WO2020000643A1 - Ct图像肺结节检测方法、设备及可读存储介质 - Google Patents

Ct图像肺结节检测方法、设备及可读存储介质 Download PDF

Info

Publication number
WO2020000643A1
WO2020000643A1 PCT/CN2018/104189 CN2018104189W WO2020000643A1 WO 2020000643 A1 WO2020000643 A1 WO 2020000643A1 CN 2018104189 W CN2018104189 W CN 2018104189W WO 2020000643 A1 WO2020000643 A1 WO 2020000643A1
Authority
WO
WIPO (PCT)
Prior art keywords
image
sample data
convolutional neural
dimensional convolutional
candidate nodule
Prior art date
Application number
PCT/CN2018/104189
Other languages
English (en)
French (fr)
Inventor
窦琪
刘权德
陈浩
Original Assignee
深圳视见医疗科技有限公司
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by 深圳视见医疗科技有限公司 filed Critical 深圳视见医疗科技有限公司
Publication of WO2020000643A1 publication Critical patent/WO2020000643A1/zh

Links

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
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B6/00—Apparatus for radiation diagnosis, e.g. combined with radiation therapy equipment
    • A61B6/52—Devices using data or image processing specially adapted for radiation diagnosis
    • A61B6/5211—Devices using data or image processing specially adapted for radiation diagnosis involving processing of medical diagnostic data
    • A61B6/5217—Devices using data or image processing specially adapted for radiation diagnosis involving processing of medical diagnostic data extracting a diagnostic or physiological parameter from medical diagnostic data
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B6/00—Apparatus for radiation diagnosis, e.g. combined with radiation therapy equipment
    • A61B6/02—Devices for diagnosis sequentially in different planes; Stereoscopic radiation diagnosis
    • A61B6/03—Computerised tomographs
    • A61B6/032—Transmission computed tomography [CT]
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B6/00—Apparatus for radiation diagnosis, e.g. combined with radiation therapy equipment
    • A61B6/50—Clinical applications
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B6/00—Apparatus for radiation diagnosis, e.g. combined with radiation therapy equipment
    • A61B6/52—Devices using data or image processing specially adapted for radiation diagnosis
    • A61B6/5205—Devices using data or image processing specially adapted for radiation diagnosis involving processing of raw data to produce diagnostic data
    • 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
    • G06F18/217—Validation; Performance evaluation; Active pattern learning techniques
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G06F18/2413—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on distances to training or reference patterns
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G06F18/2415—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on parametric or probabilistic models, e.g. based on likelihood ratio or false acceptance rate versus a false rejection rate
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T3/00—Geometric image transformation in the plane of the image
    • G06T3/40—Scaling the whole image or part thereof
    • G06T3/4038—Scaling the whole image or part thereof for image mosaicing, i.e. plane images composed of plane sub-images
    • 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
    • G06T7/0014—Biomedical image inspection using an image reference approach
    • G06T7/0016—Biomedical image inspection using an image reference approach involving temporal comparison
    • 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
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/10—Segmentation; Edge detection
    • G06T7/143—Segmentation; Edge detection involving probabilistic approaches, e.g. Markov random field [MRF] modelling
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/10—Segmentation; Edge detection
    • G06T7/187—Segmentation; Edge detection involving region growing; involving region merging; involving connected component labelling
    • 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
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/40—Extraction of image or video features
    • G06V10/44—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components
    • G06V10/443—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components by matching or filtering
    • G06V10/449—Biologically inspired filters, e.g. difference of Gaussians [DoG] or Gabor filters
    • G06V10/451—Biologically inspired filters, e.g. difference of Gaussians [DoG] or Gabor filters with interaction between the filter responses, e.g. cortical complex cells
    • G06V10/454—Integrating the filters into a hierarchical structure, e.g. convolutional neural networks [CNN]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/40—Extraction of image or video features
    • G06V10/50—Extraction of image or video features by performing operations within image blocks; by using histograms, e.g. histogram of oriented gradients [HoG]; by summing image-intensity values; Projection analysis
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/764—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using classification, e.g. of video objects
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/77—Processing image or video features in feature spaces; using data integration or data reduction, e.g. principal component analysis [PCA] or independent component analysis [ICA] or self-organising maps [SOM]; Blind source separation
    • G06V10/774—Generating sets of training patterns; Bootstrap methods, e.g. bagging or boosting
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/82—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using neural networks
    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
    • G16H50/00—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics
    • G16H50/30—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics for calculating health indices; for individual health risk assessment
    • 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
    • G06T2207/20076—Probabilistic image processing
    • 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
    • G06T2207/20084—Artificial neural networks [ANN]
    • 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
    • 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

Definitions

  • the present invention relates to the technical field of image processing, and in particular, to a method, a device, a device, and a readable storage medium for detecting CT nodules in lung nodules.
  • Lung nodules are one of the most important early signs of lung cancer. According to the characteristics of lung nodules, the lesion characteristics of lung lesions can be inferred. Due to the uncertainty of nodule size, shape and density, Existing detection methods are difficult to meet the market demand for the accuracy of pulmonary nodule detection.
