A method and system for classifying lung nodule images

Abstract

The application discloses a kind of lung nodule image classification method and system, it is related to image recognition field.The method includes: input includes the pixel block of lung nodule image and the segmentation mask of lung nodule, the first cross-sectional layer of the pixel block and the second cross-sectional layer of the segmentation mask are respectively obtained by pre-processing to the pixel block and segmentation mask, the foreground pixel of the first cross-sectional layer and the second cross-sectional layer is extracted, and target feature matrix is obtained by pre-set processing to the foreground pixel, according to the target feature matrix, the classifier is trained, the nodule type classification is carried out to the lung nodule image by trained classifier, by the application, high requirement hardware equipment condition is not needed, a large number of labeled data is not needed, reduce artificial loss, combined with the information of nodule segmentation, starting from the method that doctor reads film when judging lung nodule type, it has very strong explainability, and classification effect is good.

Description

Technical Field

[0001] The present invention relates to the field of image recognition, and in particular to a method and system for classifying lung nodule images. Background Art

[0002] Pulmonary nodule density is generally classified into four types: solid, partially solid, ground glass, and calcification. Doctors determine the nodule density type mainly based on the HU value near and inside the nodule. However, manual judgment is affected by work experience and is prone to misjudgment. Summary of the Invention

[0003] The technical problem to be solved by the present invention is to provide a method and system for classifying lung nodule images in response to the deficiencies of the existing technology.

[0004] The technical solution of the present invention to solve the above technical problems is as follows:

[0005] A lung nodule image classification method, comprising:

[0006] The input includes pixel patches of the lung nodule image and the segmentation mask of the lung nodule;

[0007] Preprocessing the pixel block and the segmentation mask to obtain a first cross-sectional layer of the pixel block and a second cross-sectional layer of the segmentation mask respectively;

[0008] extracting foreground pixels of the first cross-sectional layer and the second cross-sectional layer, and performing a preset process on the foreground pixels to obtain a target feature matrix;

[0009] A classifier is trained according to the target feature matrix, and the lung nodule image is classified into nodule types using the trained classifier.

[0010] The beneficial effects of the present invention are: this scheme obtains the first cross-sectional layer of the pixel block and the second cross-sectional layer of the segmentation mask respectively by pre-processing the pixel block and the segmentation mask, extracts the foreground pixels of the first cross-sectional layer and the second cross-sectional layer, and performs preset processing on the foreground pixels to obtain a target feature matrix, trains the classifier according to the target feature matrix, and classifies the nodule type of the lung nodule image by the trained classifier. The present invention does not require high-demand hardware equipment conditions, does not require a large amount of labeled data, reduces manual loss, combines the information of nodule segmentation, and starts from the method of judging the type of lung nodules when doctors read the film. It has strong interpretability and good classification effect.

[0011] Furthermore, the preprocessing of the pixel block and the segmentation mask to obtain the first cross-sectional layer of the pixel block and the second cross-sectional layer of the segmentation mask respectively specifically includes:

[0012] extracting a layer of a cross section with the largest area of ​​the pixel block as a first cross section layer;

[0013] A layer of the segmentation mask with the largest cross-section area is extracted as the second cross-section layer.

[0014] The beneficial effect of adopting the above further solution is that: through the preprocessing operation, this solution only captures the image range that the doctor is concerned about, and effectively avoids the interference of irrelevant factors such as background.

[0015] Furthermore, extracting foreground pixels of the first cross-section layer and the second cross-section layer further includes:

[0016] Extracting foreground pixels of the first cross-sectional layer and the second cross-sectional layer, and processing them into one dimension to obtain a foreground pixel matrix;

[0017] A windowing operation is performed on the foreground pixel matrix to obtain a foreground pixel matrix after the windowing operation.

[0018] Furthermore, performing preset processing on the foreground pixels to obtain a target feature matrix specifically includes:

[0019] Sorting the foreground pixel matrix to obtain a sorted foreground pixel matrix;

[0020] A first characteristic matrix obtained by processing the sorted foreground pixel matrix using an interpolation method;

[0021] Performing grayscale normalization processing on the foreground pixel matrix to obtain a second characteristic matrix;

[0022] The first feature matrix and the second feature matrix are concatenated to obtain the target feature matrix.

