CN111784638B - Convolutional neural network-based lung nodule false positive screening method and system - Google Patents
Convolutional neural network-based lung nodule false positive screening method and system Download PDFInfo
- Publication number
- CN111784638B CN111784638B CN202010502186.6A CN202010502186A CN111784638B CN 111784638 B CN111784638 B CN 111784638B CN 202010502186 A CN202010502186 A CN 202010502186A CN 111784638 B CN111784638 B CN 111784638B
- Authority
- CN
- China
- Prior art keywords
- image data
- lung
- candidate
- lung nodule
- original
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 206010056342 Pulmonary mass Diseases 0.000 title claims abstract description 125
- 238000013527 convolutional neural network Methods 0.000 title claims abstract description 46
- 238000012216 screening Methods 0.000 title claims abstract description 39
- 238000000034 method Methods 0.000 title claims abstract description 38
- 238000012549 training Methods 0.000 claims abstract description 42
- 238000012545 processing Methods 0.000 claims abstract description 36
- 210000004072 lung Anatomy 0.000 claims abstract description 28
- 238000011176 pooling Methods 0.000 claims description 24
- 238000010606 normalization Methods 0.000 claims description 7
- 239000011295 pitch Substances 0.000 claims description 7
- 238000000605 extraction Methods 0.000 claims description 6
- 230000015572 biosynthetic process Effects 0.000 claims description 5
- 238000004364 calculation method Methods 0.000 claims description 5
- 230000004927 fusion Effects 0.000 claims description 5
- 238000003786 synthesis reaction Methods 0.000 claims description 5
- 239000000203 mixture Substances 0.000 claims description 4
- 238000001514 detection method Methods 0.000 abstract description 9
- 230000006870 function Effects 0.000 description 8
- 238000013135 deep learning Methods 0.000 description 5
- 230000007547 defect Effects 0.000 description 4
- 206010058467 Lung neoplasm malignant Diseases 0.000 description 3
- 201000005202 lung cancer Diseases 0.000 description 3
- 208000020816 lung neoplasm Diseases 0.000 description 3
- 230000004913 activation Effects 0.000 description 2
- 238000002591 computed tomography Methods 0.000 description 2
- 238000001914 filtration Methods 0.000 description 2
- 238000002372 labelling Methods 0.000 description 2
- PXFBZOLANLWPMH-UHFFFAOYSA-N 16-Epiaffinine Natural products C1C(C2=CC=CC=C2N2)=C2C(=O)CC2C(=CC)CN(C)C1C2CO PXFBZOLANLWPMH-UHFFFAOYSA-N 0.000 description 1
- 206010028980 Neoplasm Diseases 0.000 description 1
- 201000011510 cancer Diseases 0.000 description 1
- 238000012937 correction Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 230000036210 malignancy Effects 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 230000002685 pulmonary effect Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 230000004083 survival effect Effects 0.000 description 1
- 238000013519 translation Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
- G06T7/0012—Biomedical image inspection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10072—Tomographic images
- G06T2207/10081—Computed x-ray tomography [CT]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20212—Image combination
- G06T2207/20221—Image fusion; Image merging
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30004—Biomedical image processing
- G06T2207/30061—Lung
- G06T2207/30064—Lung nodule
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- Health & Medical Sciences (AREA)
- General Physics & Mathematics (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Biophysics (AREA)
- Molecular Biology (AREA)
- Biomedical Technology (AREA)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Computational Linguistics (AREA)
- Software Systems (AREA)
- Life Sciences & Earth Sciences (AREA)
- Medical Informatics (AREA)
- Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
- Radiology & Medical Imaging (AREA)
- Quality & Reliability (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Apparatus For Radiation Diagnosis (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a method and a system for screening false positive of lung nodules based on a convolutional neural network, wherein the method comprises the following steps: acquiring coordinate positions and maximum radius values of the lung nodule candidates from lung CT image data; extracting original 3D image data of the candidate lung nodule from the lung CT image data according to the coordinate position and the maximum radius value, and carrying out interpolation processing on the original 3D image data; obtaining sample data of three planes corresponding to the 3D image data of the lung nodule candidate obtained through interpolation, and scaling the sample data of the three planes to form a training set; and training the convolutional neural network based on the training set, and performing false positive screening on the candidate lung nodule through a convolutional neural network model obtained through training. The embodiment of the invention can solve the problem of high false positive rate in the process of identifying the lung nodule in the existing end-to-end network, and improves the accuracy of computer-aided automatic detection of the lung nodule.
