CN107507197B - Lung parenchyma extraction method based on clustering algorithm and convolutional neural network - Google Patents
Lung parenchyma extraction method based on clustering algorithm and convolutional neural network Download PDFInfo
- Publication number
- CN107507197B CN107507197B CN201710712015.4A CN201710712015A CN107507197B CN 107507197 B CN107507197 B CN 107507197B CN 201710712015 A CN201710712015 A CN 201710712015A CN 107507197 B CN107507197 B CN 107507197B
- Authority
- CN
- China
- Prior art keywords
- image
- lung
- layer
- neural network
- convolutional neural
- 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
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/11—Region-based segmentation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
- G06F18/232—Non-hierarchical techniques
- G06F18/2321—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions
- G06F18/23213—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions with fixed number of clusters, e.g. K-means clustering
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/26—Segmentation of patterns in the image field; Cutting or merging of image elements to establish the pattern region, e.g. clustering-based techniques; Detection of occlusion
- G06V10/267—Segmentation of patterns in the image field; Cutting or merging of image elements to establish the pattern region, e.g. clustering-based techniques; Detection of occlusion by performing operations on regions, e.g. growing, shrinking or watersheds
-
- 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V2201/00—Indexing scheme relating to image or video recognition or understanding
- G06V2201/03—Recognition of patterns in medical or anatomical images
- G06V2201/031—Recognition of patterns in medical or anatomical images of internal organs
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Bioinformatics & Computational Biology (AREA)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Probability & Statistics with Applications (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
- Apparatus For Radiation Diagnosis (AREA)
Abstract
The invention provides a lung parenchyma extraction method based on a clustering algorithm and a convolutional neural network, which comprises the steps of preprocessing a lung CT image by adopting the clustering algorithm to obtain data sets of a lung parenchyma region and a non-lung parenchyma region of the CT image, dividing the known data set of the lung CT image into a training set and a verification set, and taking the unknown data set of the lung CT image as a test set; establishing a convolutional neural network model, and training the convolutional neural network model by adopting a training set and a verification set to obtain a trained convolutional neural network model; the test set is input into the trained convolutional neural network model to obtain a lung parenchymal region of the CT image, so that the function of extracting the lung parenchymal region of the lung CT image of an unknown patient is realized, the basis of automatically searching the lung cancer region in the next stage is built, and the extraction and classification of the lung cancer in the next stage are facilitated.
Description
Technical Field
The invention belongs to the technical field of medical image processing, and particularly relates to a lung parenchyma extraction method based on a clustering algorithm and a convolutional neural network.
Background
In histology, lung tissue is divided into two parts, lung parenchyma, i.e., branches of each level of bronchus in lung and a large number of alveolar structures at the end, and lung interstitium is connective tissue, blood vessels, lymphatic vessels, nerves, etc. In order to evaluate and study lung volume and lung cancer, doctors often need to know the lung parenchyma condition for the first time, and accurate segmentation of the lung parenchyma plays an important role in further studying tissues and lesions of functions of organs in the lung.
During the study and application of images, people often are interested in only certain parts of the images. These parts, often referred to as targets, generally correspond to specific areas of the image having unique properties. In order to identify and analyze the target, these regions need to be separated and extracted, and further utilized on the basis of the separated regions. Image segmentation is a technique and a process for dividing an image into regions with characteristics and extracting an interested target. Classical image segmentation methods such as thresholding, region growing, edge detection, clustering, and neural network techniques.
Threshold segmentation is the oldest segmentation technique and is also the simplest and most practical. In many cases, the gray scale value of the target area in the image is different from that of the background area or different areas, and the target area and the background area or different areas in the image can be divided according to the uniformity of the gray scale, which is generally greatly influenced by noise.
The edge detection algorithm is more suitable for the segmentation of simple images with more remarkable edge gray value transition and less noise. For images with complex edges and strong noise, the contradiction between noise immunity and detection accuracy is faced. If the detection accuracy is improved, false edges generated by noise can result in unreasonable contours: if the noise immunity is improved, missing detection of the contour and positional deviation may occur.
