Information processing method, device and equipment for pulmonary nodule malignant risk assessment

Abstract

The invention discloses an information processing method, device and equipment for pulmonary nodule malignant risk assessment, and relates to the technical field of pulmonary nodule malignant risk assessment. According to the method, the intricate relation among various characteristics of cfDNA is considered, four key characteristics and four comprehensive characteristics are extracted, and the length, the tail end motif, the sequence motif, the methylation rate and the relevance among the four characteristics of the cfDNA are covered. The features are coded and converted to obtain eight images, and the features are fused through a convolution fusion strategy. The pulmonary nodule malignant risk assessment method based on the cfDNA can break through the limitation of a traditional method, and can effectively learn the complex mode of the cfDNA on the pulmonary nodule malignant risk assessment, thereby realizing the accurate identification of the pulmonary nodule property, and improving the accuracy of predicting the malignant risk by means of the deep learning identification capability. The method provided by the invention is of great significance to comprehensive and accurate analysis of cfDNA characteristics.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of lung nodule malignant risk assessment, in particular to a lung nodule malignant risk assessment information processing method, device and equipment. BACKGROUND

[0002] Cell-free DNA (cfDNA) in plasma is composed of a mixture of fragmented DNA molecules released from various tissues in the body. When cells die, free nucleic acids enter the circulation as part of normal cell turnover or pathology. In healthy humans, most cfDNA molecules in plasma are derived from blood cells. When an organ or tissue undergoes apoptosis or necrosis, more cfDNA from the organ or tissue appears in the plasma.

[0003] Each cfDNA fragment has the molecular characteristics of the tissue cells from which it originated.

[0004] 1) Since DNA methylation is spatiotemporal specific in different tissues and cells, each cfDNA fragment has the DNA methylation state characteristics of the tissue cells from which it originated; 2) Since the expression of nuclease genes in different tissues is spatiotemporal specific, fragmentomics, such as fragment length, end motif, sequence motif, etc., can represent the state characteristics of the tissue cells from which they originated to some extent.

[0005] Deep learning technology, with its powerful automatic feature extraction and complex pattern recognition capabilities, has shown outstanding advantages in the analysis of cfDNA, providing a powerful tool for unlocking these molecular codes. Kim M et al. developed a deep learning model integrating cfDNA methylation and fragment size maps in 2024, opening up new ideas for lung cancer risk prediction. Shen et al. (2024) trained a deep learning model for cancer based on cfDNA end motif features, which provides new possibilities for understanding early cancer characteristics.

[0006] The existing patent CN202310112740.3 uses cfDNA methylation information to predict lung nodule malignant risk. However, this patent only predicts the possibility of lung nodule malignancy based on the methylation of 6 genes (ARL9, SHOX2, RASSF1A, APC, PCDHGC5, PTGER4, etc. 6 gene sequences or any combination of fragments), without using the methylation information of the whole cfDNA, therefore, it cannot comprehensively and accurately analyze the cfDNA characteristics.

[0007] The methylation information of cfDNA is closely related to the fragmentomics of cfDNA. Because of the spatiotemporal specificity of DNA methylation, the DNA methylation characteristics of cells of the same tissue in different states (normal / abnormal) are greatly different. It is well known that DNA methylation seriously affects the action of nucleases on DNA, so there is a biological link between the state of DNA and the length and end motif of cfDNA. The spatiotemporal expression of these nucleases is closely related, so the sequence characteristics of cfDNA are closely related to diseases, and the sequence motif and end motif can represent the sequence characteristics of genes to some extent. The complex relationship between these characteristics is difficult for non-biological circles to capture.

[0008] In view of this, the present application is proposed. SUMMARY

[0009] The purpose of the present application is to provide an information processing method, device and equipment for lung nodule malignant risk assessment to solve the above technical problems.