Description
Technical Field
The invention relates to the technical field of medical treatment, in particular to a method and a system for screening false positive of lung nodules based on a convolutional neural network.
Background
Lung cancer is the most common malignancy in the world, early discovery of early treatment is the only effective way to increase survival, and low dose CT (Computed Tomography, i.e., computed tomography) is the only early lung cancer screening tool currently available. Early lung cancer is mainly manifested as asymptomatic lung nodules, and due to its complex morphology, even experienced doctors have difficulty in making accurate decisions.
In the related research field of lung nodule detection, a computer-aided lung nodule automatic detection method based on lung CT images is proposed, and generally comprises the following two key steps: first, a lung nodule candidate region is obtained through preliminary detection of a lung nodule, and a correct recognition result containing a lung nodule and a false positive object not containing the lung nodule exist in the lung nodule candidate region at the same time; secondly, a proper lung nodule classifier is trained to screen the detection result of the lung nodule candidate region so as to remove false positive lung nodules. The detection method developed according to the two steps has the defect of solving the following problems: the deep learning network cannot distinguish the characteristics of the false positive lung nodules, so that a large number of false positive samples are generated, and the possibility of relying on manual identification is extremely low; the classification performance of the existing two-dimensional classifier is similar to the recognition capability of the deep learning network, and further screening work cannot be performed on false positive samples generated by the deep learning network; the existing three-dimensional classifier has the problems of difficult model training, over-fitting of the model, large calculated amount and the like.
Disclosure of Invention
The invention aims to overcome the defects of the prior art, provides a method and a system for screening false positive of lung nodules based on a convolutional neural network, can solve the problem of high false positive rate in the process of identifying the lung nodules by the existing end-to-end network, and improves the accuracy of computer-aided automatic detection of the lung nodules.
In order to solve the problems, the invention provides a convolutional neural network-based lung nodule false positive screening method, which comprises the following steps:
Acquiring coordinate positions and maximum radius values of the lung nodule candidates from lung CT image data;
Extracting original 3D image data of the lung nodule candidate from the lung CT image data according to the coordinate position and the maximum radius value, and carrying out interpolation processing on the original 3D image data to obtain final lung nodule candidate 3D image data;
Obtaining sample data of three planes corresponding to the 3D image data of the lung nodule candidate, and scaling the sample data of the three planes to form a training set;
And training the convolutional neural network based on the training set, and performing false positive screening on the candidate lung nodule through a convolutional neural network model obtained through training.
Optionally, the extracting the original 3D image data of the candidate lung nodule from the lung CT image data according to the coordinate position and the maximum radius value includes:
Setting a treatment region size of the lung nodule candidate based on the maximum radius value, the treatment region being centered on the coordinate location;
Calculating the number of slices contained in the original 3D image data of the lung nodule candidate to 2n+1 based on a given slice interval;
and taking the processing area as an intermediate layer, acquiring continuous N slices right above the processing area and continuous N slices right below the processing area, and combining the continuous N slices into the original 3D image data according to the original slice sequence.
Optionally, the interpolating the original 3D image data includes:
And carrying out interpolation operation on 2N slice pitches existing on the Z axis of the original 3D image data by using a Lagrange interpolation method with the XY pixel pitch of the original 3D image data as a reference.
Optionally, the acquiring sample data of three planes corresponding to the lung nodule candidate 3D image data includes:
Obtaining maximum projection data of each of three planes corresponding to the lung nodule candidate 3D image data;
And respectively carrying out normalization processing on the maximum projection data of each of the three planes to obtain sample data of each plane.
Optionally, the normalizing the maximum projection data of each of the three planes includes:
Based on the human tissue density range, screening and replacing maximum projection data of each of the three planes;
And carrying out normalization processing on the maximum projection data of each plane after screening and replacing.
