Non-small cell lung cancer patient lifetime prediction method and device based on deep learning
Through deep learning technology, combining clinical and multimodal gene data, extracting and fusion characteristics, the problem of insufficient survival prediction in traditional methods is solved, and more accurate and reliable survival prediction is achieved, providing strong clinical decision-making. data support.
Patent Information
- Application Number
- CN202410925141.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-11
- Publication Date
- 2024-11-12
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
The traditional non-small cell lung cancer survival prediction method is based on clinical and pathological characteristics analysis, and cannot fully reveal the complexity of the disease, resulting in insufficient prediction accuracy and reliability. In addition, deep learning technology insufficient utilization of data modes, which can only predict patient survival results. And the survival time is relatively short.
Using a deep learning-based method, the clinical data set and multimodal gene data set of patients with non-small cell lung cancer are obtained through the collector, and the features are extracted using matching neural network models, and input them into the multi-interactive memory network for feature fusion. Finally, survival prediction results are generated using the survival prediction model.
Improve the accuracy and reliability of survival prediction, can express patient characteristics more accurately, capture complex structures and patterns in the data, reveal dynamic changes and key drivers of survival, provide strong data support for clinical decision-making, and optimize treatment Strategy and significantly improve patient outcomes.
Smart Images

Figure CN118942690A_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the field of survival prediction for patients with non-small cell lung cancer, and in particular to a method and device for predicting survival for patients with non-small cell lung cancer based on deep learning. Background Art
[0002] Non-small cell lung cancer has the characteristics of poor prognosis and high mortality, so it is of great significance to predict the survival of patients with non-small cell lung cancer. Traditional cancer survival prediction is mainly based on the analysis of clinical and pathological characteristics, but this information often cannot fully reveal the complexity of the disease, thus limiting the accuracy and reliability of survival prediction. Current deep learning technology faces several challenges, including insufficient utilization of data modalities, only being able to predict patient survival outcomes, and predicting shorter patient survival. Summary of the invention
[0003] The purpose of this application is to provide a method and device for predicting the survival period of patients with non-small cell lung cancer based on deep learning, which can improve the accuracy and reliability of survival prediction and provide strong data support for the clinical decision-making process.
[0004] To achieve the above objectives, this application provides the following solutions:
[0005] In the first aspect, the present application provides a device for predicting the survival of patients with non-small cell lung cancer based on deep learning, comprising: a collector, a processor and a display. The collector is connected to the processor; the collector is used to collect clinical data sets and multimodal gene data sets of patients with non-small cell lung cancer. The processor is connected to the display; the processor is used to obtain clinical data sets and multimodal gene data sets of patients with non-small cell lung cancer; each modal gene data set in the clinical data set and the multimodal gene data set is respectively extracted using a matching neural network model; all extracted features are spliced together and input into a multi-interactive memory network to obtain a fused feature vector; based on the fused feature vector, a survival prediction model is used to generate a survival prediction result for patients with non-small cell lung cancer. The display is used to display the survival prediction result of patients with non-small cell lung cancer.
[0006] In the second aspect, the present application provides a method for predicting the survival of patients with non-small cell lung cancer based on deep learning, including: obtaining a clinical data set and a multimodal gene data set of patients with non-small cell lung cancer; extracting features from each modal gene data set in the clinical data set and the multimodal gene data set using a matching neural network model; concatenating all extracted features and inputting them into a multi-interactive memory network to obtain a fused feature vector; and generating a survival prediction result for patients with non-small cell lung cancer using a survival prediction model based on the fused feature vector.
[0007] According to the specific embodiments provided in this application, this application discloses the following technical effects:
[0008] The present application provides a method and device for predicting the survival of patients with non-small cell lung cancer based on deep learning. The processor integrates clinical data and multimodal genetic data. The complementarity of multimodal data provides more accurate and reliable information, and can more accurately express the characteristics of patients. The processor uses a neural network that matches each gene feature to extract features, automatically learns abstract features in the data, and can adapt well to the features of the gene dimension and capture deeper feature representations in the data. The processor uses a multi-interactive memory network for feature fusion, which can better capture the complex structures and patterns in the data. The processor finally uses a survival prediction model to generate survival prediction results for patients with non-small cell lung cancer and displays them on a display, effectively revealing the dynamic changes and key driving factors of the survival of patients with non-small cell lung cancer, thereby providing strong data support for the clinical decision-making process. BRIEF DESCRIPTION OF THE DRAWINGS
[0009] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.
[0010] Figure 1 This is a schematic diagram of the structure of a device for predicting the survival period of patients with non-small cell lung cancer based on deep learning in one embodiment of the present application;
[0011] Figure 2 A schematic diagram of the structure of an interaction module provided in one embodiment of the present application;
[0012] Figure 3 This is a flowchart of a method for predicting the survival period of patients with non-small cell lung cancer based on deep learning in one embodiment of the present application;
[0013] Figure 4 This is a schematic diagram of the principle of a method for predicting the survival period of patients with non-small cell lung cancer based on deep learning in one embodiment of the present application. DETAILED DESCRIPTION
[0014] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.
[0015] In order to make the above-mentioned objects, features and advantages of the present application more obvious and easy to understand, the present application is further described in detail below with reference to the accompanying drawings and specific implementation methods.
[0016] The present application embodiment provides a device for predicting the survival period of patients with non-small cell lung cancer based on deep learning, such as Figure 1 As shown, it includes: a collector 102, a processor 104 and a display 106. The collector 102 is connected to the processor 104; the collector 102 is used to collect clinical data sets and multimodal gene data sets of patients with non-small cell lung cancer. The processor 104 is connected to the display 106; the processor 104 is used to obtain clinical data sets and multimodal gene data sets of patients with non-small cell lung cancer; each modal gene data set in the clinical data set and the multimodal gene data set is respectively extracted using a matching neural network model; all extracted features are spliced together and input into a multi-interactive memory network to obtain a fused feature vector; according to the fused feature vector, a survival prediction model is used to generate a survival prediction result for patients with non-small cell lung cancer. The display 106 is used to display the survival prediction result of patients with non-small cell lung cancer.
[0017] The collector 102 may be, but is not limited to, various desktop computers, laptops, smart phones, tablet computers, IoT devices, and portable wearable devices. IoT devices may be smart speakers, smart TVs, smart air conditioners, smart car-mounted devices, etc. Portable wearable devices may be smart watches, smart bracelets, head-mounted devices, etc.
[0018] The survival prediction results displayed on the display 106 can provide a basis for doctors to select clinical treatment plans and evaluate patient prognosis.
[0019] In another exemplary embodiment of the present application, in order to further improve the accuracy of the survival prediction results, after the processor 104 obtains the clinical data set and the multimodal gene data set of the non-small cell lung cancer patient, it is also necessary to perform data processing on the clinical data set and the multimodal gene data set to extract key data. The specific process is: extract the fields that affect the survival period from the clinical data set, and convert them into a 256-dimensional feature vector through the embedding layer; use principal component analysis to extract the features with the largest gene feature variance value for each modal gene data set in the multimodal gene data set, and convert them into matrices respectively, to obtain the feature matrix corresponding to each modal gene data set. The multimodal gene data set includes a miRNA data set, an mRNA data set, a CNV data set, and a DNAm data set.
[0020] Take a specific example to illustrate the process and results of extracting key data: In the clinical data set of patients with non-small cell lung cancer, the following fields are extracted:
[0021] submitter_id,project_id,days_to_death,gender,race,vital_status,age_at_diagnosis,days_to_last_follow_up,tumor_stage,additional_pharmaceutical_therapy,additional_r adiation_therapy,additional_surgery_locoregional_procedure,new_tumor_event_after_initial_treatment,pathologic_M,pathologic_N,pathologic_T,person_neoplasm_cance r_status,postoperative_rx_tx,primary_therapy_outcome_success,tissue_or_organ_of_origin,time,and converted into a total 256-dimensional feature vector through the embedding layer.Wherein, submitter_id represents the submitter ID, project_id represents the project ID, days_to_death represents the number of days to death, gender represents gender, race represents race, vital_status represents survival status, age_at_diagnosis represents age at diagnosis, days_to_last_follow_up represents the number of days from the last follow-up to diagnosis, tumor_stage represents tumor stage, additional_pharmaceutical_therapy represents additional drug therapy, additional_radiation_therapy represents additional radiotherapy, additional_surgery_locoregional_ procedure indicates additional local surgery, new_tumor_event_after_initial_treatment indicates new tumor event after initial treatment, pathologic_M indicates pathologic M stage, pathologic_N indicates pathologic N stage, pathologic_T indicates pathologic T stage, person_neoplasm_cancer_status indicates individual tumor cancer status, postoperative_rx_tx indicates postoperative treatment, primary_therapy_outcome_success indicates successful primary treatment outcome, tissue_or_organ_of_origin indicates tissue or organ origin, and time indicates survival time (in years).
[0022] For multimodal gene datasets, principal component analysis was used to extract features with the largest gene feature variance values for miRNA datasets, mRNA datasets, CNV datasets, and DNAm datasets, which were 576, 30976, 16900, and 48841 features, respectively. These features were converted into matrices with dimensions of 24*24, 176*176, 130*130, and 221*221.
[0023] In another exemplary embodiment of the present application, in order to adapt well to the characteristics of the gene dimension and capture the deeper feature representation in the data, it is necessary to use a neural network that matches each gene feature for training and feature extraction. The processor uses a matching neural network model to extract features for each modality gene data set in the clinical data set and the multimodal gene data set, respectively. Specifically, the following method is used: the 256-dimensional feature vector is sequentially passed through the first input layer, normalization layer, random inactivation layer, linear layer, activation function and the first output layer to obtain a 512-dimensional feature vector of the clinical data set; the neural network ResNet101 is used to extract features for the feature matrix corresponding to the miRNA data set; the neural network EffientNet_b0 is used to extract features for the feature matrix corresponding to the mRNA data set; the neural network EffientNet_b4 is used to extract features for the feature matrix corresponding to the CNV data set; the neural network EffientNet_b5 is used to extract features for the feature matrix corresponding to the DNAm data set; wherein the dimensions of the output data of the output layers of the neural network ResNet101, the neural network EffientNet_b0, the neural network EffientNet_b4 and the neural network EffientNet_b5 are all 512. The feature results extracted above are all 512-dimensional feature vectors.
[0024] In another exemplary embodiment of the present application, the multi-interaction memory network is mainly composed of multiple interaction modules through multiple rounds of interaction, and the multiple interaction modules are connected in sequence. The structure of the interaction module is as follows Figure 2 As shown in the figure. Each interaction module is mainly composed of an attention module and a gated recurrent unit (GRU). The memory information obtained by the interaction module is used as a query vector, and the attention weight is calculated through the attention mechanism. Finally, these attention weights are used to weight the input features to obtain the attention vector. Finally, the vector and the input features are interacted using GRU to obtain the memory information of the next interaction module. The memory information of the first interaction module is copied from the input feature vector. The attention vector formula is as follows:
[0025]
[0026] m is the dimension of the input sequence memory, and v is the attention vector.
[0027] The formula for GRU is as follows:
[0028] z t =σ(w z ·[ν,x])
[0029] r t =σ(w r ·[ν,x])
[0030]
[0031] Among them, h(t) is the updated memory vector memory, z t is the update gate, r t is the reset gate. Here, the number of interactions of the memory module is set to 3. σ is the sigmoid activation function, w z is the weight matrix of the update gate, x is the input of the current network, w r is the weight matrix of the reset gate, It is a candidate memory cell used to update the hidden state at the current moment.
[0032] In another exemplary embodiment of the present application, the survival period prediction module is a multi-layer perceptron, and the specific structure is:
[0033] (0):Dropout(p=0.5,inplace=False)
[0034] (1):Linear(in_features=512, out_features=2048, bias=True)
[0035] (2): ReLU (inplace = True)
[0036] (3):BatchNorm1d(2048,eps=1e-05,momentum=0.1,affine=True,track_running_s tats=True)
[0037] (4):Dropout(p=0.5,inplace=False)
[0038] (5):Linear(in_features=2048, out_features=1024, bias=True)
[0039] (6): ReLU (inplace = True)
[0040] (7):BatchNorm1d(1024,eps=1e-05,momentum=0.1,affine=True,track_running_s tats=True)
[0041] (8):Dropout(p=0.5,inplace=False)
[0042] (9):Linear(in_features=1024, out_features=512, bias=True)
[0043] (10): ReLU (inplace = True)
[0044] (11):BatchNorm1d(512,eps=1e-05,momentum=0.1,affine=True,track_running_s tats=True)
[0045] (12):Dropout(p=0.5,inplace=False)
[0046] (13):Linear(in_features=512, out_features=512, bias=True)
[0047] (14): ReLU (inplace = True)
[0048] (15):Linear(in_features=512, out_features=30, bias=True)
[0049] (16):Sigmoid()
[0050] The multi-layer perceptron (MLP) includes a second input layer, four hidden layers and a second output layer connected in sequence; the dimension of the input data of the second input layer is 2048; the number of neurons in the four hidden layers is 2048, 1024, 512 and 512 respectively, and the activation function of the hidden layer is ReLU; the number of neurons in the output layer is 30, and the activation function of the output layer is Sigmoid. The final risk assessment result (survival prediction result) is generated through Sigmoid.
[0051] The survival prediction result is the survival probability for each year within the preset period. If the preset period is 30 years, the 30 output points are the survival probabilities of 1-30 years respectively. These survival probabilities together constitute the patient's survival period.
[0052] Based on the same inventive concept, the present application embodiment also provides a method for predicting the survival period of patients with non-small cell lung cancer based on deep learning, such as Figure 3 As shown, the steps include steps 201 to 204. Among them:
[0053] Step 201: Obtain a clinical dataset and a multimodal gene dataset of patients with non-small cell lung cancer.
[0054] Step 202: extract features from the clinical dataset and each modality gene dataset in the multimodal gene dataset using a matching neural network model.
[0055] Step 203: concatenate all extracted features and input them into a multi-interactive memory network to obtain a fused feature vector.
[0056] Step 204: Generate a survival prediction result for non-small cell lung cancer patients using a survival prediction model according to the fused feature vector.
[0057] The implementation of the above steps 201 to 204 integrates clinical data and multimodal genetic data. The complementarity of multimodal data provides more accurate and reliable information and can more accurately express the characteristics of patients. A neural network that matches each gene feature is used to extract features and automatically learn abstract features in the data, which can well adapt to the features of the gene dimension and capture deeper feature representations in the data. A multi-interactive memory network is used for feature fusion to better capture the complex structure and pattern in the data. Finally, a survival prediction model is used to generate survival prediction results for patients with non-small cell lung cancer, which effectively reveals the dynamic changes and key driving factors of the survival of patients with non-small cell lung cancer, and can provide strong data support for the clinical decision-making process. The results can not only optimize the treatment strategy, but also significantly improve the prognosis of patients.
[0058] In another exemplary embodiment of the present application, in order to make full use of data of different modalities, the multimodal gene dataset includes a miRNA dataset, an mRNA dataset, a CNV dataset, and a DNAm dataset.
[0059] In another exemplary embodiment of the present application, in order to further improve the accuracy of the survival prediction results, after obtaining the clinical data set and the multimodal gene data set of non-small cell lung cancer patients, it is also necessary to process the clinical data set and the multimodal gene data set to extract key data. Then after the above step 201, the method may also include the following steps 301 to 302. Among them:
[0060] Step 301: extract fields that affect survival from the clinical data set and convert them into 256-dimensional feature vectors through an embedding layer;
[0061] Step 302: Use principal component analysis to extract the features with the largest gene feature variance value from each modal gene data set in the multimodal gene data set, and convert them into matrices to obtain the feature matrix corresponding to each modal gene data set. The multimodal gene data set includes miRNA data set, mRNA data set, CNV data set and DNAm data set.
[0062] In another exemplary embodiment of the present application, in order to adapt well to the characteristics of the gene dimension and capture the deeper feature representation in the data, it is necessary to use a neural network matching each gene feature to train and extract features, then the above step 202 is replaced by the following steps 401 to 405:
[0063] Step 401: The 256-dimensional feature vector is sequentially passed through the first input layer, the normalization layer, the random dropout layer, the linear layer, the activation function and the first output layer to obtain a 512-dimensional feature vector of the clinical data set.
[0064] Step 402: Extract features from the feature matrix corresponding to the miRNA dataset using the neural network ResNet101.
[0065] Step 403: Extract features from the feature matrix corresponding to the mRNA data set using the neural network EffientNet_b0.
[0066] Step 404: Extract features from the feature matrix corresponding to the CNV data set using the neural network EffientNet_b4.
[0067] Step 405: The neural network EffientNet_b5 is used to extract features from the feature matrix corresponding to the DNAm data set; wherein the dimensions of the output layer output data of the neural network ResNet101, the neural network EffientNet_b0, the neural network EffientNet_b4 and the neural network EffientNet_b5 are all 512. The above extracted feature results are all 512-dimensional feature vectors.
[0068] In another exemplary embodiment of the present application, the survival period prediction module is a multilayer perceptron, which includes a second input layer, four hidden layers and a second output layer connected in sequence; the dimension of the input data of the second input layer is 2048; the number of neurons in the four hidden layers is 2048, 1024, 512 and 512 respectively, and the activation function of the hidden layer is ReLU; the number of neurons in the output layer is 30, and the activation function of the output layer is Sigmoid.
[0069] The principle of this method is Figure 4 shown. Figure 4 In which clinical means clinical, NeuralNetwork means neural network, MMN means multiple interactive memory network, and survivalprobability means survival probability.
[0070] In order to verify the comparison between the existing survival prediction methods and this method, the classic CPH model, random survival forest RSF, proportional hazard method DeepSurv, non-proportional DeepHit, multi-task logistic regression MTLR and Nnet-surival were considered for comparison. Four evaluation indicators, C-index, Ctd, IBS and INBLL, were used for comparison, as shown in Table 1.
[0071] Table 1 Evaluation indicators
[0072] C-index Ctd IBS INBLL CPH 0.677 0.677 0.217 0.645 RSF 0.654 0.701 0.185 0.548 DeepSurv 0.459 0.460 0.222 0.641 CoxTime 0.767 0.789 0.136 0.435 DeepHit 0.603 0.652 0.163 0.536 MTLR 0.735 0.785 0.136 0.474 Net-survival 0.496 0.495 0.378 1.626 This method 0.748 0.757 0.172 0.513
[0073] This method achieved the best C-index and Ctd scores, and also had excellent performance in IBS and INBLL scores. This method has high accuracy and superiority.
[0074] The method consists of three key parts: feature extraction, feature fusion, and survival prediction. In the feature extraction stage, the model uses convolutional neural networks to extract features from clinical information, miRNA, mRNA, CNV, and DNAm. Then, in the feature fusion stage, a multi-interactive memory network is used to integrate and fuse these multimodal features. Finally, in the survival prediction stage, a multi-layer perceptron is used to perform long-term survival prediction on the fused features. Experiments on LUAD (lung adenocarcinoma) and LUSC (lung squamous cell carcinoma) datasets and in-depth analysis of survival curves have demonstrated its excellent predictive performance. This model successfully integrates multiple data sources, effectively reveals the dynamic changes and key drivers of patient survival, and can provide strong data support for the clinical decision-making process. The results not only optimize the treatment strategy, but also significantly improve the patient's prognosis.
[0075] The advantages of this application are as follows:
[0076] 1. Compared with a single modality, multimodal genetic data fusion integrates different types of information, can provide a more comprehensive and multi-angle vision, and more accurately express the characteristics of patients. The complementarity of multimodal data provides more accurate and reliable information. Multimodal data fusion can significantly improve the performance of the system and enhance the robustness and stability of the system.
[0077] 2. A neural network that matches the characteristics of each gene is used to train and extract features. End-to-end learning can be performed, and the abstract features in the data can be automatically learned. It can adapt well to the characteristics of the gene dimension and capture deeper feature representations in the data.
[0078] 3. Using multi-interaction networks for feature fusion can make the effect of feature fusion better and enhance the ability of multimodal data processing. Through the fusion of interactive memory units, the complex structures and patterns in the data can be better captured.
[0079] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, 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 specification.
[0080] This article uses specific examples to illustrate the principles and implementation methods of this application. The description of the above embodiments is only used to help understand the method and core ideas of this application. At the same time, for those skilled in the art, according to the ideas of this application, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting this application.
Claims
1. A device for predicting survival of patients with non-small cell lung cancer based on deep learning, characterized in that: include: Collector, processor and display; The collector is connected to the processor; The collector is used to collect clinical data sets and multimodal gene data sets of patients with non-small cell lung cancer; The processor is connected to the display; The processor is used to obtain a clinical dataset and a multimodal gene dataset of a patient with non-small cell lung cancer; A matching neural network model is used to extract features from each modality of the clinical dataset and the multimodal gene dataset; All the extracted features are concatenated and input into a multi-interactive memory network to obtain a fused feature vector; and a survival prediction model is used to generate a survival prediction result of a patient with non-small cell lung cancer according to the fused feature vector; The display is used to display the survival prediction results of patients with non-small cell lung cancer.
2. The device for predicting survival period of patients with non-small cell lung cancer based on deep learning according to claim 1, characterized in that: The processor is further configured to: Extracting fields that affect survival from the clinical dataset and converting them into 256-dimensional feature vectors through an embedding layer; Principal component analysis is used to extract features with the largest gene feature variance value from each modal gene dataset in a multimodal gene dataset, and the features are converted into matrices respectively to obtain a feature matrix corresponding to each modal gene dataset; the multimodal gene dataset includes a miRNA dataset, an mRNA dataset, a CNV dataset and a DNAm dataset.
3. The device for predicting survival period of patients with non-small cell lung cancer based on deep learning according to claim 1, characterized in that: A matching neural network model is used to extract features from each modality of the clinical dataset and the multimodal gene dataset, including: The 256-dimensional feature vector is sequentially passed through a first input layer, a normalization layer, a random dropout layer, a linear layer, an activation function, and a first output layer to obtain a 512-dimensional feature vector of the clinical data set; The neural network ResNet101 was used to extract features from the feature matrix corresponding to the miRNA dataset; The neural network EffientNet_b0 is used to extract features from the feature matrix corresponding to the mRNA dataset; The neural network EffientNet_b4 is used to extract features from the feature matrix corresponding to the CNV data set; The neural network EffientNet_b5 is used to extract features from the feature matrix corresponding to the DNAm data set; Among them, the dimensions of the output layer output data of the neural network ResNet101, the neural network EffientNet_b0, the neural network EffientNet_b4 and the neural network EffientNet_b5 are all 512.
4. The device for predicting survival period of patients with non-small cell lung cancer based on deep learning according to claim 1, characterized in that: The survival period prediction model is a multi-layer perceptron; The multilayer perceptron comprises a second input layer, four hidden layers and a second output layer connected in sequence; The dimension of the input data of the second input layer is 2048; the number of neurons in the four hidden layers is 2048, 1024, 512 and 512 respectively, and the activation function of the hidden layer is ReLU; the number of neurons in the output layer is 30, and the activation function of the output layer is Sigmoid.
5. The device for predicting survival period of patients with non-small cell lung cancer based on deep learning according to claim 1, characterized in that: The survival period prediction result is the survival probability for each year within a preset number of years.
6. A method for predicting survival of patients with non-small cell lung cancer based on deep learning, characterized in that: include: Obtain clinical datasets and multimodal genetic datasets of patients with non-small cell lung cancer; A matching neural network model is used to extract features from each modality of the clinical dataset and the multimodal gene dataset; All extracted features are concatenated and input into the multi-interactive memory network to obtain the fused feature vector; According to the fused feature vector, a survival prediction model is used to generate a survival prediction result for patients with non-small cell lung cancer.
7. The method for predicting survival of patients with non-small cell lung cancer based on deep learning according to claim 6, characterized in that: Obtain clinical datasets and multimodal genetic datasets of patients with non-small cell lung cancer, and later include: Extracting fields that affect survival from the clinical dataset and converting them into 256-dimensional feature vectors through an embedding layer; Principal component analysis is used to extract features with the largest gene feature variance value from each modal gene dataset in a multimodal gene dataset, and the features are converted into matrices respectively to obtain a feature matrix corresponding to each modal gene dataset; the multimodal gene dataset includes a miRNA dataset, an mRNA dataset, a CNV dataset and a DNAm dataset.
8. The method for predicting survival of patients with non-small cell lung cancer based on deep learning according to claim 7, characterized in that: A matching neural network model is used to extract features from each modality of the clinical dataset and the multimodal gene dataset, including: The 256-dimensional feature vector is sequentially passed through a first input layer, a normalization layer, a random dropout layer, a linear layer, an activation function, and a first output layer to obtain a 512-dimensional feature vector of the clinical data set; The neural network ResNet101 was used to extract features from the feature matrix corresponding to the miRNA dataset; The neural network EffientNet_b0 is used to extract features from the feature matrix corresponding to the mRNA dataset; The neural network EffientNet_b4 is used to extract features from the feature matrix corresponding to the CNV data set; The neural network EffientNet_b5 is used to extract features from the feature matrix corresponding to the DNAm data set; Among them, the dimensions of the output layer output data of the neural network ResNet101, the neural network EffientNet_b0, the neural network EffientNet_b4 and the neural network EffientNet_b5 are all 512.
9. The method for predicting survival of patients with non-small cell lung cancer based on deep learning according to claim 6, characterized in that: The survival period prediction model is a multi-layer perceptron; The multilayer perceptron comprises a second input layer, four hidden layers and a second output layer connected in sequence; The dimension of the input data of the second input layer is 2048; the number of neurons in the four hidden layers is 2048, 1024, 512 and 512 respectively, and the activation function of the hidden layer is ReLU; the number of neurons in the output layer is 30, and the activation function of the output layer is Sigmoid.
10. The method for predicting survival of patients with non-small cell lung cancer based on deep learning according to claim 6, characterized in that: The survival period prediction result is the survival probability for each year within a preset number of years.
Citation Information
Cited By
Lung cancer lifetime prediction model training method based on feature selection
CN120636850A
Lung cancer lifetime prediction system based on prognosis factor multi-data fusion
CN121812154A
Lung cancer survival prediction system based on prognosis factor multi-data fusion
CN121812154B
