Refined pulmonary nodule detection method, system, equipment and medium

By pre-processing and RESIZE processing of CT images, combined with lung nodule screening and false positive reduction network, the problem of low accuracy of CT image data is solved, the accuracy and speed of lung nodule detection is improved, and more efficient lung Nodules screening and classification.

CN113658162BActive Publication Date: 2024-12-03YANGSHI TECHNOLOGY (GUANGDONG) CO LTD +1
View PDF 1 Cites 0 Cited by

Patent Information

Application Number
CN202110975714.4
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2021-08-24
Publication Date
2024-12-03
Estimated Expiration
2041-08-24

Smart Images

  • Figure CN113658162B_ABST
    Figure CN113658162B_ABST
Patent Text Reader

Abstract

The present invention discloses a refined lung nodule detection method, system, device and medium. The refined lung nodule detection method of the present invention comprises: obtaining a lung CT image and performing preprocessing and RESIZE processing on the image; using a lung nodule screening network to locate the position of the candidate lung nodules in the lung CT image after RESIZE processing, and obtaining a first candidate lung nodule set P; according to the candidate lung nodules in the first candidate lung nodule set P, according to the same position, obtaining an original candidate lung nodule set SP from the preprocessed lung CT image; using a lung nodule detection network to extract the original candidate lung nodules from the original candidate lung nodule set SP, and obtaining a second candidate lung nodule set N; using a lung nodule false positive reduction network to detect the original candidate lung nodules in the second candidate lung nodule set N, and outputting the detection result of the true positive lung nodules. The refined lung nodule detection method of the present invention has high precision of the original lung data, and strong phase quality and resolution of the lung CT image.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of medical imaging computer-aided diagnosis, and in particular to a refined lung nodule detection method, system, device and computer-readable storage medium. Background Art

[0002] Lung cancer is the cancer with the highest morbidity and mortality in the world. Lung cancer exists in the form of lung nodules in its early stages. Early diagnosis and treatment of lung nodules can improve the patient's survival rate. Computed tomography (CT) is currently the most commonly used image for lung nodule detection. The current computer-aided detection system for lung nodules mainly consists of two stages: candidate nodule detection and false positive screening. The goal of the candidate nodule detection stage is to detect all suspected nodules in the CT images as much as possible to improve the sensitivity of lung nodule detection; the goal of the false positive screening stage is to classify the detected candidate nodules into true positive nodules and false positive nodules to exclude false positive nodules from the candidate nodules, thereby improving the accuracy of lung nodule detection.

[0003] However, because patients undergoing computed tomography scans have different body shapes, the resulting CT images show great differences in lung shape, size, texture, etc. In addition, the CT images may also contain a lot of noise and motion artifacts caused by the patient's movement during the scan, resulting in low data accuracy, poor phase quality, and low resolution of the CT images, which in turn affects the later screening and detection of lung nodules. Summary of the invention

[0004] Based on this, the purpose of the present invention is to provide a refined lung nodule detection method, system, device and computer-readable storage medium, which can provide lung original data with extremely high data accuracy, improve the phase quality and resolution of lung CT images, and ensure the effective implementation of the entire lung nodule detection method.

[0005] The purpose of the present invention is achieved through the following technical solutions:

[0006] In a first aspect, a refined lung nodule detection method comprises the following steps:

[0007] Acquire lung CT images and preprocess them;

[0008] Perform RESIZE processing on the preprocessed lung CT images;

[0009] Using the lung nodule screening network, the positions of candidate lung nodules are located in the lung CT images processed by RESIZE, and the first candidate lung nodule set P is obtained;

[0010] According to the candidate lung nodules in the first candidate lung nodule set P, an original candidate lung nodule set SP is obtained from the preprocessed lung CT image according to the same position;

[0011] The pulmonary nodule detection network is used to extract original candidate pulmonary nodules from the original candidate pulmonary nodule set SP to obtain a second candidate pulmonary nodule set N;

[0012] The pulmonary nodule false positive reduction network is used to detect the original candidate pulmonary nodules in the second candidate pulmonary nodule set N, and the detection results of the true positive pulmonary nodules are output.

