Fusion gene data and clinical image generated ai diagnosis and treatment platform

Abstract

This invention discloses a generative AI diagnostic and treatment platform that integrates gene data and clinical images, belonging to the field of medical technology. It includes modules for data acquisition, desensitization and protection, data processing, feature extraction, gene analysis, fusion analysis, predictive analysis, treatment plan generation, result display, model management, self-iteration, and user management. This invention enables non-invasive and accurate diagnosis, reducing potential harm to newborns, minimizing unnecessary repeat examinations, lowering overall diagnostic costs, and improving the stability and accuracy of disease prediction results. Simultaneously, this invention enhances the foresight and scientific rigor of treatment plan development, providing strong support for early intervention and precision medicine, improving diagnostic efficiency and risk assessment accuracy, and saving hospitals and patients significant time and effort.

Description

Technical Field

[0001] This invention relates to the field of medical technology, and in particular to a generative AI diagnosis and treatment platform that integrates gene data and clinical images. Background Technology

[0002] Against the backdrop of a continued decline in fertility rates and total population, birth defects have gradually become a significant factor affecting population health. Related studies indicate that the incidence of birth defects in my country has long been at a high level. Insufficient screening or delayed diagnosis of some congenital diseases places a heavy economic and psychological burden on affected families and exerts sustained pressure on public health. Traditional prenatal screening technologies have limitations in sensitivity and disease coverage, relying excessively on single testing methods and human experience, making it difficult to achieve early and accurate identification of high-risk groups. This results in unsatisfactory birth defect prevention and control, further impacting fertility confidence and population health.

[0003] In practice, the neonatal birth defect intervention system in grassroots and remote areas still faces many challenges. Due to factors such as uneven distribution of medical resources, insufficient professional and technical personnel, and limited awareness of eugenics, some areas struggle to provide systematic and standardized prenatal screening and genetic counseling services, leaving high-risk pregnant women without timely and effective assessment and intervention. This not only increases maternal and infant health risks but also, to some extent, hinders the improvement of overall population quality.

[0004] Precise prevention and control of birth defects in newborns heavily relies on the comprehensive analysis of multimodal prenatal screening information and genetic data. Relying solely on traditional testing methods and human experience can easily lead to missed or misdiagnosed cases, failing to meet the practical needs of early intervention. Addressing the technical bottlenecks in screening and intervention, our team has developed "Intelligent Defect Recognition – A Pioneer in Early Screening and Intervention for Newborn Birth Defects" by integrating genomics, medical imaging, and clinical data resources. This aims to achieve early identification and precise intervention of birth defects, reduce the risk of their occurrence, and provide strong technical support for improving population health and promoting balanced development of medical services. To this end, we propose a generative AI diagnostic platform that integrates genetic data and clinical imaging. Summary of the Invention

[0005] The purpose of this invention is to address the shortcomings of existing technologies by proposing a generative AI diagnostic and treatment platform that integrates gene data and clinical images.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: The generative AI diagnosis and treatment platform that integrates gene data and clinical images includes a data acquisition module, a desensitization and protection module, a data processing module, a feature extraction module, a gene analysis module, a fusion analysis module, a predictive analysis module, a treatment plan generation module, a results display module, a model management module, a self-iteration module, and a user management module. The data acquisition module is used to collect multimodal data from each newborn and simultaneously upload it to the corresponding mobile terminal and medical terminal. The desensitization protection module is used to desensitize sensitive information of each newborn. The data processing module is used to clean and standardize the collected modal data. The feature extraction module is used to extract features from medical image data in each modality of data; The gene analysis module analyzes the genetic information data of each newborn based on bioinformatics analysis methods. The fusion analysis module is used to construct a multimodal data representation system and simultaneously analyze multimodal features from different sources; The predictive analysis module is used to jointly model the extracted modal features to predict the probability of disease occurrence and risk classification results for each newborn. The treatment plan generation module generates corresponding individualized treatment plans based on the prediction results of each newborn. The results display module is used to centrally display the analysis results, prediction results, and individualized treatment plans for each newborn. The model management module is used to manage the disease prediction models built within the platform and to continuously evaluate the performance of the corresponding models using different datasets. The self-iteration module is used to continuously collect the diagnosis and treatment data of each newborn, and to continuously train and correct the parameters of the disease prediction model based on the real-time diagnosis and treatment data. The user management module is used to uniformly configure the usage permissions, project size, sample size, and data storage cycle of different users.