  • the main purpose of the present invention is to provide a CT image lung nodule detection method, device, device and readable storage medium, which are aimed at solving the technical problem of poor accuracy of automatic detection of lung nodules based on CT images in the prior art.
  • the CT image lung nodule detection method includes:
  • the CT image is subjected to pixel segmentation processing through a pre-stored three-dimensional convolutional neural pixel segmentation network to obtain a probability map corresponding to the CT image, and the connected map analysis is performed on the probability map to obtain a candidate nodule region.
  • the steps include:
  • the CT image is divided into regions of a preset size to obtain sub-regions of the CT image.
  • the sub-regions are subjected to a predetermined number of down-sampling processes through a pre-stored three-dimensional pixel segmentation network.
  • the sub-region performs the same upsampling process for a preset number of times;
  • Connected-field labeling is performed on the probability map corresponding to the CT image to obtain candidate nodule regions.
  • the downsampling processing includes convolution, activation, batch normalization, and pooling processing of the subregion
  • the upsampling processing includes deconvolution, activation, batch normalization of the subregion after the downsampling processing And pooling.
  • the step of fusing processing each probability prediction value to obtain the classification result of the candidate nodule region includes:
  • the step of predicting the candidate nodule region by each prediction model corresponding to a different three-dimensional convolutional neural network classifier pre-stored to obtain each probability prediction value corresponding to the candidate nodule region includes:
  • the step of downsampling the candidate nodule region includes convolution, activation, batch normalization, and pooling of the candidate nodule region.
  • the method includes:
  • the method before the acquiring the CT image of the to-be-detected computer tomography and performing the prediction on the CT image through a three-dimensional convolutional neural pixel segmentation network, the method includes:
  • Multi-process online collection of CT image sample data using the collected CT image sample data as first sample data, and using the first sample data as input data to perform corresponding three-dimensional convolutional neural pixel segmentation networks and three-dimensional volume Training of a product neural network classifier and obtaining difficult-to-separate sample data, wherein the difficult-to-separate sample data is data that fails to predict in the first sample data;
  • the CT image of the to-be-detected computer tomography scan is executed. A step of.
  • the present invention also provides a CT image lung nodule detection device.
  • the CT image lung detection device includes:
  • An acquisition module configured to acquire an electronic computer tomography CT image to be detected, and perform pixel segmentation processing on the CT image through a pre-stored three-dimensional convolutional neural pixel segmentation network to obtain a probability map corresponding to the CT image; Graph connected region labeling to obtain candidate nodule regions;
  • a model fusion module is configured to predict the candidate nodule region through each prediction model corresponding to a different three-dimensional convolutional neural network classifier that is stored, to obtain each probability prediction value corresponding to the candidate nodule region, and to fusion process the Each probability prediction value obtains a classification result of the candidate nodule region.
  • the obtaining module includes:
  • a sub-probability map obtaining unit configured to perform a segmentation process of a preset size on the CT image to obtain a sub-area of the CT image, and perform a predetermined number of down-sampling processes on the sub-area through a pre-stored three-dimensional pixel segmentation network , Performing the same preset number of upsampling processing on the sub-region after the downsampling processing;
  • a feature fusion unit configured to perform bridge feature fusion processing on the sub-regions obtained after the down-sampling processing and the up-sampling processing, respectively, to obtain a sub-probability map equal to the size of the sub-region;
  • a mosaic restoration unit configured to perform mosaic restoration on each sub-probability map to obtain a probability map corresponding to the CT image
  • the connected domain processing module is configured to label the connected domain of the probability map corresponding to the CT image to obtain candidate nodule regions.
  • the sub-probability map obtaining unit is configured to perform convolution, activation, batch normalization, and pooling processing on the sub-region, and is further configured to deconvolve, activate, Batch standardization and pooling.
  • the model fusion module includes:
  • An averaging unit configured to averagely process the respective probability prediction values to obtain a target probability prediction value of the candidate nodule region
  • a comparison unit configured to compare the target probability prediction value with a pre-stored threshold and obtain a comparison result, and obtain a classification result of the candidate nodule region based on the comparison result, wherein the pre-stored threshold is based on a corresponding three-dimensional
  • the corresponding ROC curve in the convolutional neural network classifier is determined.
  • model fusion module further includes:
  • a model fusion unit is configured to perform down-sampling processing and full connection processing on the candidate nodule regions through the respective prediction models corresponding to different three-dimensional convolutional neural network classifiers, to obtain the respective probabilities corresponding to the candidate nodule regions. Predictive value;
  • the model fusion unit is further configured to perform convolution, activation, batch normalization, and pooling processing on the candidate nodule region.
  • the CT image lung nodule detection device further includes:
  • An output module is configured to output the target probability prediction value and generate target prompt information.