Clustering analysis is an important branch of unsupervised pattern recognition in pattern recognition. The data sets are classified into different categories according to their internal structure, so that the features of the samples within the same category are as similar as possible, while the differences of the sample points belonging to the different categories are as large as possible.
The basic idea of the neural network-based segmentation method is to obtain a linear decision function by training a multilayer perceptron, and then classify pixels by using the decision function to achieve the purpose of segmentation. This approach requires a large amount of training data. The neural network has huge connection, is easy to introduce spatial information, and can better solve the problems of noise and non-uniformity in images. The choice of which network architecture is the main problem to be solved by this approach.
In summary, the segmentation methods proposed at present are mostly specific to problems, and have long respective lengths, and do not have a general standard. Aiming at lung parenchyma segmentation, the invention provides a convolutional neural network trained by clustering generated data, which can automatically segment lung parenchyma areas without threshold values and is efficient and accurate.
Disclosure of Invention
Aiming at the defects of the prior art, the invention provides a lung parenchyma extraction method based on a clustering algorithm and a convolutional neural network.
A lung parenchyma extraction method based on a clustering algorithm and a convolutional neural network comprises the following steps:
step 1: preprocessing the lung CT image by adopting a clustering algorithm to obtain data sets of a lung parenchymal region and a non-lung parenchymal region of the CT image, dividing the known data set of the lung CT image into a training set and a verification set, and taking the unknown data set of the lung CT image as a test set;
step 1.1: standardizing the lung CT image, and segmenting the standardized lung CT image into small image blocks with the size of A;
step 1.2: clustering the average value and the minimum value of the CT value of the small image blocks with the size of A by adopting a Kmeans algorithm respectively, and clustering the small image blocks into a low-density tissue and a high-density tissue;
step 1.3: performing cross inspection on the clustering result of the average value of the CT values of the small image blocks and the clustering result of the minimum value of the CT values of the small image blocks to remove a background region of the CT image;
step 1.4: extracting the intersection of the lung parenchymal region in the clustering result of the average value of the CT values of the image small blocks and the lung parenchymal region of the clustering result of the minimum value of the CT values of the image small blocks;
step 1.5: and (3) performing maximum communication body operation on the intersection of the lung parenchymal regions obtained in the step (1.4) to obtain data sets of the lung parenchymal regions and non-lung parenchymal regions of the CT images, dividing the data set of the known lung CT image into a training set and a verification set, and taking the data set of the lung CT image to be divided as a test set.
Step 2: establishing a convolutional neural network model, and training the convolutional neural network model by adopting a training set and a verification set to obtain a trained convolutional neural network model;
step 2.1: expanding the image small blocks with the size of A in the training set and the verification set into image small blocks with the size of B;
step 2.2: establishing a convolutional neural network model, inputting the expanded small image blocks into the convolutional neural network model, and training the weight and the deviation of each layer of the convolutional neural network model;
step 2.3: and inputting the verification set into the convolutional neural network model for classification, and determining optimized training parameters through the running time loss and the classification accuracy of the verification set to obtain the trained convolutional neural network model.
And step 3: and inputting the test set into the trained convolutional neural network model to obtain the lung parenchymal region of the CT image.
The convolutional neural network model structure is as follows: the first layer is an image small block input layer, the second layer is a convolution layer, the third layer is a maximum value pooling layer, and the fourth layer is a full-connection layer;
the convolutional layer comprises a convolutional layer ReLU layer and a Norm layer;
the full connection layer comprises a full connection layer ReLU layer, a random forgetting layer, a full connection layer classifier and a Softmax function layer.
The division standard with the size A is as follows: the divided image small blocks with the size A all contain lung tissues in the CT images, and the automatic segmentation time of the image small blocks of each CT image is within 50 MS.
The training set and the verification set each have 50% of image patches of size a of the lung parenchyma and image patches of size a of the non-lung parenchyma.
