Lung cancer classification method, apparatus, device, and storage medium

Abstract

The present disclosure relates to a lung cancer classification method, device, equipment and storage medium, wherein the method comprises: acquiring a sensor array signal of a sampling gas; inputting the sensor array signal into a plurality of feature extraction network branches respectively for feature extraction to generate a plurality of feature vectors corresponding to the plurality of feature extraction network branches respectively, wherein the plurality of feature extraction network branches have different convolution kernel scales, and each feature extraction network branch has a one-dimensional convolution kernel; performing feature fusion on the plurality of feature vectors to generate a base feature vector of the sensor array signal; inputting the base feature vector into a classification model to generate a lung cancer classification result corresponding to the sensor array signal. According to the technical solution of the present disclosure, the accuracy of lung cancer classification can be improved.

Description

TECHNICAL FIELD

[0001] The present disclosure relates to the technical field of artificial intelligence, and particularly relates to a lung cancer classification method and device, equipment and a storage medium. BACKGROUND

[0002] Gas sensors have been widely used in modern industry, especially in the field of medical detection. In the lung cancer classification scene, an electronic nose gas sensor array is used to sample the signal of the patient's respiratory gas, and the content of different component gas markers contained in the gas data is analyzed through deep learning, so as to realize lung cancer classification, which has important clinical significance for diagnosis and treatment.

[0003] In the related art, when lung cancer classification is realized based on an electronic nose, a feature extractor is combined with a shallow learning mode, the feature extractor converts the original signal into a low-dimensional vector, and the shallow learning mode is, for example, an artificial neural network or a support vector machine. Common feature extractors include Fourier transform, wavelet transform, empirical mode decomposition, spectral kurtosis, and sparse representation. However, the construction of the feature extractor requires relevant prior knowledge. Due to the increasing complexity and diversity of sensor array signals composed of different types of samples, it is very time-consuming to construct a feature extractor for each type of sensor array. In addition, due to changes in components, systems, or sampling frequencies, the working environment of the electronic nose instrument, and other factors, the signal may change, and the difference between gas types is small, making it difficult to distinguish the signal itself, which is not suitable for artificial intervention guidance. The accuracy of this method needs to be improved. SUMMARY

[0004] To solve the above technical problems, the present disclosure provides a lung cancer classification method, device, equipment and storage medium.

[0005] In a first aspect, the embodiments of the present disclosure provide a lung cancer classification method, comprising:

[0006] obtaining a sensor array signal of a sampled gas;

[0007] inputting the sensor array signal into a plurality of feature extraction network branches respectively for feature extraction, to generate a plurality of feature vectors respectively corresponding to the plurality of feature extraction network branches, wherein the plurality of feature extraction network branches have different convolution kernel scales, and each feature extraction network branch has a one-dimensional convolution kernel;

[0008] performing feature fusion on the plurality of feature vectors to generate a base feature vector of the sensor array signal;

[0009] inputting the base feature vector into a classification model to generate a lung cancer classification result corresponding to the sensor array signal.

[0010] In a second aspect, the embodiments of the present disclosure provide a lung cancer classification device, comprising:

[0011] an acquisition module configured to acquire a sensor array signal of a sampling gas;

[0012] an extraction module configured to input the sensor array signal into a plurality of feature extraction network branches respectively for feature extraction, and generate a plurality of feature vectors corresponding to the plurality of feature extraction network branches respectively, wherein the plurality of feature extraction network branches have different convolution kernel scales, and each feature extraction network branch has a one-dimensional convolution kernel;

[0013] a fusion module configured to perform feature fusion on the plurality of feature vectors, and generate a base feature vector of the sensor array signal;

[0014] a classification module configured to input the base feature vector into a classification model, and generate a lung cancer classification result corresponding to the sensor array signal.

[0015] In a third aspect, the embodiments of the present disclosure provide an electronic device, comprising: a processor; a memory configured to store executable instructions of the processor; and the processor configured to read the executable instructions from the memory and execute the instructions to implement the lung cancer classification method of the first aspect.