  • the CT image lung nodule detection device further includes:
  • An acquisition module configured to collect CT image sample data in multiple processes online, use the acquired CT image sample data as first sample data, and use the first sample data as input data to perform corresponding three-dimensional convolutional neural pixels Segmentation network and three-dimensional convolutional neural network classifier training, and obtain hard-to-separate sample data, where the hard-to-separate sample data is data that fails prediction in the first sample data;
  • the hard-to-separate sample data enhancement module is used to perform multiple translation and horizontal flip operations on the hard-to-separate sample data to increase the proportion of hard-to-separate sample data in the first sample data and increase the hard-to-separate sample data.
  • the data after the sample data ratio is used as the second sample data;
  • a training module configured to use the second sample data as input data to train a corresponding three-dimensional convolutional neural pixel segmentation network and a three-dimensional convolutional neural network classifier multiple times;
  • An execution module configured to execute when the predicted accuracy of the three-dimensional convolutional neural pixel segmentation network and the three-dimensional convolutional neural network classifier reach a first target accuracy rate and a second target accuracy rate, respectively, to obtain the electrons to be detected Steps for computed tomography CT images.
  • the present invention also provides a CT image lung nodule detection device.
  • the CT image lung nodule detection device includes: a memory, a processor, a communication bus, and a CT image lung stored in the memory. Nodule detection procedure,
  • the communication bus is used to implement a communication connection between the processor and the memory
  • the processor is configured to execute the CT image lung nodule detection program to implement the following steps:
  • the CT image is subjected to pixel segmentation processing through a pre-stored three-dimensional convolutional neural pixel segmentation network to obtain a probability map corresponding to the CT image, and the connected map analysis is performed on the probability map to obtain a candidate nodule region.
  • the steps include:
  • the CT image is divided into regions of a preset size to obtain sub-regions of the CT image.
  • the sub-regions are subjected to a predetermined number of down-sampling processes through a pre-stored three-dimensional pixel segmentation network.
  • the sub-region performs the same upsampling process for a preset number of times;
  • Connected-field labeling is performed on the probability map corresponding to the CT image to obtain candidate nodule regions.
  • the downsampling processing includes convolution, activation, batch normalization, and pooling processing of the subregion
  • the upsampling processing includes deconvolution, activation, and batch normalization of the subregion after the downsampling processing. And pooling.
  • the step of fusing processing each probability prediction value to obtain the classification result of the candidate nodule region includes:
  • the step of predicting the candidate nodule region by each prediction model corresponding to a different three-dimensional convolutional neural network classifier pre-stored to obtain each probability prediction value corresponding to the candidate nodule region includes:
  • the step of downsampling the candidate nodule region includes convolution, activation, batch normalization, and pooling of the candidate nodule region.
  • the method includes:
  • the method before the acquiring the CT image of the to-be-detected computer tomography and performing the prediction on the CT image through a three-dimensional convolutional neural pixel segmentation network, the method includes:
  • Multi-process online collection of CT image sample data using the collected CT image sample data as first sample data, and using the first sample data as input data to perform corresponding three-dimensional convolutional neural pixel segmentation networks and three-dimensional volume Training of a product neural network classifier and obtaining difficult-to-separate sample data, wherein the difficult-to-separate sample data is data that fails to predict in the first sample data;
  • the CT image of the to-be-detected computer tomography scan is executed. A step of.
  • the present invention further provides a readable storage medium, where the readable storage medium stores one or more programs, and the one or more programs can be executed by one or more processors to Used for:
  • the present invention obtains a probability map corresponding to the CT image by acquiring a CT image of an electronic computer tomography scan to be detected, and performing pixel segmentation processing on the CT image through a pre-stored three-dimensional convolutional neural pixel segmentation network. Connected domain labeling to obtain candidate nodule regions; perform prediction processing on the candidate nodule regions through each of the prediction models corresponding to the different three-dimensional convolutional neural network classifiers stored, to obtain the respective probability prediction values corresponding to the candidate nodule regions, Fusion processing the respective probability prediction values to obtain a classification result of the candidate nodule region.
  • the pre-stored 3D convolutional neural pixel segmentation network and the 3D convolutional neural network classifier have been trained and have a certain degree of accuracy in the processing network.
  • candidates are obtained through the pre-stored 3D convolutional neural network. Nodule area.
  • multi-model fusion is used to compensate for the contingency of different models, and the target probability prediction value of the candidate nodule area is obtained, thus avoiding the uncertainty due to the characteristics of lung nodule size, shape, and density.
  • the phenomenon of low accuracy of lung nodule detection is caused, and the technical problem of poor accuracy of automatic detection of lung nodules based on CT images in the prior art is solved.
  • FIG. 1 is a schematic flowchart of a first embodiment of a method for detecting a pulmonary nodule in a CT image according to the present invention
  • FIG. 2 is a schematic flowchart of a second embodiment of a method for detecting a pulmonary nodule in a CT image according to the present invention
  • FIG. 3 is a schematic diagram of a device structure of a hardware operating environment involved in a method according to an embodiment of the present invention
  • FIG. 4 is a schematic diagram of a method for detecting a lung nodule in a CT image according to the present invention.