[0023] The beneficial effect of adopting this further solution is that, by concatenating the first and second feature matrices to obtain the target feature matrix, this solution not only addresses the significant size differences among nodules but also effectively reflects the proportion of grayscale values ​​within each nodule. During the feature extraction step, two different methods are used to obtain nodule grayscale features, extracting the characteristics of the nodule's HU range. When doctors determine nodule type, this determination of the HU value range improves classification results and reduces manual error.

[0024] Furthermore, the method further includes: performing an etching operation on the second cross-section layer to obtain an etched second cross-section layer.

[0025] The beneficial effects of adopting the above further scheme are as follows: this scheme removes pixels at the edge of the nodule affected by the partial volume effect through the corrosion operation, thereby optimizing the segmentation mask data;

[0026] The obtaining of the foreground pixel matrix specifically includes:

[0027] Foreground pixels of the first cross-section layer and the corroded second cross-section layer are extracted and processed into one dimension to obtain a foreground pixel matrix.

[0028] Another technical solution of the present invention to solve the above technical problems is as follows:

[0029] A lung nodule image classification system includes: an input module, a preprocessing module, a feature extraction module and a classification module;

[0030] The input module is used to input pixel blocks including a lung nodule image and a segmentation mask of the lung nodule;

[0031] The pre-processing module is used to perform pre-processing on the pixel block and the segmentation mask to obtain a first cross-sectional layer of the pixel block and a second cross-sectional layer of the segmentation mask respectively;

[0032] The feature extraction module is used to extract foreground pixels of the first cross-sectional layer and the second cross-sectional layer, and perform preset processing on the foreground pixels to obtain a target feature matrix;

[0033] The classification module is used to train a classifier according to the target feature matrix, and classify the lung nodule image into nodule types using the trained classifier.

[0034] The beneficial effects of the present invention are: this scheme obtains the first cross-sectional layer of the pixel block and the second cross-sectional layer of the segmentation mask respectively by pre-processing the pixel block and the segmentation mask, extracts the foreground pixels of the first cross-sectional layer and the second cross-sectional layer, and performs preset processing on the foreground pixels to obtain a target feature matrix, trains the classifier according to the target feature matrix, and classifies the nodule type of the lung nodule image by the trained classifier. The present invention does not require high-demand hardware equipment conditions, does not require a large amount of labeled data, reduces manual loss, combines the information of nodule segmentation, and starts from the method of judging the type of lung nodules when doctors read the film. It has strong interpretability and good classification effect.

[0035] Furthermore, the pre-processing module is specifically used to extract the layer of the largest cross-section of the pixel block as the first cross-section layer;

[0036] A layer of the segmentation mask with the largest cross-section area is extracted as the second cross-section layer.

[0037] The beneficial effect of adopting the above further solution is that: through the preprocessing operation, this solution only captures the image range that the doctor is concerned about, and effectively avoids the interference of irrelevant factors such as background.

[0038] Furthermore, the feature extraction module is specifically used to extract foreground pixels of the first cross-sectional layer and the second cross-sectional layer, and process them into one dimension to obtain a foreground pixel matrix;

[0039] A windowing operation is performed on the foreground pixel matrix to obtain a foreground pixel matrix after the windowing operation.

[0040] Furthermore, the feature extraction module is specifically used to sort the foreground pixel matrix to obtain a sorted foreground pixel matrix;

[0041] A first characteristic matrix obtained by processing the sorted foreground pixel matrix using an interpolation method;

[0042] Performing grayscale normalization processing on the foreground pixel matrix to obtain a second characteristic matrix;

[0043] The first feature matrix and the second feature matrix are concatenated to obtain the target feature matrix.

[0044] The beneficial effect of adopting this further solution is that, by concatenating the first and second feature matrices to obtain the target feature matrix, this solution not only addresses the significant size differences among nodules but also effectively reflects the proportion of grayscale values ​​within each nodule. During the feature extraction step, two different methods are used to obtain nodule grayscale features, extracting the characteristics of the nodule's HU range. When doctors determine nodule type, this determination of the HU value range improves classification results and reduces manual error.