Optionally, the scaling the sample data of the three planes includes: the sample data for each of the three planes is set to be equal in size.
Optionally, the network structure of the convolutional neural network includes four convolutional pooling layers, a feature synthesis layer, a full connection layer and a Softmax function layer.
Optionally, the training the convolutional neural network based on the training set, and performing false positive screening on the candidate lung nodule through a convolutional neural network model obtained by training includes:
Sequentially inputting sample data of each of three planes corresponding to the candidate lung nodule 3D image data to the four convolution pooling layers, and obtaining a first feature map corresponding to each plane;
performing feature fusion on the first feature map corresponding to each plane through the feature composition layer to obtain a second feature map of the candidate lung nodule 3D image data;
And classifying the result of the second feature map based on the full connection layer, and outputting the false positive probability of the candidate lung nodule according to the classified result based on the Softmax function layer.
The embodiment of the invention also provides a pulmonary nodule false positive screening system based on the convolutional neural network, which comprises:
The parameter acquisition module is used for acquiring the coordinate position and the maximum radius value of the lung nodule candidate from the lung CT image data;
the image extraction module is used for extracting original 3D image data of the lung nodule candidates from the lung CT image data according to the coordinate positions and the maximum radius values, and carrying out interpolation processing on the original 3D image data to obtain final lung nodule candidate 3D image data;
The data processing module is used for acquiring sample data of three planes corresponding to the 3D image data of the lung nodule candidate, and scaling the sample data of the three planes to form a training set;
And the sample training module is used for training the convolutional neural network based on the training set and carrying out false positive screening on the candidate lung nodule through the convolutional neural network model obtained through training.
In the embodiment of the invention, considering the defect that the deep learning network cannot distinguish the characteristics of the false positive lung nodules in the prior art, a novel convolutional neural network is provided for carrying out rapid and efficient characteristic extraction and classification screening on the candidate lung nodules, so that the accuracy of computer-aided automatic detection of the lung nodules can be improved; in addition, the embodiment of the invention can ensure the integrity and sufficiency of the input sample in the convolutional neural network by extracting the 3D image of the candidate lung nodule from the original lung CT image and increasing the sample data of the 3D image by utilizing the Lagrange interpolation method.
Drawings
In order to more clearly illustrate the embodiments of the invention or the technical solutions in the prior art, the drawings which are required in the description of the embodiments or the prior art will be briefly described, it being obvious that the drawings in the description below are only some embodiments of the invention, and that other drawings may be obtained according to these drawings without inventive effort for a person skilled in the art.
FIG. 1 is a schematic flow chart of a method for screening false positive lung nodules based on a convolutional neural network, which is disclosed in the embodiment of the invention;
Fig. 2 is a schematic structural diagram of a convolutional neural network-based lung nodule false positive screening system according to an embodiment of the present invention.
Detailed Description
The following description of the embodiments of the present invention will be made clearly and completely with reference to the accompanying drawings, in which it is apparent that the embodiments described are only some embodiments of the present invention, but not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.
Specifically, fig. 1 shows a flow chart of a method for screening false positive of lung nodules based on a convolutional neural network in an embodiment of the invention, and the method comprises the following steps:
S101, acquiring coordinate positions and maximum radius values of lung nodule candidates from lung CT image data;
In the embodiment of the invention, because each of all the lung nodule candidates contained in the lung CT image data has unique labeling information, the labeling information includes a coordinate position and a maximum radius value, and the required parameters can be obtained by direct search.
S102, extracting original 3D image data of the candidate lung nodule from the lung CT image data according to the coordinate position and the maximum radius value, and carrying out interpolation processing on the original 3D image data to obtain final candidate lung nodule 3D image data;
The specific implementation process comprises the following steps:
(1) Setting a treatment region size of the lung nodule candidate based on the maximum radius value, the treatment region being centered on the coordinate location;
Specifically, in the embodiment of the present invention, according to the known maximum radius value R, the length and width of the treatment area of the candidate lung nodule are set to be 1.5rx2.
