward model's score is used to guide the training of the orbital imaging diagnostic report generation model, improving its strategy toward generating higher reward reports.

[0141] While the pre-training of the orbital imaging diagnostic report generation model is constrained and enhanced by the "expert knowledge graph for orbital imaging reports," the vast amount of pre-training data can still contain harmful, misleading, and biased representations. In this embodiment of the present invention, a reinforcement learning method based on human feedback and the "knowledge graph for orbital imaging diagnosis" can further standardize and refine the output of the orbital imaging diagnostic report generation model.

[0142] In some optional specific implementations, obtaining the quality assessment result of the second orbital imaging diagnostic report includes:

[0143] Obtaining a quality assessment model for orbital imaging diagnostic reports, where positive and negative samples used to train the quality assessment model are manually screened orbital imaging diagnostic reports or orbital imaging diagnostic reports classified according to the seniority of the doctors who wrote the reports;

[0144] The quality assessment model is used to perform quality assessment on the second orbital imaging diagnosis report.

[0145] In some optional specific implementations, the method for constructing a model for generating an orbital imaging diagnostic report provided by an embodiment of the present invention further includes:

[0146] Obtain third orbital imaging diagnostic report;

[0147] Rewriting the third orbital image diagnosis report using an orbital image diagnosis report generation model to obtain a fourth orbital image diagnosis report;

[0148] Obtaining a fifth orbital imaging diagnostic report obtained by manually modifying the fourth orbital imaging diagnostic report;

[0149] The third orbital image diagnosis report is used as an optimization training sample, and the fifth orbital image diagnosis report is used as an optimization training label to perform optimization training on the orbital image diagnosis report generation model.

[0150] In this embodiment of the present invention, experts can manually evaluate the text reports and modification suggestions automatically generated by the orbital imaging diagnosis report generation model. Then, unqualified reports are collected, and expert feedback information is stored, and the orbital imaging diagnosis report generation model is iteratively optimized.

[0151] The method of the embodiment of the present invention combines two core components of artificial intelligence technology - large language model (LLM) and knowledge graph (KG), breaking the boundary between two different types of knowledge representation of LLM and KG, enhancing each other in knowledge expression and reasoning, improving the efficiency and accuracy of orbital imaging diagnosis report generation, and making orbital imaging diagnosis reports more standardized, with strong interpretability and credibility.

[0152] In this embodiment, a method for generating an orbital imaging diagnosis report is provided, which can be used in various computer devices, such as terminals, servers, etc. Figure 2 FIG. 1 is a flow chart of a method for generating an orbital imaging diagnosis report according to an embodiment of the present invention. Figure 2 As shown, the process includes the following steps:

[0153] Step S201, obtaining a sixth orbital imaging diagnosis report;

[0154] Step S202: rewrite the sixth orbital image diagnosis report using the orbital image diagnosis report generation model to obtain a seventh orbital image diagnosis report; wherein the orbital image diagnosis report generation model is constructed using any of the above-mentioned orbital image diagnosis report generation model construction methods.

[0155] In an embodiment of the present invention, after the orbital imaging diagnosis report generation model is constructed using the orbital imaging diagnosis report generation model construction method provided in the above embodiment, the model can be deployed on a computer device such as a server, and begin to receive and understand the imaging reports of newly added junior physicians in the non-training set, and generate standardized orbital imaging diagnosis reports through reasoning.

[0156] In some optional specific implementations, the method for generating an orbital imaging diagnostic report further includes:

[0157] Comparing the sixth orbital imaging diagnostic report and the seventh orbital imaging diagnostic report to generate modification suggestions; and / or,

[0158] Output the review result of the sixth orbital imaging diagnostic report.

[0159] In an embodiment of the present invention, the orbital imaging diagnostic report input by the orbital imaging diagnostic report generation model and the orbital imaging diagnostic report output can also be compared, and modification opinions and AI review results can be automatically generated to assist junior doctors in learning the methods and techniques of diagnosing orbital diseases and writing reports.

[0160] In this embodiment, a device for constructing a model for generating an orbital image diagnostic report is also provided. The device is used to implement the above-mentioned embodiments and preferred embodiments of the method for constructing a model for generating an orbital image diagnostic report. Details that have already been described will not be repeated. As used below, the term "module" can refer to a combination of software and / or hardware that implements a predetermined function. Although the devices described in the following embodiments are preferably implemented in software, implementation using hardware, or a combination of software and hardware, is also possible and contemplated.