[0016] In a fourth aspect, the embodiments of the present disclosure provide a computer-readable storage medium, which stores a computer program, and the computer program is executed by a processor to implement the lung cancer classification method of the first aspect.

[0017] Compared with the prior art, the technical scheme provided by the embodiments of the present disclosure has the following advantages: the sensor array signal is input into a plurality of feature extraction network branches with different scales respectively for feature extraction, a plurality of feature vectors are generated, the plurality of feature vectors are fused for feature fusion, a base feature vector of the sensor array signal is generated, and then the base feature vector is input into a classification model to generate a lung cancer classification result corresponding to the sensor array signal. Thus, for the lung cancer classification scene based on the electronic nose gas sensor array, the feature extraction network used in the image field is improved to realize the application of the feature extraction network in the lung cancer classification scene. Compared with the way of constructing a feature extractor, the technical scheme reduces manual intervention and improves processing efficiency. Moreover, the multi-scale feature extraction network is used for signal down-sampling feature extraction, and the features obtained by multi-scale sampling are fused for feature fusion. The base features of the multi-element time series signal are extracted from the multi-scale perspective, so as to accurately classify the breathomics gas signal of a patient and improve the accuracy of lung cancer classification. BRIEF DESCRIPTION OF DRAWINGS

[0018] The accompanying drawings, which are incorporated into and constitute a part of the specification, illustrate embodiments consistent with the present disclosure and serve to explain the principles of the present disclosure together with the specification.

[0019] In order to more clearly illustrate the technical solutions in the embodiments of the present disclosure or the prior art, the accompanying drawings required by the embodiments or the prior art description will be briefly introduced. Obviously, those skilled in the art can obtain other drawings according to these drawings without any creative effort.

[0020] Figure 1 A flowchart of a lung cancer classification method provided by the embodiments of the present disclosure;

[0021] Figure 2 A sensor array signal diagram provided by the embodiments of the present disclosure;

[0022] Figure 3 A network structure diagram provided by the embodiments of the present disclosure;

[0023] Figure 4 A schematic diagram of another lung cancer classification method provided by the embodiments of the present disclosure;

[0024] Figure 5 A structure diagram of a lung cancer classification device provided by the embodiments of the present disclosure. DETAILED DESCRIPTION

[0025] In order to more clearly illustrate the technical solutions in the embodiments of the present disclosure or the prior art, the accompanying drawings required by the embodiments or the prior art description will be briefly introduced. Obviously, those skilled in the art can obtain other drawings according to these drawings without any creative effort.

[0026] In the following description, many specific details are set forth in order to fully understand the present disclosure, but the present disclosure can also be implemented in other ways different from those described herein; obviously, the embodiments in the specification are only some of the embodiments of the present disclosure, not all the embodiments.

[0027] Figure 1 A flowchart of a lung cancer classification method provided by the embodiments of the present disclosure, the method provided by the embodiments of the present disclosure can be executed by a lung cancer classification device, which can be implemented by software and / or hardware, and can be integrated on any electronic device with computing capability.

[0028] As Figure 1 shown, the lung cancer classification method provided by the embodiments of the present disclosure can include:

[0029] Step 101, acquiring a sensor array signal of a sampling gas.

[0030] In this embodiment, the electronic nose gas sensor array is used to sample the patient's respiratory gas signal, and the sensor array signal is obtained. The gas sensor array can be composed of multiple sub-sensors, and the sensor array signal includes multiple sub-sensor signals, which are used for lung cancer classification of patient samples.

[0031] Optionally, the gas sensor array is composed of 22 sub-sensors. In this example, the 22 sub-sensor signals in the sensor array are used. After the electronic nose gas sensor array collects data from the patient's respiratory gas, each gas sensor array generates a breathomics multivariate time series with a dimension of 22. As an example, the sensor array signal is as shown in Figure 2 .

[0032] In step 102, the sensor array signal is input into multiple feature extraction network branches for feature extraction, and multiple feature vectors corresponding to the multiple feature extraction network branches are generated.

[0033] Among them, the multiple feature extraction network branches have different convolution kernel scales, and each feature extraction network branch has a one-dimensional convolution kernel.