(2) Based on a given slice interval, the number of slices contained in the original 3D image data of the lung nodule candidate is calculated to be 2N+1, wherein the calculation formula of the number of slices is as follows:
；
In the method, in the process of the invention, For the number of slices,/>For the height of the original 3D image data,/>Is the slice spacing. It should be noted that, since the embodiment of the present invention specifies that the original 3D image of the candidate lung nodule is a cube, the height of the original 3D image is 1.5rx2, and the slice interval is obtained by the lung CT image data.
(3) Taking the processing area as an intermediate layer, acquiring continuous N slices right above the processing area and continuous N slices right below the processing area, and combining the continuous N slices into the original 3D image data according to an original slice sequence, wherein the size of the 2N slices is equal to the size of the processing area of the candidate lung nodule, and the original slice sequence is obtained through the lung CT image data;
(4) And taking the XY pixel pitch of the original 3D image data as a reference, and carrying out interpolation operation on 2N slice pitches existing on the Z axis of the original 3D image data by using a Lagrange interpolation method so as to ensure that the pixel pitches of the original 3D image data on three coordinate axes of the X axis, the Y axis and the Z axis are equal.
S103, obtaining sample data of three planes corresponding to the 3D image data of the lung nodule candidate, and scaling the sample data of the three planes to form a training set;
The specific implementation process comprises the following steps:
(1) Obtaining maximum projection data of each of three planes corresponding to the lung nodule candidate 3D image data;
specifically, maximum projection processing is performed on the 3D image data of the lung nodule candidate on an XY plane, an XZ plane and a YZ plane, to obtain maximum projection data of the XY plane corresponding to the 3D image data of the lung nodule candidate, where the maximum projection data on the XY plane is: wherein/> For the X-axis coordinate value of the projection data,/>For the Y-axis coordinate value of the projection data,/>Is the 1 st pixel point/>, through which the light passes when projected onto the XY planeN is the total number of pixels through which the light passes. The HU value is Hunter unit, and is used to represent the relative density of tissue structure on the CT image of the lung.
(2) Respectively carrying out normalization processing on maximum projection data of each of the three planes to obtain sample data of each plane;
Specifically, based on the human tissue density range, screening and replacing maximum projection data of each of the three planes; then, normalization processing is carried out on the maximum value projection data of each plane after screening and replacement, so that the pixel value corresponding to the maximum value projection data of each plane is at the moment Within the range.
The screening replacement process is further described as follows: since the relative density of air is-1000 HU and the relative density of human skeleton is 400HU, the embodiment of the invention limits the density range of human tissues toThe filtering and replacing the maximum projection data of the XY plane corresponding to the 3D image data of the lung nodule candidate comprises the following three cases: when the maximum projection data of the XY plane is at/>When the range is within, no replacement processing is performed; when the maximum projection data of the XY plane is smaller than-1000 HU, replacing the maximum projection data with-1000 HU; when the maximum projection data of the XY plane is more than 400HU, the maximum projection data is replaced by 400HU.
It should be noted that, for the sample data of each plane obtained after normalization processing, one or more of affine, scaling, rotation, mirror image, overturning, translation, filtering and other expansion modes are used to expand the data, so as to ensure that the data volume of the input convolutional neural network is sufficient.
(3) The sample data of each of the three planes is scaled, that is, the size of the sample data of each of the three planes is set to be equal, and the size of the sample data of each of the three planes is unified to 128×128 pixels, where the sample data of each of the three planes also includes the sample data obtained by expansion.
And S104, training the convolutional neural network based on the training set, and performing false positive screening on the candidate lung nodule through a convolutional neural network model obtained through training.
In the embodiment of the invention, the network structure of the convolutional neural network comprises four convolutional pooling layers, a characteristic synthesis layer, a full connection layer and a Softmax function layer, wherein the four convolutional pooling layers are respectively:
a first convolution pooling layer: comprising 64 convolution kernels of size 3 x 3;
A second convolution pooling layer: comprising 128 convolution kernels of size 3 x 3;
third convolution pooling layer: comprising 156 convolution kernels of size 3 x 3;
fourth convolution pooling layer: comprising 512 convolution kernels of size 3 x 3.