[0013] The refined lung nodule detection method of the present invention gradually refines the lung nodule detection process. It first performs preprocessing such as data enhancement, data standardization, data noise reduction, and data normalization on the lung CT image, and then unifies the size of the lung CT image. This provides original data with extremely high data accuracy for subsequent positioning, screening, extraction, and classification detection of lung nodules, improves the phase quality and resolution of the lung CT image, and ensures the effective implementation of the entire lung nodule detection method.

[0014] Further preferably, the preprocessing in obtaining and preprocessing the lung CT image comprises one or any combination of standardization, normalization, noise removal, data enhancement, and image segmentation, wherein the image segmentation is to segment the lung area from the obtained lung CT image; and

[0015] The RESIZE processing of the preprocessed lung CT image specifically includes cropping the preprocessed lung CT image to a preset size.

[0016] Further preferably, the refined pulmonary nodule detection method further comprises the steps of:

[0017] Extracting one or any combination of grayscale features, shape features, and texture features of candidate pulmonary nodules from the first candidate pulmonary nodule set P, and filtering out some non-pulmonary nodules in the candidate pulmonary nodules according to a preset classification rule;

[0018] The grayscale features include one or any combination of grayscale mean, grayscale variance, grayscale minimum, skewness, and kurtosis; the shape features include one or any combination of area, radius, perimeter, roundness, compactness, concavity, slenderness, and rectangularity; the texture features include one or any combination of energy, contrast, entropy, and inverse moment; and

[0019] The preset classification rule is composed of a preset threshold corresponding to each data included in the grayscale feature, a preset threshold corresponding to each data included in the shape feature, and a preset threshold corresponding to each data included in the texture feature.

[0020] Further preferably, the refined lung nodule detection method also includes the step of backing up and saving the preprocessed lung CT image so that when there is an objection to the detection result of the true positive lung nodule, the lung CT image can be traced and re-detected.

[0021] In the second aspect, a refined lung nodule detection system includes:

[0022] A lung CT image preprocessing module, used for obtaining and preprocessing lung CT images;

[0023] A lung CT image RESIZE processing module is used to perform RESIZE processing on the pre-processed lung CT image;

[0024] A first candidate pulmonary nodule set P obtaining module, used to locate the positions of candidate pulmonary nodules in the lung CT image processed by RESIZE by using the pulmonary nodule screening network, and obtain the first candidate pulmonary nodule set P;

[0025] An original candidate pulmonary nodule set SP obtaining module is used to obtain an original candidate pulmonary nodule set SP from the preprocessed lung CT image according to the candidate pulmonary nodules in the first candidate pulmonary nodule set P and at the same position;

[0026] A second candidate pulmonary nodule set N obtaining module is used to extract original candidate pulmonary nodules from the original candidate pulmonary nodule set SP using a pulmonary nodule detection network to obtain a second candidate pulmonary nodule set N;

[0027] The lung nodule true positive detection module is used to detect the original candidate lung nodules in the second candidate lung nodule set N using the lung nodule false positive reduction network, and output the detection results of the true positive lung nodules.

[0028] Further preferably, the preprocessing in the lung CT image preprocessing module includes one or any combination of standardization processing, normalization processing, noise removal processing, data enhancement processing, and image segmentation processing, wherein the image segmentation processing is to segment the lung area from the obtained lung CT image; and

[0029] The RESIZE processing of the lung CT image RESIZE processing module is specifically to crop the pre-processed lung CT image to a preset size.