The cross test is carried out on the clustering result of the average value of the CT values of the image small blocks and the clustering result of the minimum value of the CT values of the image small blocks, and the specific process of removing the background area of the CT image is as follows:
and checking whether the image small blocks of the high-density tissues exist in four radial directions of each image small block of the low-density tissues, if so, the image small blocks of the low-density tissues are the suspected lung parenchyma area, otherwise, the image small blocks of the low-density tissues are the background area.
The specific process of expanding the image patches with the size of a in the training set and the verification set into the image patches with the size of B is as follows:
the image patch with the size of A is used as a center and is expanded to an image patch with the size of B at a position in the original CT image.
The training parameters comprise: learning rate, convolution kernel size, convolution kernel number, Norm layer normalized channel number, fully connected first layer output number, Dropout layer forgetting rate, pooling layer type, Batch number, Epochs value.
The invention has the beneficial effects that:
the invention provides a lung parenchyma extraction method based on a clustering algorithm and a convolutional neural network, which automatically divides a lung parenchyma region and a non-lung parenchyma region of a CT image of a patient through an unsupervised learning algorithm, namely the clustering algorithm, so as to achieve the functions of automatically generating a training set and a verification set required by training the convolutional neural network; an optimal convolutional neural network model is designed aiming at the lung parenchymal region classification through a supervised learning algorithm, namely a convolutional neural network, so that the lung parenchymal region classification effect with high accuracy is achieved; the lung parenchymal region extraction function of the lung CT image of an unknown patient is realized, the basis of the next stage of automatically searching the lung cancer region is built, and the extraction and classification of the lung cancer of the next stage are facilitated.
Drawings
FIG. 1 is a flow chart of a method for extracting lung parenchyma based on a clustering algorithm and a convolutional neural network according to an embodiment of the present invention;
FIG. 2 is a schematic illustration of a segmentation process for a normalized lung CT image according to an embodiment of the present invention;
wherein, (a) is a lung CT image with a segmentation size of 64 x 64;
(b) a lung CT image with a segmentation size of 32 x 32 is obtained;
(c) segmenting the lung into a CT image with the size of 16 x 16;
(d) segmenting the lung into CT images with the size of 8 x 8;
(e) segmenting the lung into CT images with the size of 4 x 4;
(f) segmenting the lung into CT images with the size of 2 x 2;
FIG. 3 is a schematic diagram of a process for obtaining a lung parenchymal region and a non-lung parenchymal region of a CT image according to an embodiment of the present invention;
the method comprises the following steps that (a) a Kmeans algorithm is adopted to cluster CT value average values of small image blocks with the size of A respectively;
(b) low-density cross-like inspection results which are average values of CT values;
(c) clustering the minimum value of the CT value of the small image blocks with the size of A by adopting a Kmeans algorithm;
(d) a low-density cross-like inspection result which is the minimum value of the CT value;
(e) the schematic diagram is the intersection of the lung parenchymal region in the clustering result of the average value of the CT values of the extracted image small blocks and the lung parenchymal region of the clustering result of the minimum value of the CT values of the image small blocks;
(f) a schematic diagram of a lung parenchymal region and a non-lung parenchymal region of a CT image;
FIG. 4 is a diagram illustrating expanding an image tile of size A to an image tile of size B in accordance with an embodiment of the present invention;
wherein (a) is an image patch of size a;
(b) is an image patch of size B;
FIG. 5 is a schematic diagram of a convolutional neural network model in accordance with an embodiment of the present invention;
wherein, (a) is an image tile input layer;
(b) is a convolutional layer;
(c) a maximum pooling layer;
(d) is a full connection layer;
FIG. 6 is a schematic diagram of three-dimensional modeling of segmented lung parenchymal regions in accordance with an embodiment of the present invention;
wherein, (a) is the classified three-dimensional reconstruction result of the patient with chronic obstructive pulmonary disease;
(b) classifying three-dimensional reconstruction results of lung cancer patients under common CT scanning image data;
(c) the method is a classified three-dimensional reconstruction result of the lung cancer patient under a whole-body tumor imaging scanning mode of a PET/CT device.