[0010] The present application is implemented as follows: In a first aspect, the present application provides a method for constructing a lung nodule malignant risk assessment model based on cfDNA multi-features, comprising the following steps: (1) Extracting the features of cfDNA of prior data or new data, and encoding each extracted feature to obtain a feature map; the features include cfDNA length, end motif, sequence motif, methylation rate, cfDNA length and methylation rate comprehensive feature, cfDNA length and end motif comprehensive feature, end motif and methylation rate comprehensive feature, and end motif and sequence motif comprehensive feature; each sample of the prior data or new data has a binary classification label of lung nodule property; (2) Convolution fusion: the fusion of the feature map comprises the following steps: inputting each feature map into a neural network structure, extracting representative features through a convolution layer sliding window, and then reducing the dimension through a pooling layer to retain key features; the key features are spliced along a preset dimension to form a composite feature; (3) Constructing a lung nodule malignant risk assessment model based on a Transformer architecture: The composite feature is input into the Transformer for feature extraction and mapping to lung nodule malignant risk, thereby constructing a lung nodule malignant risk assessment model.

[0011] In a second aspect, the present application provides an information processing method for lung nodule malignant risk assessment, comprising the following steps: The lung nodule malignant risk assessment model is constructed according to the construction method of the lung nodule malignant risk assessment model based on the cfDNA multi-features, and then the cfDNA sequencing data of the sample to be detected is input; the features of the cfDNA of the sample to be detected are extracted in sequence, and the features after convolution fusion are output, so that the lung nodule malignant risk assessment result is obtained.

[0012] In a third aspect, the present application provides a device for lung nodule malignant risk assessment, comprising an input module, a control module and an output module. The input module is configured to input cfDNA sequencing data of prior data or new data. The control module comprises a cfDNA feature extraction module, a convolution fusion module and a lung nodule malignant risk assessment model based on Transformer. The output module is configured to output lung nodule malignant risk. The cfDNA feature extraction module is used to extract the features of the cfDNA of the prior data or the new data, and to encode each extracted feature to obtain a feature map; the features include cfDNA length, terminal motif, sequence motif, methylation rate, cfDNA length and methylation rate comprehensive feature, cfDNA length and terminal motif comprehensive feature, terminal motif and methylation rate comprehensive feature, and terminal motif and sequence motif comprehensive feature; each sample of the prior data or the new data has a binary classification label of lung nodule property. The convolution fusion module is used to fuse the feature map: inputting each feature map into a neural network structure, extracting representative features through a convolution layer sliding window, and then reducing dimensions through a pooling layer to retain key features; the key features are spliced along a preset dimension to form a composite feature. The lung nodule malignant risk assessment model based on Transformer is used to input the composite feature into Transformer for feature extraction, map to lung nodule malignant risk, thereby constructing a lung nodule malignant risk assessment model; and input the integrated feature after convolution fusion of the sample to be detected into the lung nodule malignant risk assessment model to output a lung nodule malignant risk assessment result.

[0013] In a fourth aspect, the present application provides a device for lung nodule malignant risk assessment, comprising a processor and a memory, wherein the memory stores at least one instruction, at least one program, a code set or an instruction set, and the processor loads and executes the at least one instruction, the at least one program, the code set or the instruction set to realize the information processing method for lung nodule malignant risk assessment as described above.

[0014] In a fifth aspect, the present application provides a computer-readable storage medium, wherein the storage medium stores at least one instruction, at least one program, a code set or an instruction set, and the at least one instruction, the at least one program, the code set or the instruction set is loaded and executed by a processor to implement the information processing method for lung nodule malignancy risk assessment.

[0015] The present application has the following beneficial effects: Currently, there are limitations in predicting lung nodule malignancy risk using cfDNA characteristics, and the complex relationship between cfDNA length and methylation rate, terminal motifs, and terminal motifs and overall motifs is not fully considered, resulting in defects in the prediction model and the inability to deeply explore disease information.

[0016] The present application considers the complex relationship between various characteristics of cfDNA, extracts four key features and four comprehensive features, covers cfDNA length, terminal motifs, sequence motifs, methylation rate, and the correlation between the four features. Eight maps are obtained by encoding and converting the above features, and the features are fused by convolution fusion strategy. This convolution fusion method can effectively capture the spatial correlation between feature maps and tightly integrate the features to provide a highly structured input for subsequent lung nodule deep learning models, which is crucial for improving the model's understanding of complex data and prediction performance. The application of the Transformer model to lung nodule malignancy risk assessment can break through the limitations of traditional methods and effectively learn the complex patterns of cfDNA in lung nodule malignancy risk assessment, thereby achieving accurate identification of lung nodule properties and improving the accuracy of predicting malignant risk through deep learning recognition capabilities. BRIEF DESCRIPTION OF DRAWINGS

[0017] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following will briefly introduce the drawings needed in the embodiments. It should be understood that the following drawings only show some embodiments of the present application, and therefore should not be considered as limiting the scope. For those skilled in the art, other related drawings can also be obtained without creative labor.