It should be noted that, when the sample data is subjected to convolution calculation by the first convolution pooling layer, 64 different feature maps are obtained, after the partial feature maps in the 64 different feature maps are subjected to secondary convolution by the second convolution pooling layer, 128 different feature maps are obtained, after the partial feature maps in the 128 different feature maps are subjected to tertiary convolution by the third convolution pooling layer, 156 different feature maps are obtained, and after the partial feature maps in the 156 different feature maps are subjected to quaternary convolution by the fourth convolution pooling layer, 512 different feature maps are obtained; and each convolution pooling layer takes LeakyRELU (linear unit with leakage correction) as an activation function, the corresponding convolution result is required to be input into the activation function for mapping processing, 2 x 2 maximum pooling processing is executed, the original convolution result is subjected to feature extraction, and the calculation complexity of data is reduced by deleting unimportant feature graphs.
Specifically, the training process for the convolutional neural network based on the training set includes:
(1) Sequentially inputting sample data of each of three planes corresponding to the candidate lung nodule 3D image data to the four convolution pooling layers, and obtaining a first feature map corresponding to each plane;
Further, sample data of an XY plane corresponding to the 3D image data of the lung nodule candidate is gradually input to the first convolution pooling layer, the second convolution pooling layer, the third convolution pooling layer and the fourth convolution pooling layer for processing, and then an XY plane feature map of 8 multiplied by 512 is output; similarly, after the sample data of the XZ plane and the YZ plane corresponding to the 3D image data of the lung nodule candidate are processed by the four convolution pooling layers, an XZ plane feature map of 8×8×512 and a YZ plane feature map of 8×8×512 are correspondingly output.
(2) Performing feature fusion on the first feature map corresponding to each plane through the feature composition layer to obtain a second feature map of the candidate lung nodule 3D image data;
Further, the XY plane feature map, the XZ plane feature map and the YZ plane feature map are input to the feature synthesis layer to perform feature fusion, and a convolution operation is performed by using an internal 1×1 convolution kernel to obtain a second feature map of 8×8×1024 corresponding to the candidate lung nodule 3D image data.
(3) And classifying the result of the second feature map based on the full connection layer, and outputting the false positive probability of the candidate lung nodule according to the classified result based on the Softmax function layer.
Further, since 1024 feature results included in the second feature map are all highly purified feature results, the basic features of the candidate lung nodule can be accurately expressed, 1024 feature results are classified by using the full-connection layer with the core size of 1024×2, similar feature results are gathered into one class, and finally the above classification results are mapped to one by using the Softmax function layerAnd selecting a node value with the highest probability from the false positive categories as the false positive probability of the candidate lung nodule.
It should be noted that, the binary cross entropy loss function is adopted in the convolutional neural network model to determine the proximity degree of the classification result output by the fully connected layer and the expected output result, so as to feed back the prediction accuracy of the false positive probability of the candidate lung nodule.
Specifically, fig. 2 shows a convolutional neural network-based lung nodule false positive screening system in an embodiment of the present invention, the system comprising:
A parameter obtaining module 201, configured to obtain a coordinate position and a maximum radius value of a lung nodule candidate from lung CT image data;
An image extraction module 202, configured to extract original 3D image data of the candidate lung nodule from the lung CT image data according to the coordinate position and the maximum radius value, and perform interpolation processing on the original 3D image data to obtain final candidate lung nodule 3D image data;
The data processing module 203 is configured to obtain sample data of three planes corresponding to the 3D image data of the lung nodule candidate, and perform scaling processing on the sample data of the three planes to form a training set;
And the sample training module 204 is configured to train the convolutional neural network based on the training set, and perform false positive screening on the candidate lung nodule through a convolutional neural network model obtained by training.
The system is configured to execute the above-mentioned method for screening false positive of lung nodule based on convolutional neural network, and for the specific implementation of each module in the system, please refer to the flowchart of the method shown in fig. 1 and specific implementation, which will not be repeated here.
In the embodiment of the invention, considering the defect that the deep learning network cannot distinguish the characteristics of the false positive lung nodules in the prior art, a novel convolutional neural network is provided for carrying out rapid and efficient characteristic extraction and classification screening on the candidate lung nodules, so that the accuracy of computer-aided automatic detection of the lung nodules can be improved; in addition, the embodiment of the invention can ensure the integrity and sufficiency of the input sample in the convolutional neural network by extracting the 3D image of the candidate lung nodule from the original lung CT image and increasing the sample data of the 3D image by utilizing the Lagrange interpolation method.