[0030] Further preferably, it further comprises a non-pulmonary nodule filtering module, which is used to extract one or any combination of grayscale features, shape features, and texture features of candidate pulmonary nodules from the first candidate pulmonary nodule set P, and filter out some non-pulmonary nodules in the candidate pulmonary nodules according to a preset classification rule;

[0031] The grayscale features include one or any combination of grayscale mean, grayscale variance, grayscale minimum, skewness, and kurtosis; the shape features include one or any combination of area, radius, perimeter, roundness, compactness, concavity, slenderness, and rectangularity; the texture features include one or any combination of energy, contrast, entropy, and inverse moment; and

[0032] The preset classification rule is composed of a preset threshold corresponding to each data included in the grayscale feature, a preset threshold corresponding to each data included in the shape feature, and a preset threshold corresponding to each data included in the texture feature.

[0033] Further preferably, a lung CT image backup module is included, which is used to back up and save the preprocessed lung CT images so that when there is an objection to the detection result of the true positive lung nodule, the lung CT images can be traced and re-detected.

[0034] In a third aspect, a refined lung nodule detection device is provided, the device comprising a storage device and a processor, the storage device being used to store one or more programs;

[0035] When the one or more programs are executed by the processor, the processor implements the refined pulmonary nodule detection method as described in any one of the above items.

[0036] In a fourth aspect, a computer-readable storage medium stores at least one program, which, when executed by a processor, implements the refined lung nodule detection method as described in any one of the above items.

[0037] Compared with the prior art, the refined lung nodule detection method, system, device and computer-readable storage medium of the present invention have high data accuracy, no unnecessary calculation steps, simple and fast calculation, and gradually refine the lung nodule detection process, accurate detection results, and strong practicality.

[0038] For better understanding and implementation, the present invention is described in detail below with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 It is a flow chart of the refined lung nodule detection method of the present invention.

[0040] Figure 2 It is a module schematic diagram of the refined lung nodule detection system of the present invention. DETAILED DESCRIPTION

[0041] The directional terms such as up, down, left, right, front, back, front, back, top, bottom, etc. mentioned or may be mentioned in this specification are defined relative to their structure and are relative concepts. Therefore, they may change accordingly according to their different locations and different usage conditions. Therefore, these or other directional terms should not be interpreted as restrictive terms.

[0042] The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present disclosure. Instead, they are merely examples of methods consistent with some aspects of the present disclosure as detailed in the appended claims.

[0043] The terms used in this disclosure are for the purpose of describing specific embodiments only and are not intended to limit the disclosure. The singular forms of "a", "said" and "the" used in this disclosure and the appended claims are also intended to include plural forms unless the context clearly indicates otherwise. It should also be understood that the term "and / or" used herein refers to and includes any or all possible combinations of one or more associated listed items.

[0044] A refined lung nodule detection method, such as Figure 1 As shown, the following steps are included:

[0045] S1. Obtain lung CT images and preprocess them.

[0046] Specifically, the lung CT image is obtained and preprocessed, and the preprocessing includes one or any combination of standardization, normalization, noise removal, data enhancement, and image segmentation to make the quality of the lung CT image more refined and improve the subsequent detection accuracy.

[0047] Among them, the image segmentation processing of the lung CT image is to segment the lung area from the obtained lung CT image, which can be specifically the lung parenchyma area, that is, the bronchi at all levels in the lungs and the alveolar structures at their terminals, that is, the cavities and tube walls in contact with the air in the lungs.

[0048] The normalization processing of lung CT images is performed by the following formula:

[0049]

[0050] where x i represents the pixel value of the lung CT image, max(x) represents the maximum value of the pixel value of the lung CT image, min(x) represents the minimum value of the pixel value of the lung CT image, and norm represents the normalized pixel value, which falls in the interval [0, 1].

[0051] The normalization of lung CT images is performed by the following formula:

[0052]

[0053] Where μ represents the mean of the lung CT image, x represents the lung CT image matrix, σ represents the standard deviation of all pixels of the lung CT image, and N represents the number of pixels in the lung CT image.