  • the present invention provides a CT image lung nodule detection method.
  • the CT image lung nodule detection method includes:
  • Obtain a tomographic CT image to be detected perform pixel segmentation on the CT image through a pre-stored three-dimensional convolutional neural pixel segmentation network, obtain a probability map corresponding to the CT image, and label the probability map with connected domains
  • Each probability prediction value obtains a classification result of the candidate nodule region.
  • Step S10 Obtain a tomographic CT image to be detected, perform pixel segmentation processing on the CT image through a pre-stored three-dimensional convolutional neural pixel segmentation network to obtain a probability map corresponding to the CT image, and perform the probability map on the CT image. Connected domain labeling to obtain candidate nodule regions;
  • the CT image Before performing the test, first obtain or receive a CT image of the computer tomography to be detected, that is, the CT image, wherein the CT image is obtained from a hospital by collecting lung nodules through CT, and the CT image may include a doctor
  • the labeled lung nodule area is the nodule marked area.
  • the CT nodule image is first predicted through a pre-stored three-dimensional convolutional neural pixel segmentation network to obtain a CT.
  • the image corresponds to a probability map, and the connected map analysis is performed on the probability map to obtain candidate nodule regions.
  • 3D-Unet is a three-dimensional convolutional neural pixel segmentation network.
  • stage one candidate nodule regions (candidate).
  • the 3D convolutional neural pixel segmentation network is a trained neural network model.
  • the sigmoid function used in the last convolutional layer in the pixel segmentation network is activated, and the pixel segmentation network is trained using dice.
  • the loss and focal loss loss functions can alleviate the imbalance of positive and negative samples.
  • step S10 includes:
  • Step S11 Perform segmentation processing on the CT image with a preset size to obtain a sub-region of the CT image, and perform a down-sampling process on the sub-regions a predetermined number of times through a pre-stored three-dimensional pixel segmentation network. Performing subsequent up-sampling processing on the sub-areas for the same preset number of times;
  • Step S12 performing bridge feature fusion processing on the sub-regions obtained after the down-sampling processing and the up-sampling processing, respectively, to obtain a sub-probability map equal to the size of the sub-regions;
  • Step S13 stitch and restore each sub-probability map to obtain a probability map corresponding to the CT image
  • Step S14 label the probability map corresponding to the CT image to obtain a candidate nodule region.
  • the storage space of the GPU of the graphics processor is generally difficult to meet the simultaneous calculation of the entire CT image. Therefore, the CT image needs to be segmented with a preset size to meet the computing power of the GPU.
  • the image is a subregion of the CT image. For example, during the detection process, the CT image is divided into small regions of 128 * 128 * 128, and the small region of 128 * 128 * 128 is a subregion of the CT image.
  • the sub-regions of the CT image are respectively predicted through a three-dimensional convolutional neural pixel segmentation network to obtain each sub-probability map. Specifically, the sub-regions are separately performed through a pre-stored three-dimensional pixel segmentation network.
  • a predetermined number of downsampling processes, the same predetermined number of upsampling processes are performed on the subregions after the downsampling process, and the subregions obtained after the downsampling process and after the upsampling process are bridged, respectively Fusion processing to obtain a sub-probability map equal to the size of the sub-region to compensate for the loss of information in the sub-region caused by down-sampling, where the same number of down-sampling processes and up-sampling processes are used, and the original CT can be obtained Sub-probability maps with the same shape of the sub-regions of the image.
  • the bridge feature fusion processing refers to the addition of a bridge structure between the sub-regions of the same size after the down-sampling and the up-sampling process. Image features are fused, which can avoid the loss of possible sub-region information.
  • each sub-probability map can obtain the probability map corresponding to the CT image, and by accurately obtaining the probability map, it can lay a foundation for finding and obtaining candidate nodule regions.
  • the downsampling processing includes convolution, activation, batch normalization, and pooling processing of the subregion
  • the upsampling processing includes deconvolution, activation, batch normalization, and pooling of the subregion after the downsampling processing.
  • the sub-region needs to be convoluted, activated, batch normalized, and pooled a preset number of times.
  • the convolution process can be understood as: the statistical characteristics of one part of the image are the same as the other parts, that is, the features learned in this part can also be used in the corresponding other part, so the learned features are used as detectors and applied Go anywhere in this image. That is, the features learned from the small-scale image are convolved with the original large-size image, and an activation value of a different feature is obtained for any pixel position on the image.
  • the corresponding batch of standardized processing is performed. Correction processing.
  • the corresponding correction parameters are pre-stored.
  • pooling processing is performed.
  • the maximum pooling processing can be used to obtain the maximum activation value to extract local features.
  • the sub-region can be obtained after the down-sampling process.
  • De-convolution, activation, batch normalization, and pooling are performed on the sub-region after the down-sampling process for the same preset number of times to obtain a sub-probability map equal to the size of the sub-region.
  • the deconvolution process is the reverse process of the convolution process, and will not be described in detail here.