Those of ordinary skill in the art will appreciate that all or part of the steps in the various methods of the above embodiments may be implemented by a program to instruct related hardware, the program may be stored in a computer readable storage medium, and the storage medium may include: read Only Memory (ROM), random access Memory (RAM, random Access Memory), magnetic or optical disk, and the like.
The above describes in detail a method and a system for screening false positive of lung nodule based on convolutional neural network, and specific examples are adopted herein to illustrate the principle and implementation of the present invention, and the above description of the examples is only used to help understand the method and core idea of the present invention; meanwhile, as those skilled in the art will have variations in the specific embodiments and application scope in accordance with the ideas of the present invention, the present description should not be construed as limiting the present invention in view of the above.
Claims (6)
1. A convolutional neural network-based lung nodule false positive screening method, the method comprising:
Acquiring coordinate positions and maximum radius values of the lung nodule candidates from lung CT image data;
Extracting original 3D image data of the lung nodule candidate from the lung CT image data according to the coordinate position and the maximum radius value, and carrying out interpolation processing on the original 3D image data to obtain final lung nodule candidate 3D image data;
Obtaining sample data of three planes corresponding to the 3D image data of the lung nodule candidate, and scaling the sample data of the three planes to form a training set;
training a convolutional neural network based on the training set, and performing false positive screening on the candidate lung nodules through a convolutional neural network model obtained through training;
The extracting raw 3D image data of the candidate lung nodule from the lung CT image data according to the coordinate locations and the maximum radius values comprises:
Setting a treatment region size of the lung nodule candidate based on the maximum radius value, the treatment region being centered on the coordinate location;
When the maximum radius value is R, setting the length and the width of a treatment area of the lung nodule candidate to be 1.5Rx2;
Calculating the number of slices contained in the original 3D image data of the lung nodule candidate to 2n+1 based on a given slice interval; wherein the calculation formula of the slice number is as follows:
；
In the method, in the process of the invention, For the number of slices,/>For the height of the original 3D image data,/>Is the slice spacing; providing the original 3D image of the candidate lung nodule as a cube, the height of the original 3D image being 1.5R x 2, and the slice spacing being derived from the lung CT image data;
Taking the processing area as an intermediate layer, acquiring continuous N slices right above the processing area and continuous N slices right below the processing area, and combining the continuous N slices into the original 3D image data according to the original slice sequence; wherein the size of 2N slices is equal to the size of the treatment area of the lung nodule candidate, and the original sequential combination of the slices is obtained through the lung CT image data;
The network structure of the convolutional neural network comprises four convolutional pooling layers, a characteristic synthesis layer, a full connection layer and a Softmax function layer; the training of the convolutional neural network based on the training set and the false positive screening of the candidate lung nodule through the convolutional neural network model obtained by training comprise the following steps:
Sequentially inputting sample data of each of three planes corresponding to the candidate lung nodule 3D image data to the four convolution pooling layers, and obtaining a first feature map corresponding to each plane;
performing feature fusion on the first feature map corresponding to each plane through the feature composition layer to obtain a second feature map of the candidate lung nodule 3D image data;
And classifying the result of the second feature map based on the full connection layer, and outputting the false positive probability of the candidate lung nodule according to the classified result based on the Softmax function layer.
2. The lung nodule false positive screening method of claim 1, wherein the interpolating the raw 3D image data comprises:
And carrying out interpolation operation on 2N slice pitches existing on the Z axis of the original 3D image data by using a Lagrange interpolation method with the XY pixel pitch of the original 3D image data as a reference.
3. The method of claim 1, wherein the obtaining sample data for three planes corresponding to the candidate lung nodule 3D image data comprises:
Obtaining maximum projection data of each of three planes corresponding to the lung nodule candidate 3D image data;
And respectively carrying out normalization processing on the maximum projection data of each of the three planes to obtain sample data of each plane.