[0045] Furthermore, it also includes: a mask optimization module, which is used to perform an etching operation on the second cross-sectional layer to obtain the second cross-sectional layer after etching.

[0046] The beneficial effect of adopting the above further solution is that: this solution removes pixels at the edge of the nodule that are affected by the partial volume effect through the corrosion operation, thereby optimizing the segmentation mask data.

[0047] The feature extraction module is specifically used to extract foreground pixels of the first cross-section layer and the corroded second cross-section layer, and process them into one dimension to obtain a foreground pixel matrix.

[0048] Advantages of additional aspects of the present invention will be given in part in the following description and in part will be obvious from the following description, or will be learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figure 1 A schematic flow chart of a method for classifying lung nodules images provided in an embodiment of the present invention;

[0050] Figure 2 A structural block diagram of a lung nodule image classification system provided by an embodiment of the present invention;

[0051] Figure 3 A schematic diagram of the processing flow of a classification algorithm provided for other embodiments of the present invention. DETAILED DESCRIPTION

[0052] The principles and features of the present invention are described below with reference to the accompanying drawings. The embodiments given are only used to explain the present invention and are not used to limit the scope of the present invention.

[0053] like Figure 1 As shown, a lung nodule image classification method provided by an embodiment of the present invention includes: S1, inputting a pixel block of a lung nodule image and a segmentation mask of the lung nodule;

[0054] S2, preprocessing the pixel block and the segmentation mask to obtain a first cross-sectional layer of the pixel block and a second cross-sectional layer of the segmentation mask respectively;

[0055] It should be noted that, in one embodiment, preprocessing to obtain the largest cross-section of the lung nodule may include calculating the largest cross-section based on the area of ​​each cross-section of the lung nodule mask. Specifically, the calculation method involves traversing each cross-section of the lung nodule segmentation mask, calculating the number of foreground pixels in that cross-section, and treating this number as the foreground area of ​​that layer; the layer with the largest number of pixels is the cross-section with the largest area.

[0056] Take out the layer of the largest cross-section corresponding to the image pixel block of the nodule, which is recorded as IMG slice At the same time, the layer with the largest cross-section area corresponding to the lung nodule segmentation mask is taken out and recorded as MASK slice When doctors determine the type of lung nodules, they usually view the nodule image in cross-section. Taking cross-sectional images of nodules is more in line with doctors' nodule identification methods.

[0057] S3, extracting foreground pixels of the first cross-sectional layer and the second cross-sectional layer, and performing a preset process on the foreground pixels to obtain a target feature matrix;

[0058] S4, training a classifier based on the target feature matrix, and classifying the lung nodule image by nodule type using the trained classifier. The classifier may include a random forest classifier or other types of classifiers, such as SVM.

[0059] In one embodiment, if Figure 3 As shown, the input is a lung CT image pixel block IMG containing nodules origin , and the corresponding lung nodule segmentation mask Mask originIn the nodule segmentation mask, the pixel with a value of 1 is the foreground, representing IMG origin The corresponding position is the nodule area; the pixel with a value of 0 is the background, representing IMG origin The corresponding position is the non-nodule area. The mask can be manually marked or automatically segmented by an automatic segmentation algorithm, such as Unet.

[0060] This solution obtains the first cross-sectional layer of the pixel block and the second cross-sectional layer of the segmentation mask respectively by pre-processing the pixel block and the segmentation mask, extracts the foreground pixels of the first cross-sectional layer and the second cross-sectional layer, and performs preset processing on the foreground pixels to obtain a target feature matrix, trains a classifier according to the target feature matrix, and classifies the nodule type of the lung nodule image by the trained classifier. The present invention does not require high-demand hardware equipment conditions, does not require a large amount of labeled data, reduces manual loss, combines the nodule segmentation information, and starts from the method of judging the type of lung nodules when doctors read the film. It has strong interpretability and good classification effect.

[0061] Preferably, in any of the above embodiments, the preprocessing of the pixel block and the segmentation mask to respectively obtain the first cross-sectional layer of the pixel block and the second cross-sectional layer of the segmentation mask specifically includes:

[0062] extracting a layer of a cross section with the largest area of ​​the pixel block as a first cross section layer;

[0063] A layer of the segmentation mask with the largest cross-section area is extracted as the second cross-section layer.