[0161] This embodiment provides a device for constructing a model for generating an orbital image diagnosis report. Figure 3 Shown, including:

[0162] A knowledge graph acquisition module 301 is configured to acquire a knowledge graph for orbital imaging diagnosis, wherein the entities corresponding to the nodes in the knowledge graph include one or more of anatomical sites, pathological features, disease considerations, and recommended examination methods;

[0163] A triple acquisition module 302 is configured to acquire a triple corresponding to each target edge in the knowledge graph and the two target entities corresponding to the target edge; the target edge indicates a target relationship between the two entities corresponding to the two target edges;

[0164] A covering module 303 is configured to cover one of the two target entities and the target relationship corresponding to the triple;

[0165] The target sample construction module 304 is configured to obtain a target training sample based on the masked triples; correspondingly, to obtain a target training label based on the uncovered triples;

[0166] A model building module 305 is used to build a model for generating an orbital imaging diagnosis report;

[0167] The model training module 306 is used to train the constructed orbital image diagnosis report generation model using the target training samples and the corresponding target training labels.

[0168] In an optional embodiment, the orbital imaging diagnostic report generation model construction device further includes:

[0169] An adding module is used to create a low-order matrix for the network layer in the orbital imaging diagnosis report generation model, and add the low-order matrix to the corresponding network layer weight;

[0170] A first fine-tuning training sample acquisition module, configured to acquire a first fine-tuning training sample;

[0171] The first fine-tuning module is used to use the first fine-tuning training sample to fine-tune the orbital imaging diagnosis report generation model, and only update the parameters of the low-order matrix during the fine-tuning training process.

[0172] In an optional embodiment, the orbital imaging diagnostic report generation model construction device further includes:

[0173] A first orbital imaging diagnostic report acquisition module is configured to acquire a first orbital imaging diagnostic report, where the first orbital imaging diagnostic report is an orbital imaging diagnostic report that is screened according to a preset rule and is determined to need to be rewritten;

[0174] a rewriting module, configured to rewrite the first orbital image diagnosis report using the orbital image diagnosis report generation model to obtain a second orbital image diagnosis report;

[0175] A quality assessment result acquisition module, configured to acquire a quality assessment result of the second orbital imaging diagnostic report;

[0176] A reward model training module, configured to train a reward model based on the second orbital image diagnosis report and the corresponding quality assessment result;

[0177] a reward output module, configured to use the reward model to perform a quality assessment on the orbital image diagnosis report output by the orbital image diagnosis report generation model based on the second fine-tuning training sample, and output a corresponding reward;

[0178] The second fine-tuning module is used to fine-tune and optimize the orbital imaging diagnosis report generation model based on the reward output by the reward model using a proximal strategy optimization method.

[0179] In an optional implementation, the quality assessment result acquisition module includes:

[0180] a quality assessment model acquisition unit, configured to acquire a quality assessment model for orbital imaging diagnostic reports, wherein the positive and negative samples for training the quality assessment model are manually screened orbital imaging diagnostic reports or orbital imaging diagnostic reports classified according to the seniority of the doctors who wrote the reports;

[0181] A quality assessment unit is used to perform quality assessment on the second orbital image diagnosis report using the quality assessment model.

[0182] In an optional implementation, the knowledge graph acquisition module 301 includes:

[0183] a medical knowledge data acquisition unit, configured to acquire medical knowledge data related to orbital imaging diagnosis, wherein the medical knowledge data includes data in an expert knowledge base of orbital imaging reports;

[0184] An extraction unit, configured to extract entities and relations between entities from the medical knowledge data using a general macro model;

[0185] A knowledge graph construction unit is used to construct the knowledge graph of orbital imaging diagnosis based on the entities extracted by the general large model and the relationships between the entities.

[0186] In an optional embodiment, the orbital imaging diagnostic report generation model construction device further includes:

[0187] A third orbital imaging diagnostic report acquisition module, used to obtain a third orbital imaging diagnostic report;

[0188] a fourth orbital image diagnosis report acquisition module, configured to rewrite the third orbital image diagnosis report using an orbital image diagnosis report generation model to obtain a fourth orbital image diagnosis report;

[0189] a fifth orbital imaging diagnostic report acquisition module, configured to acquire a fifth orbital imaging diagnostic report obtained by manually modifying the fourth orbital imaging diagnostic report;

[0190] The optimization training module is used to use the third orbital image diagnosis report as an optimization training sample and the fifth orbital image diagnosis report as an optimization training label to optimize the orbital image diagnosis report generation model.