4. A lung nodule false positive screening method according to claim 3, wherein said normalizing the maximum projection data for each of said three planes respectively comprises:
Based on the human tissue density range, screening and replacing maximum projection data of each of the three planes;
And carrying out normalization processing on the maximum projection data of each plane after screening and replacing.
5. A method of screening out false positives for lung nodules as claimed in claim 3 wherein said scaling the sample data for the three planes comprises: the sample data for each of the three planes is set to be equal in size.
6. A convolutional neural network-based lung nodule false positive screening system, the system comprising:
The parameter acquisition module is used for acquiring the coordinate position and the maximum radius value of the lung nodule candidate from the lung CT image data;
the image extraction module is used for extracting original 3D image data of the lung nodule candidates from the lung CT image data according to the coordinate positions and the maximum radius values, and carrying out interpolation processing on the original 3D image data to obtain final lung nodule candidate 3D image data;
The data processing module is used for acquiring sample data of three planes corresponding to the 3D image data of the lung nodule candidate, and scaling the sample data of the three planes to form a training set;
the sample training module is used for training the convolutional neural network based on the training set and carrying out false positive screening on the candidate lung nodules through a convolutional neural network model obtained through training;
The extracting raw 3D image data of the candidate lung nodule from the lung CT image data according to the coordinate locations and the maximum radius values comprises:
Setting a treatment region size of the lung nodule candidate based on the maximum radius value, the treatment region being centered on the coordinate location;
When the maximum radius value is R, setting the length and the width of a treatment area of the lung nodule candidate to be 1.5Rx2;
Calculating the number of slices contained in the original 3D image data of the lung nodule candidate to 2n+1 based on a given slice interval; wherein the calculation formula of the slice number is as follows:
；
In the method, in the process of the invention, For the number of slices,/>For the height of the original 3D image data,/>Is the slice spacing; providing the original 3D image of the candidate lung nodule as a cube, the height of the original 3D image being 1.5R x 2, and the slice spacing being derived from the lung CT image data;
Taking the processing area as an intermediate layer, acquiring continuous N slices right above the processing area and continuous N slices right below the processing area, and combining the continuous N slices into the original 3D image data according to the original slice sequence; wherein the size of 2N slices is equal to the size of the treatment area of the lung nodule candidate, and the original sequential combination of the slices is obtained through the lung CT image data;
The network structure of the convolutional neural network comprises four convolutional pooling layers, a characteristic synthesis layer, a full connection layer and a Softmax function layer; the training of the convolutional neural network based on the training set and the false positive screening of the candidate lung nodule through the convolutional neural network model obtained by training comprise the following steps:
Sequentially inputting sample data of each of three planes corresponding to the candidate lung nodule 3D image data to the four convolution pooling layers, and obtaining a first feature map corresponding to each plane;
performing feature fusion on the first feature map corresponding to each plane through the feature composition layer to obtain a second feature map of the candidate lung nodule 3D image data;
And classifying the result of the second feature map based on the full connection layer, and outputting the false positive probability of the candidate lung nodule according to the classified result based on the Softmax function layer.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010502186.6A CN111784638B (en) | 2020-06-04 | 2020-06-04 | Convolutional neural network-based lung nodule false positive screening method and system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010502186.6A CN111784638B (en) | 2020-06-04 | 2020-06-04 | Convolutional neural network-based lung nodule false positive screening method and system |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN111784638A CN111784638A (en) | 2020-10-16 |
| CN111784638B true CN111784638B (en) | 2024-06-25 |
Family
ID=72754624
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010502186.6A Active CN111784638B (en) | 2020-06-04 | 2020-06-04 | Convolutional neural network-based lung nodule false positive screening method and system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111784638B (en) |