Detailed Description
The following detailed description of embodiments of the invention refers to the accompanying drawings.
A method for extracting lung parenchyma based on a clustering algorithm and a convolutional neural network, as shown in fig. 1, includes the following steps:
step 1: preprocessing the lung CT image by adopting a clustering algorithm to obtain data sets of a lung parenchyma region and a non-lung parenchyma region of the CT image, dividing the known data set of the lung CT image into a training set and a verification set, and taking the unknown data set of the lung CT image as a test set.
Step 1.1: and standardizing the lung CT image, and segmenting the standardized lung CT image into small image blocks with the size of A.
In the present embodiment, the division criteria of the size a are: the divided image small blocks with the size A all contain lung tissues in the CT images, and the automatic segmentation time of the image small blocks of each CT image is within 50 MS.
Fig. 2 is a schematic diagram of a segmentation process of the normalized lung CT image, and the time consumption and the segmentation characteristics of the normalized lung CT image are shown in table 1.
TABLE 1 time consuming and segmentation characteristics for segmenting normalized CT images of the lungs
In this embodiment, in combination with the evaluation of the characteristics of the consumption time and the segmentation in table 1, it can be seen that as the size of the patch is reduced, the time consumption increases exponentially, and the difference in the characteristics between each patch is increased, so that the present invention finally selects 8 × 8 as the optimal size of the image patch a.
Step 1.2: and respectively clustering the average value and the minimum value of the CT value of the small image blocks with the size of A by adopting a Kmeans algorithm, and clustering the small image blocks into low-density tissues and high-density tissues.
In the present embodiment, the result of clustering the average CT values of the small image blocks of size a using the Kmeans algorithm is shown in fig. 3 (a), and the result of clustering the minimum CT values of the small image blocks of size a using the Kmeans algorithm is shown in fig. 3 (c).
Step 1.3: and performing cross inspection on the clustering result of the average value of the CT values of the small image blocks and the clustering result of the minimum value of the CT values of the small image blocks to remove the background area of the CT image.
In this embodiment, the cross test is performed on the clustering result of the average value of the CT values of the image patches and the clustering result of the minimum value of the CT values of the image patches, and a specific process of removing the background region of the CT image is as follows:
and checking whether the image small blocks of the high-density tissues exist in four radial directions of each image small block of the low-density tissues, if so, the image small blocks of the low-density tissues are the suspected lung parenchyma area, otherwise, the image small blocks of the low-density tissues are the background area.
The low-density cross-like test result of the average CT value is shown in fig. 3 (b), and the low-density cross-like test result of the minimum CT value is shown in fig. 3 (d).
Step 1.4: the intersection of the lung parenchymal region in the clustering result of the average value of the CT values of the image patches and the lung parenchymal region of the clustering result of the minimum value of the CT values of the image patches is extracted, as shown in (e) in fig. 3.
Step 1.5: performing maximum communication volume operation on the intersection of the lung parenchymal regions obtained in step 1.4 to obtain data sets of the lung parenchymal regions and the non-lung parenchymal regions of the CT image, as shown in (f) in fig. 3, dividing the data set of the known lung CT image into a training set and a verification set, and taking the data set of the lung CT image to be divided as a test set.
In the present embodiment, the number of image patches of a size a of the lung parenchyma and the number of image patches of a size a of the non-lung parenchyma in the training set and the verification set each account for 50%. The ratio of the data set of known lung CT images divided into training and validation sets was set to 7: 1.