[0034] As an example, the feature extraction network is a residual network, and the number of residual network branches is six. The convolution kernel scales of the six residual network branches are 1x3, 1x5, 1x7, 1x9, 1x11, and 1x15. It should be noted that the number and specific convolution kernel scale of the above residual network branches are a selection, and can be adjusted according to actual needs, which is not limited here.

[0035] In this embodiment, each residual network branch extracts features from the sensor array signal, and the result of the sensor array signal being extracted by the residual network with multiple different scale convolution kernels can be obtained. In related technologies, the residual network is used in the image processing field, and in this embodiment, the residual network is improved based on the characteristics of the sensor array signal to be used in the lung cancer classification scene based on the electronic nose gas sensor array. The feature extraction of the multivariate time series is further realized for lung cancer classification.

[0036] The implementation of the residual network branch is illustrated below.

[0037] Two CNN (Convolutional Neural Networks) blocks are stacked to construct a sub-block, and a residual learning structure is explored between each CNN block to construct a deep network for complex feature extraction, where the CNN block is constructed as follows: s = BN(y), h = LeakyRelu(s), where is a convolution operator; BN is a batch normalization operation, which can reduce the gradient vanishing phenomenon and allow a higher learning rate to achieve fast convergence; LeakyRelu is an activation function. Among them, the sub-block form is as follows: h1 = BasicCNN(x), h2 = BasicCNN(h1), y = h2 + x, where BasicCNN represents a CNN block.

[0038] Further, a residual network branch is constructed by stacking multiple sub-blocks, and each residual network branch has 4 stages (Stage), and each stage has the following form: where BasicBlock represents a sub-block, and i represents the number of sub-blocks. As an example, the number of sub-blocks of the 4 stages is 3, 4, 6, and 3 in turn.

[0039] The multiple residual network branches share the weight parameters of the head network, and the head network is composed of convolution layers, normalization layers, activation layers and pooling layers. For example, the head network is composed of a convolution layer with a 1x7 convolution kernel, a normalization layer, an activation layer and a maximum pooling layer, which can be represented as follows: s h = BN(y h ), h h = LeakyRelu(s h ), As an example, taking the aforementioned six residual network branches as an example, the model structure diagram is shown in Figure 3 .

[0040] In step 103, the multiple feature vectors are fused to generate a base feature vector of the sensor array signal.

[0041] In this embodiment, the base feature vector of the sensor array signal is generated by fusing multiple feature vectors of different scales. The feature fusion includes addition operation on the feature vectors.

[0042] In an embodiment of the present disclosure, for multiple sub-sensor signals included in the sensor array signal, feature extraction is performed by the following steps: each sub-sensor signal is input into multiple feature extraction network branches for feature extraction to generate multiple feature vectors corresponding to each sub-sensor signal. Further, the multiple feature vectors corresponding to each sub-sensor signal are fused to generate a fusion feature corresponding to each sub-sensor signal, and a base feature vector of the sensor array signal is generated according to the fusion feature corresponding to each sub-sensor signal.

[0043] As an example, taking 22 sub-sensor signals as an example, each sub-sensor signal is input into six residual network branches for feature extraction to generate six feature vectors of different scales (1x512 dimensions), and then the six feature vectors of different scales are fused to generate a fusion feature (1x512 dimensions) of the sub-sensor signal. For 22 sub-sensor signals, 22 fusion features are obtained as base feature vectors of the sensor array signal.

[0044] In step 104, the base feature vector is input into a classification model to generate a lung cancer classification result corresponding to the sensor array signal.

[0045] In this embodiment, the base feature vector of the sensor array signal is input into a pre-trained classification model to generate a lung cancer classification result corresponding to the sensor array signal. The classification model can be implemented based on a neural network, and the lung cancer classification result is used to indicate a lung cancer category, which can be determined according to a relevant standard.