[0018] Figure 1 Flowchart for lung nodule malignancy risk assessment based on cfDNA multi-features. DETAILED DESCRIPTION

[0019] The embodiments of the present application will now be described in detail with reference to the accompanying drawings, one or more examples of which are described below. Each example is provided as an explanation and not as a limitation of the present application. It is obvious to those skilled in the art that various modifications and changes can be made to the present application without departing from the scope or spirit of the present application. For example, features described or illustrated as part of one embodiment can be used in another embodiment to produce further embodiments.

[0020] Noun explanation: Prior data: data that already exists or is available before the current research, experiment, analysis or decision-making process begins.

[0021] New data refers to data collected specifically for the current research / experiment.

[0022] The present invention aims at the deficiencies of the prior art, integrates the key characteristics of cfDNA, innovates the algorithm and model, and is expected to improve the prediction accuracy and open up a new way for accurate prediction of lung nodule malignant risk. The deep learning model based on cfDNA multi-features has great significance for understanding lung nodules. It can improve the prediction accuracy of lung nodule malignant risk, and the lung nodule malignant risk assessment model provided by the present invention can comprehensively and accurately analyze the cfDNA features, and accurately predict the malignant risk with the deep learning recognition ability.

[0023] In the first aspect, the present invention provides a method for constructing a lung nodule malignant risk assessment model based on cfDNA multi-features, which comprises the following steps: (1) Extract the features of cfDNA of prior data or new data, and encode each extracted feature to obtain a feature map; the features include cfDNA length, terminal motif, sequence motif, methylation rate, cfDNA length and methylation rate comprehensive feature, cfDNA length and terminal motif comprehensive feature, terminal motif and methylation rate comprehensive feature, and terminal motif and sequence motif comprehensive feature; each sample of prior data or new data has a binary classification label of lung nodule property; (2) Convolution fusion: the feature map is fused by the following steps: inputting each feature map into a neural network structure, extracting representative features through a convolution layer sliding window, and then reducing dimension through a pooling layer to retain key features; the key features are spliced along a preset dimension to form a composite feature; (3) Constructing a lung nodule malignant risk assessment model based on a Transformer architecture: The composite feature is input into the Transformer for feature extraction and mapping to the lung nodule malignant risk, thereby constructing a lung nodule malignant risk assessment model.

[0024] In the above extracted features, cfDNA length, terminal motif, sequence motif and methylation rate are key features, and cfDNA length and methylation rate comprehensive feature, cfDNA length and terminal motif comprehensive feature, terminal motif and methylation rate comprehensive feature, and terminal motif and sequence motif comprehensive feature are all comprehensive features. The above features are fused by convolution fusion strategy to retain the complex relationship between the features, and the model for lung nodule malignant risk assessment is constructed by the Transformer algorithm.

[0025] The output result of the convolutional neural network structure is the key feature, that is, the output result of the neural network structure is the key feature. The key feature is the visual information that is automatically screened by the convolutional neural network through learning, which is the most effective and most representative visual information for completing a specific task (classification, identification, etc.), and can ignore irrelevant details and focus on the essential differences between different categories. It is the core basis for the convolutional neural network to realize accurate image analysis.

[0026] In a preferred embodiment of the application, the dimension of the feature map of the cfDNA length is (100-250) x 1 x 1, wherein 100-250 represents the cfDNA length. The dimension of the feature map of the terminal motif is 256 x 1 x 1, the dimension of the feature map of the sequence motif is 256 x 1 x 1, and the dimension of the feature map of the methylation rate is (50-1000) x 1 x 1; the dimension of the feature map of the comprehensive feature of cfDNA length and methylation rate is (100-250) x (50-1000) x 1; the dimension of the feature map of the comprehensive feature of cfDNA length and terminal motif is (100-250) x 256 x 1; the dimension of the feature map of the comprehensive feature of terminal motif and methylation rate is (50-1000) x 256 x 1; and the dimension of the comprehensive feature of terminal motif and sequence motif is 256 x 256 x 1.