  • the sub-probability maps are stitched and restored to obtain the probability maps corresponding to the CT images. It should be noted that the sub-probability maps are stitched in a preset order to obtain the corresponding CT images. In the probability map, the preset order is associated with a segmentation order mapping of each sub-region in the pixel-segmented CT image.
  • Step S20 Prediction of the candidate nodule region through each prediction model corresponding to a different three-dimensional convolutional neural network classifier that is pre-stored to obtain each probability prediction value corresponding to the candidate nodule region, and fusion processing the respective probability prediction Value to obtain the classification result of the candidate nodule region.
  • a plurality of three-dimensional convolutional neural network classifiers are trained.
  • the prediction models in the different three-dimensional convolutional neural network classifiers are different.
  • the two three-dimensional convolutional neural network classifiers may be used.
  • the two prediction models corresponding to the network classifier that is, the prediction of the candidate nodule region is performed by using the fusion of the two prediction models to obtain the respective probability prediction values corresponding to the candidate nodule region, and the fusion is used to process the respective probability prediction values.
  • the second stage is state2. Because multiple prediction models are passed, the chance before the model can be eliminated, and the detection accuracy and accuracy can be improved. After obtaining the predicted values of each probability, a classification result of the candidate nodule region can be obtained through the predicted values of each probability.
  • the step of predicting the candidate nodule region by using the respective prediction models corresponding to different pre-stored different three-dimensional convolutional neural network classifiers to obtain each probability prediction value corresponding to the candidate nodule region includes:
  • Step S21 Perform downsampling processing and full connection processing on the candidate nodule region respectively through the respective prediction models corresponding to different three-dimensional convolutional neural network classifiers that are pre-stored to obtain each probability prediction value corresponding to the candidate nodule region;
  • Step S22 wherein the step of downsampling the candidate nodule region includes convolution, activation, batch normalization, and pooling of the candidate nodule region.
  • the 3D convolutional neural network classifier includes multiple downsampling layers and the last fully connected layer to achieve each candidate node.
  • the node area performs downsampling processing and full connection processing.
  • the process of performing downsampling processing on each candidate nodule area includes convolution, activation, batch normalization, and pooling processing of the candidate nodule area.
  • the full connection processing is The nodes obtained after the downsampling process are connected to comprehensively process the image features corresponding to each node to finally obtain the respective probability prediction values corresponding to the candidate nodule area.
  • each candidate nodule region is predicted, and each candidate nodule region corresponds to each three-dimensional convolutional neural network classifier to obtain a probability prediction value. Therefore, each candidate nodule region corresponds to multiple probability prediction values.
  • the step of the fusion processing the respective probability prediction values to obtain the classification result of the candidate nodule region includes:
  • Step S23 averagely process the respective probability prediction values to obtain a target probability prediction value of the candidate nodule area
  • Step S24 Compare the target probability prediction value with a pre-stored threshold and obtain a comparison result. Based on the comparison result, obtain a classification result of the candidate nodule region, where the pre-stored threshold is based on a three-dimensional convolutional neural network. The corresponding ROC curve in the network classifier is determined.
  • the plurality of probability prediction values are averaged, and the averaged probability prediction value is used as a target of the candidate nodule region.
  • Probability prediction value After obtaining the target probability prediction value, obtain the pre-stored threshold value, and compare the target probability prediction value with the pre-stored threshold value to obtain the comparison result.
  • the pre-stored threshold value can be adjusted. Specifically, the pre-stored value is adjustable. The threshold is determined based on the overall ROC curves of different models in the 3D convolutional neural network classifier. After a comparison result is obtained, based on the comparison result, a classification result of the candidate nodular region can be obtained, and the classification result includes a nodular region and a non-nodular region.
  • the target probability prediction value may be output processed, and target prompt information may be generated, and the corresponding prompt manner of the target prompt information is not specifically limited.
  • the present invention obtains a probability map corresponding to the CT image by acquiring a CT image of an electronic computer tomography scan to be detected, and performing pixel segmentation processing on the CT image through a pre-stored three-dimensional convolutional neural pixel segmentation network. Connected domain labeling to obtain candidate nodule regions; perform prediction processing on the candidate nodule regions through each of the prediction models corresponding to the different three-dimensional convolutional neural network classifiers stored, to obtain the respective probability prediction values corresponding to the candidate nodule regions, Fusion processing the respective probability prediction values to obtain a classification result of the candidate nodule region.
  • the three-dimensional convolutional neural pixel segmentation network and the three-dimensional convolutional neural network classifier have been trained and processed with a certain degree of accuracy.
  • the candidate nodule region is obtained through the three-dimensional convolutional neural network.
  • multi-model fusion is used to compensate for the contingency of different models and obtain the target probability prediction value of the candidate nodule region, thereby avoiding the pulmonary nodules due to the uncertainty of the characteristics such as the size, shape and density of the pulmonary nodules.
  • the phenomenon of low detection accuracy solves the technical problem that the existing detection methods can hardly meet the market demand for the accuracy of lung nodule detection.