[0054] The denoising of lung CT images can be achieved by combining local denoising and non-local denoising to comprehensively filter out Gaussian noise, Poisson noise, multiplicative noise, and salt and pepper noise that may exist in lung CT images.

[0055] S2. Perform RESIZE processing on the preprocessed lung CT image.

[0056] Specifically, the RESIZE processing of the pre-processed lung CT image is specifically to crop the pre-processed lung CT image to a preset size to unify the sizes of a plurality of lung CT images. The RESIZE processing may adopt the INTER_AREA bilinear interpolation method to reduce the pre-processed lung CT image to a size of 96*96*z.

[0057] S3. Using the lung nodule screening network, the positions of the candidate lung nodules are located in the lung CT image processed by RESIZE to obtain the first candidate lung nodule set P.

[0058] Specifically, it is preferred to use an artificial intelligence three-dimensional visualization tool to locate candidate lung nodules so as to identify candidate lung nodules more accurately and comprehensively.

[0059] More specifically, S3 also includes step S31: extracting one or any combination of the grayscale features, shape features, and texture features of the candidate lung nodules from the first candidate lung nodule set P, and filtering out some non-pulmonary nodules in the candidate lung nodules according to preset classification rules, so as to further select the candidate lung nodule data within the first candidate lung nodule set P.

[0060] Among them, the grayscale features include one or any combination of grayscale mean, grayscale variance, grayscale minimum, skewness, and kurtosis, the shape features include one or any combination of area, radius, circumference, roundness, compactness, concavity, slenderness, and rectangularity, and the texture features include one or any combination of energy, contrast, entropy, and inverse moment.

[0061] Furthermore, the preset classification rule is composed of preset thresholds corresponding to each data included in the grayscale feature, preset thresholds corresponding to each data included in the shape feature, and preset thresholds corresponding to each data included in the texture feature. For example, the preset threshold of the grayscale mean included in the grayscale feature is A, the preset threshold of the area included in the shape feature is B, and the preset threshold of the energy included in the texture feature is C. If at least two of the grayscale mean, area, and grayscale mean of the candidate lung nodules extracted from the first candidate lung nodule set P exceed the corresponding thresholds, then they are filtered out from the first candidate lung nodule set P. In addition, the classification rule can be modified according to actual applications into any composition form of various data in the grayscale feature, shape feature, and texture feature.

[0062] S4. According to the candidate lung nodules in the first candidate lung nodule set P, an original candidate lung nodule set SP is obtained from the preprocessed lung CT image according to the same position.

[0063] S5. Use the lung nodule detection network to extract original candidate lung nodules from the original candidate lung nodule set SP to obtain a second candidate lung nodule set N.

[0064] The lung CT image is first preprocessed and RESIZE processed to obtain the first candidate lung nodule set P. The quality and accuracy of the source image of the candidate lung nodules in the first candidate lung nodule set P are first guaranteed, and then the original candidate lung nodule set SP is obtained from the preprocessed lung CT image. The original candidate lung nodules, that is, the second candidate lung nodule set N, can be further obtained in turn, thereby ensuring the originality of the data, reducing the complexity of the system calculation and improving the smoothness of the system operation.

[0065] S6. Use the lung nodule false positive reduction network to detect the original candidate lung nodules in the second candidate lung nodule set N, and output the detection results of the true positive lung nodules.

[0066] Specifically, it is preferred to use a lung nodule false positive reduction network with a hybrid attention module to detect the original candidate lung nodules in the second candidate lung nodule set N, so as to more comprehensively extract the features of the original candidate lung nodules in the second candidate lung nodule set N and output more accurate detection results.

[0067] S7. Back up and save the pre-processed lung CT image so that when there is any objection to the detection result of the true positive lung nodule, the lung CT image can be traced back and re-tested.

