Intelligent rectal cancer prognosis analysis method based on two-stage 3D feature extraction

Abstract

The invention discloses a rectal cancer prognosis intelligent analysis method based on two-stage 3D feature extraction. The method comprises the steps of obtaining a prognostic MRI image of a target patient, constructing a 3D convolutional neural network based on a 3D-Unet architecture, performing staged training on the 3D convolutional neural network, and generating a rectal cancer tumor extraction model; taking the preprocessed MRI 3D image as a model input, segmenting a rectal cancer tumor image, and outputting a 3D tumor image and a risk score in stages; and performing prognosis evaluation by combining the risk score with clinical data of the target patient to obtain the survival rate of the target patient. According to the invention, the 3D tumor information in the MRI image can be extracted, the accuracy of prognosis analysis is further enhanced, a more accurate and more reliable prognosis result is provided, and the clinical decision is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of image segmentation processing, and more specifically, to an intelligent analysis method for rectal cancer prognosis based on two-stage 3D feature extraction. Background Art

[0002] Magnetic resonance imaging (MRI) images are an important evaluation tool in the prognosis analysis of rectal cancer. Traditional image analysis methods usually rely on manually selected ROI layers, which only utilize the maximum cross-sectional information of the tumor. However, this single-layer ROI analysis ignores the overall morphology and spatial distribution of the tumor, resulting in deviations in the prognostic evaluation results. In recent years, deep learning technology has gradually been applied to medical image analysis, improving the accuracy of analysis through automated feature extraction, but most methods are still limited to single-layer ROI analysis.

[0003] Existing prognostic analysis methods for rectal cancer MRI images mainly focus on processing single-layer ROIs, usually using convolutional neural networks (CNNs) for feature extraction and analysis. These methods can improve the efficiency of prognostic analysis to a certain extent, but have limitations in processing three-dimensional information. Typical implementation schemes include processing single-layer images through two-dimensional convolution, but this method cannot fully utilize the three-dimensional information in MRI images, resulting in insufficient grasp of the spatial structure and overall morphology of the tumor, which in turn affects the accuracy and reliability of the prognostic results. Summary of the invention

[0004] In order to solve the above technical problems, the present invention proposes an intelligent analysis method for rectal cancer prognosis based on two-stage 3D feature extraction, the purpose of which is to extract 3D tumor information in MRI images, so as to perform prognosis analysis more accurately, provide more precise and reliable prognosis evaluation results, and thus improve clinical decision-making.

[0005] The present invention provides a rectal cancer prognosis intelligent analysis method based on two-stage 3D feature extraction, comprising:

[0006] Step S102, obtaining an MRI image of the target patient's prognosis, and preprocessing the MRI image;

[0007] Step S104, constructing a three-dimensional convolutional neural network based on a 3D-Unet architecture, training the three-dimensional convolutional neural network in stages, and generating a rectal cancer tumor extraction model;

[0008] Step S106, using the preprocessed 3D MRI image as a model input, segmenting the rectal cancer tumor image using the rectal cancer tumor extraction model, and obtaining a 3D tumor image and a risk score;

[0009] Step S108, combining the risk score with the clinical data of the target patient to perform a prognosis assessment to obtain the survival rate of the target patient.

[0010] In this solution, an MRI image of the target patient's prognosis is obtained, and the MRI image is preprocessed, specifically:

[0011] Extracting the MRI image of the target patient's prognosis to obtain an MRI grayscale image, performing floating-point image conversion on the MRI grayscale image to obtain a converted MRI grayscale image;

[0012] Correcting the converted MRI grayscale image using N4 bias correction to obtain a corrected training MRI grayscale image;

[0013] Histogram matching is performed on the corrected training MRI grayscale image to obtain a preprocessed 3D MRI image.