It is found from fig. 3 that the lung parenchymal region of the result obtained by clustering the minimum CT values of the image patches with the size a respectively by using the Kmeans algorithm is larger than the lung parenchymal region of the result obtained by clustering the average CT values of the image patches with the size a respectively by using the Kmeans algorithm, that is, the minimum CT value clustering method exceeds the edge of the real lung parenchyma, and the effect of the minimum CT value clustering method is better than that of the average CT value clustering method in terms of removing the background noise under the human body. Therefore, the invention can keep the lung parenchyma region which is accurate by the CT value average value clustering method by taking the intersection region of the two, and can also control the background noise by the CT value minimum value clustering method. After the steps, the background noise below a few parts of the body is not removed, so that the background can be completely removed by adopting the operation of the communicating body, and a more accurate lung parenchymal area is obtained.
Step 2: and establishing a convolutional neural network model, and training the convolutional neural network model by adopting a training set and a verification set to obtain the trained convolutional neural network model.
Step 2.1: and expanding the image small blocks with the size of A in the training set and the verification set into image small blocks with the size of B.
In the present embodiment, as shown in fig. 4, the image patch with the size a is expanded to the image patch with the size B at the position in the original CT image with the image patch with the size a as the center, so that it is possible to avoid the convolution effect on the small patch with the too small size due to the large size of the convolution kernel in the convolution process from being insignificant.
In the present embodiment, the dimension B is 32 × 32.
Step 2.2: and establishing a convolutional neural network model, inputting the expanded small image blocks into the convolutional neural network model, and training the weight and the deviation of each layer of the convolutional neural network model.
In the present embodiment, a simplified AlexNet structure adopted by the convolutional neural network model structure is shown in fig. 5, and includes: the first layer is an image small block input layer, the second layer is a convolution layer, the third layer is a maximum value pooling layer, and the fourth layer is a full-connection layer.
The convolutional layer comprises a convolutional layer ReLU layer and a Norm layer;
the full connection layer comprises a full connection layer ReLU layer, a random forgetting layer, a full connection layer classifier and a Softmax function layer.
In the embodiment, the maximum consideration is that the classification category number only includes 2 categories, but not one thousand categories of AlexNet, so that the convolutional neural network model only reserves one convolutional layer with 6 convolutional kernels in 5 convolutional layers of AlexNet, and the addition of the ReLU active layer and the Norm normalization layer can accelerate the random gradient descent and prevent overfitting; the pooling layer selects maximum pooling instead of average pooling, so that the network complexity can be reduced; the fully-connected layer comprises 120 distribution characteristics, and a ReLU activation layer and a Dropout forgetting layer are added next to accelerate the convergence of random gradient and prevent overfitting; the second layer of the full connection layer is a 2-classification classifier which is used for classifying the lung parenchymal region and the non-lung parenchymal region; finally, the Softmax function is used as an output layer, and the probability distribution of the output of the Softmax function approximately represents the output distribution.
Step 2.3: and inputting the verification set into the convolutional neural network model for classification, and determining optimized training parameters through the running time loss and the classification accuracy of the verification set to obtain the trained convolutional neural network model.
In this embodiment, the training parameters include: learning rate, convolution kernel size, convolution kernel number, Norm layer normalized channel number, fully connected first layer output number, Dropout layer forgetting rate, pooling layer type, Batch number, Epochs value.
In this embodiment, the convolutional neural network model optimization parameters and the experimental results are shown in table 2:
TABLE 2 convolutional neural network model optimization parameters and Experimental results
As can be seen from table 2, the first row is taken as a reference standard, and it can be seen that the size of the convolution kernel in the convolutional layer is increased from 5 × 5 to 10 × 10, the verification accuracy is not improved, but the corresponding time consumption is improved by 27%; when the value of Epochs is set to 50 to 80, the verification accuracy is not improved and the time consumption is sharply increased by three times; when the learning rate is set to be 0.0001, the time consumption is doubled, but the accuracy rate reaches 99.17%, and the time cost is sacrificed within a certain range to improve the accuracy rate; we are not close to 100% accuracy when the learning rate is set to 0.00001, but 85% of the time cost. In summary, the optimal parameters determined by the present invention are shown in the last row of table 2.
The optimized training parameters are: the learning rate is 0.0001, the convolution kernel size is 5 x 5, the number of convolution kernels is 6, the number of normalized channels of the Norm layer is 3, the number of outputs of the fully connected first layer is 120, the Dropout layer forgetting rate is 0.5 (50%), the Max type of pooling layer, the number of batchs is 128, and the Epochs value is 50.
And step 3: and inputting the test set into the trained convolutional neural network model to obtain the lung parenchymal region of the CT image.
In this embodiment, the test set is input into the trained convolutional neural network model, and the segmented lung parenchymal region is three-dimensionally modeled to test the accuracy and the universality of the convolutional neural network model.
Inputting the test set into the trained convolutional neural network model, performing three-dimensional modeling on the segmented lung parenchymal region, performing golden standard demarcation on the lung parenchymal region according to a lung CT image to be segmented by a doctor, overlapping the golden standard and the image of the classification result according to the classification result of the test set in the neural network, and calculating the accuracy (Daiss similarity coefficient), the sensitivity, the specificity and the like. As shown in fig. 6, the classified three-dimensional reconstruction result of the patient with chronic obstructive pulmonary disease is shown in fig. 6 (a), the classified three-dimensional reconstruction result of the patient with lung cancer under the normal CT scan image data is shown in fig. 6 (b), and the classified three-dimensional reconstruction result of the patient with lung cancer under the whole-body tumor imaging scan mode of the PET/CT apparatus is shown in fig. 6 (c).
Claims (5)
1. A lung parenchyma extraction method based on a clustering algorithm and a convolutional neural network is characterized by comprising the following steps:
step 1: preprocessing the lung CT image by adopting a clustering algorithm to obtain a data set of a lung parenchymal region and a non-lung parenchymal region of the lung CT image, dividing the known data set of the lung CT image into a training set and a verification set, and taking the unknown data set of the lung CT image as a test set, wherein the method comprises the following steps:
step 1.1: standardizing the lung CT image, and segmenting the standardized lung CT image into small image blocks with the size of A;
step 1.2: clustering the average value and the minimum value of the CT value of the small image blocks with the size of A by adopting a Kmeans algorithm respectively, and clustering the small image blocks into a low-density tissue and a high-density tissue;
step 1.3: performing cross inspection on the clustering result of the average value of the CT values of the small image blocks and the clustering result of the minimum value of the CT values of the small image blocks to remove a background region of the CT image;
step 1.4: extracting the intersection of the lung parenchymal region in the clustering result of the average value of the CT values of the image small blocks and the lung parenchymal region of the clustering result of the minimum value of the CT values of the image small blocks;
step 1.5: performing maximum connector operation on the intersection of the lung parenchymal regions obtained in the step 1.4 to obtain data sets of the lung parenchymal regions and non-lung parenchymal regions of the CT images, dividing the data set of the known lung CT images into a training set and a verification set, and taking the data set of the unknown lung CT images as a test set;
the division standard of the size A is as follows: the divided image small blocks with the size of A all contain lung tissues in the CT images, and the automatic segmentation time of the image small blocks of each CT image is within 50 MS;
the number of image small blocks with the size of A of the lung parenchyma and the number of image small blocks with the size of A of the non-lung parenchyma in the training set and the verification set respectively account for 50 percent;
the cross test is carried out on the clustering result of the average value of the CT values of the image small blocks and the clustering result of the minimum value of the CT values of the image small blocks, and the specific process of removing the background area of the CT image is as follows:
checking whether the four radial directions of the image small blocks of each low-density tissue have the image small blocks of the high-density tissue, if so, the image small blocks of the low-density tissue are a suspected lung parenchyma area, otherwise, the image small blocks of the low-density tissue are a background area;
step 2: establishing a convolutional neural network model, and training the convolutional neural network model by adopting a training set and a verification set to obtain a trained convolutional neural network model;
and step 3: and inputting the test set into the trained convolutional neural network model to obtain the lung parenchymal region of the CT image.
2. The method for extracting lung parenchyma based on a clustering algorithm and a convolutional neural network as claimed in claim 1, wherein the step 2 comprises the steps of:
step 2.1: expanding the image small blocks with the size of A in the training set and the verification set into image small blocks with the size of B;
step 2.2: establishing a convolutional neural network model, inputting the expanded small image blocks into the convolutional neural network model, and training the weight and the deviation of each layer of the convolutional neural network model;
step 2.3: and inputting the verification set into the convolutional neural network model for classification, and determining optimized training parameters through the running time loss and the classification accuracy of the verification set to obtain the trained convolutional neural network model.
3. The method for extracting lung parenchyma based on a clustering algorithm and a convolutional neural network as claimed in claim 1, wherein the convolutional neural network model structure is: the first layer is an image small block input layer, the second layer is a convolution layer, the third layer is a maximum value pooling layer, and the fourth layer is a full-connection layer;
the convolutional layer comprises a convolutional layer ReLU layer and a Norm layer;
the full connection layer comprises a full connection layer ReLU layer, a random forgetting layer, a full connection layer classifier and a Softmax function layer.
4. The method for extracting lung parenchyma based on a clustering algorithm and a convolutional neural network as claimed in claim 3, wherein the specific process of expanding the image patches with size A in the training set and the verification set into the image patches with size B is as follows:
the image patch with the size of A is used as a center and is expanded to an image patch with the size of B at a position in the original CT image.
5. The method of claim 3, wherein the training parameters comprise: learning rate, convolution kernel size, convolution kernel number, Norm layer normalized channel number, fully connected first layer output number, Dropout layer forgetting rate, pooling layer type, Batch number, Epochs value.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710712015.4A CN107507197B (en) | 2017-08-18 | 2017-08-18 | Lung parenchyma extraction method based on clustering algorithm and convolutional neural network |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710712015.4A CN107507197B (en) | 2017-08-18 | 2017-08-18 | Lung parenchyma extraction method based on clustering algorithm and convolutional neural network |
Publications (2)
Publication Number | Publication Date |
---|---|
CN107507197A CN107507197A (en) | 2017-12-22 |
CN107507197B true CN107507197B (en) | 2020-11-24 |
Family
ID=60691358
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201710712015.4A Active CN107507197B (en) | 2017-08-18 | 2017-08-18 | Lung parenchyma extraction method based on clustering algorithm and convolutional neural network |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN107507197B (en) |
Families Citing this family (11)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108009594B (en) * | 2017-12-25 | 2018-11-13 | 北京航空航天大学 | A kind of image-recognizing method based on change grouping convolution |
KR102172213B1 (en) * | 2018-06-04 | 2020-10-30 | 주식회사 딥바이오 | System and method for two phase diagnosis using neural network |
CN109065165B (en) * | 2018-07-25 | 2021-08-17 | 东北大学 | Chronic obstructive pulmonary disease prediction method based on reconstructed airway tree image |
CN109087296B (en) * | 2018-08-07 | 2021-08-10 | 东北大学 | Method for extracting human body region in CT image |
CN109300122B (en) | 2018-09-14 | 2023-04-07 | 东软医疗***股份有限公司 | Image processing and threshold determining method, device and equipment |
CN109508650A (en) * | 2018-10-23 | 2019-03-22 | 浙江农林大学 | A kind of wood recognition method based on transfer learning |
CN109636802A (en) * | 2019-01-18 | 2019-04-16 | 天津工业大学 | Pulmonary parenchyma based on depth convolutional neural networks is through CT image partition method |
CN110796652B (en) * | 2019-10-30 | 2023-03-21 | 上海联影智能医疗科技有限公司 | Image processing method, computer device, and storage medium |