Families Citing this family (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112541909B (en) * | 2020-12-22 | 2022-05-17 | 南开大学 | Lung nodule detection method and system based on three-dimensional neural network of slice perception |
| CN113658162A (en) * | 2021-08-24 | 2021-11-16 | 中山仰视科技有限公司 | Refined lung nodule detection method, system, equipment and medium |
| CN116934757B (en) * | 2023-09-18 | 2023-11-21 | 电子科技大学 | Method, equipment and storage medium for lung nodule false positive pruning |
Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107154043A (en) * | 2017-06-05 | 2017-09-12 | 杭州健培科技有限公司 | A kind of Lung neoplasm false positive sample suppressing method based on 3DCNN |
Family Cites Families (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| KR101587719B1 (en) * | 2014-06-10 | 2016-01-22 | 원광대학교산학협력단 | Apparatus for analysing medical image and method for classifying pulmonary vessel and pulmonary nodule |
| CN107464234B (en) * | 2017-07-24 | 2020-09-01 | 上海交通大学 | Lung nodule image deep learning identification system based on RGB channel superposition method and method thereof |
| CN108230323B (en) * | 2018-01-30 | 2021-03-23 | 浙江大学 | Pulmonary nodule false positive screening method based on convolutional neural network |
| CN108765369B (en) * | 2018-04-20 | 2023-05-02 | 平安科技（深圳）有限公司 | Method, apparatus, computer device and storage medium for detecting lung nodule |
| CN110807788B (en) * | 2019-10-21 | 2023-07-21 | 腾讯科技（深圳）有限公司 | Medical image processing method, medical image processing device, electronic equipment and computer storage medium |
| CN110675398B (en) * | 2019-10-22 | 2022-05-17 | 深圳瀚维智能医疗科技有限公司 | Mammary gland ultrasonic screening method and device and computer equipment |
-
2020
- 2020-06-04 CN CN202010502186.6A patent/CN111784638B/en active Active
Patent Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107154043A (en) * | 2017-06-05 | 2017-09-12 | 杭州健培科技有限公司 | A kind of Lung neoplasm false positive sample suppressing method based on 3DCNN |
Also Published As
| Publication number | Publication date |
|---|---|
| CN111784638A (en) | 2020-10-16 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111476292B (en) | Small sample element learning training method for medical image classification processing artificial intelligence | |
| CN109102502B (en) | Pulmonary nodule detection method based on three-dimensional convolutional neural network | |
| CN109584252B (en) | Lung lobe segment segmentation method and device of CT image based on deep learning | |
| Wang et al. | A multi-view deep convolutional neural networks for lung nodule segmentation | |
| CN111784638B (en) | Convolutional neural network-based lung nodule false positive screening method and system | |
| CN110889852B (en) | Liver segmentation method based on residual error-attention deep neural network | |
| CN111445478B (en) | Automatic intracranial aneurysm region detection system and detection method for CTA image | |
| Wu et al. | Stereo matching with fusing adaptive support weights | |
| CN110503630B (en) | Cerebral hemorrhage classifying, positioning and predicting method based on three-dimensional deep learning model | |
| CN110008962B (en) | Weak supervision semantic segmentation method based on attention mechanism | |
| CN110689547A (en) | Pulmonary nodule segmentation method based on three-dimensional CT image | |
| JP2023550844A (en) | Liver CT automatic segmentation method based on deep shape learning | |
| CN112699869A (en) | Rib fracture auxiliary detection method based on deep learning and image identification method | |
| CN109035300B (en) | Target tracking method based on depth feature and average peak correlation energy | |
| CN112329871B (en) | Pulmonary nodule detection method based on self-correction convolution and channel attention mechanism | |
| CN112734755A (en) | Lung lobe segmentation method based on 3D full convolution neural network and multitask learning | |
| CN110853049A (en) | Abdominal ultrasonic image segmentation method | |
| CN113744183A (en) | Pulmonary nodule detection method and system | |
| CN111860587A (en) | Method for detecting small target of picture | |
| CN112820399A (en) | Method and device for automatically diagnosing benign and malignant thyroid nodules | |
| CN111667491A (en) | Breast mass image generation method with marginal landmark annotation information based on depth countermeasure network | |
| CN115661509A (en) | Surgical instrument identification and classification method based on three-dimensional point cloud ICP (inductively coupled plasma) registration algorithm | |
| CN116228792A (en) | Medical image segmentation method, system and electronic device | |
| CN112381762A (en) | CT rib fracture auxiliary diagnosis system based on deep learning algorithm | |
| CN111259733A (en) | Point cloud image-based ship identification method and device |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant |