Disease prediction method and apparatus, device, storage medium and program product

By dual training of the initial base model and the prediction model, the problem of insufficient accuracy in disease prediction methods is solved, and efficient and accurate disease prediction is achieved.

WO2026237121A1PCT designated stage Publication Date: 2026-11-19TSINGHUA UNIVERSITY +1
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
PCT/CN2025/099807
Authority / Receiving Office
WO · WO
Patent Type
Applications
Current Assignee / Owner
Priority Date
2025-05-14
Filing Date
2025-06-09
Publication Date
2026-11-19

AI Technical Summary

Technical Problem

Existing disease prediction methods lack accuracy in predicting disease progression.

Method used

By training the initial pedestal model, medical features of medical images are extracted. The initial prediction model is then trained based on the trained pedestal model and a second set of medical images carrying time information, resulting in a prediction model with disease prediction capabilities. This reduces the reliance on disease type labels carried by medical images and improves the training efficiency and prediction accuracy of the model.

Benefits of technology

This greatly improves the accuracy and training efficiency of disease prediction, reduces the workload of labeling, and enhances the model's generalization ability and prediction stability.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN2025099807_19112026_PF_FP_ABST
    Figure CN2025099807_19112026_PF_FP_ABST
Patent Text Reader

Abstract

A disease prediction method and apparatus, a device, a storage medium and a program product. The method comprises: on the basis of a first medical image set, training an initial foundation model to obtain a trained foundation model; on the basis of the trained foundation model and a second medical image set, training an initial prediction model to obtain a prediction model, the prediction model being used for predicting a disease, the second medical image set comprising: for the same subject, a second medical image collected at a first moment, a third medical image collected at a second moment, and a time interval between the first moment and the second moment, the trained foundation model being used for extracting medical features of medical images, and the first medical image set comprising a plurality of first medical images; acquiring longitudinal health data of a target subject and, on the basis of the longitudinal health data and the prediction model, predicting a development trend of a disease of the target subject.
Need to check novelty before this filing date? Find Prior Art

Description

Disease prediction methods, devices, equipment, storage media, and program products

[0001] Related applications

[0002] This application claims priority to Chinese patent application filed on May 14, 2025, with application number 202510621904.4, entitled “Training method, apparatus, device, storage medium and program product for disease prediction model”, the entire contents of which are incorporated herein by reference. Technical Field

[0003] This application relates to the field of artificial intelligence technology, and in particular to a disease prediction method, apparatus, device, storage medium, and program product. Background Technology

[0004] Longitudinal prediction refers to the technique of predicting disease progression and patient survival time by analyzing a patient's longitudinal health data (such as medical records, examination results, treatment process, etc.).

[0005] In recent years, the development of artificial intelligence has brought hope for addressing these complex challenges in the global healthcare field. Combining AI-trained models to enable longitudinal disease prediction capabilities has become an important research direction in medicine. Typical models include OpenAI's Generative Pre-trained Transformer (GPT) based on the Transformer architecture, Google's Bidirectional Encoder Representations from Transformers (BERT) and Text-to-Text Transfer Transformer (T5) pre-trained language models, etc.

[0006] However, current disease prediction methods still have limitations in predicting disease progression and suffer from inaccurate predictions. Summary of the Invention

[0007] Therefore, it is necessary to provide a disease prediction method, device, equipment, storage medium, and program product that can improve the accuracy of disease prediction in response to the above-mentioned technical problems.

[0008] In a first aspect, this application provides a disease prediction method, comprising: training an initial base model based on a first medical image set to obtain a trained base model; the trained base model is used to extract medical features from the medical images; the first medical image set includes multiple first medical images; training an initial prediction model based on the trained base model and a second medical image set to obtain a prediction model; the prediction model is used to predict diseases; the second medical image set includes: a second medical image acquired at a first time moment for the same object, a third medical image acquired at a second time moment, and the time interval between the first and second time moments; acquiring longitudinal health data of the target object, and predicting the disease development trend of the target object based on the longitudinal health data and the prediction model.

[0009] In one embodiment, the above-mentioned training of the initial pedestal model based on the first medical image set to obtain the trained pedestal model includes: performing a first masking process on the first medical image in the first medical image set to obtain a masked first medical image; and training the initial pedestal model based on the first medical image and the masked first medical image to obtain the trained pedestal model.

[0010] In one embodiment, the initial base model includes a first initial encoder and a first initial decoder. The initial base model is trained based on a first medical image and a masked first medical image to obtain a trained base model. The training process includes: inputting the masked first medical image into the first initial encoder for encoding to obtain a first encoded feature; inputting the first encoded feature into the first initial decoder for decoding to obtain a first decoded feature; and training the first initial encoder and the first initial decoder based on a first difference between the image features of the first medical image and the first decoded feature to obtain the trained base model.

[0011] In one embodiment, the above-mentioned training of the initial prediction model based on the trained base model and the second medical image set to obtain the prediction model includes: inputting a third medical image from the second medical image set into the trained base model for feature extraction to obtain the features of the third medical image; and training the initial prediction model based on the features of the third medical image and the second medical image to obtain the prediction model.

[0012] In one embodiment, the above-mentioned training of the initial prediction model based on the features of the third medical image and the second medical image to obtain the prediction model includes: performing a second masking process on the second medical image to obtain a masked second medical image; and training the initial prediction model based on the features of the masked second medical image and the third medical image, as well as the time interval between the first time and the second time to obtain the prediction model.

[0013] In one embodiment, the initial prediction model includes a second initial encoder and a second initial decoder. The initial prediction model is trained based on the features of the masked second medical image and the third medical image, as well as the time interval between the first and second time moments, to obtain the prediction model. This includes: inputting the masked second medical image and the time interval into the second initial encoder for encoding to obtain second encoded features; inputting the second encoded features into the second initial decoder for decoding to obtain second decoded features; and training the second initial encoder and the second initial decoder based on the features of the third image and the second decoded features to obtain the prediction model.

[0014] In one embodiment, the initial base model is a masked autoencoder (MAE) model, and the initial prediction model is a RETFound model.

[0015] In one embodiment, the first medical image is a medical image that does not carry information about the image acquisition time or the disease type corresponding to the image.

[0016] In one embodiment, training an initial pedestal model based on a first medical image set to obtain a trained pedestal model includes: pre-training the initial pedestal model on an ImageNet dataset to enable the initial pedestal model to learn common image structures and features, thereby obtaining a pre-trained initial pedestal model; and training the pre-trained initial pedestal model based on the first medical image set to obtain a trained pedestal model.

[0017] In one embodiment, the first medical image in the first medical image set is subjected to a first masking process to obtain a masked first medical image, including: segmenting each first medical image in the first medical image set to obtain multiple segmented images corresponding to the first medical image; randomly performing a first masking process on a portion of the multiple segmented images to obtain each masked first medical image.

[0018] In one embodiment, the first encoded feature refers to the features of the unmasked portion of the masked first medical image; the first decoded feature refers to all the features of the first medical image inferred from the first encoded feature.

[0019] In one embodiment, training the first initial encoder and the first initial decoder based on the first difference between the image features of the first medical image and the first decoding features to obtain the trained base model includes: determining whether the first difference is less than or equal to a first preset difference threshold based on the first difference between the image features of the first medical image and the first decoding features; if the first difference is greater than the first preset difference threshold, adjusting the model parameters of the first initial encoder and the first initial decoder to obtain a new first initial encoder and a new first initial decoder, and returning to execute the input of the masked first medical image into the first initial encoder for encoding processing to obtain the first encoded features, until the first difference is less than or equal to the first preset difference threshold to obtain the trained base model.

[0020] Secondly, this application also provides a disease prediction device, including a first training module, a second training module, and a prediction module.

[0021] The first training module is used to train the initial pedestal model based on the first medical image set to obtain the trained pedestal model; the trained pedestal model is used to extract medical features from the medical images; the first medical image set includes multiple first medical images.

[0022] The second training module is used to train the initial prediction model based on the trained base model and the second medical image set to obtain the prediction model; the prediction model is used to predict diseases; the second medical image set includes: a second medical image acquired at a first time point for the same object, a third medical image acquired at a second time point, and the time interval between the first and second time points.

[0023] The prediction module is used to acquire longitudinal health data of the target object and predict the disease development trend of the target object based on the longitudinal health data and the prediction model.

[0024] Thirdly, this application also provides a computer device, including a memory and a processor. The memory stores a computer program, and the processor executes the computer program to perform the following steps: training an initial base model based on a first set of medical images to obtain a trained base model; the trained base model is used to extract medical features from the medical images; the first set of medical images includes multiple first medical images; training an initial prediction model based on the trained base model and a second set of medical images to obtain a prediction model; the prediction model is used to predict diseases; the second set of medical images includes: a second medical image acquired at a first time moment for the same object, a third medical image acquired at a second time moment, and the time interval between the first and second times moment; acquiring longitudinal health data of the target object, and predicting the disease development trend of the target object based on the longitudinal health data and the prediction model.

[0025] Fourthly, this application also provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, performs the following steps: training an initial pedestal model based on a first medical image set to obtain a trained pedestal model; the trained pedestal model is used to extract medical features from the medical images; the first medical image set includes multiple first medical images; training an initial prediction model based on the trained pedestal model and a second medical image set to obtain a prediction model; the prediction model is used to predict diseases; the second medical image set includes: a second medical image acquired at a first time moment for the same object, a third medical image acquired at a second time moment, and the time interval between the first and second time moments; acquiring longitudinal health data of the target object, and predicting the disease development trend of the target object based on the longitudinal health data and the prediction model.

[0026] Fifthly, this application also provides a computer program product, including a computer program that, when executed by a processor, performs the following steps: training an initial base model based on a first medical image set to obtain a trained base model; the trained base model is used to extract medical features from the medical images; the first medical image set includes multiple first medical images; training an initial prediction model based on the trained base model and a second medical image set to obtain a prediction model; the prediction model is used to predict diseases; the second medical image set includes: a second medical image acquired at a first time moment for the same object, a third medical image acquired at a second time moment, and the time interval between the first and second time moments; acquiring longitudinal health data of the target object, and predicting the disease development trend of the target object based on the longitudinal health data and the prediction model.

[0027] The aforementioned disease prediction method, apparatus, device, storage medium, and program product train an initial base model based on a first medical image set to obtain a base model with simple medical image recognition features. Then, based on the trained base model and a second medical image set carrying time information, the initial prediction model is trained to obtain a prediction model with disease prediction function. Longitudinal health data of the target object is acquired, and based on the longitudinal health data and the prediction model, the disease development trend of the target object is predicted. Compared with existing methods that train disease prediction models based on image labels to predict diseases, this scheme eliminates the need for medical images to carry disease type labels during the entire model training process, reducing the workload of label annotation and greatly improving the model training efficiency. Furthermore, the two training iterations significantly improve the accuracy of the model in predicting diseases. Attached Figure Description

[0028] To more clearly illustrate the technical solutions in the embodiments of this application or related technologies, the drawings used in the description of the embodiments of this application or related technologies will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0029] Figure 1 is an application environment diagram of a disease prediction method in one embodiment;

[0030] Figure 2 is a flowchart illustrating a disease prediction method in one embodiment;

[0031] Figure 3 is a flowchart illustrating the disease prediction method in another embodiment;

[0032] Figure 4 is a flowchart illustrating the disease prediction method in another embodiment;

[0033] Figure 5 is a flowchart illustrating the disease prediction method in another embodiment;

[0034] Figure 6 is a flowchart illustrating a disease prediction method in another embodiment;

[0035] Figure 7 is a flowchart illustrating the disease prediction method in another embodiment;

[0036] Figure 8 is a flowchart illustrating the disease prediction method in another embodiment;

[0037] Figure 9 is a structural block diagram of a disease prediction device in one embodiment. Detailed Implementation

[0038] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0039] Longitudinal prediction refers to the technique of predicting disease progression and patient survival time by analyzing a patient's longitudinal health data (such as medical records, examination results, treatment process, etc.).

[0040] In recent years, the development of artificial intelligence has brought hope for addressing these complex challenges in the global healthcare field. Combining AI-trained models to enable longitudinal disease prediction capabilities has become an important research direction in medicine. Typical models include OpenAI's Generative Pre-trained Transformer (GPT) based on the Transformer architecture, Google's Bidirectional Encoder Representations from Transformers (BERT) and Text-to-Text Transfer Transformer (T5) pre-trained language models, etc.

[0041] However, current methods for predicting disease progression still have limitations and suffer from inaccurate predictions. This application aims to address this problem.