  • the present invention provides another embodiment of a CT nodule detection method.
  • the computerized tomography CT image to be detected is obtained, and the CT image is predicted by a three-dimensional convolutional neural pixel segmentation network.
  • the steps include:
  • Step S01 Collect CT image sample data in multiple processes online, use the collected CT image sample data as first sample data, and use the first sample data as input data to perform a corresponding three-dimensional convolutional neural pixel segmentation network And training of a three-dimensional convolutional neural network classifier, and obtaining difficult-to-separate sample data, where the difficult-to-separate sample data is data that fails prediction in the first sample data;
  • Step S02 performing multiple panning and horizontal flip operations on the difficult-to-separate sample data to increase the proportion of the hard-to-separate sample data in the first sample data, and increasing the data after the proportion of the hard-to-separate sample data is increased.
  • the second sample data As the second sample data;
  • Step S03 using the second sample data as input data to train the corresponding 3D convolutional neural pixel segmentation network and the 3D convolutional neural network classifier multiple times;
  • step S04 when the predicted accuracy of the three-dimensional convolutional neural pixel segmentation network and the three-dimensional convolutional neural network classifier reaches the first target accuracy rate and the second target accuracy rate, respectively, the acquisition of the to-be-detected electronic computer tomography is performed. Steps for scanning CT images.
  • FIG. 3 is a schematic structural diagram of a device in a hardware operating environment according to an embodiment of the present invention.
  • the CT image lung nodule detection device may be a PC, a portable computer, or a terminal device such as a mobile terminal.
  • the CT image lung nodule detection device may include: a processor 1001, such as a CPU, a memory 1005, and a communication bus 1002.
  • the communication bus 1002 is used to implement connection and communication between the processor 1001 and the memory 1005.
  • the memory 1005 may be a high-speed RAM memory or a non-volatile memory. memory), such as disk storage.
  • the memory 1005 may optionally be a storage device independent of the foregoing processor 1001.
  • the CT image lung nodule detection device may further include a user interface, a network interface, a camera, an RF (Radio Frequency) circuits, sensors, audio circuits, WiFi modules, etc.
  • the user interface may include a display, an input unit such as a keyboard, and the optional user interface may also include a standard wired interface and a wireless interface.
  • the network interface can optionally include a standard wired interface and a wireless interface (such as a WI-FI interface).
  • CT image lung nodule detection device does not constitute a limitation on the CT image lung nodule detection device, and may include more or fewer parts than the illustration, or a combination Certain components, or different component arrangements.
  • the memory 1005 as a computer storage medium may include an operating system, a network communication module, and a CT image lung nodule detection program.
  • the operating system is a program that manages and controls the hardware and software resources of the CT image lung nodule detection device, and supports the operation of CT image lung nodule detection programs and other software and / or programs.
  • the network communication module is used to implement communication between components in the memory 1005 and to communicate with other hardware and software in the CT image lung nodule detection device.
  • the processor 1001 is configured to execute a CT image lung nodule detection program stored in the memory 1005 to implement the CT image lung nodule detection method described in any one of the above. step.
  • CT image lung nodule detection device of the present invention is basically the same as each embodiment of the CT image lung nodule detection method described above, and details are not described herein again.
  • the present invention also provides a CT image lung nodule detection device.
  • the CT image lung nodule detection device includes:
  • An acquisition module configured to acquire an electronic computer tomography CT image to be detected, and perform pixel segmentation processing on the CT image through a pre-stored three-dimensional convolutional neural pixel segmentation network to obtain a probability map corresponding to the CT image; Graph connected region labeling to obtain candidate nodule regions;
  • a model fusion module is configured to predict the candidate nodule region through each prediction model corresponding to a different three-dimensional convolutional neural network classifier that is stored, to obtain each probability prediction value corresponding to the candidate nodule region, and to fusion process the Each probability prediction value obtains a classification result of the candidate nodule region.
  • CT image lung nodule detection device of the present invention is basically the same as each embodiment of the CT image lung nodule detection method described above, and details are not described herein again.
  • the present invention provides a readable storage medium, where the readable storage medium stores one or more programs, and the one or more programs can also be executed by one or more processors to implement any of the foregoing. Steps of the CT image pulmonary nodule detection method described in item 3.
  • the specific implementation manner of the readable storage medium of the present invention is basically the same as each embodiment of the CT image lung nodule detection method described above, and details are not described herein again.