[0027] The terminal motif refers to a conserved unit with specific structural or sequence characteristics located in the terminal region (N-terminal / C-terminal or 5'-terminal / 3'-terminal) of a nucleic acid.

[0028] The sequence motif refers to a conserved fragment with specific functions in a nucleic acid sequence, which realizes recognition, binding or catalysis and other biological functions through a conserved nucleotide arrangement pattern.

[0029] The methylation rate refers to the proportion of methylation-modified sites in a specific molecule (DNA, etc.) to the total potential modification sites, which is a core indicator for quantifying the degree of epigenetic modification or chemical modification. The present application specifically refers to the proportion of the number of methylated cytosines in each cfDNA sequence to the total number of cytosines.

[0030] Since the methylation rate is between 0 and 1, in one embodiment, the present application subdivides this range into 100 intervals, counts the number of cfDNA in each interval, and finally generates a 100 x 1 x 1 map for each sample for subsequent modeling. In other embodiments, the division interval of the methylation rate can also be adjusted as needed.

[0031] In the present application, the comprehensive feature of cfDNA length and methylation rate is the comprehensive feature including the cfDNA length feature and the methylation rate feature.

[0032] In a preferred embodiment of the application, the neural network structure is selected from a CNN, an RNN, a GNN, an MLP, a GAN, a diffusion model, a capsule network, or a sparse network. In a preferred embodiment, the neural network structure is a CNN.

[0033] In a second aspect, the present application provides an information processing method for lung nodule malignancy risk assessment, comprising the following steps: A lung nodule malignancy risk assessment model is constructed according to the method for constructing a lung nodule malignancy risk assessment model based on cfDNA multi-features, and then the cfDNA sequencing data and bioinformatics analysis results of the sample to be tested are input. The features of the cfDNA of the sample to be tested are extracted in sequence, and the features after convolution fusion are output. The lung nodule malignancy risk assessment model is used to obtain the lung nodule malignancy risk assessment result.

[0034] In a third aspect, the present application provides a device for lung nodule malignancy risk assessment, comprising an input module, a control module and an output module. The input module is configured to input the cfDNA sequencing data of the prior data or new data. The control module comprises a cfDNA feature extraction module, a convolution fusion module and a lung nodule malignancy risk assessment model based on Transformer. The output module is configured to output the lung nodule malignancy risk. The cfDNA feature extraction module is used to extract the features of the cfDNA of the prior data or new data, and to encode each extracted feature to obtain a feature map. The features include cfDNA length, terminal motif, sequence motif, methylation rate, cfDNA length and methylation rate comprehensive feature, cfDNA length and terminal motif comprehensive feature, terminal motif and methylation rate comprehensive feature, and terminal motif and sequence motif comprehensive feature. Each sample of the prior data or new data has a binary classification label of lung nodule property. The convolution fusion module is used to fuse the feature map: input each feature map into a neural network structure, extract representative features through a convolution layer sliding window, and then reduce the dimension through a pooling layer to retain key features. The key features are spliced along a preset dimension to form a composite feature. The lung nodule malignancy risk assessment model based on Transformer is used to input the comprehensive feature into a Transformer for feature extraction, map to lung nodule malignancy risk, thereby constructing a lung nodule malignancy risk assessment model. The composite feature after convolution fusion of the sample to be tested is input into the lung nodule malignancy risk assessment model, and the lung nodule malignancy risk assessment result is output.

[0035] In a preferred embodiment of the application, the dimension of the feature map of the cfDNA length is (100-250) x 1 x 1 dimension, wherein 100-250 represents the cfDNA length. The dimension of the feature map of the terminal motif is 256 x 1 x 1 dimension, the dimension of the feature map of the sequence motif is 256 x 1 x 1 dimension, the dimension of the feature map of the methylation rate is (50-1000) x 1 x 1 dimension; the dimension of the feature map of the comprehensive characteristics of cfDNA length and methylation rate is (100-250) x (50-1000) x 1 dimension; the dimension of the feature map of the comprehensive characteristics of cfDNA length and terminal motif is (100-250) x 256 x 1 dimension; the dimension of the feature map of the comprehensive characteristics of terminal motif and methylation rate is (50-1000) x 256 x 1 dimension; the dimension of the comprehensive characteristics of terminal motif and sequence motif is 256 x 256 x 1 dimension.