[0046] According to the technical scheme of the embodiment of the present disclosure, the sensor array signal is input into multiple feature extraction network branches of different scales for feature extraction to generate multiple feature vectors, the multiple feature vectors are fused to generate a base feature vector of the sensor array signal, and then the base feature vector is input into a classification model to generate a lung cancer classification result corresponding to the sensor array signal. Thus, for the lung cancer category detection scene based on the electronic nose gas sensor array, the feature extraction network used in the image field is improved to realize the application of the feature extraction network in the lung cancer classification scene. Compared with the way of constructing a feature extractor, the processing efficiency is improved by reducing human intervention. Moreover, the multi-scale feature extraction network is used for signal down-sampling feature extraction, and the features obtained by multi-scale sampling are fused. The base feature of the multi-element time series signal is extracted from the perspective of multi-scale to accurately classify the breathomics gas signal of the patient. The problem that the signal analysis ability needs to be improved due to changes in operating conditions such as components, systems, or sampling frequency when applying an electronic nose for gas classification is solved, and the accuracy of lung cancer classification is improved.

[0047] Based on the above embodiment, in order to further improve the accuracy of lung cancer classification, the classification model in this application is implemented by an improved neural network based on a multi-head self-attention mechanism. The transformer neural network is a deep learning network for the image field, and in this embodiment, the multi-element time series signal in the lung cancer classification scene is improved by modifying the transformer neural network to realize the classification model. Compared with the traditional lung cancer classification method, the accuracy is further improved.

[0048] Figure 4This is a schematic diagram of another lung cancer classification method provided in an embodiment of the present disclosure, as shown below. Figure 4 As shown, in this method, the basic feature vector is input into the classification model to generate lung cancer classification results corresponding to the sensor array signals, including:

[0049] Step 401: Input the base feature vector into the target convolutional layer for convolution to obtain the convolutional base feature vector.

[0050] The target convolutional layer is a 1×1 convolutional layer.

[0051] In this embodiment, in order to adapt to the learning of multivariate time-series data, the image segmentation, convolution and flattening operations of the network front end are removed. The N vectors (tokens) output by the aforementioned feature extraction network are used as the input of the classification model. These N vectors represent the basic features extracted from the sensor array signals at different scales.

[0052] In the classification model, a 1×1 convolutional layer is constructed to convolve the base feature vectors, achieving cross-channel aggregation and realizing linear combination and transformation of information between channels. Furthermore, it achieves feature dimensionality enhancement by expanding the dimensions of the base feature vectors, increasing the expressive power of the features. This allows features that are difficult to distinguish in low dimensions to be accurately distinguished in high dimensions, making the feature structure sparser, which is beneficial for training and learning, and increases semantic generalization. The convolutional features are then processed through a dropout layer. In this step, the activation function used is LeakyReLU.

[0053] Step 402: Insert the pre-trained classification vector into the base feature vector after convolution to generate the first vector.

[0054] In this embodiment, the classification model is implemented based on a transformer neural network, and the position embedding layer has been removed. Feature vectors at different scales are obtained by repeatedly extracting the entire signal using different methods, and there is no spatial relationship between them. Furthermore, classification information is added by adding a new classification vector and inserting it into the base feature vectors. Here, the classification vector is a trainable parameter, and its data format is a vector. Optionally, the classification vector is concatenated with the base feature vectors; for example, the classification vector is concatenated with the N vectors output from the previous steps to generate a first vector, which includes N+1 vectors.

[0055] Step 403: Encode the first vector using an encoder to generate the second vector.

[0056] In this embodiment, the encoder adopts an encoder block of a transformer neural network. In the plurality of serially connected encoder blocks of the transformer neural network, the feature vector is normalized by a normalization (LayerNorm) layer, is subjected to multi-head self-attention learning and dropout processing, and the transformer adopts the same structure as the feature extraction network to avoid gradient disappearance. The implementation of the encoder can refer to the existing structure of the encoder block in the transformer neural network, which will not be described here. Thus, the first vector is input into the encoder block of the transformer neural network for encoding processing to generate the second vector.

[0057] In step 404, the second vector is subjected to layer normalization processing to extract a classification vector.

[0058] In step 405, the classification vector is classified to generate a lung cancer classification result.