[0042] After introducing the background technology of the disease prediction method provided in the embodiments of this application, the implementation environment involved in the disease prediction method provided in the embodiments of this application will be briefly described below. The disease prediction method provided in the embodiments of this application can be applied to a computer device as shown in FIG1. ​​The computer device includes a processor, a memory, an input / output interface, a communication interface, a display unit, and an input device. The processor, memory, and input / output interface are connected through a system bus, and the communication interface, display unit, and input device are connected to the system bus through the input / output interface. The processor of the computer device provides computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores an operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The input / output interface of the computer device is used for exchanging information between the processor and external devices. The communication interface of the computer device is used for wired or wireless communication with external terminals; wireless communication can be achieved through WIFI, mobile cellular networks, NFC (Near Field Communication), or other technologies. When the computer program is executed by the processor, it implements a disease prediction method. The display unit of this computer device is used to form a visually visible image and can be a display screen, a projection device, or a virtual reality imaging device. The display screen can be an LCD screen or an e-ink screen. The input device of this computer device can be a touch layer covering the display screen, or buttons, a trackball, or a touchpad set on the casing of the computer device, or an external keyboard, touchpad, or mouse, etc.

[0043] Those skilled in the art will understand that the structure shown in Figure 1 is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. A specific terminal may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0044] After introducing the application scenarios of the disease prediction method provided in the embodiments of this application above, the disease prediction method described in this application will be described in detail below.

[0045] In one embodiment, as shown in FIG2, a disease prediction method is provided, which is illustrated by taking the application of the method to the computer device in FIG1 as an example, and includes the following steps: S201 to S203.

[0046] S201. Based on the first medical image set, train the initial pedestal model to obtain the trained pedestal model.

[0047] The first medical image set includes multiple first medical images, which are medical images that do not carry information about the image acquisition time or the corresponding disease type. The initial pedestal model can be a masked autoencoder (MAE) model, a visual model based on self-supervised learning (SSL). The trained pedestal model can extract medical features such as edge detection, color contrast, and texture from the medical images.

[0048] In this embodiment, when it is necessary to train a prediction model to enable it to predict diseases, a base model can be trained first, enabling the base model to extract medical features from medical images. Before training the base model, a first medical image set including multiple first medical images can be obtained, and the initial base model can be trained based on the first medical image set to obtain the trained base model.

[0049] In one embodiment, the initial pedestal model can be trained based on a first medical image set using methods such as self-training, co-training, image-based semi-supervised learning, generative adversarial networks, variational autoencoders, cluster analysis, autoencoders, principal component analysis, and masking to obtain the trained pedestal model.

[0050] In this embodiment, before processing the first medical images in the first medical image set, it is also necessary to obtain the first medical image set. For example, the first medical image set can be millions of fundus image data.

[0051] It should be noted that, based on the first medical image set, the initial pedestal model is trained to obtain the trained pedestal model. This includes first pre-training the initial pedestal model (i.e., the MAE model) on the ImageNet dataset, allowing the MAE model to learn general image structures and features, and then training the pre-trained initial pedestal model based on the first medical image set to obtain the trained pedestal model.

[0052] S202. Based on the trained pedestal model and the second medical image set, the initial prediction model is trained to obtain the prediction model.

[0053] The second medical image set includes: a second medical image of the same object acquired at a first time point, a third medical image acquired at a second time point, and the time interval between the first and second time points. It should be noted that the second medical image set may also include: a fourth medical image of the same object acquired at a third time point, the time interval between the first and third time points, and the time interval between the second and third time points.

[0054] The prediction model is used to predict diseases. It should be noted that when training the prediction model, if the second and third medical images from the second medical image set are used, the trained prediction model will be able to predict diseases occurring after the time interval between the first and second time points. For example, if the currently acquired medical image is input into the prediction model for disease prediction, the type of disease occurring after the time interval between the first and second time points and the probability of that type of disease occurring can be obtained.

[0055] It should be noted that when training the prediction model, if the second and fourth medical images from the second medical image set are used, the trained prediction model will have the function of predicting diseases after the time interval between the first and third time points. For example, if the currently acquired medical image is input into the prediction model for disease prediction, the type of disease that occurs after the time interval between the first and third time points and the probability of that type of disease occurring can be obtained.

[0056] In one embodiment, after obtaining the trained base model, the second and third medical images in the second medical image set can be input into the initial prediction model to train the initial prediction model and obtain the trained prediction model.

[0057] In one embodiment, the initial prediction model can be trained based on a second set of medical images using methods such as self-training, co-training, image-based semi-supervised learning, generative adversarial networks, variational autoencoders, cluster analysis, autoencoders, principal component analysis, and masking to obtain the trained prediction model.

[0058] S203, Obtain longitudinal health data of the target object, and predict the disease development trend of the target object based on the longitudinal health data and the prediction model.

[0059] The target group is patients. Longitudinal health data of patients is acquired, such as medical records, examination results, and treatment processes. This data is then input into a predictive model to obtain predictions of the disease progression trend of the target group.

[0060] In the embodiments of this application, an initial base model is trained based on a first set of medical images to obtain a base model with simple medical features for recognizing medical images. Then, based on the trained base model and a second set of medical images carrying time information, an initial prediction model is trained to obtain a prediction model with disease prediction capabilities. Finally, longitudinal health data of the target object is acquired, and the disease development trend of the target object is predicted based on the longitudinal health data and the prediction model. Compared with existing methods of training disease prediction models based on image labels, this scheme eliminates the need for medical images to carry disease type labels during the entire model training process, reducing the workload of label annotation and greatly improving the training efficiency of the model. In addition, training the model twice greatly improves the accuracy of the model in predicting diseases.

[0061] In one embodiment, based on the embodiment shown in Figure 2, the process of training the pedestal model can be specifically implemented. As shown in Figure 3, the above-mentioned S201 "training the initial pedestal model according to the first medical image set to obtain the trained pedestal model" includes steps S301 and S302.

[0062] S301. Perform a first masking process on the first medical image in the first medical image set to obtain the masked first medical image.

[0063] Masking refers to the process of occluding a portion of an image.

[0064] In this embodiment, the initial pedestal model is trained based on a self-training method and the first medical images in the first medical image set. Before training, the first medical images need to be masked to obtain the masked first medical images.