[0191] The further functional description of each of the above modules and units is the same as that of the above corresponding embodiments and will not be repeated here.

[0192] The orbital imaging diagnostic report generation model construction device in this embodiment is presented in the form of a functional unit, where the unit refers to an ASIC (Application Specific Integrated Circuit) circuit, a processor and memory that executes one or more software or fixed programs, and / or other devices that can provide the above functions.

[0193] This embodiment also provides an orbital imaging diagnostic report generation device, which is used to implement the above-mentioned orbital imaging diagnostic report generation method embodiment and preferred implementation method. Details that have already been described will not be repeated. As used below, the term "module" can refer to a combination of software and / or hardware that implements a predetermined function. Although the devices described in the following embodiments are preferably implemented in software, implementation using hardware, or a combination of software and hardware, is also possible and contemplated.

[0194] This embodiment provides a device for generating an orbital image diagnosis report. Figure 4 Shown, including:

[0195] A report acquisition module 401 is used to obtain a sixth orbital imaging diagnosis report;

[0196] The rewriting module 402 is used to rewrite the sixth orbital image diagnosis report using the orbital image diagnosis report generation model to obtain a seventh orbital image diagnosis report; wherein the orbital image diagnosis report generation model is constructed using any of the above-mentioned orbital image diagnosis report generation model construction methods.

[0197] In an optional embodiment, the orbital imaging diagnostic report generating device further includes:

[0198] a modification suggestion generating module, configured to compare the sixth orbital imaging diagnostic report with the seventh orbital imaging diagnostic report to generate modification suggestions; and / or,

[0199] The audit result output module is used to output the audit result of the sixth orbital imaging diagnosis report.

[0200] The further functional description of each of the above modules and units is the same as that of the above corresponding embodiments and will not be repeated here.

[0201] The orbital imaging diagnostic report generating device in this embodiment is presented in the form of a functional unit, where the unit refers to an ASIC (Application Specific Integrated Circuit) circuit, a processor and memory that executes one or more software or fixed programs, and / or other devices that can provide the above functions.

[0202] The embodiment of the present invention also provides a computer device having the above Figure 3 The orbital imaging diagnosis report generation model building device shown or the one having the above Figure 4 The orbital imaging diagnostic report generating device shown.

[0203] See also Figure 5 , Figure 5 is a structural diagram of a computer device provided by an optional embodiment of the present invention, such as Figure 5 As shown, the computer device includes: one or more processors 10, memory 20, and interfaces for connecting various components, including high-speed interfaces and low-speed interfaces. Various components utilize different buses to communicate with each other and can be installed on a common mainboard or installed in other ways as needed. The processor can process the instructions executed in the computer device, including instructions stored in the memory or on the memory to display the graphical information of the GUI on an external input / output device (such as, a display device coupled to the interface). In some optional embodiments, if necessary, multiple processors and / or multiple buses can be used together with multiple memories and multiple memories. Equally, multiple computer devices can be connected, and each device provides part of the necessary operations (for example, as a server array, a group of blade servers, or a multi-processor system). Figure 5 A processor 10 is taken as an example.

[0204] The processor 10 may be a central processing unit, a network processor, or a combination thereof. The processor 10 may further include a hardware chip. The hardware chip may be an application-specific integrated circuit, a programmable logic device, or a combination thereof. The programmable logic device may be a complex programmable logic device, a field programmable gate array, a general purpose array logic, or any combination thereof.

[0205] The memory 20 stores instructions that can be executed by at least one processor 10, so that the at least one processor 10 executes the method shown in the above embodiment.

[0206] The memory 20 may include a program storage area and a data storage area, wherein the program storage area may store an operating system and application programs required for at least one function; the data storage area may store data created based on the use of the computer device, etc. In addition, the memory 20 may include a high-speed random access memory, and may also include a non-transient memory, such as at least one disk storage device, a flash memory device, or other non-transient solid-state storage device. In some optional embodiments, the memory 20 may optionally include a memory remotely located relative to the processor 10, and these remote memories may be connected to the computer device via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0207] The memory 20 may include a volatile memory, such as a random access memory; the memory may also include a non-volatile memory, such as a flash memory, a hard disk or a solid-state drive; the memory 20 may also include a combination of the above types of memory.

[0208] The computer device further includes an input device 30 and an output device 40. The processor 10, the memory 20, the input device 30 and the output device 40 may be connected via a bus or other means. Figure 5 The bus connection is taken as an example.