[0036] In a preferred embodiment of the application, the neural network structure is selected from CNN, RNN, GNN, MLP, GAN, diffusion model, capsule network or sparse network.

[0037] In a preferred embodiment of the application, the lung nodule malignant risk assessment result of the lung nodule malignant risk assessment model module based on the Transformer is lung nodule malignant high risk or lung nodule malignant low risk.

[0038] In a fourth aspect, the present application provides a lung nodule malignant risk assessment device, which comprises a processor and a memory, and the memory stores at least one instruction, at least one program, a code set or an instruction set, and the at least one instruction, at least one program, code set or instruction set is loaded and executed by the processor to realize the information processing method for lung nodule malignant risk assessment as described above.

[0039] In a fifth aspect, the present application provides a computer readable storage medium, which stores at least one instruction, at least one program, a code set or an instruction set, and the at least one instruction, at least one program, code set or instruction set is loaded and executed by the processor to realize the information processing method for lung nodule malignant risk assessment as described above.

[0040] Specifically, the electronic device can include a memory, a processor, a bus and a communication interface, which are directly or indirectly electrically connected to each other to realize the transmission or interaction of data. For example, these elements can be electrically connected to each other through one or more buses or signal lines. The processor can process information and / or data related to target identification to perform one or more functions described in the present application.

[0041] The memory can be, but is not limited to, random access memory (RAM), read only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), etc.

[0042] The processor can be an integrated circuit chip with signal processing capability. The processor can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it can also be a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field programmable gate array (FPGA) or other programmable logic device, a discrete gate or transistor logic device, a discrete hardware component.

[0043] In order to make the purpose, technical scheme and advantages of the embodiments of the present application more clear, the technical scheme in the embodiments of the present application will be clearly and completely described below. If the specific conditions are not specified in the embodiments, the conventional conditions or the conditions recommended by the manufacturer are used. If the reagents or instruments used are not specified by the manufacturer, they are all conventional products that can be purchased on the market.

[0044] The features and performances of the present application will be further described in detail below in combination with the embodiments.

[0045] Embodiment 1 This embodiment provides a model construction method for lung nodule malignant risk assessment. The flow chart of lung nodule malignant risk assessment is shown in Figure 1 .

[0046] It includes the following steps: (1) Extract cfDNA of the test set plasma sample, and sequence and bioinformatics analysis; each sample has a binary classification label of lung nodule property (high risk or low risk); then extract the following features respectively: cfDNA length, end motif, sequence motif, methylation rate, cfDNA length-methylation rate, cfDNA length-end motif comprehensive feature, end motif-methylation rate comprehensive feature, end motif-sequence motif comprehensive feature; wherein the first four are key features, and the last four are associated features.

[0047] a. End motif The present application encodes the end motif using the 4-mer method, which can obtain 256 different motifs. For each sample, the number of cfDNA corresponding to each motif is counted to obtain the data set . Finally, a 256x1x1 image is generated for each sample for subsequent model use.

[0048]

[0049] b. Sequence motif The present application performs 4-mer statistical analysis on cfDNA sequences, which is consistent with the end motif encoding, and can also obtain 256 motifs. Taking a sequence of length 166 bp as an example, according to the 4-mer statistical rule, (166-4+1), i.e. 163 4-mer fragments can be obtained. For each sample, the present application counts the number of each of the 256 motifs, to obtain the data set . Finally, a 256x1x1 image is constructed for each sample for subsequent model use.

[0050] .

[0051] c. cfDNA length cfDNA length is mostly distributed in the 80 to 300 bp interval. The present application carefully counts the number of cfDNA of each length in this interval to obtain the data set . Finally, a 221x1x1 image is generated for each sample for subsequent model analysis. It should be noted that cfDNA sequences longer than 300 bp or shorter than 80 bp are not within the scope of the present application.

[0052]

[0053] d. Methylation rate The present application processes cfDNA data. First, the number of cytosines in each cfDNA sequence is counted, and then the number of methylated cytosines is counted, from which the cytosine methylation rate is calculated. If the proportion of cytosines in a sequence is less than 20%, its methylation rate is not calculated. Since the methylation rate is between 0 and 1, the present application subdivides this range into 100 intervals, and in other embodiments, the range can also be subdivided into 50 intervals or 200 intervals. The number of cfDNA in each interval is counted to obtain the data set A 100x1x1 graph is finally generated for each sample for subsequent models.