[0064] Through the preprocessing operation, this solution only captures the image range that the doctor is concerned about, and effectively avoids interference from irrelevant factors such as the background.

[0065] Preferably, in any of the above embodiments, extracting foreground pixels of the first cross-section layer and the second cross-section layer further comprises:

[0066] Extracting foreground pixels of the first cross-sectional layer and the second cross-sectional layer, and processing them into one dimension to obtain a foreground pixel matrix;

[0067] A windowing operation is performed on the foreground pixel matrix to obtain a foreground pixel matrix after the windowing operation.

[0068] It should be noted that, in a certain embodiment, obtaining the foreground pixel matrix may include: taking out IMG slice Mask slice-erode The pixels corresponding to the foreground are straightened into one dimension and recorded as Array foreground .

[0069] Since the range of HU values ​​in CT images is large, doctors will set fixed window widths and positions when reading films. foreground Perform windowing operation. Set the HU value range you want to see [H min , H max ], then Array foreground Smaller than H min The pixel is set to H min , Array foreground Medium is greater than H max The pixel is set to H max In one embodiment of the present invention, H min and H max Set to lung window, the setting value can be -1200, 600.

[0070] Preferably, in any of the above embodiments, performing preset processing on the foreground pixels to obtain a target feature matrix specifically includes:

[0071] Sorting the foreground pixel matrix to obtain a sorted foreground pixel matrix;

[0072] A first characteristic matrix obtained by processing the sorted foreground pixel matrix using an interpolation method;

[0073] Performing grayscale normalization processing on the foreground pixel matrix to obtain a second characteristic matrix;

[0074] The first feature matrix and the second feature matrix are concatenated to obtain the target feature matrix.

[0075] It should be noted that the preset processing of the foreground pixels may include:

[0076] First, Array foreground Sort the pixel values ​​in ascending order and get the sorted foreground pixel matrix Array foreground-sorted .

[0077] Then, an interpolation method, such as linear interpolation, is used to interpolate the foreground pixel matrix into a feature matrix Feature of length N. N can be set to a suitable positive integer according to the data set. In one embodiment of the present invention, N can be set to 100.

[0078] In another embodiment, Array foreground The normalized grayscale histogram Feature2. The specific calculation method can be:

[0079] Set the gray level L. In this invention, set L=18. Calculate Array foregrounda gray histogram GH of the Array foreground , and then divide each element in GH by the length of the Array

[0080] Finally, Feature1 and Feature2 are spliced together to obtain the Feature matrix.

[0081] The scheme solves the problem of different nodule sizes and well reflects the proportion of gray values in the nodule by splicing the first feature matrix and the second feature matrix to obtain the target feature matrix. In the feature extraction step, two different methods are used to obtain the nodule gray scale feature, and the HU range feature of the nodule is extracted. When doctors judge the type of the nodule, they judge the HU value range in the nodule, which has good classification effect and reduces the error of manual operation.

[0082] Preferably, in any of the above embodiments, further comprising: performing an etching operation on the second cross-sectional layer to obtain an etched second cross-sectional layer;

[0083] The obtaining of the foreground pixel matrix specifically includes:

[0084] The foreground pixels of the first cross-sectional layer and the etched second cross-sectional layer are extracted and processed into one dimension to obtain a foreground pixel matrix.

[0085] It should be noted that in an embodiment, the foreground pixel matrix is obtained by performing an etching operation on the lung nodule segmentation mask Mask slice to obtain an etched mask Mask slice-erode , and removing the pixels affected by the partial volume effect at the edge of the nodule.

[0086] The scheme removes the pixels affected by the partial volume effect at the edge of the nodule by etching operation, and optimizes the segmentation mask data.

[0087] In an embodiment, as shown in Figure 2 , a lung nodule image classification system includes an input module 1101, a preprocessing module 1102, a feature extraction module 1103, and a classification module 1104.

[0088] The input module 1101 is configured to input a pixel block and a segmentation mask of a lung nodule image.

[0089] The preprocessing module 1102 is configured to perform preprocessing on the pixel block and the segmentation mask to obtain a first cross-sectional layer of the pixel block and a second cross-sectional layer of the segmentation mask.

[0090] The feature extraction module 1103 is used to extract foreground pixels of the first cross-sectional layer and the second cross-sectional layer, and perform a preset process on the foreground pixels to obtain a target feature matrix;

[0091] The classification module 1104 is used to train a classifier according to the target feature matrix, and classify the lung nodule image into nodule types using the trained classifier.

[0092] This solution obtains the first cross-sectional layer of the pixel block and the second cross-sectional layer of the segmentation mask respectively by pre-processing the pixel block and the segmentation mask, extracts the foreground pixels of the first cross-sectional layer and the second cross-sectional layer, and performs preset processing on the foreground pixels to obtain a target feature matrix, trains a classifier according to the target feature matrix, and classifies the nodule type of the lung nodule image by the trained classifier. The present invention does not require high-demand hardware equipment conditions, does not require a large amount of labeled data, reduces manual loss, combines the nodule segmentation information, and starts from the method of judging the type of lung nodules when doctors read the film. It has strong interpretability and good classification effect.

[0093] Preferably, in any of the above embodiments, the pre-processing module 1102 is specifically configured to extract a layer of a largest cross-section of the pixel block as the first cross-section layer;

[0094] A layer of the segmentation mask with the largest cross-section area is extracted as the second cross-section layer.

[0095] Through the preprocessing operation, this solution only captures the image range that the doctor is concerned about, and effectively avoids interference from irrelevant factors such as the background.

[0096] Preferably, in any of the above embodiments, the feature extraction module 1103 is specifically used to extract foreground pixels of the first cross-sectional layer and the second cross-sectional layer, and process them into one dimension to obtain a foreground pixel matrix;

[0097] A windowing operation is performed on the foreground pixel matrix to obtain a foreground pixel matrix after the windowing operation.

[0098] Preferably, in any of the above embodiments, the feature extraction module 1103 is specifically configured to perform sorting processing on the foreground pixel matrix to obtain a sorted foreground pixel matrix;

[0099] A first characteristic matrix obtained by processing the sorted foreground pixel matrix using an interpolation method;

[0100] Performing grayscale normalization processing on the foreground pixel matrix to obtain a second characteristic matrix;

[0101] The first feature matrix and the second feature matrix are concatenated to obtain the target feature matrix.

[0102] This solution combines the first and second feature matrices to create the target feature matrix. This not only addresses the significant size differences among nodules but also effectively reflects the proportion of grayscale values ​​within each nodule. During the feature extraction step, two different methods were used to determine the grayscale features of the nodules, extracting the HU range within the nodules. This HU range is crucial for determining nodule type, resulting in effective classification and reducing manual error.

[0103] Preferably, in any of the above embodiments, the method further comprises: a mask optimization module, configured to perform an etching operation on the second cross-section layer to obtain an etched second cross-section layer;

[0104] The feature extraction module is specifically used to extract foreground pixels of the first cross-section layer and the corroded second cross-section layer, and process them into one dimension to obtain a foreground pixel matrix.

[0105] This scheme uses corrosion operation to remove pixels at the edge of nodules affected by partial volume effect and optimize the segmentation mask data.

[0106] It can be understood that in some embodiments, some or all of the optional implementation methods in the above embodiments may be included.

[0107] It should be noted that the above embodiments are product embodiments corresponding to the previous method embodiments. For the description of the optional implementation methods in the product embodiments, please refer to the corresponding description in the above method embodiments, which will not be repeated here.

[0108] The reader should understand that, in the description of this specification, reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art can combine and combine different embodiments or examples described in this specification and features of different embodiments or examples without contradiction.

[0109] In the several embodiments provided in this application, it should be understood that the disclosed apparatus and method can be implemented in other ways. For example, the method embodiments described above are merely illustrative. For example, the division of steps is merely a logical function division. In actual implementation, other division methods may be used. For example, multiple steps may be combined or integrated into another step, or some features may be ignored or not performed.

[0110] If the above method is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the various embodiments of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, and other media that can store program code.