[0068] Specifically, the secondary detection preferably repeats the above S1-S6 to avoid detection abnormalities caused by improper system operation, etc. Of course, the secondary detection can also be performed by other conventional detection methods, the purpose of which is to re-detect abnormal pulmonary nodules to improve the final detection accuracy and not miss any suspected true positive pulmonary nodules.

[0069] The present invention also discloses a refined pulmonary nodule detection system, such as Figure 2 As shown, including:

[0070] A lung CT image preprocessing module, used for obtaining and preprocessing lung CT images;

[0071] A lung CT image RESIZE processing module is used to perform RESIZE processing on the pre-processed lung CT image;

[0072] A first candidate pulmonary nodule set P obtaining module, used to locate the positions of candidate pulmonary nodules in the lung CT image processed by RESIZE by using the pulmonary nodule screening network, and obtain the first candidate pulmonary nodule set P;

[0073] An original candidate pulmonary nodule set SP obtaining module is used to obtain an original candidate pulmonary nodule set SP from the preprocessed lung CT image according to the candidate pulmonary nodules in the first candidate pulmonary nodule set P and at the same position;

[0074] A second candidate pulmonary nodule set N obtaining module is used to extract original candidate pulmonary nodules from the original candidate pulmonary nodule set SP using a pulmonary nodule detection network to obtain a second candidate pulmonary nodule set N;

[0075] The lung nodule true positive detection module is used to detect the original candidate lung nodules in the second candidate lung nodule set N using the lung nodule false positive reduction network, and output the detection results of the true positive lung nodules.

[0076] Further preferably, the preprocessing in the lung CT image preprocessing module includes one or any combination of standardization processing, normalization processing, noise removal processing, data enhancement processing, and image segmentation processing, wherein the image segmentation processing is to segment the lung area from the obtained lung CT image; and

[0077] The RESIZE processing of the lung CT image RESIZE processing module is specifically to crop the pre-processed lung CT image to a preset size.

[0078] Further preferably, it further comprises a non-pulmonary nodule filtering module, which is used to extract one or any combination of grayscale features, shape features, and texture features of candidate pulmonary nodules from the first candidate pulmonary nodule set P, and filter out some non-pulmonary nodules in the candidate pulmonary nodules according to a preset classification rule;

[0079] The grayscale features include one or any combination of grayscale mean, grayscale variance, grayscale minimum, skewness, and kurtosis; the shape features include one or any combination of area, radius, perimeter, roundness, compactness, concavity, slenderness, and rectangularity; the texture features include one or any combination of energy, contrast, entropy, and inverse moment; and

[0080] The preset classification rule is composed of a preset threshold corresponding to each data included in the grayscale feature, a preset threshold corresponding to each data included in the shape feature, and a preset threshold corresponding to each data included in the texture feature.

[0081] Further preferably, a lung CT image backup module is included, which is used to back up and save the preprocessed lung CT images so that when there is an objection to the detection result of the true positive lung nodule, the lung CT images can be traced and re-detected.

[0082] The present invention also discloses a refined lung nodule detection device, the device comprising a storage device and a processor, the storage device being used to store one or more programs;

[0083] When the one or more programs are executed by the processor, the processor implements the refined pulmonary nodule detection method as described above.

[0084] The device may also preferably include a communication interface, and the communication interface is used for communicating with an external device and transmitting data interactively.

[0085] It should be noted that the memory may include a high-speed RAM memory, and may also include a nonvolatile memory (nonvolatile memory), such as at least one disk memory.

[0086] In specific implementation, if the memory, processor and communication interface are integrated on a chip, the memory, processor and communication interface can communicate with each other through an internal interface. If the memory, processor and communication interface are implemented independently, the memory, processor and communication interface can be connected to each other through a bus and communicate with each other.

[0087] The present invention also discloses a computer-readable storage medium storing at least one program, which, when executed by a processor, implements the refined lung nodule detection method as described above.