[0007] As a further aspect of the present invention, the specific steps for the gene analysis module to analyze the genetic information data of each newborn are as follows: S1.1: Collect multimodal data of each newborn, classify and manage the medical imaging data in each modality according to the unique identifier of each newborn, and unify the grayscale range of each medical imaging data. Then, locate and segment the preset target regions in each medical imaging data. After completing the region segmentation, extract features from the medical imaging data of each target region. At the same time, obtain the corresponding clinical data and medical history data in each modality and extract the corresponding features. Among them, the multimodal data includes gene and genetic information data, medical imaging data, clinical data, medical history and perinatal data, and developmental assessment and follow-up data. S1.2: Obtain genetic information data from each newborn multimodal data, and obtain the corresponding gene sequence data, gene expression data, and chromosome abnormality data from the genetic information data. Then, collect the base quality of each gene sequence data, remove gene sequence data with base quality lower than the preset quality, and unify the dimensions and numerical range of each gene expression data. Finally, map each chromosome abnormality data to the preset structure. S1.3: Extract the features of various types of genetic data after processing, encode all genetic features into structured feature vectors to form a corresponding set of genetic features, and then identify genetic features with diagnostic value lower than the preset standard by analyzing the correlation between different genetic features and the degree of association with disease phenotypes. At the same time, construct and train a genetic inference model. S1.4: Input all extracted genetic features into the trained genetic inference model. The genetic inference model outputs the corresponding disease risk assessment results based on the input genetic features. Then, based on all genetic features, analysis results and corresponding disease phenotype information, the model is integrated to construct a disease-specific gene database.

[0008] As a further aspect of the present invention, the specific steps of the fusion analysis module in simultaneously parsing multimodal features from different sources are as follows: S2.1: Extract each modal feature and confirm the structural form of each type of feature. Then, divide the semantic level of each modal feature and assign a clear modal identifier to each type of feature. Then, unify the numerical range of each modal feature while retaining the relative change relationship of each modal feature. Then, align each modal feature according to the unique identifier of the corresponding newborn and introduce time synchronization rules to map different modal features to a preset time window. S2.2: Mine the potential correlations between various modal features, and establish corresponding mapping relationships based on the mining results. Assign corresponding initial weights according to the importance of different modal features in disease diagnosis, and perform weighted combination of various modal features. S2.3: Check the feature dimensions, numerical range, missing information, and modality identification completeness of each modality feature. Then, based on the modality features that have passed the test, construct a multimodal data representation system and form the corresponding multimodal feature vector.

[0009] As a further aspect of the present invention, the specific steps of the predictive analysis module in predicting the probability of disease occurrence and risk grading results for each newborn are as follows: S3.1: Based on various features, form a corresponding multimodal feature sample set and establish a mapping relationship with the corresponding real diagnostic results or disease labels. Then, divide each modal feature sample set into a training set, a validation set, and a test set according to a preset ratio. Then, according to the feature dimension, sample size, and task type, select the corresponding model structure to build a disease prediction model and set the initial parameters of the disease prediction model. S3.2: Input training samples into the disease prediction model in batches. The model predicts the samples based on the current parameters and compares the prediction results with the true disease labels. Then, calculate the adjustment direction of the model based on the prediction error and update the model parameters step by step. During the training process, after each round of training, the validation set performs prediction analysis on the prediction results of the disease prediction model, compares the difference between the prediction results of the validation set and the true labels, evaluates the generalization ability of the model, and adjusts the key parameters of the model based on the evaluation results. S3.3: When the model's performance degrades on the validation set, the model's adaptability to new data is adjusted by limiting model complexity or adjusting training strategies. After multiple rounds of training and validation, the disease prediction model is evaluated on the test set until the model meets the preset requirements. Then, the performance of disease prediction models built with different algorithms is evaluated, and the performance of each model in different diagnostic tasks is analyzed. Finally, the prediction results of disease prediction models built with different algorithms are weighted and fused to generate the corresponding comprehensive prediction results. S3.4: Input the multimodal feature vectors of each newborn into the trained disease prediction model. Based on the feature patterns learned during training, the disease prediction model calculates the probability value of disease occurrence for each newborn and classifies different types of defects during the calculation process. At the same time, based on the prediction results of the disease prediction model, it identifies defect types with matching degrees exceeding the preset range and marks the corresponding confidence information. After obtaining the corresponding disease probability value and defect type judgment results, it classifies the risk of each newborn according to preset rules.

[0010] As a further aspect of the present invention, the specific steps of the solution generation module in generating the corresponding individualized treatment plan are as follows: S4.1: Compare the genetic characteristics of each newborn with known pathogenic genes in the disease-specific gene database to identify whether each newborn has genetic susceptibility factors or special genetic background. At the same time, combine the family genetic history information of each newborn to assess the potential risk of genetic transmission. Then, interpret the imaging characteristics of each newborn with known clinical test indicators to analyze whether the structural or morphological abnormalities reflected in each imaging characteristic are consistent with the clinical test indicators and determine the stage of disease development. S4.2: Obtain the current status of each newborn and match it with the built-in diagnosis and treatment rule base and medical knowledge base. Based on the defect type judgment result, risk level and current status of each newborn, select diagnosis and treatment paths and intervention plan frameworks with matching degree exceeding the preset threshold from the diagnosis and treatment rule base. Combine the genetic risk results and imaging results of each newborn to adjust the priority, intensity and timing of intervention measures to form a corresponding individualized intervention plan. S4.3: Based on the intervention plan generated for each newborn, formulate corresponding follow-up strategies, and determine the follow-up frequency, follow-up content and key monitoring indicators according to the disease risk level and current status of each newborn, and clarify the time node for the next assessment. For patients with the highest risk level, increase the corresponding follow-up density. S4.4: Based on the intervention plan and follow-up strategy for each newborn, assess the corresponding prognosis. By analyzing the current multimodal characteristics of each newborn and the outcome data of similar historical cases, predict the disease development trend of each newborn under different intervention conditions. After all analyses are completed, the final diagnosis and treatment recommendations are structured and organized.

[0011] As a further aspect of the present invention, the specific steps of the self-iterative module continuously training and correcting the parameters of the disease prediction model are as follows: S5.1: Continuously collect clinical feedback data from each newborn, filter out and remove clinical feedback data that is below the preset standard, then associate and match each clinical feedback data with the historical data of the corresponding newborn, and then, with the unique identifier and treatment timeline of each newborn as the core, match each clinical feedback data with the corresponding original multimodal features, prediction results and treatment plans to form a complete data link; S5.2: The clinical feedback data are used as new label information and included in the training sample pool. While retaining the original training data, the newly added feedback samples are integrated with the historical samples to construct an incremental training dataset. After the incremental training dataset is prepared, the disease prediction model is trained again to fine-tune the original parameters of the disease prediction model. At the same time, the prediction bias of the model is corrected through the new training samples. S5.3: After completing an incremental training, the updated model is evaluated for performance. By comparing the prediction results, stability and error changes before and after the model update, it is determined whether the model has achieved performance improvement in real clinical scenarios and whether there is a risk of performance decline or instability. After that, when the model performance evaluation results meet the preset standards, the updated model is stored and identified as a new version. While retaining the historical model version, the new model is deployed to the actual application environment.

[0012] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. This invention collects multimodal data from newborns, categorizes and manages medical imaging data based on each newborn's unique identifier, unifies the grayscale range of each medical imaging data, locates and segments preset target regions in each medical imaging data, extracts corresponding image features from the segmented target regions, and simultaneously acquires corresponding clinical data and medical history data and extracts corresponding features. It also collects gene sequences, gene expression, and chromosomal abnormality data from each newborn's multimodal data, performs base quality control on gene sequences, unifies the dimensions of gene expression data, performs structured mapping on chromosomal abnormality data, extracts genetic features from various types of genetic data and encodes them into structured vectors, constructs a genetic inference model to output corresponding disease risk assessment results, and forms a disease-specific gene database. Subsequently, it confirms the structure, semantically layers, and unifies the numerical values ​​of each modality feature. Alignment of modal features is performed based on the unique identifier of each newborn. Time synchronization rules are introduced and correlations between modal features are explored. Weights are assigned according to importance and features are combined to form corresponding multimodal feature vectors. A unified multimodal data representation system is constructed. A mapping relationship is established between multimodal feature samples and real labels. Training, validation, and test sets are divided. A disease prediction model is constructed and trained. The model performance is optimized through multiple rounds of training, validation, and testing. The results of multiple models are weighted and fused. Finally, the multimodal features of each newborn are input into the disease prediction model, which outputs the corresponding disease occurrence probability, defect type, and confidence level. According to preset rules, the corresponding risk level is divided. This can achieve non-invasive and accurate diagnosis, reduce potential harm to newborns, reduce unnecessary repeated examinations, lower overall diagnostic costs, and improve the stability and accuracy of disease prediction results.

[0013] 2. This invention compares the genetic characteristics of each newborn with known pathogenic genes in a specialized disease gene database to identify whether each newborn has genetic susceptibility factors or a special genetic background. Combined with corresponding family genetic history information, a comprehensive assessment of potential genetic transmission risks is conducted. Subsequently, the imaging characteristics of each newborn are jointly interpreted with threshold clinical test indicators to analyze whether the structural or morphological abnormalities reflected in the imaging characteristics corroborate the test results, determining the stage of disease development. Simultaneously, the current health status of each newborn is obtained and matched with a built-in diagnostic and treatment rule base and medical knowledge base. Based on the defect type, risk level, and current status, highly matching diagnostic and treatment pathways and intervention plans are selected. Then, combined with genetic risk and imaging manifestations, the priority, intensity, and timing of intervention measures are adjusted to form an individualized intervention plan. Follow-up is then conducted based on the intervention plan. The strategy determines the corresponding follow-up frequency, follow-up content, and key monitoring indicators based on the disease risk level of each newborn. For high-risk newborns, the follow-up density is increased, and the next assessment time is clearly defined. Simultaneously, by combining multimodal characteristics and historical similar case outcome data, the disease development trend under different intervention conditions is predicted, forming a prognostic assessment result. Finally, the treatment recommendations are output in a structured manner. Subsequently, clinical feedback data from each newborn is continuously collected, and after quality screening and correlation integration, a complete data link is formed. Each clinical feedback data point is used as a new label and incorporated into the training sample pool to construct an incremental training dataset. The disease prediction model is continuously trained, evaluated, and updated. This approach improves the foresight and scientific rigor of treatment plan development, provides strong support for early intervention and precision medicine, helps improve diagnostic efficiency and risk assessment accuracy, and saves hospitals and patients significant time and effort. Attached Figure Description

[0014] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used together with the embodiments of the invention to explain the invention and do not constitute a limitation thereof.

[0015] Figure 1 This is a system block diagram of the generative AI diagnosis and treatment platform that integrates gene data and clinical images proposed in this invention. Figure 2 This is a flowchart of the disease prediction model algorithm for the generative AI diagnosis and treatment platform that integrates gene data and clinical images proposed in this invention. Figure 3 This diagram illustrates the generation of personalized treatment plans using the generative AI diagnostic and treatment platform that integrates gene data and clinical images, as proposed in this invention. Detailed Implementation

[0016] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0017] Example 1, referring to Figure 1 and Figure 2 The generative AI diagnosis and treatment platform that integrates gene data and clinical images includes a data acquisition module, a desensitization and protection module, a data processing module, a feature extraction module, a gene analysis module, a fusion analysis module, a predictive analysis module, a treatment plan generation module, a result display module, a model management module, a self-iteration module, and a user management module. The data acquisition module is used to collect multimodal data from each newborn and upload it synchronously to the corresponding mobile and medical terminals; the desensitization protection module is used to desensitize sensitive information of each newborn; the data processing module is used to clean and standardize the collected modal data; and the feature extraction module is used to extract features from the medical imaging data in each modal data.

[0018] The gene analysis module uses bioinformatics analysis methods to analyze the genetic information data of each newborn.

[0019] Specifically, multimodal data from each newborn is collected. Medical imaging data within each modality is categorized and managed based on a unique identifier for each newborn, and the grayscale range of each medical imaging dataset is standardized. Then, preset target regions within each medical imaging dataset are located and segmented. After region segmentation, features are extracted from the medical imaging data of each target region. Simultaneously, corresponding clinical and medical history data from each modality are acquired, and relevant features are extracted. Genetic information data from each newborn's multimodal data is obtained, including corresponding gene sequence data, gene expression data, and chromosomal abnormality data. The base quality of each gene sequence is then assessed, and gene sequences with base quality below a preset quality are removed. The system unifies the dimensions and numerical ranges of gene expression data, maps chromosomal abnormality data to a predefined structure, extracts features from various genetic data, and encodes all genetic features into structured feature vectors to form corresponding sets of genetic features. Then, by analyzing the correlation between different genetic features and their association with disease phenotypes, it identifies genetic features with diagnostic value lower than the predefined standard. At the same time, it constructs and trains a genetic inference model, inputting all extracted genetic features into the trained model. The genetic inference model outputs corresponding disease risk assessment results based on the input genetic features. Finally, it integrates all genetic features, analysis results, and corresponding disease phenotype information to construct a disease-specific gene database.

[0020] It should be further explained that multimodal data includes gene and genetic information data, medical imaging data, clinical data, medical history and perinatal data, as well as developmental assessment and follow-up data.

[0021] The fusion analysis module is used to construct a multimodal data representation system and simultaneously analyze multimodal features from different sources.

[0022] Specifically, the modal features are extracted, and the structural form of each type of feature is confirmed. Then, the semantic levels of each modal feature are divided, and a clear modal identifier is assigned to each type of feature. Next, the numerical range of each modal feature is unified, while the relative change relationship of each modal feature is preserved. Then, each modal feature is aligned according to the unique identifier of the corresponding newborn. Time synchronization rules are introduced to map different modal features to a preset time window, and potential associations between modal features are explored. Based on the mining results, corresponding mapping relationships are established. According to the importance of different modal features in disease diagnosis, corresponding initial weights are assigned, and each modal feature is weighted and combined. The feature dimension, numerical range, missing status, and modal identifier integrity of each modal feature are checked. Finally, based on each modal feature that passes the test, a multimodal data expression system is constructed to form the corresponding multimodal feature vector.

[0023] The predictive analysis module is used to jointly model the extracted modal features to predict the probability of disease occurrence and risk classification results for each newborn.

[0024] Specifically, based on various features, corresponding multimodal feature sample sets are formed, and mapping relationships are established with corresponding real diagnostic results or disease labels. Then, each modal feature sample set is divided into training, validation, and test sets according to a preset ratio. Next, based on feature dimensions, sample size, and task type, a corresponding model structure is selected to build a disease prediction model, and initial parameters are set. Training samples are input into the disease prediction model in batches. The model predicts the samples based on the current parameters and compares the prediction results with the real disease labels. The adjustment direction of the model is calculated based on the prediction error, and the model parameters are gradually updated. During training, after each round of training, the validation set performs prediction analysis on the disease prediction model's prediction results, comparing the differences between the validation set prediction results and the real labels to evaluate the model's generalization ability. Based on the evaluation results, key model parameters are adjusted. When the model's performance deteriorates on the validation set, adjustments are made through limiting... To improve model complexity or adjust training strategies and adaptability to new data, after multiple rounds of training and validation, the disease prediction model is evaluated using a test set until it meets preset requirements. Then, the performance of disease prediction models built with different algorithms is evaluated, and the performance of each model in different diagnostic tasks is analyzed. The prediction results of disease prediction models built with different algorithms are then weighted and fused to generate a comprehensive prediction result. The multimodal feature vectors of each newborn are input into the trained disease prediction model. Based on the feature patterns learned during training, the disease prediction model calculates the probability of disease occurrence for each newborn and classifies different types of defects during the calculation process. Simultaneously, based on the prediction results of the disease prediction model, defect types with matching degrees exceeding a preset range are identified and their corresponding confidence information is marked. After obtaining the corresponding disease probability values ​​and defect type judgment results, the risk of each newborn is classified according to preset rules.

[0025] Example 2, Figure 1 and Figure 3 The generative AI diagnosis and treatment platform that integrates gene data and clinical images includes a data acquisition module, a desensitization and protection module, a data processing module, a feature extraction module, a gene analysis module, a fusion analysis module, a predictive analysis module, a treatment plan generation module, a result display module, a model management module, a self-iteration module, and a user management module. The treatment plan generation module generates corresponding individualized treatment plans based on the prediction results of each newborn.

[0026] Specifically, the genetic characteristics of each newborn are compared with known pathogenic genes in a specialized disease gene database to identify whether each newborn has genetic susceptibility factors or a special genetic background. Simultaneously, family genetic history information is used to assess potential genetic transmission risks. Then, the imaging characteristics of each newborn are jointly interpreted with known clinical laboratory indicators to analyze whether structural or morphological abnormalities reflected in the imaging characteristics corroborate clinical laboratory indicators, determine the disease development stage, obtain the current status of each newborn, and match it with a built-in diagnostic and treatment rule base and medical knowledge base. Based on the defect type, risk level, and current status of each newborn, diagnostic and treatment pathways and intervention frameworks with a matching degree exceeding a preset threshold are selected from the diagnostic and treatment rule base, and combined with… The genetic risk outcomes and imaging results of each newborn are used to adjust the priority, intensity, and timing of interventions, resulting in corresponding individualized intervention plans. Based on these intervention plans, corresponding follow-up strategies are developed, and the follow-up frequency, content, and key monitoring indicators are determined according to each newborn's disease risk level and current status. The next assessment time is also specified, with increased follow-up density for patients with the highest risk level. Based on the intervention plans and follow-up strategies for each newborn, the corresponding prognoses are assessed. By analyzing the multimodal characteristics of current newborns and the outcome data of similar historical cases, the disease progression trend of each newborn under different intervention conditions is predicted. After all analyses are completed, the final treatment recommendations are structured and organized.

[0027] The results display module is used to centrally display the analysis results, prediction results, and individualized treatment plans for each newborn; the model management module is used to manage the disease prediction models built within the platform and continuously evaluate the performance of the corresponding models through different datasets.

[0028] The self-iteration module is used to continuously collect the diagnosis and treatment data of each newborn, and to continuously train and correct the parameters of the disease prediction model based on the real-time diagnosis and treatment data.

[0029] Specifically, clinical feedback data from each newborn is continuously collected. Data below the preset standard is filtered and removed. Then, each clinical feedback data is correlated and matched with the corresponding newborn's historical data. Using each newborn's unique identifier and treatment timeline as the core, each clinical feedback data is mapped to the corresponding original multimodal features, prediction results, and treatment plans, forming a complete data link. Each clinical feedback data is used as new label information and incorporated into the training sample pool. While retaining the original training data, the newly added feedback samples are integrated with historical samples to construct an incremental training dataset. After the incremental training dataset is prepared, the disease prediction model is trained again, fine-tuning the original parameters and correcting prediction biases using new training samples. After completing one incremental training iteration, the updated model's performance is evaluated. By comparing the prediction results, stability, and error changes before and after the model update, it is determined whether the model achieves performance improvement in real clinical scenarios and whether there are risks of performance degradation or instability. Once the model performance evaluation results meet the preset standards, the updated model is stored and labeled as a new version. While retaining historical model versions, the new model is deployed to the actual application environment.

[0030] The user management module is used to uniformly configure the usage permissions, project size, sample size, and data storage cycle of different users.

Claims

1. A generative AI diagnosis and treatment platform that fuses gene data with clinical images, characterized in that, The system comprises a data collection module, a desensitization protection module, a data processing module, a feature extraction module, a gene analysis module, a fusion analysis module, a prediction analysis module, a scheme generation module, a result display module, a model management module, a self-iteration module, and a user management module. The data collection module is configured to collect multi-modal data of each newborn and synchronously upload the multi-modal data to a corresponding mobile terminal and a medical terminal. The desensitization protection module is configured to perform desensitization processing on sensitive information of each newborn. The data processing module is configured to perform cleaning and standardization processing on the collected multi-modal data. The feature extraction module is configured to perform feature extraction on medical image data in the multi-modal data. The gene analysis module is configured to analyze genetic information data of each newborn based on a bioinformatics analysis method. The fusion analysis module is configured to construct a multi-modal data expression system and synchronously analyze multi-modal features of different sources. The prediction analysis module is configured to jointly model the extracted multi-modal features, predict a disease occurrence probability of each newborn, and obtain a risk classification result. The scheme generation module is configured to generate a corresponding individualized diagnosis and treatment scheme according to the prediction result of each newborn. The result display module is configured to centrally display analysis results, prediction results, and individualized diagnosis and treatment schemes of each newborn. The model management module is configured to manage a disease prediction model constructed in the platform and continuously evaluate performance of the corresponding model through different data sets. The self-iteration module is configured to continuously collect diagnosis and treatment data of each newborn, and continuously train and correct parameters of the disease prediction model according to real-time diagnosis and treatment data. The user management module is configured to uniformly configure usage permissions, project sizes, sample capacities, and data storage periods of different users. 2.The fusion gene data and clinical image generative AI diagnosis and treatment platform of claim 1, wherein, The gene analysis module analyzes genetic information data of each newborn in the following specific steps: S1.1: Collect multi-modal data of each newborn, classify and manage medical image data in the multi-modal data according to a unique identifier corresponding to each newborn, uniformly set an image gray scale range of each medical image data, then locate and segment a preset target region in each medical image data, complete the region segmentation, extract features of each target region, obtain corresponding clinical data and medical history data in the multi-modal data, and extract corresponding features, wherein the multi-modal data comprises genetic and genetic information data, medical image data, clinical data, medical history and perinatal data, and development evaluation and follow-up data; S1.2: Obtain genetic information data in the multi-modal data of each newborn, obtain corresponding gene sequence data, gene expression data, and chromosome abnormality data in the genetic information data, collect base quality of each gene sequence data, remove gene sequence data with base quality lower than a preset quality, uniformly set a dimension and a value range of each gene expression data, and then uniformly map each chromosome abnormality data to a preset structure. S1.3: Extract the features of each type of genetic data after processing, and encode all genetic features into a structured feature vector to form a corresponding genetic feature set. Then, by analyzing the correlation between different genetic features and the degree of association with disease phenotypes, identify genetic features with a diagnostic value below a predetermined standard, and simultaneously construct and train a genetic reasoning model; S1.4: Input all extracted genetic features into the trained genetic reasoning model. The genetic reasoning model outputs the corresponding disease risk assessment results based on the input genetic features. Then, based on all genetic features, analysis results, and corresponding disease phenotype information, construct a disease-specific gene database. 3.The fusion gene data and clinical image generative AI diagnosis and treatment platform of claim 2, wherein, The specific steps of the fusion analysis module for synchronously analyzing multi-modal features from different sources are as follows: S2.1: Extract each modality feature and confirm the structure of each type of feature. Then, divide the semantic levels of each modality feature and assign a clear modality identifier to each type of feature. After that, unify the numerical range of each modality feature while preserving the relative change relationship of each modality feature. Then, align each modality feature according to the unique identifier of the corresponding newborn, and map different modality features to the preset time window by introducing time synchronization rules; S2.2: Mine the potential associations between each modality feature, and establish the corresponding mapping relationship based on the mining results. Assign the corresponding initial weight based on the importance of different modality features in disease diagnosis, and combine each modality feature with a weight; S2.3: Check the feature dimension, numerical range, missing condition, and modality identifier integrity of each modality feature. Then, based on the modality features that pass the test, construct a multi-modal data expression system to form the corresponding multi-modal feature vector. 