[0065] In one embodiment, the first medical image in the first medical image set can be segmented to obtain multiple segmented images corresponding to the first medical image. Then, a portion of the segmented images from the multiple segmented images can be randomly subjected to a first masking process to obtain the masked first medical image. For example, randomly occluding 75% of the segmented images from the multiple segmented images yields a first medical image that retains 25% of the image information.

[0066] S302. Based on the first medical image and the masked first medical image, train the initial pedestal model to obtain the trained pedestal model.

[0067] In this embodiment, after obtaining the masked first medical image, the initial pedestal model can be trained based on the first medical image and the masked first medical image to obtain the trained pedestal model.

[0068] In one embodiment, the initial pedestal model includes a first initial encoder and a first initial decoder. Referring to Figure 4, a method is provided for training the initial pedestal model based on a first medical image and a masked first medical image, namely, the above-mentioned S302 "training the initial pedestal model based on the first medical image and the masked first medical image to obtain the trained pedestal model", including steps S3021 to S3023.

[0069] S3021. The masked first medical image is input into the first initial encoder for encoding processing to obtain the first encoded feature.

[0070] In this context, encoding refers to a method of encoding images, with different images corresponding to different encoding features. Furthermore, different masked first medical images will have different corresponding first encoding features.

[0071] The first encoded feature refers to the feature of the unmasked portion in the first medical image after masking.

[0072] In this embodiment, after obtaining the masked first medical image, the masked first medical image can be input into the first initial encoder in the initial base model for encoding processing to obtain the first encoded feature.

[0073] S3022. Input the first encoded feature into the first initial decoder for decoding processing to obtain the first decoded feature.

[0074] The first decoding feature refers to all the features of the first medical image inferred from the first encoding feature.

[0075] In this embodiment, after obtaining the first encoded feature, the first encoded feature can be input into the first initial decoder for decoding processing, and all features of the first medical image can be inferred, namely the first decoded feature.

[0076] S3023. Based on the first difference between the image features and the first decoding features of the first medical image, train the first initial encoder and the first initial decoder to obtain the trained base model.

[0077] The image features of the first medical image refer to the image features after feature extraction of the first medical image.

[0078] The first difference can be the pixel-level reconstruction loss.

[0079] In this embodiment, after obtaining the first decoded feature, feature extraction can be performed on the first medical image to obtain the image features of the first medical image. A first difference between the image features of the first medical image and the first decoded feature is determined, and it is determined whether the first difference is less than or equal to a first preset difference threshold. If the first difference is greater than the first preset difference threshold, the model parameters of the first initial encoder and the first initial decoder are adjusted to obtain a new first initial encoder and a new first initial decoder. The process then returns to the step of inputting the masked first medical image into the first initial encoder for encoding processing to obtain the first encoded feature, until the first difference is less than or equal to the first preset difference threshold, thus obtaining the trained base model.

[0080] In one embodiment, the method of extracting features from the first medical image may include: inputting the first medical image into a pre-trained image feature extraction network for feature extraction to obtain the image features of the first medical image.

[0081] In the embodiments of this application, by training the first initial encoder and the first initial decoder in the initial pedestal model, the trained pedestal model can learn the global and local features of the image more efficiently, while reducing the computational burden. Furthermore, this pedestal model training method does not require pre-annotation of the image; it only requires partial masking of the image, followed by sequential input of the masked image into the first initial encoder and the first initial decoder to train the pedestal model. This significantly reduces the workload of image annotation and improves the training efficiency of the pedestal model.

[0082] In one embodiment, based on the embodiments shown in Figures 2-4, the training process of the prediction model can be specifically implemented. As shown in Figure 5, the above-mentioned S202 "training the initial prediction model based on the trained base model and the second medical image set to obtain the prediction model" includes steps S401 and S402.

[0083] S401. Input the third medical image from the second medical image set into the trained base model for feature extraction to obtain the features of the third medical image.

[0084] In the embodiments of this application, to optimize the quality of input data, the images in the second medical image set undergo a series of preprocessing steps before entering the model. These include background removal (removing black backgrounds to increase the focus on lesion areas), downsampling (uniformly adjusting the image size to 256×256 to improve computational efficiency), and a series of data augmentation techniques, such as random cropping, resizing, horizontal flipping (to improve the robustness of the model), color jittering (to enhance the model's adaptability to images captured by different devices), and Gaussian blur (to simulate low-quality images in reality). These preprocessing steps effectively improve the model's generalization ability, enabling it to better adapt to image data from different medical institutions and improve the stability and accuracy of predictions.

[0085] In this embodiment, after obtaining the trained base model, the third medical image in the second medical image set can be input into the trained base model for feature extraction to obtain the features of the third medical image.

[0086] In embodiments of this application, before processing the second medical image set, it is necessary to acquire the second medical image set. For example, the second medical image set may be a longitudinal image set containing 1,304,292 fundus images, covering 304,345 patients, with the images collected over a period of 9 years. In one embodiment, after acquiring the second medical image set, the individual medical images in the second medical image set can be grouped according to patient identifiers, and the acquisition time of each medical image can be labeled. In one embodiment, an automated quality control (QC) process can also be used to remove low-quality images (e.g., blurry images, overexposed images, images with excessive artifacts, etc.) and unify the size of all medical images in the second medical image set (e.g., adjust the size of all medical images in the second medical image set to 256×256 pixels).

[0087] S402. Based on the features of the third medical image and the second medical image, train the initial prediction model to obtain the prediction model.

[0088] Among them, the third medical image and the second medical image are fundus images of the same subject acquired at different times.

[0089] In one embodiment, before obtaining the features of the third medical image, a second medical image can be obtained and processed. Based on the difference between the features of the processed second medical image and the features of the third medical image, an initial prediction model can be trained to obtain a prediction model.

[0090] In this embodiment, an initial prediction model is trained based on medical images of the same object at different times, enabling the trained prediction model to predict future disease trends. Furthermore, feature extraction is performed on a third medical image based on the trained base model, making the features of the acquired third medical image more accurate, thereby improving the accuracy of the subsequently trained prediction model.

[0091] In one embodiment, based on the embodiment shown in Figure 5, the training process of the prediction model can be further implemented in detail. As shown in Figure 6, the above-mentioned S402 "training the initial prediction model according to the features of the third medical image and the second medical image to obtain the prediction model" includes steps S501 and S502.

[0092] S501. Perform a second masking process on the second medical image to obtain the masked second medical image.