[0088] It should be understood that the computer-readable storage medium is any data storage device that can store data or programs, which can then be read by a computer system. Examples of computer-readable storage media include read-only memory, random access memory, CD-ROM, HDD, DVD, magnetic tape, and optical data storage devices, etc.

[0089] The computer readable storage medium can also be distributed over network coupled computer systems so that the computer readable code is stored and executed in a distributed fashion.

[0090] The program code contained on the computer-readable storage medium may be transmitted using any appropriate medium, including but not limited to: wireless, wire, optical cable, radio frequency (RF), etc., or any suitable combination of the above.

[0091] In some embodiments, the computer-readable storage medium may also be non-transitory.

[0092] Compared with the prior art, the refined lung nodule detection method, system, device and computer-readable storage medium of the present invention have high data accuracy, no unnecessary calculation steps, simple and fast calculation, and gradually refine the lung nodule detection process, accurate detection results, and strong practicality.

[0093] The above-mentioned embodiments only express several implementation methods of the present invention, and the description is relatively specific and detailed, but it cannot be understood as limiting the scope of the invention patent. It should be pointed out that for ordinary technicians in this field, several modifications and improvements can be made without departing from the concept of the present invention, which all belong to the protection scope of the present invention.

Claims

1. A refined lung nodule detection method, It is characterized in that The following steps are involved: A lung CT image is obtained and preprocessed. The preprocessing includes one or any combination of standardization, normalization, noise removal, data enhancement, and image segmentation. The normalization of the lung CT image is performed by the following formula: , where x i represents the pixel value of the lung CT image, max(x) represents the maximum value of the pixel value of the lung CT image, min(x) represents the minimum value of the pixel value of the lung CT image, and norm represents the normalized pixel value, which falls in the interval [0, 1]. The normalization of lung CT images is performed by the following formula: , ; Where μ represents the mean of the lung CT image, x represents the lung CT image matrix, σ represents the standard deviation of all pixels in the lung CT image, and N represents the number of pixels in the lung CT image. The denoising of lung CT images uses a combination of local denoising and non-local denoising to filter out Gaussian noise, Poisson noise, multiplicative noise, and salt and pepper noise that may exist in lung CT images; Performing RESIZE processing on the preprocessed lung CT image, wherein the performing RESIZE processing on the preprocessed lung CT image specifically includes cropping the preprocessed lung CT image to a preset size; Using the lung nodule screening network, the positions of candidate lung nodules are located in the lung CT images processed by RESIZE, and the first candidate lung nodule set P is obtained; According to the candidate lung nodules in the first candidate lung nodule set P, an original candidate lung nodule set SP is obtained from the preprocessed lung CT image according to the same position; The pulmonary nodule detection network with a hybrid attention module is used to extract original candidate pulmonary nodules from the original candidate pulmonary nodule set SP to obtain a second candidate pulmonary nodule set N; The original candidate lung nodules in the second candidate lung nodule set N are detected using the lung nodule false positive reduction network, and the detection results of the true positive lung nodules are output; The refined lung nodule detection method further comprises the steps of: Extracting one or any combination of grayscale features, shape features, and texture features of candidate pulmonary nodules from the first candidate pulmonary nodule set P, and filtering out some non-pulmonary nodules in the candidate pulmonary nodules according to a preset classification rule; The grayscale features include one or any combination of grayscale mean, grayscale variance, grayscale minimum, skewness, and kurtosis; the shape features include one or any combination of area, radius, perimeter, roundness, compactness, concavity, slenderness, and rectangularity; the texture features include one or any combination of energy, contrast, entropy, and inverse moment; and The preset classification rules are composed of preset thresholds corresponding to each data included in the grayscale feature, preset thresholds corresponding to each data included in the shape feature, and preset thresholds corresponding to each data included in the texture feature, wherein the preset threshold of the grayscale mean included in the grayscale feature is A, the preset threshold of the area included in the shape feature is B, and the preset threshold of the energy included in the texture feature is C. If at least two of the grayscale mean, area, and grayscale mean of the candidate lung nodule extracted from the first candidate lung nodule set P exceed the corresponding thresholds, then it will be filtered out from the first candidate lung nodule set P.

2. The refined lung nodule detection method according to claim 1, It is characterized in that The refined lung nodule detection method also includes the step of backing up and saving the pre-processed lung CT image so that when there is an objection to the detection result of the true positive lung nodule, the lung CT image can be traced back and re-detected.

3. A refined lung nodule detection system, It is characterized in that include: The lung CT image preprocessing module is used to obtain and preprocess the lung CT image. The preprocessing includes one or any combination of standardization, normalization, denoising, data enhancement, and image segmentation. The normalization of the lung CT image is performed by the following formula: , where x i represents the pixel value of the lung CT image, max(x) represents the maximum value of the pixel value of the lung CT image, min(x) represents the minimum value of the pixel value of the lung CT image, and norm represents the normalized pixel value, which falls in the interval [0, 1]. The normalization of lung CT images is performed by the following formula: , ; Where μ represents the mean of the lung CT image, x represents the lung CT image matrix, σ represents the standard deviation of all pixels in the lung CT image, and N represents the number of pixels in the lung CT image. The denoising of lung CT images uses a combination of local denoising and non-local denoising to filter out Gaussian noise, Poisson noise, multiplicative noise, and salt and pepper noise that may exist in lung CT images; Performing RESIZE processing on the preprocessed lung CT image, wherein the performing RESIZE processing on the preprocessed lung CT image specifically includes cropping the preprocessed lung CT image to a preset size; A first candidate pulmonary nodule set P obtaining module, used to locate the positions of candidate pulmonary nodules in the lung CT image processed by RESIZE by using the pulmonary nodule screening network, and obtain the first candidate pulmonary nodule set P; An original candidate pulmonary nodule set SP obtaining module is used to obtain an original candidate pulmonary nodule set SP from the preprocessed lung CT image according to the candidate pulmonary nodules in the first candidate pulmonary nodule set P and at the same position; A second candidate pulmonary nodule set N obtaining module is used to extract original candidate pulmonary nodules from the original candidate pulmonary nodule set SP using a pulmonary nodule detection network with a hybrid attention module to obtain a second candidate pulmonary nodule set N; A lung nodule true positive detection module is used to detect the original candidate lung nodules in the second candidate lung nodule set N using a lung nodule false positive reduction network, and output the detection result of the true positive lung nodule; A non-pulmonary nodule filtering module is used to extract one or any combination of grayscale features, shape features, and texture features of candidate pulmonary nodules from the first candidate pulmonary nodule set P, and filter out some non-pulmonary nodules in the candidate pulmonary nodules according to a preset classification rule; The grayscale features include one or any combination of grayscale mean, grayscale variance, grayscale minimum, skewness, and kurtosis; the shape features include one or any combination of area, radius, perimeter, roundness, compactness, concavity, slenderness, and rectangularity; the texture features include one or any combination of energy, contrast, entropy, and inverse moment; and The preset classification rule is composed of a preset threshold corresponding to each data included in the grayscale feature, a preset threshold corresponding to each data included in the shape feature, and a preset threshold corresponding to each data included in the texture feature; and, The preset classification rules are composed of preset thresholds corresponding to each data included in the grayscale feature, preset thresholds corresponding to each data included in the shape feature, and preset thresholds corresponding to each data included in the texture feature, wherein the preset threshold of the grayscale mean included in the grayscale feature is A, the preset threshold of the area included in the shape feature is B, and the preset threshold of the energy included in the texture feature is C. If at least two of the grayscale mean, area, and grayscale mean of the candidate lung nodule extracted from the first candidate lung nodule set P exceed the corresponding thresholds, then it will be filtered out from the first candidate lung nodule set P.

Citation Information

Patent Citations

  • Pulmonary nodule detection method based on machine learning

    CN108549912A