4.The fusion gene data and clinical image generative AI diagnosis and treatment platform of claim 3, wherein, The specific steps of the prediction analysis module for predicting the disease occurrence probability and risk classification results of each newborn are as follows: S3.1: Based on each type of feature, form a corresponding multi-modal feature sample set, and establish a mapping relationship with the corresponding true diagnosis result or disease label. Then, divide the multi-modal feature sample set into training set, validation set, and test set according to the preset proportion. According to the feature dimension, sample size, and task type, select the corresponding model structure to establish a disease prediction model, and set the initial parameters of the disease prediction model; S3.2: Input the training samples into the disease prediction model in batches. The model predicts the samples based on the current parameters, and compares the prediction results with the true disease labels. Then, calculate the adjustment direction of the model based on the prediction error, and update the model parameters step by step. During the training process, after each round of training, the validation set analyzes the prediction results of the disease prediction model, compares the differences between the validation set prediction results and the true labels, evaluates the generalization ability of the model, and adjusts the key parameters of the model based on the evaluation results; S3.3: When the performance of the model is found to be declining on the validation set, adjust the model's ability to adapt to new data by limiting the model's complexity or adjusting the training strategy. After completing multiple rounds of training and validation, evaluate the disease prediction model on the test set until the model meets the preset requirements. Then, evaluate the performance of the disease prediction models constructed by different algorithms, analyze the performance of each model in different diagnosis tasks, and then weight and fuse the prediction results of the disease prediction models constructed by different algorithms to generate the corresponding comprehensive prediction results; S3.4: Input the multi-modal feature vectors of each newborn into the trained disease prediction model. The disease prediction model calculates the probability value of each newborn's disease occurrence based on the feature patterns learned during training. During the calculation, different types of defects are classified and judged. According to the prediction results of the disease prediction model, identify defects whose matching degree exceeds the preset range and mark the corresponding confidence information. After obtaining the corresponding disease probability value and defect type judgment result, classify the risk of each newborn according to the preset rules. 5.The fusion gene data and clinical image generative AI diagnosis and treatment platform of claim 4, wherein, The specific steps for the scheme generation module to generate the corresponding individualized diagnosis and treatment scheme are as follows: S4.1: Compare the genetic characteristics of each newborn with the known pathogenic genes in the disease-specific gene database to identify whether each newborn has genetic predisposing factors or special genetic backgrounds. Meanwhile, combined with the family genetic history information of each newborn, evaluate the potential genetic transmission risk. Then, jointly interpret the image features of each newborn with known clinical test indicators to analyze whether the structural or morphological abnormalities reflected in each image feature are mutually confirmed with the clinical test indicators, and determine the development stage of the disease; S4.2: Obtain the current state of each newborn and match it with the built-in diagnosis and treatment rule library and medical knowledge base. According to the defect type judgment result, risk level and current state of each newborn, select the diagnosis and treatment path and intervention scheme framework with a matching degree exceeding the preset threshold from the diagnosis and treatment rule library. Adjust the priority, intensity and timing of intervention measures based on the genetic risk results and image performance results of each newborn to form the corresponding individualized intervention scheme; S4.3: Based on the intervention scheme generation of each newborn, develop the corresponding follow-up strategy, and determine the follow-up frequency, follow-up content and key monitoring indicators according to the disease risk level and current state of each newborn. Determine the next evaluation time node, and increase the follow-up density for patients with the highest risk level; S4.4: Based on the intervention scheme and follow-up strategy of each newborn, evaluate the corresponding prognosis. By analyzing the multi-modal features of each newborn and the outcome data of similar historical cases, predict the disease development trend of each newborn under different intervention conditions. After all the analysis is completed, structure the final diagnosis and treatment recommendations. 6.The fusion gene data and clinical image generative AI diagnosis and treatment platform of claim 5, wherein, The specific steps for the self-iteration module to continuously train and correct the disease prediction model are as follows: S5.1: Continuously collect clinical feedback data from each newborn, filter out and remove clinical feedback data that is below the preset standard, then associate and match each clinical feedback data with the historical data of the corresponding newborn, and then, with the unique identifier and treatment timeline of each newborn as the core, match each clinical feedback data with the corresponding original multimodal features, prediction results and treatment plans to form a complete data link; S5.2: The clinical feedback data are used as new label information and included in the training sample pool. While retaining the original training data, the newly added feedback samples are integrated with the historical samples to construct an incremental training dataset. After the incremental training dataset is prepared, the disease prediction model is trained again to fine-tune the original parameters of the disease prediction model. At the same time, the prediction bias of the model is corrected through the new training samples. S5.3: After completing an incremental training, the updated model is evaluated for performance. By comparing the prediction results, stability and error changes before and after the model update, it is determined whether the model has achieved performance improvement in real clinical scenarios and whether there is a risk of performance decline or instability. After that, when the model performance evaluation results meet the preset standards, the updated model is stored and identified as a new version. While retaining the historical model version, the new model is deployed to the actual application environment.

Images