Abstract

一种CT图像肺结节检测方法、装置、设备及可读存储介质，CT图像肺结节检测方法包括：获取待检测的电子计算机断层扫描CT图像，通过预存的三维卷积神经像素分割网络对CT图像进行像素分割处理，得到CT图像对应的概率图，对概率图进行连通域标记得到候选结节区域（S10）；通过预存的不同三维卷积神经网络分类器对应的多个模型对候选结节区域进行预测，得到候选结节区域对应的各个概率预测值，融合处理所述各个概率预测值，得到候选结节区域的分类结果（S20）。由此，本方法提高了CT图像自动化检测肺结节的精确度。

Description

CT图像肺结节检测方法、设备及可读存储介质
本申请要求于2018年06月28日提交中国专利局、申请号为201810695264.8、发明名称为“CT图像肺结节检测方法、装置、设备及可读存储介质”的中国专利申请的优先权，其全部内容通过引用结合在申请中。
技术领域
本发明涉及图像处理技术领域，尤其涉及一种CT图像肺结节检测方法、装置、设备及可读存储介质。
背景技术
肺结节是肺癌最重要的早期征象之一，根据肺结节的病变特征能够推断出肺病灶的病变特性，由于结节的大小、形状以及密度等特征的不确定性, 现有检测方法难以满足市场对肺结节检测精确度的需求。
发明内容
本发明的主要目的在于提供一种CT图像肺结节检测方法、装置、设备及可读存储介质，旨在解决现有技术中基于CT图像自动化检测肺结节精确度欠佳的技术问题。
为实现上述目的，本发明提供一种CT图像肺结节检测方法，所述CT图像肺结节检测方法包括：
获取待检测的电子计算机断层扫描CT图像，通过预存的三维卷积神经像素分割网络对所述CT图像进行像素分割处理，得到所述CT图像对应的概率图，对所述概率图进行连通域标记得到候选结节区域；
通过预存的不同三维卷积神经网络分类器对应的各个预测模型对所述候选结节区域进行预测，得到所述候选结节区域对应的各个概率预测值，融合处理所述各个概率预测值，得到所述候选结节区域的分类结果。
可选地，所述通过预存的三维卷积神经像素分割网络对所述CT图像进行像素分割处理，得到所述CT图像对应的概率图，对所述概率图进行连通域分析得到候选结节区域步骤包括：
对所述CT图像进行预设尺寸的区域分割处理，得到CT图像的子区域，通过预存的三维像素分割网络对所述子区域分别进行预设次数的下采样处理，对下采样处理后的所述子区域进行同样预设次数的上采样处理；
对所述下采样处理后以及上采样处理后分别得到的所述子区域进行桥接特征融合处理，得到与所述子区域尺寸相等的子概率图；
对所述各个子概率图进行拼接还原得到所述CT图像对应的概率图；
对所述CT图像对应的概率图进行连通域标记，得到候选结节区域。
可选地，所述下采样处理包括对所述子区域的卷积、激活、批标准化以及池化处理，上采样处理包括对所述下采样处理后子区域的反卷积、激活、批标准化以及池化处理。
可选地，所述融合处理所述各个概率预测值，得到所述候选结节区域的分类结果步骤包括：
平均处理所述各个概率预测值，得到所述候选结节区域的目标概率预测值；
将所述目标概率预测值与预存阀值进行比较，并得到比较结果，基于该比较结果，得到所述候选结节区域的分类结果，其中，所述预存阀值根据对应三维卷积神经网络分类器中对应的ROC曲线确定。
可选地，所述通过预存的不同三维卷积神经网络分类器对应的各个预测模型对所述候选结节区域进行预测，得到所述候选结节区域对应的各个概率预测值步骤包括：
通过预存的不同三维卷积神经网络分类器对应的各个预测模型分别对所述候选结节区域进行下采样处理以及全连接处理，得到所述候选结节区域对应的各个概率预测值；
其中，所述对所述候选结节区域进下采样处理步骤包括对所述候选结节区域的卷积、激活、批标准化以及池化处理。
可选地，所述得到所述候选结节区域的目标概率预测值步骤之后包括：
输出所述目标概率预测值，并生成目标提示信息。
可选地，所述获取待检测电子计算机断层扫描CT图像，通过三维卷积神经像素分割网络对所述CT图像进行预测步骤之前包括：
线上多进程采集CT图像样本数据，将所述采集的CT图像样本数据作为第一样本数据，将所述第一样本数据作为输入数据以进行对应三维卷积神经像素分割网络以及三维卷积神经网络分类器的训练，并得到难分样本数据，其中，所述难分样本数据是第一样本数据中预测失败的数据；
对所述难分样本数据进行多次平移和水平翻转操作，以增大难分样本数据在所述第一样本数据中的比例，并将增大难分样本数据比例后的数据作为第二样本数据；
将所述第二样本数据作为输入数据以多次进行对应三维卷积神经像素分割网络以及三维卷积神经网络分类器的训练；
当训练后的所述三维卷积神经像素分割网络以及三维卷积神经网络分类器的预测准确率分别达到第一目标准确率、第二目标准确率时，执行获取待检测电子计算机断层扫描CT图像的步骤。
本发明还提供一种CT图像肺结节检测装置，所述CT图像肺检测装置包括：