[0093] In this embodiment, the initial prediction model is trained based on the self-training method and the second medical images in the second medical image set. Before training, the second medical images need to be processed by a second masking to obtain the masked second medical images.

[0094] In one embodiment, the second medical images in the second medical image set can be segmented first to obtain multiple segmented images corresponding to the second medical image. Then, a portion of the segmented images from the multiple segmented images can be randomly subjected to a second masking process to obtain a masked second medical image. For example, randomly occluding 75% of the segmented images from the multiple segmented images yields a second medical image that retains 25% of the image information.

[0095] S502. Based on the features of the masked second and third medical images, and the time interval between the first and second time moments, the initial prediction model is trained to obtain the prediction model.

[0096] In this embodiment, after obtaining the features of the masked second and third medical images, the initial prediction model can be trained based on the features of the masked second and third medical images and the time interval between the first and second time moments to obtain the prediction model.

[0097] In one embodiment, the initial prediction model includes a second initial encoder and a second initial decoder. Referring to Figure 7, this application provides a method for training an initial prediction model based on the features of a masked second medical image and a third medical image, as well as the time interval between a first time moment and a second time moment. Specifically, step S502, "training the initial prediction model based on the features of the masked second medical image and the third medical image, as well as the time interval between the first time moment and the second time moment, to obtain a prediction model," includes steps S5021 to S5023.

[0098] S5021. Input the masked second medical image and time interval into the second initial encoder for encoding processing to obtain the second encoded feature.

[0099] The second medical images after masking are different, and the corresponding second coding features are also different.

[0100] The second encoded feature refers to the feature of the unmasked portion of the second medical image after the mask carrying the time interval.

[0101] In this embodiment, after obtaining the masked second medical image, the masked second medical image and the time interval can be input into the second initial encoder in the initial prediction model for encoding processing to obtain the second encoded feature.

[0102] In one embodiment, to enable the prediction model to learn the characteristics of disease changes over time, we introduce the sinusoidal timestep embedding method. This method can generate time information codes corresponding to time intervals based on sine and cosine functions, map the time information codes to a high-dimensional space, and then input the masked second medical image and the time information codes into the second initial encoder in the initial prediction model for encoding processing to obtain the second encoded features.

[0103] In one embodiment, to enhance the training model's understanding of temporal information encoding, RETFound Plus employs sinusoidal timestep embedding to map the time difference into a high-dimensional representation during the training process of the prediction model. This representation is then transformed using a multi-layer perceptron (MLP) to align with the hidden dimension of the prediction model. This temporal information encoding is incorporated into class labels (CLS), enabling the prediction model to effectively capture the characteristics of fundus images changing over time and improving the accuracy of disease progression prediction.

[0104] S5022. Input the second encoded feature into the second initial decoder for decoding processing to obtain the second decoded feature.

[0105] The second decoding feature refers to all the features of the second medical image inferred from the second encoding feature.

[0106] In this embodiment, after obtaining the second encoded feature, the second encoded feature can be input into the second initial decoder for decoding processing, and all features of the second medical image can be inferred, namely the second decoded feature.

[0107] S5023. Based on the second difference between the features of the third image and the second decoded features, train the second initial encoder and the second initial decoder to obtain the prediction model.

[0108] In this embodiment, after obtaining the features of the third image and the second decoding features, a second difference between the features of the third image and the second decoding features is determined, and it is determined whether the second difference meets a second preset difference threshold. If the second difference does not meet the second preset difference threshold, the model parameters of the second initial encoder and the second initial decoder are adjusted to obtain a new second initial encoder and a new second initial decoder. Then, the process of inputting the difference between the masked second image and the time information into the second initial encoder for encoding processing to obtain the second encoded features is returned until the second difference meets the second preset difference threshold, and the prediction model is obtained.

[0109] In one embodiment, the second difference may be obtained by comparing the cosine similarity between the features of the third image and the second decoded features.

[0110] In this embodiment, by training the second initial encoder and the second initial decoder in the initial prediction model, the trained prediction model is able to have disease prediction capabilities. Furthermore, this disease prediction method does not require pre-annotation of the images; it only requires partial masking of the images, followed by sequential input of the masked images into the second initial encoder and the second initial decoder to train the prediction model. This significantly reduces the workload of image annotation and improves the training efficiency of the prediction model.

[0111] In one embodiment, referring to Figure 8, a disease prediction method is also provided, including steps S10 to S19.

[0112] S10. Perform a first masking process on the first medical image in the first medical image set to obtain the masked first medical image.

[0113] S11. Input the masked first medical image into the first initial encoder for encoding processing to obtain the first encoded feature.

[0114] S12. Input the first encoded feature into the first initial decoder for decoding to obtain the first decoded feature.

[0115] S13. Based on the first difference between the image features and the first decoding features of the first medical image, train the first initial encoder and the first initial decoder to obtain the trained base model.

[0116] S14. Input the third medical image from the second medical image set into the trained base model for feature extraction to obtain the features of the third medical image.

[0117] S15. Perform a second masking process on the second medical image to obtain a masked second medical image; wherein, the third medical image and the second medical image are medical images of the same object acquired at different times.

[0118] S16. Input the masked second medical image and the time interval between the second and third medical images into the second initial encoder for encoding processing to obtain the second encoded feature.

[0119] S17. Input the second encoded feature into the second initial decoder for decoding to obtain the second decoded feature.

[0120] S18. Based on the second difference between the features of the third image and the second decoded features, train the second initial encoder and the second initial decoder to obtain the prediction model.

[0121] S19. Obtain longitudinal health data of the target object, and predict the disease development trend of the target object based on the longitudinal health data and the prediction model.

[0122] It should be noted that the descriptions in S1-T18 above can be found in the relevant descriptions in the above embodiments, and their effects are similar, so they will not be repeated here.

[0123] In one embodiment, to adapt to downstream tasks of disease risk and progression prediction, the prediction model (i.e., RETFound Plus) needs to be fine-tuned using a pseudo-survival model after training. Each data point is represented as a triple (x, t, e), where x is the baseline retinal image, t is the event occurrence time or truncation time, and e is the event indicator (1 if the event occurred before truncation, 0 otherwise). The model estimates a survival probability function S(t), representing the probability that the event time exceeds t. This survival function is modeled using a log-logarithmic distribution and includes patient-specific location parameters μ and scale parameters σ. These parameters are obtained by processing a high-dimensional representation of the model encoder output through a fully connected layer.

[0124] It should be noted that the fine-tuning process is optimized using a negative log-likelihood loss function, which is specifically designed to handle truncated data. The survival function S(t) is expressed in log-logarithmic form, and its corresponding log-likelihood function is used to estimate the parameters μ and σ. The loss function combines the likelihood of untruncated events with the survival probability of truncated observations, thus comprehensively considering both truncated and untruncated data.

[0125] Finally, the model's performance was evaluated using two main metrics: Harrell's concordance index (c-index) and the Integrated Brier score (IBS). The c-index measures the model's ability to rank patient event risks, while the IBS assesses the calibration and discrimination ability of the predicted survival probability. Both metrics were calculated by inverse probability weighting to adjust for censoring effects and over a follow-up period of up to 10 years, consistent with the longest event time in the dataset. Finally, RETFound Plus was tested on several external datasets to evaluate its generalization ability across different populations and disease cohorts. These datasets included multi-ethnic cohorts from different geographic regions, ensuring the model's robustness in diverse clinical settings. The model's predictive performance was compared with existing base models such as MAE and the original RETFound model, further validating its superiority.

[0126] In one embodiment, the predictive model can be fine-tuned on the SDPP cohort. The SDPP cohort is a community-based longitudinal cohort comprising 79,284 participants who underwent physical examinations at the East China Sanatorium and Shanghai Sixth People's Hospital between December 2015 and November 2022. Baseline data included demographic information, anthropometric indices, biochemical measurements, and retinal images. Following the baseline survey, 25,231 participants completed at least four years of annual follow-up. Diagnosis of DR grading and DME was based on macular and central fundus images of each eye during baseline and follow-up periods.

[0127] In addition, external validation includes multi-country datasets, for example:

[0128] (1) CUHK-STDR cohort: a prospective observational study involving 337 diabetic patients. Participants were recruited from the Eye Centre of the Chinese University of Hong Kong between July 2015 and November 2016 and followed up for at least 5 years.

[0129] (2) SEED cohort: a multi-ethnic longitudinal population study including Malay, Indian and Chinese adults in Singapore, with a total of 1,699 diabetic patients completing a 5-year follow-up.

[0130] (3) OUH cohort: a retrospective hospital cohort that included 535 patients at Odense University Hospital in Denmark between 2015 and 2022;

[0131] (4) AREDS cohort: a randomized, placebo-controlled clinical trial based at 11 retinal specialty clinics in the United States, which recruited 4,757 participants between November 1992 and November 2005.

[0132] (5) MEH-AlzEye cohort: A retrospective cohort study that linked ophthalmological data of 353,157 patients who visited Moorfields Eye Hospital between 2008 and 2018 with inpatient health data across England.

[0133] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0134] Based on the same inventive concept, this application also provides a disease prediction device for implementing the disease prediction method described above. The solution provided by this device is similar to the implementation described in the above method; therefore, the specific limitations in one or more disease prediction device embodiments provided below can be found in the limitations of the disease prediction method described above, and will not be repeated here.

[0135] In an exemplary embodiment, as shown in FIG9, a disease prediction device is provided, including: a first training module 10, a second training module 11, and a prediction module 12.

[0136] The first training module 10 is used to train the initial pedestal model based on the first medical image set to obtain the trained pedestal model; the trained pedestal model is used to extract medical features from the medical images; the first medical image set includes multiple first medical images.

[0137] The second training module 11 is used to train the initial prediction model based on the trained base model and the second medical image set to obtain the prediction model; the prediction model is used to predict diseases; the second medical image set includes: a second medical image acquired at a first time for the same object, a third medical image acquired at a second time, and the time interval between the first time and the second time.

[0138] The prediction module 12 is used to acquire longitudinal health data of the target object and predict the development trend of the target object's disease based on the longitudinal health data and the prediction model.

[0139] In an exemplary embodiment, the first training module 10 described above includes a processing unit and a first training unit.

[0140] The processing unit is specifically used to perform a first masking process on the first medical image in the first medical image set to obtain the masked first medical image.

[0141] The first training unit is specifically used to train the initial pedestal model based on the first medical image and the masked first medical image to obtain the trained pedestal model.

[0142] In an exemplary embodiment, the aforementioned initial base model includes a first initial encoder and a first initial decoder. The aforementioned first training unit is further configured to input the masked first medical image into the first initial encoder for encoding processing to obtain a first encoded feature; input the first encoded feature into the first initial decoder for decoding processing to obtain a first decoded feature; and train the first initial encoder and the first initial decoder based on the first difference between the image features of the first medical image and the first decoded feature to obtain the trained base model.

[0143] In an exemplary embodiment, the second training module 11 described above includes: an extraction unit and a second training unit.

[0144] The extraction unit is specifically used to input the third medical image from the second medical image set into the trained base model for feature extraction, thereby obtaining the features of the third medical image.

[0145] The second training unit is specifically used to train the initial prediction model based on the features of the third medical image and the second medical image to obtain the prediction model.

[0146] In an exemplary embodiment, the second training unit is further configured to perform a second masking process on the second medical image to obtain a masked second medical image; and to train an initial prediction model based on the features of the masked second medical image and the third medical image, as well as the time interval between the first and second time moments, to obtain a prediction model.

[0147] In an exemplary embodiment, the initial prediction model includes a second initial encoder and a second initial decoder. The second training unit is further configured to input the masked second medical image and time interval into the second initial encoder for encoding processing to obtain second encoded features; input the second encoded features into the second initial decoder for decoding processing to obtain second decoded features; and train the second initial encoder and the second initial decoder based on the second difference between the features of the third image and the second decoded features to obtain a prediction model.

[0148] In an exemplary embodiment, the initial pedestal model is a masked autoencoder (MAE) model, and the initial prediction model is a RETFound model.

[0149] Each module in the aforementioned disease prediction device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the memory of a computer device as software, so that the processor can call and execute the corresponding operations of each module.

[0150] In one exemplary embodiment, a computer device is provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to perform the following steps:

[0151] Based on the first medical image set, the initial pedestal model is trained to obtain the trained pedestal model; the trained pedestal model is used to extract medical features from the medical images; the first medical image set includes multiple first medical images;

[0152] Based on the trained base model and the second medical image set, the initial prediction model is trained to obtain the prediction model; the prediction model is used to predict diseases; the second medical image set includes: a second medical image acquired at the first time point for the same object, a third medical image acquired at the second time point, and the time interval between the first time point and the second time point;

[0153] Obtain longitudinal health data of the target object, and predict the disease development trend of the target object based on the longitudinal health data and the prediction model.

[0154] In one embodiment, the processor, when executing a computer program, also performs the following steps:

[0155] The first medical image in the first medical image set is subjected to a first masking process to obtain the masked first medical image.

[0156] The initial pedestal model is trained based on the first medical image and the masked first medical image to obtain the trained pedestal model.

[0157] In one embodiment, the processor, when executing a computer program, also performs the following steps:

[0158] The masked first medical image is input into the first initial encoder for encoding processing to obtain the first encoded feature;

[0159] The first encoded feature is input into the first initial decoder for decoding processing to obtain the first decoded feature;

[0160] Based on the first difference between the image features and the first decoding features of the first medical image, a first initial encoder and a first initial decoder are trained to obtain the trained pedestal model.

[0161] In one embodiment, the processor, when executing a computer program, also performs the following steps:

[0162] The third medical image in the second medical image set is input into the trained base model for feature extraction to obtain the features of the third medical image;

[0163] The initial prediction model is trained based on the features of the third medical image and the second medical image to obtain the prediction model.

[0164] In one embodiment, the processor, when executing a computer program, also performs the following steps:

[0165] The second medical image is subjected to a second masking process to obtain the masked second medical image.

[0166] Based on the features of the masked second and third medical images, as well as the time interval between the first and second time points, the initial prediction model is trained to obtain the prediction model.

[0167] In one embodiment, the processor, when executing a computer program, also performs the following steps:

[0168] The masked second medical image and time interval are input into the second initial encoder for encoding processing to obtain the second encoded features;

[0169] The second encoded feature is input into the second initial decoder for decoding processing to obtain the second decoded feature;

[0170] Based on the second difference between the features of the third image and the second decoded features, a second initial encoder and a second initial decoder are trained to obtain a prediction model.

[0171] The initial base model mentioned above is the Masked Autoencoder (MAE) model, and the initial prediction model mentioned above is the RETFound model.

[0172] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, the computer program performing the following steps when executed by a processor:

[0173] Based on the first medical image set, the initial pedestal model is trained to obtain the trained pedestal model; the trained pedestal model is used to extract medical features from the medical images; the first medical image set includes multiple first medical images;

[0174] Based on the trained base model and the second medical image set, the initial prediction model is trained to obtain the prediction model; the prediction model is used to predict diseases; the second medical image set includes: a second medical image acquired at the first time point for the same object, a third medical image acquired at the second time point, and the time interval between the first time point and the second time point;

[0175] Obtain longitudinal health data of the target object, and predict the disease development trend of the target object based on the longitudinal health data and the prediction model.

[0176] In one embodiment, when the computer program is executed by a processor, it further performs the following steps:

[0177] The first medical image in the first medical image set is subjected to a first masking process to obtain the masked first medical image.

[0178] The initial pedestal model is trained based on the first medical image and the masked first medical image to obtain the trained pedestal model.

[0179] In one embodiment, when the computer program is executed by a processor, it further performs the following steps:

[0180] The masked first medical image is input into the first initial encoder for encoding processing to obtain the first encoded feature;

[0181] The first encoded feature is input into the first initial decoder for decoding processing to obtain the first decoded feature;

[0182] Based on the first difference between the image features and the first decoding features of the first medical image, a first initial encoder and a first initial decoder are trained to obtain the trained pedestal model.

[0183] In one embodiment, when the computer program is executed by a processor, it further performs the following steps:

[0184] The third medical image in the second medical image set is input into the trained base model for feature extraction to obtain the features of the third medical image;

[0185] The initial prediction model is trained based on the features of the third medical image and the second medical image to obtain the prediction model.

[0186] In one embodiment, when the computer program is executed by a processor, it further performs the following steps:

[0187] The second medical image is subjected to a second masking process to obtain the masked second medical image.

[0188] Based on the features of the masked second and third medical images, as well as the time interval between the first and second time points, the initial prediction model is trained to obtain the prediction model.

[0189] In one embodiment, when the computer program is executed by a processor, it further performs the following steps:

[0190] The masked second medical image and time interval are input into the second initial encoder for encoding processing to obtain the second encoded features;

[0191] The second encoded feature is input into the second initial decoder for decoding processing to obtain the second decoded feature;

[0192] Based on the second difference between the features of the third image and the second decoded features, a second initial encoder and a second initial decoder are trained to obtain a prediction model.

[0193] The initial base model mentioned above is the Masked Autoencoder (MAE) model, and the initial prediction model mentioned above is the RETFound model.

[0194] In one embodiment, a computer program product is provided, including a computer program that, when executed by a processor, performs the following steps:

[0195] Based on the first medical image set, the initial pedestal model is trained to obtain the trained pedestal model; the trained pedestal model is used to extract medical features from the medical images; the first medical image set includes multiple first medical images;

[0196] Based on the trained base model and the second medical image set, the initial prediction model is trained to obtain the prediction model; the prediction model is used to predict diseases; the second medical image set includes: a second medical image acquired at the first time point for the same object, a third medical image acquired at the second time point, and the time interval between the first time point and the second time point;

[0197] Obtain longitudinal health data of the target object, and predict the disease development trend of the target object based on the longitudinal health data and the prediction model.

[0198] In one embodiment, when the computer program is executed by a processor, it further performs the following steps:

[0199] The first medical image in the first medical image set is subjected to a first masking process to obtain the masked first medical image.

[0200] The initial pedestal model is trained based on the first medical image and the masked first medical image to obtain the trained pedestal model.

[0201] In one embodiment, when the computer program is executed by a processor, it further performs the following steps:

[0202] The masked first medical image is input into the first initial encoder for encoding processing to obtain the first encoded feature;

[0203] The first encoded feature is input into the first initial decoder for decoding processing to obtain the first decoded feature;

[0204] Based on the first difference between the image features and the first decoding features of the first medical image, a first initial encoder and a first initial decoder are trained to obtain the trained pedestal model.

[0205] In one embodiment, when the computer program is executed by a processor, it further performs the following steps:

[0206] The third medical image in the second medical image set is input into the trained base model for feature extraction to obtain the features of the third medical image;

[0207] The initial prediction model is trained based on the features of the third medical image and the second medical image to obtain the prediction model.