[0059] In this embodiment, the output of the encoder is normalized by a normalization (LayerNorm) layer, and a classification vector for class learning is extracted. Then, the classification vector is classified by a classification layer full connection to obtain a predicted lung cancer class. The classification layer can adopt an MLP Head (a layer structure for classification) layer, and the feature extraction network structure is also used in the full connection processing to avoid gradient disappearance.

[0060] The model training process will be described below.

[0061] In one embodiment of the present disclosure, a sample sensor array signal is obtained, which is labeled with a lung cancer class. Then, the sample sensor array signal is input into a plurality of feature extraction network branches for feature extraction to generate a plurality of sample feature vectors. The plurality of sample feature vectors are fused to generate a base feature vector of the sample sensor array signal. The classification model and the classification vector are trained according to the base feature vector of the sample sensor array signal and the labeled lung cancer class.

[0062] In the training process, the classification model part performs the following steps: the sample base feature vector is input into a target convolution layer for convolution to obtain a convolved sample base feature vector. A preset vector is inserted into the convolved sample base feature vector to generate a third vector. The third vector is encoded by an encoder to generate a fourth vector. Then, the fourth vector is subjected to layer normalization processing, and a preset vector is extracted based on the processing result. The classification model and the classification vector are trained based on the preset vector and the labeled lung cancer class.

[0063] In the embodiments of the present disclosure, for the multi-element time sequence signal in the lung cancer classification scene, the transformer neural network is improved to realize the application in the lung cancer classification scene. The transformer neural network based on the multi-head self-attention mechanism performs differentiated attention learning on the features, learns the partial features in the base feature vector that are conducive to classification, ignores the redundant features that are not conducive to or interfere with classification, enhances the extraction capability of the multi-element time sequence features, thereby realizes the downstream lung cancer classification task, improves the generalization capability of the model, and further improves the lung cancer classification accuracy.

[0064] Figure 5 A structural schematic diagram of a lung cancer classification device provided by the embodiments of the present disclosure is shown in FIG. 1. Figure 5 As shown in the figure, the lung cancer classification device comprises an acquisition module 51, an extraction module 52, a fusion module 53, and a classification module 54.

[0065] The acquisition module 51 is configured to acquire a sensor array signal of a sampling gas.

[0066] The extraction module 52 is configured to input the sensor array signal into a plurality of feature extraction network branches respectively for feature extraction, and generate a plurality of feature vectors respectively corresponding to the plurality of feature extraction network branches. The plurality of feature extraction network branches have different convolution kernel scales, and each feature extraction network branch has a one-dimensional convolution kernel.

[0067] The fusion module 53 is configured to perform feature fusion on the plurality of feature vectors, and generate a base feature vector of the sensor array signal.

[0068] The classification module 54 is configured to input the base feature vector into a classification model, and generate a lung cancer classification result corresponding to the sensor array signal.

[0069] In an embodiment of the present disclosure, the classification model is implemented based on a transformer neural network, and the classification model does not include a position encoding layer. The classification module 54 is specifically configured to: input the base feature vector into a target convolution layer for convolution to obtain a convolved base feature vector, wherein the target convolution layer is a 1x1 convolution layer; insert a pre-trained classification vector into the convolved base feature vector to generate a first vector; encode the first vector through an encoder to generate a second vector; perform layer normalization processing on the second vector to extract a classification vector; and perform classification based on the classification vector to generate the lung cancer classification result.

[0070] In an embodiment of the present disclosure, the device further comprises a training module configured to: obtain a sample sensor array signal, wherein the sample sensor array signal is labeled with a lung cancer category; input the sample sensor array signal into a plurality of feature extraction network branches to perform feature extraction and generate a plurality of sample feature vectors; perform feature fusion on the plurality of sample feature vectors to generate a base feature vector of the sample sensor array signal; and train the classification model and the classification vector according to the base feature vector of the sample sensor array signal and the labeled lung cancer category.

[0071] In an embodiment of the present disclosure, the sensor array signal comprises a plurality of sub-sensor signals, and the extraction module 52 is specifically configured to: input each sub-sensor signal into a plurality of feature extraction network branches respectively to perform feature extraction and generate a plurality of feature vectors corresponding to each sub-sensor signal.

[0072] The fusion module 53 is specifically configured to: perform feature fusion on the plurality of feature vectors corresponding to each sub-sensor signal to generate a fusion feature corresponding to each sub-sensor signal; and generate a base feature vector of the sensor array signal according to the fusion feature corresponding to each sub-sensor signal.

[0073] In an embodiment of the present disclosure, the plurality of feature extraction network branches share weight parameters of a head network, and the head network is composed of a convolution layer, a normalization layer, an activation layer and a pooling layer.

[0074] In an embodiment of the present disclosure, the number of feature extraction network branches is six, and the convolution kernel scales of the feature extraction network branches are 1x3, 1x5, 1x7, 1x9, 1x11 and 1x15 respectively.

[0075] The lung cancer classification device provided in the embodiments of the present disclosure can execute any lung cancer classification method provided in the embodiments of the present disclosure, has the corresponding function modules and beneficial effects of the execution method. The contents not described in detail in the device embodiments of the present disclosure can be referred to the description in any method embodiment of the present disclosure.

[0076] The electronic device according to an embodiment of the disclosure can include one or more processors and a memory. The processor can be a central processing unit (CPU) or other form of processing unit having data processing and / or instruction execution capabilities, and can control other components in the electronic device to perform desired functions. The memory can include one or more computer program products, which can include various forms of computer-readable storage media, such as volatile memory and / or non-volatile memory. The volatile memory, for example, can include random access memory (RAM), cache memory, and / or the like. The non-volatile memory, for example, can include read-only memory (ROM), hard disk, flash memory, and / or the like. One or more computer program instructions can be stored on the computer-readable storage media, and the processor can execute the program instructions to implement the method of the above-mentioned embodiments of the disclosure and / or other desired functions. Various contents such as input signals, signal components, noise components, and the like can also be stored in the computer-readable storage media.

[0077] In one example, the electronic device can further include an input device and an output device, which are interconnected through a bus system and / or other forms of connection mechanism. In addition, the input device can include, for example, a keyboard, a mouse, and the like. The output device can output various information to the outside, including the determined distance information, direction information, and the like. The output device can include, for example, a display, a speaker, a printer, a communication network and a remote output device connected thereto, and the like. In addition, the electronic device can include any other appropriate components such as a bus, an input / output interface, and the like, according to specific application cases.

[0078] In addition to the above-mentioned method and device, the embodiments of the disclosure can also be a computer program product including computer program instructions, which, when executed by a processor, cause the processor to perform any method provided by the embodiments of the disclosure.

[0079] The computer program product can be written in any combination of one or more programming languages, including an object-oriented programming language such as Java, C++, and the like, and a conventional procedural programming language such as "C" language or similar programming languages. The program code can be executed entirely on a user computing device, partially on a user device, as an independent software package, partially on a user computing device and partially on a remote computing device, or entirely on a remote computing device or server.

[0080] In addition, an embodiment of the present disclosure can also be a computer-readable storage medium, having stored thereon computer program instructions, which, when executed by a processor, cause the processor to perform any of the methods provided by the embodiments of the present disclosure.

[0081] The computer-readable storage medium can be any combination of one or more computer-readable media. The computer-readable medium can be a computer-readable storage medium or a computer-readable signal medium. The computer-readable storage medium can include, for example, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any suitable combination of the above. More specific examples (a non-exhaustive list) of the computer-readable storage medium include an electrical connection having one or more wires, a portable disc, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above.

[0082] It should be noted that, in this document, the terms such as "first" and "second" are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between the entities or operations. Also, the terms "include", "contain" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements not only includes those elements, but also includes other elements not explicitly listed or inherent to such a process, method, article or device. Without more limitations, the element defined by the statement "including one" does not exclude the presence of another identical element in the process, method, article or device including the element.

[0083] The above description is only a specific implementation of the present disclosure, enabling those skilled in the art to understand or implement the present disclosure. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the present disclosure. Therefore, the present disclosure will not be limited to these embodiments described herein, but will conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for classifying lung cancer, characterized in that, include: Acquire the sensor array signal of the sampled gas; The sensor array signal is input into multiple feature extraction network branches for feature extraction, generating multiple feature vectors corresponding to the multiple feature extraction network branches respectively. The multiple feature extraction network branches have different convolution kernel scales, and each feature extraction network branch has a one-dimensional convolution kernel. The multiple feature vectors are fused to generate the base feature vector of the sensor array signal; The base feature vector is input into the classification model to generate lung cancer classification results corresponding to the sensor array signals; The classification model is based on a transformer neural network and does not include a position encoding layer. The step of inputting the base feature vector into the classification model to generate a lung cancer classification result corresponding to the sensor array signal includes: The base feature vector is input into the target convolutional layer for convolution to obtain the convolutional base feature vector, wherein the target convolutional layer is a 1×1 convolutional layer; The pre-trained classification vector is inserted into the base feature vector after convolution to generate the first vector; The first vector is encoded by an encoder to generate the second vector; Perform layer normalization on the second vector to extract the classification vector; The lung cancer classification result is generated based on the classification vector.

2. The method as described in claim 1, characterized in that, Also includes: Acquire sample sensor array signals, wherein the sample sensor array signals are labeled with lung cancer category; The sample sensor array signal is input into multiple feature extraction network branches for feature extraction, generating multiple sample feature vectors; The feature vectors of the multiple samples are fused to generate the base feature vector of the sample sensor array signal; The classification model and the classification vector are trained based on the base feature vector of the sample sensor array signal and the labeled lung cancer category.

3. The method as described in claim 1, characterized in that, The sensor array signal includes multiple sub-sensor signals. The step of inputting the sensor array signal into multiple feature extraction network branches for feature extraction, generating multiple feature vectors corresponding to each of the multiple feature extraction network branches, includes: Each of the sub-sensor signals is input into multiple feature extraction network branches for feature extraction, generating multiple feature vectors corresponding to each of the sub-sensor signals; The step of fusing the multiple feature vectors to generate the base feature vector of the sensor array signal includes: The multiple feature vectors corresponding to each sub-sensor signal are fused to generate a fused feature corresponding to each sub-sensor signal; Based on the fusion features corresponding to each of the sub-sensor signals, a base feature vector of the sensor array signal is generated.

4. The method as described in claim 1, characterized in that, The multiple feature extraction network branches share the weight parameters of the head network, which consists of convolutional layers, normalization layers, activation layers, and pooling layers.

5. The method as described in claim 1, characterized in that, The feature extraction network has six branches, and the convolutional kernel sizes of the feature extraction network branches are 1×3, 1×5, 1×7, 1×9, 1×11, and 1×15, respectively.

6. A lung cancer classification device, characterized in that, include: The acquisition module is used to acquire the sensor array signal of the sampled gas; The extraction module is used to input the sensor array signal into multiple feature extraction network branches for feature extraction, and generate multiple feature vectors corresponding to the multiple feature extraction network branches respectively. The multiple feature extraction network branches have different convolution kernel scales, and each feature extraction network branch has a one-dimensional convolution kernel. The fusion module is used to fuse the multiple feature vectors to generate the base feature vector of the sensor array signal; The classification module is used to input the base feature vector into the classification model and generate a lung cancer classification result corresponding to the sensor array signal; The classification model is based on a transformer neural network and does not include a position encoding layer. The classification module is specifically used for: The base feature vector is input into the target convolutional layer for convolution to obtain the convolutional base feature vector, wherein the target convolutional layer is a 1×1 convolutional layer; The pre-trained classification vector is inserted into the base feature vector after convolution to generate the first vector; The first vector is encoded by an encoder to generate the second vector; Perform layer normalization on the second vector to extract the classification vector; The lung cancer classification result is generated based on the classification vector.

7. An electronic device, characterized in that, include: processor; Memory used to store the processor's executable instructions; The processor is configured to read the executable instructions from the memory and execute the instructions to implement the lung cancer classification method according to any one of claims 1-5.

8. A computer-readable storage medium, characterized in that, The storage medium stores a computer program, which, when executed by a processor, implements the lung cancer classification method according to any one of claims 1-5.

Images