[0209] The input device 30 can receive input digital or character information and generate key signal input related to user settings and function control of the computer device, such as a touch screen, a keypad, a mouse, a trackpad, a touch pad, an indicator stick, one or more mouse buttons, a trackball, a joystick, etc. The output device 40 can include a display device, an auxiliary lighting device (e.g., an LED), and a tactile feedback device (e.g., a vibration motor). The above-mentioned display device includes but is not limited to a liquid crystal display, a light emitting diode, a display, and a plasma display. In some optional embodiments, the display device can be a touch screen.

[0210] The computer device further includes a communication interface for the computer device to communicate with other devices or a communication network.

[0211] The embodiment of the present invention also provides a computer-readable storage medium. The above-mentioned method according to the embodiment of the present invention can be implemented in hardware, firmware, or implemented as a computer code that can be recorded in a storage medium, or implemented as a computer code that is originally stored in a remote storage medium or a non-temporary machine-readable storage medium and downloaded through a network and will be stored in a local storage medium, so that the method described herein can be stored in such software processing on a storage medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware. Among them, the storage medium can be a magnetic disk, an optical disk, a read-only storage memory, a random access memory, a flash memory, a hard disk or a solid-state drive, etc.; further, the storage medium can also include a combination of the above-mentioned types of memory. It can be understood that a computer, a processor, a microprocessor controller or programmable hardware includes a storage component that can store or receive software or computer code. When the software or computer code is accessed and executed by a computer, a processor or hardware, the method shown in the above embodiment is implemented.

[0212] A portion of the present invention may be applied as a computer program product, such as a computer program instruction, which, when executed by a computer, can call or provide the method and / or technical solution according to the present invention through the operation of the computer. Those skilled in the art should understand that the form in which the computer program instruction exists in a computer-readable medium includes, but is not limited to, a source file, an executable file, an installation package file, etc. Accordingly, the way in which the computer program instruction is executed by the computer includes, but is not limited to: the computer directly executes the instruction, or the computer compiles the instruction and then executes the corresponding compiled program, or the computer reads and executes the instruction, or the computer reads and installs the instruction and then executes the corresponding installed program. Here, the computer-readable medium may be any available computer-readable storage medium or communication medium that can be accessed by the computer.

[0213] Although the embodiments of the present invention have been described with reference to the accompanying drawings, those skilled in the art may make various modifications and variations without departing from the spirit and scope of the present invention. Such modifications and variations are all within the scope defined by the appended claims.

Claims

1. A method for constructing a model for generating an orbital imaging diagnosis report, characterized in that: The method comprises: Obtaining a knowledge graph for orbital imaging diagnosis, wherein entities corresponding to nodes in the knowledge graph include: one or more of anatomical site, pathological features, disease considerations, and recommended examination methods; Obtaining each target edge in the knowledge graph and the triples corresponding to the two target entities corresponding to the target edge; the target edge indicates a target relationship between the two entities corresponding to the two target edges; Covering one of the two target entities and the target relationship corresponding to the triple; Based on the masked triples, a target training sample is obtained; correspondingly, based on the uncovered triples, a target training label is obtained; Construct a model for generating orbital imaging diagnostic reports; The constructed orbital image diagnosis report generation model is trained using the target training samples and the corresponding target training labels.

2. The method according to claim 1, characterized in that After training the constructed orbital image diagnosis report generation model using the target training samples and the corresponding target training labels, the method further includes: For the network layer in the orbital imaging diagnostic report generation model, create a low-order matrix, and add the low-order matrix to the corresponding network layer weight; Obtain the first fine-tuning training sample; The first fine-tuning training sample is used to fine-tune the orbital imaging diagnosis report generation model, and only the parameters of the low-order matrix are updated during the fine-tuning training process.

3. The method according to claim 1, characterized in that After training the constructed orbital image diagnosis report generation model using the training samples and the corresponding training labels, the method further includes: Obtaining a first orbital imaging diagnostic report, where the first orbital imaging diagnostic report is an orbital imaging diagnostic report that is screened according to a preset rule and is determined to need to be rewritten; Rewriting the first orbital image diagnosis report using the orbital image diagnosis report generation model to obtain a second orbital image diagnosis report; Obtaining a quality assessment result of the second orbital imaging diagnostic report; Training a reward model based on the second orbital imaging diagnosis report and the corresponding quality assessment result; Using the reward model, a quality assessment is performed on the orbital imaging diagnosis report output by the orbital imaging diagnosis report generation model based on the second fine-tuning training sample, and a corresponding reward is output; Based on the rewards output by the reward model, the proximal strategy optimization method is used to fine-tune and optimize the orbital imaging diagnosis report generation model.

4. The method according to claim 3, characterized in that The obtaining of the quality assessment result of the second orbital imaging diagnostic report includes: Obtaining a quality assessment model for orbital imaging diagnostic reports, where positive and negative samples used to train the quality assessment model are manually screened orbital imaging diagnostic reports or orbital imaging diagnostic reports classified according to the seniority of the physicians who wrote the reports; The quality assessment model is used to perform quality assessment on the second orbital imaging diagnosis report.

5. The method according to claim 1, wherein The method of obtaining a knowledge graph for orbital imaging diagnosis includes: Acquiring medical knowledge data related to orbital imaging diagnosis, wherein the medical knowledge data includes data in an expert knowledge base of orbital imaging reports; Extracting entities and relationships between entities from the medical knowledge data using a general large model; The knowledge graph of orbital imaging diagnosis is constructed based on the entities extracted from the general large model and the relationships between the entities.

6. The method according to any one of claims 1 to 5, characterized in that Also includes: Obtain third orbital imaging diagnostic report; Rewriting the third orbital image diagnosis report using an orbital image diagnosis report generation model to obtain a fourth orbital image diagnosis report; Obtaining a fifth orbital imaging diagnostic report obtained by manually modifying the fourth orbital imaging diagnostic report; The third orbital image diagnosis report is used as an optimization training sample, and the fifth orbital image diagnosis report is used as an optimization training label to perform optimization training on the orbital image diagnosis report generation model.

7. A method for generating an orbital imaging diagnosis report, characterized in that: include: Obtain the sixth orbital imaging diagnostic report; The sixth orbital image diagnosis report is rewritten using the orbital image diagnosis report generation model to obtain a seventh orbital image diagnosis report; wherein the orbital image diagnosis report generation model is constructed using the orbital image diagnosis report generation model construction method described in any one of claims 1-6.

8. The method according to claim 7, characterized in that Also includes: Comparing the sixth orbital imaging diagnostic report and the seventh orbital imaging diagnostic report to generate modification suggestions; and / or, Output the review result of the sixth orbital imaging diagnostic report.

9. A device for constructing a model for generating an orbital image diagnosis report, characterized in that: The device comprises: A knowledge graph acquisition module is used to obtain a knowledge graph for orbital imaging diagnosis, wherein the entities corresponding to the nodes in the knowledge graph include: one or more of anatomical parts, pathological characteristics, disease considerations, and recommended examination methods; A triple acquisition module is configured to acquire a triple corresponding to each target edge in the knowledge graph and two target entities corresponding to the target edge; the target edge indicates a target relationship between the two entities corresponding to the two target edges; A covering module, configured to cover one of the two target entities and the target relationship corresponding to the triple; A target sample construction module is used to obtain a target training sample based on the masked triples; correspondingly, to obtain a target training label based on the uncovered triples; A model building module is used to build a model for generating orbital imaging diagnostic reports; The model training module is used to train the constructed orbital image diagnosis report generation model using the target training samples and the corresponding target training labels.

10. An orbital image diagnosis report generating device, characterized in that: The device comprises: A report acquisition module is used to obtain the sixth orbital imaging diagnosis report; A rewriting module is used to rewrite the sixth orbital image diagnosis report using the orbital image diagnosis report generation model to obtain a seventh orbital image diagnosis report; wherein the orbital image diagnosis report generation model is constructed using the orbital image diagnosis report generation model construction method according to any one of claims 1-6.

11. A computer device, characterized in that: include: A memory and a processor, wherein the memory and the processor are communicatively connected to each other, the memory stores computer instructions, and the processor executes the orbital imaging diagnostic report generation model construction method according to any one of claims 1 to 6 or the orbital imaging diagnostic report generation model construction method according to any one of claims 7 to 8 by executing the computer instructions.

12. A computer-readable storage medium, characterized in that The computer-readable storage medium stores computer instructions, which are used to enable a computer to execute the orbital imaging diagnostic report generation model construction method described in any one of claims 1 to 6 or the orbital imaging diagnostic report generation model construction method described in any one of claims 7 to 8.

13. A computer program product, characterized in that The method comprises computer instructions for causing a computer to execute the orbital imaging diagnosis report generation model construction method according to any one of claims 1 to 6 or the orbital imaging diagnosis report generation model construction method according to any one of claims 7 to 8.

Images