[0208] In one embodiment, when the computer program is executed by a processor, it further performs the following steps:

[0209] The second medical image is subjected to a second masking process to obtain the masked second medical image.

[0210] Based on the features of the masked second and third medical images, as well as the time interval between the first and second time points, the initial prediction model is trained to obtain the prediction model.

[0211] In one embodiment, when the computer program is executed by a processor, it further performs the following steps:

[0212] The masked second medical image and time interval are input into the second initial encoder for encoding processing to obtain the second encoded features;

[0213] The second encoded feature is input into the second initial decoder for decoding processing to obtain the second decoded feature;

[0214] Based on the second difference between the features of the third image and the second decoded features, a second initial encoder and a second initial decoder are trained to obtain a prediction model.

[0215] The initial base model mentioned above is the Masked Autoencoder (MAE) model, and the initial prediction model mentioned above is the RETFound model.

[0216] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile memory and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, artificial intelligence (AI) processors, etc., and are not limited to these.

[0217] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this application.

[0218] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A disease prediction method, characterized in that, The method includes: Based on a first set of medical images, an initial pedestal model is trained to obtain a trained pedestal model; the trained pedestal model is used to extract medical features from the medical images; the first set of medical images includes multiple first medical images. Based on the trained base model and the second medical image set, the initial prediction model is trained to obtain a prediction model; the prediction model is used to predict diseases; the second medical image set includes: a second medical image acquired at a first time moment for the same object, a third medical image acquired at a second time moment, and the time interval between the first time moment and the second time moment; Obtain longitudinal health data of the target object, and predict the disease development trend of the target object based on the longitudinal health data and the prediction model.

2. The method according to claim 1, characterized in that, The step of training the initial pedestal model based on the first medical image set to obtain the trained pedestal model includes: The first medical image in the first medical image set is subjected to a first masking process to obtain the masked first medical image. The initial pedestal model is trained based on the first medical image and the masked first medical image to obtain the trained pedestal model.

3. The method according to claim 2, characterized in that, The initial base model includes a first initial encoder and a first initial decoder; the step of training the initial base model based on the first medical image and the masked first medical image to obtain the trained base model includes: The masked first medical image is input into the first initial encoder for encoding processing to obtain the first encoded feature; The first encoded feature is input into the first initial decoder for decoding to obtain the first decoded feature; Based on the first difference between the image features of the first medical image and the first decoding features, the first initial encoder and the first initial decoder are trained to obtain the trained pedestal model.

4. The method according to any one of claims 1-3, characterized in that, The initial prediction model is trained based on the trained base model and the second medical image set to obtain the prediction model, including: The third medical image in the second medical image set is input into the trained base model for feature extraction to obtain the features of the third medical image; The initial prediction model is trained based on the features of the third medical image and the second medical image to obtain the prediction model.

5. The method according to claim 4, characterized in that, The step of training the initial prediction model based on the features of the third medical image and the second medical image to obtain the prediction model includes: The second medical image is subjected to a second masking process to obtain a masked second medical image; The initial prediction model is trained based on the features of the masked second medical image and the third medical image, as well as the time interval between the first and second time points, to obtain the prediction model.

6. The method according to claim 5, characterized in that, The initial prediction model includes a second initial encoder and a second initial decoder; training the initial prediction model based on the features of the masked second medical image and the third medical image, as well as the time interval between the first and second time moments, to obtain the prediction model, includes: The masked second medical image and the time interval are input into the second initial encoder for encoding processing to obtain the second encoded feature; The second encoded feature is input into the second initial decoder for decoding processing to obtain the second decoded feature; The second initial encoder and the second initial decoder are trained based on the second difference between the features of the third image and the second decoded features to obtain the prediction model.

7. The method according to any one of claims 1-6, characterized in that, The initial base model is a Masked Autoencoder (MAE) model, and the initial prediction model is a RETFound model.

8. The method according to any one of claims 1-7, characterized in that, The first medical image is a medical image that does not carry information about the image acquisition time or the disease type corresponding to the image.

9. The method according to claim 1, characterized in that, Based on the first medical image set, the initial pedestal model is trained to obtain the trained pedestal model, including: The initial pedestal model is pre-trained on the ImageNet dataset to learn general image structures and features, resulting in a pre-trained initial pedestal model. The initial pedestal model was trained based on the first set of medical images to obtain the trained pedestal model.

10. The method according to claim 2, characterized in that, The process of performing a first masking process on the first medical images in the first medical image set to obtain a masked first medical image includes: performing segmentation processing on each first medical image in the first medical image set to obtain multiple segmented images corresponding to the first medical image; and randomly performing a first masking process on a portion of the multiple segmented images to obtain each masked first medical image.

11. The method according to claim 3, characterized in that, The first encoded feature refers to the feature of the unmasked portion of the first medical image after masking; the first decoded feature refers to all the features of the first medical image inferred from the first encoded feature.

12. The method according to claim 3, characterized in that, Based on the first difference between the image features of the first medical image and the first decoded features, the first initial encoder and the first initial decoder are trained to obtain the trained base model, including: Based on the first difference between the image features of the first medical image and the first decoding features, determine whether the first difference is less than or equal to a first preset difference threshold. If the first difference is greater than the first preset difference threshold, the model parameters of the first initial encoder and the first initial decoder are adjusted to obtain a new first initial encoder and a new first initial decoder. Then, the process is returned to input the masked first medical image into the first initial encoder for encoding processing to obtain the first encoded feature, until the first difference is less than or equal to the first preset difference threshold, and the trained base model is obtained.

13. A disease prediction device, characterized in that, The device includes: The first training module is used to train the initial pedestal model based on the first medical image set to obtain the trained pedestal model; the trained pedestal model is used to extract medical features from the medical images; the first medical image set includes multiple first medical images. The second training module is used to train the initial prediction model based on the trained base model and the second medical image set to obtain a prediction model; the prediction model is used to predict diseases; the second medical image set includes: a second medical image acquired at a first time moment for the same object, a third medical image acquired at a second time moment, and the time interval between the first time moment and the second time moment; The prediction module is used to acquire longitudinal health data of the target object and predict the disease development trend of the target object based on the longitudinal health data and the prediction model.

14. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 12.

15. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 12.