[0111] The above are merely specific embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in the present invention, and such modifications or substitutions are intended to be within the scope of protection of the present invention. Therefore, the scope of protection of the present invention shall be subject to the scope of protection of the claims.

Claims

1. A method for classifying lung nodule images, characterized in that: include: The input includes pixel patches of the lung nodule image and the segmentation mask of the lung nodule; Preprocessing the pixel block and the segmentation mask to obtain a first cross-sectional layer of the pixel block and a second cross-sectional layer of the segmentation mask respectively; extracting foreground pixels of the first cross-sectional layer and the second cross-sectional layer, and performing a preset process on the foreground pixels to obtain a target feature matrix; Training a classifier according to the target feature matrix, and classifying the lung nodule image into nodule types using the trained classifier; The extracting foreground pixels of the first cross-section layer and the second cross-section layer further comprises: Extracting foreground pixels of the first cross-sectional layer and the second cross-sectional layer, and processing them into one dimension to obtain a foreground pixel matrix; Performing a windowing operation on the foreground pixel matrix to obtain a foreground pixel matrix after windowing; The performing preset processing on the foreground pixels to obtain a target feature matrix specifically includes: Sorting the foreground pixel matrix to obtain a sorted foreground pixel matrix; A first characteristic matrix obtained by processing the sorted foreground pixel matrix using an interpolation method; Performing grayscale normalization processing on the foreground pixel matrix to obtain a second characteristic matrix; The first feature matrix and the second feature matrix are concatenated to obtain the target feature matrix.

2. The method for classifying lung nodules according to claim 1, wherein: The preprocessing of the pixel block and the segmentation mask to obtain the first cross-sectional layer of the pixel block and the second cross-sectional layer of the segmentation mask respectively specifically includes: extracting a layer of a cross section with the largest area of ​​the pixel block as a first cross section layer; A layer of the segmentation mask with the largest cross-section area is extracted as the second cross-section layer.

3. The method for classifying lung nodules according to claim 1, wherein: Also includes: performing an etching operation on the second cross-sectional layer to obtain an etched second cross-sectional layer; The obtaining of the foreground pixel matrix specifically includes: Foreground pixels of the first cross-section layer and the corroded second cross-section layer are extracted and processed into one dimension to obtain a foreground pixel matrix.

4. A classification system for lung nodule images, characterized in that: include: Input module, preprocessing module, feature extraction module and classification module; The input module is used to input pixel blocks including a lung nodule image and a segmentation mask of the lung nodule; The pre-processing module is used to pre-process the pixel block and the segmentation mask to obtain a first cross-sectional layer of the pixel block and a second cross-sectional layer of the segmentation mask respectively; The feature extraction module is used to extract foreground pixels of the first cross-sectional layer and the second cross-sectional layer, and perform preset processing on the foreground pixels to obtain a target feature matrix; The classification module is used to train a classifier according to the target feature matrix, and classify the lung nodule image into nodule types using the trained classifier; The feature extraction module is specifically used to extract foreground pixels of the first cross-sectional layer and the second cross-sectional layer, and process them into one dimension to obtain a foreground pixel matrix; Performing a windowing operation on the foreground pixel matrix to obtain a foreground pixel matrix after windowing; The feature extraction module is specifically used to sort the foreground pixel matrix to obtain a sorted foreground pixel matrix; A first characteristic matrix obtained by processing the sorted foreground pixel matrix using an interpolation method; Performing grayscale normalization processing on the foreground pixel matrix to obtain a second characteristic matrix; The first feature matrix and the second feature matrix are concatenated to obtain the target feature matrix.

5. A lung nodule image classification system according to claim 4, characterized in that: The pre-processing module is specifically used to extract the layer of the largest cross-section of the pixel block as the first cross-section layer; A layer of the segmentation mask with the largest cross-section area is extracted as the second cross-section layer.

6. A lung nodule image classification system according to claim 4, characterized in that: Also includes: a mask optimization module, configured to perform an etching operation on the second cross-sectional layer to obtain an etched second cross-sectional layer; The feature extraction module is specifically used to extract foreground pixels of the first cross-section layer and the corroded second cross-section layer, and process them into one dimension to obtain a foreground pixel matrix.

Images