CN113706541B (en) * | 2020-05-20 | 2024-04-19 | 青岛海信医疗设备股份有限公司 | Image processing method and device |
CN111882560B (en) * | 2020-06-16 | 2024-05-31 | 北京工业大学 | Lung parenchyma CT image segmentation method based on weighted full convolution neural network |
CN113409306A (en) * | 2021-07-15 | 2021-09-17 | 推想医疗科技股份有限公司 | Detection device, training method, training device, equipment and medium |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106372390A (en) * | 2016-08-25 | 2017-02-01 | 姹ゅ钩 | Deep convolutional neural network-based lung cancer preventing self-service health cloud service system |
CN106651830A (en) * | 2016-09-28 | 2017-05-10 | 华南理工大学 | Image quality test method based on parallel convolutional neural network |
CN107016665A (en) * | 2017-02-16 | 2017-08-04 | 浙江大学 | A kind of CT pulmonary nodule detection methods based on depth convolutional neural networks |
-
2017
- 2017-08-18 CN CN201710712015.4A patent/CN107507197B/en active Active
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106372390A (en) * | 2016-08-25 | 2017-02-01 | 姹ゅ钩 | Deep convolutional neural network-based lung cancer preventing self-service health cloud service system |
CN106651830A (en) * | 2016-09-28 | 2017-05-10 | 华南理工大学 | Image quality test method based on parallel convolutional neural network |
CN107016665A (en) * | 2017-02-16 | 2017-08-04 | 浙江大学 | A kind of CT pulmonary nodule detection methods based on depth convolutional neural networks |
Non-Patent Citations (1)
Title |
---|
基于CT图像肺实质分割的方法;娜孜古丽•库尔曼别克等;《科技创新与应用》;20130528;第1-3章 * |
Also Published As
Publication number | Publication date |
---|---|
CN107507197A (en) | 2017-12-22 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN107507197B (en) | Lung parenchyma extraction method based on clustering algorithm and convolutional neural network | |
US20190026897A1 (en) | Brain tumor automatic segmentation method by means of fusion of full convolutional neural network and conditional random field | |
CN102509123B (en) | Brain function magnetic resonance image classification method based on complex network | |
CN102855491B (en) | A kind of central brain functional magnetic resonance image classification Network Based | |
CN109087296B (en) | Method for extracting human body region in CT image | |
CN109635846A (en) | A kind of multiclass medical image judgment method and system | |
CN103020653B (en) | Structure and function magnetic resonance image united classification method based on network analysis | |
Orobinskyi et al. | Novel approach to computer-aided detection of lung nodules of difficult location with use of multifactorial models and deep neural networks | |
Bhatele et al. | Glioma Segmentation and Classification System Based on Proposed Texture Features Extraction Method and Hybrid Ensemble Learning. | |
Paul et al. | RETRACTED ARTICLE: Computer aided diagnosis of brain tumor using novel classification techniques | |
CN111815640B (en) | Memristor-based RBF neural network medical image segmentation algorithm | |
CN110634118A (en) | Artificial intelligence-based mammary gland image recognition system and method | |
Takruri et al. | Multi-classifier decision fusion for enhancing melanoma recognition accuracy | |
CN109523484B (en) | Fractal feature-based finger vein network repair method | |
Syed et al. | Detection of tumor in MRI images using artificial neural networks | |
Afshar et al. | Lung tumor area recognition in CT images based on Gustafson-Kessel clustering | |
Najjar et al. | Histogram features extraction for edge detection approach | |
RU2668699C1 (en) | Intellectual method of diagnostics and detection of neoplasms in lungs | |
Fedorenko et al. | Semantic segmentation system of pigmented skin lesions based on convolutional neural networks | |
Peng et al. | Blood vessels segmentation by using cdnet | |
Santamaria-Pang et al. | Cell segmentation and classification via unsupervised shape ranking | |
Rahman et al. | Integrated approach of brain segmentation using neuro fuzzy k-means | |
Hinisha et al. | Lung Nodule Identification | |
CN115240014B (en) | Medical image classification system based on residual error neural network | |
CN118230074B (en) | Knowledge-aided migration learning breast cancer molecular subtype identification method and system |
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 | ||
GR01 | Patent grant |