[0014] In this solution, a three-dimensional convolutional neural network based on the 3D-Unet architecture is constructed, and the three-dimensional convolutional neural network is trained in stages, specifically:

[0015] A three-dimensional convolutional neural network is constructed based on the 3D-Unet architecture, and the encoder part is trained in the first stage, in which the three-dimensional convolution kernel is used to gradually downsample the input three-dimensional image;

[0016] The encoder part is implemented by four layers of downsampling operations, where each layer of downsampling operation uses two convolution modules, a sparse module, and a maximum pooling module to extract the feature information of the 3D image and avoid network overfitting;

[0017] The decoding part is implemented by four layers of upsampling operations, where each upsampling operation uses two convolution operations, a sparse module, and a deconvolution module to restore the size of the original image layer by layer, and finally obtains the segmentation output result through the activation function;

[0018] Reading rectal cancer diagnosis and tumor area division results from historical rectal cancer medical record data as label information, constructing a first-stage data sample based on the data label information combined with rectal cancer images, and dividing the data into first-stage training data, verification data, and test data according to a preset ratio;

[0019] Use the Dice loss function to pre-train the encoder and obtain the encoder weights.

[0020] In this scheme, the Dice loss function is used to calculate the loss, and the Dice loss function is expressed as:

[0021]

[0022] Among them, X represents the predicted output 3D tumor image, and Y represents the label information.

[0023] In this solution, the second phase of network construction and training of the three-dimensional convolutional neural network is specifically as follows:

[0024] Obtaining the encoder structure and weights of the three-dimensional convolutional neural network after the first stage of training, and filling the decoder structure with four fully connected layers to replace the decoder and skip connections trained in the first stage;

[0025] Reading whether a patient recurrence event has occurred and the follow-up time in the historical rectal cancer medical record data to generate data label information, constructing a second-stage data sample based on the data label information combined with the historical risk score, and dividing it into second-stage training data, verification data, and test data according to a preset ratio;

[0026] The weights of the encoder part of the first stage are used as the pre-trained weights of the encoder part of the second segment. The cox partial likelihood loss function is used to calculate the loss based on the training data for model training. During the training process, the validation data is used to adjust the hyperparameters and test the screening model. The screened model is tested with the test data.

[0027] In this scheme, the cox partial likelihood loss function is used to calculate the loss, and the cox partial likelihood loss function is expressed as:

[0028]

[0029] Where n represents the batch size of the three-dimensional convolutional neural network. A batch contains n training data samples, each of which contains the risk score s, follow-up time t, and whether a recurrence event e occurs. k represents the number of recurrence events in a batch sample, and i and j represent the number of items in the batch sample.

[0030] In this scheme, the risk score is combined with the clinical data of the target patient to perform prognostic evaluation and obtain the survival rate of the target patient, specifically:

[0031] Obtaining a risk score set corresponding to the second-stage test data of the three-dimensional convolutional neural network, performing ROC curve analysis on the risk score set to obtain an optimal cutoff point, and dividing the patients into a high-risk group and a low-risk group based on the optimal cutoff point;

[0032] Obtaining a risk score for the target patient, and generating a risk score feature according to a determination result of the risk score and a risk score threshold corresponding to an optimal cutoff point;

[0033] The clinical characteristics and risk score characteristics of the target patients were obtained, and KM survival analysis was used to obtain the low-risk score curve and the high-risk score curve, and the predicted survival rate of the target patients was output.

[0034] Compared with the prior art, the present invention has the following beneficial effects:

[0035] The present invention proposes a prognostic evaluation model based on 3D-UNet. The present invention realizes the ability to automatically extract ROI images through two-stage training, eliminating manual intervention and solving the defect of relying on manually intercepted ROI images in traditional tumor prognostic models.

[0036] The present invention can also extract 3D tumor information from MRI images, further enhancing the accuracy of prognostic analysis. The purpose of the present invention is to provide more accurate and reliable prognostic results through comprehensive utilization of three-dimensional image information, thereby improving clinical decision-making. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] In order to more clearly illustrate the technical solutions in the embodiments or exemplary embodiments of the present invention, the drawings required for use in the embodiments or exemplary descriptions will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained according to the drawings without paying creative work.

[0038] Figure 1 A flowchart of an intelligent analysis method for rectal cancer prognosis based on two-stage 3D feature extraction is shown;

[0039] Figure 2 A structural diagram of a three-dimensional convolutional neural network in an embodiment is shown;

[0040] Figure 3 A flow chart of a sparse module in an embodiment is shown;

[0041] Figure 4 The structural diagram of the rectal cancer tumor extraction model in the embodiment is shown;

[0042] Figure 5 A schematic diagram of the KM survival analysis of the embodiment is shown;

[0043] Figure 6 A flowchart of an embodiment for obtaining features for KM survival analysis is shown;

[0044] Figure 7 A block diagram of the rectal cancer prognosis intelligent analysis system based on two-stage 3D feature extraction is shown. DETAILED DESCRIPTION

[0045] In order to more clearly understand the above-mentioned purpose, features and advantages of the present invention, the present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments. It should be noted that the embodiments of the present application and the features in the embodiments can be combined with each other without conflict.

[0046] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways different from those described herein. Therefore, the protection scope of the present invention is not limited to the specific embodiments disclosed below.

[0047] Figure 1 A flowchart of the intelligent analysis method for rectal cancer prognosis based on two-stage 3D feature extraction is shown.

[0048] like Figure 1 As shown, this embodiment provides a rectal cancer prognosis intelligent analysis method based on two-stage 3D feature extraction, including:

[0049] S102, obtaining an MRI image of the target patient's prognosis, and preprocessing the MRI image;

[0050] S104, constructing a three-dimensional convolutional neural network based on a 3D-Unet architecture, and training the three-dimensional convolutional neural network in stages to generate a rectal cancer tumor extraction model;

[0051] S106, using the preprocessed 3D MRI image as a model input, segmenting the rectal cancer tumor image using the rectal cancer tumor extraction model, and obtaining a 3D tumor image and a risk score;

[0052] S108, combining the risk score with the clinical data of the target patient to perform prognosis assessment and obtain the survival rate of the target patient.

[0053] It should be noted that the MRI grayscale image is obtained by extracting the MRI image of the target patient's prognosis, and the MRI grayscale image is converted into a floating-point image to obtain a converted MRI grayscale image; the converted MRI grayscale image is corrected by N4 bias correction to obtain a corrected training MRI grayscale image, and the corrected training MRI grayscale image is matched with histogram to obtain a preprocessed 3DMRI image. During image preprocessing, N4 bias correction and histogram matching are performed on the MRT image to avoid noise influence and model design redundancy, and improve the efficiency of subsequent feature extraction.

[0054] According to an embodiment of the present invention, a three-dimensional convolutional neural network is constructed based on the 3D-Unet architecture, such as Figure 2As shown, the encoder part is trained in the first stage, in which a three-dimensional convolution kernel is used to gradually downsample the input three-dimensional image to extract multi-scale features and deeply mine image information, and avoid network overfitting; the decoding part is implemented by four layers of upsampling operations, in which each layer of upsampling operation uses two convolution operations, a sparse module and a deconvolution module to restore the size of the original image layer by layer, and finally the segmentation output result is obtained through the activation function; through the jump connection mechanism and the shallow feature fusion of the encoding stage, not only the rich spatial details are retained, but also the semantic expression ability of the model is significantly enhanced, ensuring the consistency of the output and input image sizes, and promoting the realization of accurate segmentation.

[0055] The sparse module effectively utilizes multi-scale convolution operations to improve the generalization ability of the network through multi-branch feature extraction, feature fusion and batch normalization. Figure 3 As shown in the figure, the input of the sparse module is a tensor C×D×H×W, which is first processed through two parallel paths. In the first path, the input first passes through a 2×2×2 3D maximum pooling layer to downsample the spatial dimension and capture the features of the salient regions. Then, a 1×1×1 3D convolution layer is used to compress the channel dimension, reducing the computational complexity and retaining important feature information. To restore the spatial dimension reduced by pooling, a 3×3×3 3D deconvolution (ConvTranspose3D) layer is used to reconstruct the spatial information. This operation can restore the spatial structure and capture fine features. A batch normalization (BN) layer is added at the end of this path to ensure the stability of training and prevent overfitting.

[0056] In the other path, the input tensor first passes through a 1×1×1 3D convolution layer to compress the number of channels and extract global features. The path then splits into three branches. In the first branch, a 3×3×3 3D convolution layer is applied to capture local features in space. In the second branch, a 1×1×1 3D convolution layer is applied to further compress the number of channels and extract features of different dimensions. The third branch directly passes through the batch normalization (BN) layer, skipping the convolution operation and performing feature normalization to ensure the stability of feature fusion at different scales of the network. The features output by these three branches are spliced ​​in the channel dimension. Through this splicing, the network can fuse multi-scale information from different branches and enhance the representation ability of complex input data. On the spliced ​​feature map, a 3×3×3 3D convolution layer is applied to further integrate and reorganize the previously extracted features, and batch normalization is performed again to ensure the stability of training.

[0057] Finally, the output feature maps from the two main branches are concatenated in the channel dimension, and then further integrated with multi-scale and multi-dimensional information through a 3×3×3 3D convolution layer to ensure that the features extracted by different branches can be effectively fused. Subsequently, the concatenated feature maps are normalized through a batch normalization (BN) layer to further improve the stability of training and prevent gradient vanishing or explosion. In this way, the continuous processing of convolution and normalization not only enhances the expressive power of the fused features, but also effectively suppresses the overfitting tendency of the model, ensuring the fine extraction of features and the generalization ability of the network.

[0058] The rectal cancer diagnosis and tumor area division results are read from the historical rectal cancer medical record data as label information, and the first-stage data samples are constructed according to the data label information and the rectal cancer images, which are divided into the first-stage training data, verification data and test data according to a preset ratio; the encoder is pre-trained using the Dice loss function to obtain the encoder weight.

[0059] The Dice loss function calculates the overlap between the predicted results and the true results and optimizes the model by minimizing the difference between the two. It is suitable for cases of class imbalance and can reduce the negative impact caused by the imbalanced ratio of foreground and background in samples. The Dice loss function is used to calculate the loss, which is expressed as:

[0060]

[0061] Among them, X represents the predicted output 3D tumor image, and Y represents the label information.

[0062] Get the encoder structure and weights of the 3D convolutional neural network after the first stage of training, such as Figure 4 As shown, four fully connected layers of different sizes are used to fill the decoder structure, replacing the decoder and jump connection trained in the first stage; the data label information is generated by reading whether the patient's recurrence event occurs and the follow-up time in the historical rectal cancer medical record data, and the second-stage data sample is constructed according to the data label information combined with the historical risk score, and divided into the second-stage training data, verification data and test data according to the preset ratio; the weight of the encoder part of the first stage is used as the pre-training weight of the second segment encoder part, and the loss is calculated based on the training data using the cox partial likelihood loss function to train the model, and the loss calculation includes all features in a batch (risk score, whether the patient's recurrence event occurs and follow-up time). During the training process, the verification data is used to adjust the hyperparameters and test the screening model, and the screened model is tested with the test data.

[0063] The loss is calculated using the cox partial likelihood loss function, which is expressed as:

[0064]

[0065] Where n represents the batch size of the three-dimensional convolutional neural network. A batch contains n training data samples, each of which contains the risk score s, follow-up time t, and whether a recurrence event e occurs. k represents the number of recurrence events in a batch sample, and i and j represent the number of items in the batch sample.

[0066] The rectal cancer tumor extraction model was configured according to the trained encoder weights and prognostic score weights. The preprocessed 3D MRI images were used as model input. The rectal cancer tumor extraction model was used to segment the rectal cancer tumor images to obtain 3D tumor images and risk scores.

[0067] According to an embodiment of the present invention, a risk score set corresponding to the second-stage test data of the three-dimensional convolutional neural network is obtained, and a ROC curve analysis is performed on the risk score set to obtain an optimal cutoff point, and patients are divided into a high-risk group and a low-risk group based on the optimal cutoff point; the risk score of the target patient is obtained, and a risk score feature is generated according to a judgment result of the risk score and the risk score threshold corresponding to the optimal cutoff point; the clinical characteristics and risk score features of the target patient are obtained, and a KM survival analysis is performed to obtain a low-risk score curve and a high-risk score curve, and a predicted survival rate of the target patient is output, and the low-risk score curve (pred0) and the high-risk score curve (pred1) are as follows: Figure 5 shown.

[0068] It should be noted that the risk score of the target patient is obtained, and a risk score feature is generated according to the judgment result of the risk score and the risk score threshold corresponding to the optimal cutoff point, and the risk score feature is a high-risk group or a low-risk group; Figure 6 As shown, features such as recurrence events and follow-up time in clinical information are extracted as clinical information features.

[0069] In another preferred embodiment of the present invention, a prognosis database for the target patient is constructed, the prognosis database is used to record the treatment response and survival rate of the target patient, a prognosis portrait of the target patient is constructed according to the treatment response and survival rate, a similarity search is performed in the historical rectal cancer medical record data based on the prognosis portrait, the historical rectal cancer medical record data that meets the preset similarity threshold standard is screened, the historical rectal cancer medical record data with a survival rate greater than the preset survival rate threshold in the screened data is obtained and marked, the patient's rehabilitation status in the marked data is analyzed, rectal cancer rehabilitation indicators are retrieved through big data methods, the patient's rehabilitation status is evaluated, the marked data whose rehabilitation evaluation results meet the health standards are selected, the corresponding prognosis method is extracted and implemented on the target patient, and the rehabilitation health results are visualized and provided to the target user to improve the target patient's confidence in treatment.

[0070] Figure 7 A block diagram of the rectal cancer prognosis intelligent analysis system based on two-stage 3D feature extraction is shown.

[0071] The second aspect of the present invention also provides a rectal cancer prognosis intelligent analysis system 7 based on two-stage 3D feature extraction, the system comprising: a memory 71, a processor 72, wherein the memory and the processor store and execute a rectal cancer prognosis intelligent analysis method program based on two-stage 3D feature extraction.

[0072] The third aspect of the present invention provides a computer-readable storage medium, which includes a program for a rectal cancer prognosis intelligent analysis method based on two-stage 3D feature extraction. When the program for the rectal cancer prognosis intelligent analysis method based on two-stage 3D feature extraction is executed by a processor, the steps of the rectal cancer prognosis intelligent analysis method based on two-stage 3D feature extraction are implemented.

[0073] In the several embodiments provided in the present application, it should be understood that the disclosed devices and methods can be implemented in other ways. The device embodiments described above are only schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation, such as: multiple units or components can be combined, or can be integrated into another system, or some features can be ignored or not executed. In addition, the coupling, direct coupling, or communication connection between the components shown or discussed can be through some interfaces, and the indirect coupling or communication connection of the devices or units can be electrical, mechanical or other forms.

[0074] The units described above as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units; they may be located in one place or distributed on multiple network units; some or all of the units may be selected according to actual needs to achieve the purpose of the present embodiment.

[0075] In addition, all functional units in the embodiments of the present invention may be integrated into one processing unit, or each unit may be separately used as a unit, or two or more units may be integrated into one unit; the above-mentioned integrated units may be implemented in the form of hardware or in the form of hardware plus software functional units.

[0076] Those skilled in the art can understand that all or part of the steps of implementing the above method embodiments can be completed by hardware related to program instructions, and the aforementioned program can be stored in a computer-readable storage medium. When the program is executed, it executes the steps of the above method embodiments; and the aforementioned storage medium includes: mobile storage devices, read-only memories (ROM, Read-Only Memory), random access memories (RAM, Random Access Memory), disks or optical disks, and other media that can store program codes.

[0077] Alternatively, if the above-mentioned integrated unit of the present invention is implemented in the form of a software function module and sold or used as an independent product, it can also be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the embodiment of the present invention can be essentially or partly reflected in the form of a software product that contributes to the prior art. The computer software product is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the methods described in each embodiment of the present invention. The aforementioned storage medium includes: various media that can store program codes, such as mobile storage devices, ROM, RAM, magnetic disks or optical disks.

[0078] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims.

Claims

1. A rectal cancer prognosis intelligent analysis method based on two-stage 3D feature extraction, characterized in that: The following steps are involved: Step S102, obtaining an MRI image of the target patient's prognosis, and preprocessing the MRI image; Step S104, constructing a three-dimensional convolutional neural network based on a 3D-Unet architecture, training the three-dimensional convolutional neural network in stages, and generating a rectal cancer tumor extraction model; Step S106, using the preprocessed 3D MRI image as a model input, segmenting the rectal cancer tumor image using the rectal cancer tumor extraction model, and obtaining a 3D tumor image and a risk score; Step S108, combining the risk score with the clinical data of the target patient to perform a prognosis assessment to obtain the survival rate of the target patient.

2. According to claim 1, a rectal cancer prognosis intelligent analysis method based on two-stage 3D feature extraction is characterized in that: Obtain an MRI image of the target patient's prognosis and preprocess the MRI image, specifically: Extracting the MRI image of the target patient's prognosis to obtain an MRI grayscale image, performing floating-point image conversion on the MRI grayscale image to obtain a converted MRI grayscale image; Correcting the converted MRI grayscale image using N4 bias correction to obtain a corrected training MRI grayscale image; Histogram matching is performed on the corrected training MRI grayscale image to obtain a preprocessed 3D MRI image.

3. The method for intelligent analysis of rectal cancer prognosis based on two-stage 3D feature extraction according to claim 1, characterized in that: A three-dimensional convolutional neural network based on the 3D-Unet architecture is constructed, and the three-dimensional convolutional neural network is trained in stages, specifically: A three-dimensional convolutional neural network is constructed based on the 3D-Unet architecture, and the encoder part is trained in the first stage, in which the three-dimensional convolution kernel is used to gradually downsample the input three-dimensional image; The encoder part is implemented by four layers of downsampling operations, where each layer of downsampling operation uses two convolution modules, a sparse module, and a maximum pooling module to extract the feature information of the 3D image and avoid network overfitting; The decoding part is implemented by four layers of upsampling operations, where each upsampling operation uses two convolution operations, a sparse module, and a deconvolution module to restore the size of the original image layer by layer, and finally obtains the segmentation output result through the activation function; Reading rectal cancer diagnosis and tumor area division results from historical rectal cancer medical record data as label information, constructing a first-stage data sample based on the data label information combined with rectal cancer images, and dividing the data into first-stage training data, verification data, and test data according to a preset ratio; Use the Dice loss function to pre-train the encoder and obtain the encoder weights.

4. The method for intelligent analysis of rectal cancer prognosis based on two-stage 3D feature extraction according to claim 3, characterized in that: The loss is calculated using the Dice loss function, which is expressed as: Among them, X represents the predicted output 3D tumor image, and Y represents the label information.

5. The method for intelligent analysis of rectal cancer prognosis based on two-stage 3D feature extraction according to claim 3, characterized in that: The second phase of network construction and training of the three-dimensional convolutional neural network is specifically as follows: Obtaining the encoder structure and weights of the three-dimensional convolutional neural network after the first stage of training, and filling the decoder structure with four fully connected layers to replace the decoder and skip connections trained in the first stage; Reading whether a patient recurrence event has occurred and the follow-up time in the historical rectal cancer medical record data to generate data label information, constructing a second-stage data sample based on the data label information combined with the historical risk score, and dividing it into second-stage training data, verification data, and test data according to a preset ratio; The weights of the encoder part of the first stage are used as the pre-trained weights of the encoder part of the second segment. The cox partial likelihood loss function is used to calculate the loss based on the training data for model training. During the training process, the validation data is used to adjust the hyperparameters and test the screening model. The screened model is tested with the test data.

6. The method for intelligent analysis of rectal cancer prognosis based on two-stage 3D feature extraction according to claim 5, characterized in that: The loss is calculated using the cox partial likelihood loss function, which is expressed as: Where n represents the batch size of the three-dimensional convolutional neural network. A batch contains n training data samples, each of which contains the risk score s, follow-up time t, and whether a recurrence event e occurs. k represents the number of recurrence events in a batch sample, and i and j represent the number of items in the batch sample.

7. The method for intelligent analysis of rectal cancer prognosis based on two-stage 3D feature extraction according to claim 1, characterized in that: The risk score is combined with the clinical data of the target patient to perform prognostic assessment and obtain the survival rate of the target patient, specifically: Obtaining a risk score set corresponding to the second-stage test data of the three-dimensional convolutional neural network, performing ROC curve analysis on the risk score set to obtain an optimal cutoff point, and dividing the patients into a high-risk group and a low-risk group based on the optimal cutoff point; Obtaining a risk score for the target patient, and generating a risk score feature according to a determination result of the risk score and a risk score threshold corresponding to an optimal cutoff point; The clinical characteristics and risk score characteristics of the target patients were obtained, and KM survival analysis was used to obtain the low-risk score curve and the high-risk score curve, and the predicted survival rate of the target patients was output.

Images