[0054] .

[0055] e. cfDNA length-methylation rate Since DNA methylation affects nuclease action, it is particularly necessary to explore the complex relationship between cfDNA length and methylation rate. Based on the cfDNA length feature and the methylation rate feature, the data set is obtained, and then a 221x100x1 dimensional graph is constructed, and in other embodiments, a 100x221x1 dimensional graph can also be constructed. Among them, "221" corresponds to the 221 length values of cfDNA in the range of 80-300bp; "100" is to divide the methylation rate of 0-1 into 100 intervals. By counting the number of cfDNA in a specific length and a specific methylation rate interval, this graph is generated for subsequent models.

[0056] .

[0057] f. cfDNA length-end motif The end motif determined by the nuclease action position directly affects the cfDNA length. The present invention comprehensively considers the complex relationship between cfDNA length and end motif, and based on the cfDNA length feature and the end motif, the data set is obtained, and a 221x256x1 graph is constructed. Among them, "221" corresponds to the 221 values of cfDNA in the length range of 80-300bp; "256" refers to the 256 end motif types encoded by the 4-mer method. By accurately counting the number of cfDNA of a specific length and a specific end motif, this graph is formed for subsequent models.

[0058] .

[0059] g. end motif-methylation rate DNA methylation significantly affects the nuclease action position, i.e., the end motif. The present invention deeply studies the correlation between the two, and based on the end motif and the methylation rate feature, the data set is obtained, and then a 100x256x1 graph is generated, and in addition, in other embodiments, a 256x100x1 graph can also be generated. Among them, "100" is to subdivide the methylation rate of 0 to 1 into 100 intervals, and "256" refers to the 256 end motifs encoded by the 4-mer method. By strictly counting the number of cfDNA in a specific methylation rate interval and a specific end motif, the graph is constructed for subsequent models.

[0060] .

[0061] h. end motif - sequence motif Both end motif and sequence motif can characterize the state characteristics of the tissue cells from which they are derived to some extent. The present application focuses on the complex relationship between the two, and based on the characteristics of end motif and sequence motif, a data set is obtained , and then a 256x256x1 graph is constructed. Among them, the former 256 corresponds to the end motif category, and the latter 256 corresponds to the sequence motif category. By counting the number of cfDNA with specific end motif and specific sequence motif, this graph is formed for subsequent models.

[0062] .

[0063] (2) Convolutional fusion of the above 8 feature maps: By extracting the key features and correlation features of cfDNA, it not only helps to capture the methylation state and fragmentomic features of cfDNA, but also provides rich inputs for the construction and use of subsequent deep learning models.

[0064] Each sample contains 8 features, including 4 key features and 4 correlation features. At the same time, each sample corresponds to a binary classification label, which is used to determine the nature of the lung nodule. Specifically, the lung nodule with low risk of malignancy is marked as 0, and the lung nodule with high risk of malignancy is marked as 1.

[0065] Each feature map is input into a carefully designed CNN structure, which uses convolutional layers to slide windows to mine local information and extract rich and representative features (i.e. representative features), and then reduces the dimension through pooling layers to retain key features. After that, the features extracted from the 8 feature maps processed by CNN are spliced along a specific dimension to form a composite feature (such as Figure 1 ). This convolutional fusion method can effectively capture the spatial correlation between feature maps and tightly integrate features, providing a highly structured input that integrates multiple aspects of information for subsequent deep learning models, which is crucial for improving the model's understanding of complex data and prediction performance.

[0066] The above "key features" are not "key features", and the meanings of the two are different. The "key features" in this embodiment only refer to important key features.

[0067] (3) Constructing a lung nodule malignancy risk assessment model based on the Transformer architecture: Input the composite feature into the Transformer for feature extraction and mapping to the lung nodule malignancy risk, thereby constructing a lung nodule malignancy risk assessment model.

[0068] Among the extracted features, cfDNA length, terminal motif, sequence motif, and methylation rate are key features, and cfDNA length and methylation rate comprehensive features, cfDNA length and terminal motif comprehensive features, terminal motif and methylation rate comprehensive features, and terminal motif and sequence motif comprehensive features are comprehensive features. By convolution fusion strategy, the complex relationship between the features is retained, and the model for evaluating the risk of lung nodules is constructed by the Transformer algorithm.

[0069] The Transformer architecture has a unique self-attention mechanism, which makes it have strong advantages in processing complex bioinformatics data. The Transformer model can effectively learn the complex patterns of cfDNA in lung nodule risk assessment by targeted training of labeled cfDNA data, thereby achieving accurate recognition of the nature of lung nodules.

[0070] Embodiment 2 The present application provides an information processing method for evaluating the risk of lung nodules, which comprises the following steps: According to the construction method of the lung nodule risk evaluation model based on the cfDNA multi-feature, a lung nodule risk evaluation model is constructed, and then the cfDNA sequencing data of the sample to be tested is input; the features of the cfDNA of the sample to be tested are extracted in sequence, and the features after convolution fusion are output, and the lung nodule risk evaluation result is obtained by the lung nodule risk evaluation model.

[0071] Embodiment 3 The present application provides a device for evaluating the risk of lung nodules, which comprises an input module, a control module and an output module. The input module is configured to input the cfDNA sequencing data of the prior data or new data; The control module comprises a cfDNA feature extraction module, a convolution fusion module, and a lung nodule risk evaluation model based on Transformer; The output module is configured to output the risk of lung nodules; The cfDNA feature extraction module is used to extract the features of the cfDNA of the prior data or new data, and to encode each extracted feature to obtain a feature map; the features include cfDNA length, terminal motif, sequence motif, methylation rate, cfDNA length and methylation rate comprehensive features, cfDNA length and terminal motif comprehensive features, terminal motif and methylation rate comprehensive features, and terminal motif and sequence motif comprehensive features; each sample of the prior data or new data has a binary classification label of lung nodule nature; The convolution fusion module is configured to: fuse the feature maps; input each feature map into a neural network structure, extract representative features through a convolution layer sliding window, and then reduce dimensions through a pooling layer to reserve key features; and splice the key features along a preset dimension to form a composite feature; The lung nodule malignancy risk assessment model module based on the Transformer is configured to: input the composite feature into the Transformer to perform feature extraction and map to a lung nodule malignancy risk, thereby constructing a lung nodule malignancy risk assessment model; and input the comprehensive feature of the convolution fusion of the to-be-tested sample into the lung nodule malignancy risk assessment model to output a lung nodule malignancy risk assessment result.

[0072] The above merely describes preferred embodiments of the present application but is not intended to limit the present application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A method for constructing a pulmonary nodule malignancy risk assessment model based on multiple features of cfDNA, characterized in that: It includes the following steps: (1) Extracting features of cfDNA from prior data or new data, and encoding each extracted feature to obtain a feature map; the features include cfDNA length, terminal motif, sequence motif, methylation rate, comprehensive features of cfDNA length and methylation rate, comprehensive features of cfDNA length and terminal motif, comprehensive features of terminal motif and methylation rate, and comprehensive features of terminal motif and sequence motif; each sample of the prior data or new data has a binary classification label of the nature of the lung nodule; (2) Convolutional fusion: The fusion of the feature maps includes the following steps: inputting each feature map into the neural network structure, extracting representative features through the sliding window of the convolution layer, and then reducing the dimension through the pooling layer to retain the key features; splicing the key features along the preset dimension to form a composite feature; (3) Building a pulmonary nodule malignancy risk assessment model based on the Transformer architecture: The composite features are input into Transformer for feature extraction and mapped to the risk of malignant lung nodules, thereby constructing a lung nodule malignant risk assessment model.

2. The method for constructing a pulmonary nodule malignancy risk assessment model based on multiple features of cfDNA according to claim 1, characterized in that: The dimension of the characteristic map of cfDNA length is (100-250)×1×1, where 100-250 represents the cfDNA length; The dimension of the feature map of the terminal motif is 256×1×1, the dimension of the feature map of the sequence motif is 256×1×1, and the dimension of the feature map of the methylation rate is (50-1000)×1×1; the dimension of the feature map of the comprehensive feature of cfDNA length and methylation rate is (100-250)×(50-1000)×1; the dimension of the feature map of the comprehensive feature of cfDNA length and terminal motif is (100-250)×256×1; the dimension of the feature map of the comprehensive feature of terminal motif and methylation rate is (50-1000)×256×1; the dimension of the comprehensive feature of terminal motif and sequence motif is 256×256×1.

3. The method for constructing a pulmonary nodule malignancy risk assessment model based on multiple features of cfDNA according to claim 1, characterized in that: The neural network structure is selected from CNN, RNN, GNN, MLP, GAN, diffusion model, capsule network or sparse network.

4. An information processing method for assessing the risk of malignancy of pulmonary nodules, characterized in that: It includes the following steps: A pulmonary nodule malignancy risk assessment model is constructed according to the method for constructing a pulmonary nodule malignancy risk assessment model based on multiple features of cfDNA as described in any one of claims 1-3, and then the cfDNA sequencing data and bioinformatics analysis results of the sample to be tested are input; the features of the cfDNA of the tested sample and the features after convolution fusion are extracted in sequence, and the pulmonary nodule malignancy risk assessment model is output to obtain the pulmonary nodule malignancy risk assessment result.

5. A device for assessing the risk of malignancy of pulmonary nodules, characterized in that: It includes: Input module, control module and output module; The input module is configured to: input cfDNA sequencing data of prior data or new data; The control module includes: a cfDNA feature extraction module, a convolution fusion module, and a Transformer-based pulmonary nodule malignancy risk assessment model module; The output module is configured to: output the risk of malignancy of pulmonary nodules; The cfDNA feature extraction module is used to extract features of cfDNA from prior data or new data, and encode each extracted feature to obtain a feature graph; the features include cfDNA length, terminal motif, sequence motif, methylation rate, combined features of cfDNA length and methylation rate, combined features of cfDNA length and terminal motif, combined features of terminal motif and methylation rate, and combined features of terminal motif and sequence motif; each sample of the prior data or new data has a binary classification label indicating the nature of a lung nodule; The convolution fusion module is used to fuse the feature maps: input each feature map into the neural network structure, extract representative features through the sliding window of the convolution layer, and then reduce the dimension and retain the key features through the pooling layer; splice the key features along the preset dimension to form a composite feature; The Transformer-based pulmonary nodule malignancy risk assessment model module is used to: input the composite features into the Transformer, perform feature extraction, and map them to the pulmonary nodule malignancy risk, thereby constructing a pulmonary nodule malignancy risk assessment model; input the comprehensive features after convolution fusion of the sample to be tested into the pulmonary nodule malignancy risk assessment model, and output the pulmonary nodule malignancy risk assessment results.

6. The device for assessing the risk of malignancy of pulmonary nodules according to claim 5, characterized in that: The dimension of the characteristic map of cfDNA length is (100-250)×1×1, where 100-250 represents the cfDNA length; The dimension of the feature map of the terminal motif is 256×1×1, the dimension of the feature map of the sequence motif is 256×1×1, and the dimension of the feature map of the methylation rate is (50-1000)×1×1; the dimension of the feature map of the comprehensive feature of cfDNA length and methylation rate is (100-250)×(50-1000)×1; the dimension of the feature map of the comprehensive feature of cfDNA length and terminal motif is (100-250)×256×1; the dimension of the feature map of the comprehensive feature of terminal motif and methylation rate is (50-1000)×256×1; the dimension of the comprehensive feature of terminal motif and sequence motif is 256×256×1.

7. The device for assessing the risk of malignancy of pulmonary nodules according to claim 5, characterized in that: The neural network structure is selected from CNN, RNN, GNN, MLP, GAN, diffusion model, capsule network or sparse network.

8. The device for assessing the risk of malignancy of pulmonary nodules according to claim 5, characterized in that: The pulmonary nodule malignancy risk assessment result of the Transformer-based pulmonary nodule malignancy risk assessment model module is a high risk of pulmonary nodule malignancy or a low risk of pulmonary nodule malignancy.

9. A device for assessing the risk of malignancy of pulmonary nodules, characterized in that: The device includes a processor and a memory, wherein the memory stores at least one instruction, at least one program, code set or instruction set, and the at least one instruction, the at least one program, the code set or instruction set is loaded and executed by the processor to implement the information processing method for assessing the malignancy risk of pulmonary nodules as described in claim 4.

10. A computer-readable storage medium, characterized in that The storage medium stores at least one instruction, at least one program, code set or instruction set, and the at least one instruction, the at least one program, the code set or instruction set is loaded and executed by the processor to implement the information processing method for assessing the malignancy risk of pulmonary nodules as described in claim 4.

Images