获取模块，用于获取待检测的电子计算机断层扫描CT图像，通过预存的三维卷积神经像素分割网络对所述CT图像进行像素分割处理，得到所述CT图像对应的概率图，对所述概率图进行连通域标记得到候选结节区域；
模型融合模块，用于通过预存的不同三维卷积神经网络分类器对应的各个预测模型对所述候选结节区域进行预测，得到所述候选结节区域对应的各个概率预测值，融合处理所述各个概率预测值，得到所述候选结节区域的分类结果。
可选地，所述获取模块包括：
子概率图获取单元，用于对所述CT图像进行预设尺寸的区域分割处理，得到CT图像的子区域，通过预存的三维像素分割网络对所述子区域分别进行预设次数的下采样处理，对下采样处理后的所述子区域进行同样预设次数的上采样处理；
特征融合单元，用于对所述下采样处理后以及上采样处理后分别得到的所述子区域进行桥接特征融合处理，得到与所述子区域尺寸相等的子概率图；
拼接还原单元，用于对所述各个子概率图进行拼接还原得到所述CT图像对应的概率图；
连通域处理模块，用于对所述CT图像对应的概率图进行连通域标记，得到候选结节区域。
可选地，所述子概率图获取单元用于对所述子区域的卷积、激活、批标准化以及池化处理，还用于对所述下采样处理后子区域的反卷积、激活、批标准化以及池化处理。
可选地，所述模型融合模块包括：
平均单元，用于平均处理所述各个概率预测值，得到所述候选结节区域的目标概率预测值；
比较单元，用于将所述目标概率预测值与预存阀值进行比较，并得到比较结果，基于该比较结果，得到所述候选结节区域的分类结果，其中，所述预存阀值根据对应三维卷积神经网络分类器中对应的ROC曲线确定。
可选地，所述模型融合模块还包括：
模型融合单元，用于通过预存的不同三维卷积神经网络分类器对应的各个预测模型分别对所述候选结节区域进行下采样处理以及全连接处理，得到所述候选结节区域对应的各个概率预测值；
其中，所述模型融合单元还用于对所述候选结节区域的卷积、激活、批标准化以及池化处理。
可选地，所述CT图像肺结节检测装置还包括：
输出模块，用于输出所述目标概率预测值，并生成目标提示信息。
可选地，所述CT图像肺结节检测装置还包括：
采集模块，用于线上多进程采集CT图像样本数据，将所述采集的CT图像样本数据作为第一样本数据，将所述第一样本数据作为输入数据以进行对应三维卷积神经像素分割网络以及三维卷积神经网络分类器的训练，并得到难分样本数据，其中，所述难分样本数据是第一样本数据中预测失败的数据；
难分样本数据加强模块，用于对所述难分样本数据进行多次平移和水平翻转操作，以增大难分样本数据在所述第一样本数据中的比例，并将增大难分样本数据比例后的数据作为第二样本数据；
训练模块，用于将所述第二样本数据作为输入数据以多次进行对应三维卷积神经像素分割网络以及三维卷积神经网络分类器的训练；
执行模块，用于当训练后的所述三维卷积神经像素分割网络以及三维卷积神经网络分类器的预测准确率分别达到第一目标准确率、第二目标准确率时，执行获取待检测电子计算机断层扫描CT图像的步骤。
此外，为实现上述目的，本发明还提供一种CT图像肺结节检测设备，所述CT图像肺结节检测设备包括：存储器、处理器，通信总线以及存储在所述存储器上的CT图像肺结节检测程序，
所述通信总线用于实现处理器与存储器间的通信连接；
所述处理器用于执行所述CT图像肺结节检测程序，以实现以下步骤：
获取待检测的电子计算机断层扫描CT图像，通过预存的三维卷积神经像素分割网络对所述CT图像进行像素分割处理，得到所述CT图像对应的概率图，对所述概率图进行连通域标记得到候选结节区域；
通过预存的不同三维卷积神经网络分类器对应的各个预测模型对所述候选结节区域进行预测，得到所述候选结节区域对应的各个概率预测值，融合处理所述各个概率预测值，得到所述候选结节区域的分类结果。
可选地，所述通过预存的三维卷积神经像素分割网络对所述CT图像进行像素分割处理，得到所述CT图像对应的概率图，对所述概率图进行连通域分析得到候选结节区域步骤包括：
对所述CT图像进行预设尺寸的区域分割处理，得到CT图像的子区域，通过预存的三维像素分割网络对所述子区域分别进行预设次数的下采样处理，对下采样处理后的所述子区域进行同样预设次数的上采样处理；
对所述下采样处理后以及上采样处理后分别得到的所述子区域进行桥接特征融合处理，得到与所述子区域尺寸相等的子概率图；
对所述各个子概率图进行拼接还原得到所述CT图像对应的概率图；
对所述CT图像对应的概率图进行连通域标记，得到候选结节区域。
可选地，所述下采样处理包括对所述子区域的卷积、激活、批标准化以及池化处理，上采样处理包括对所述下采样处理后子区域的反卷积、激活、批标准化以及池化处理。
可选地，所述融合处理所述各个概率预测值，得到所述候选结节区域的分类结果步骤包括：
平均处理所述各个概率预测值，得到所述候选结节区域的目标概率预测值；
将所述目标概率预测值与预存阀值进行比较，并得到比较结果，基于该比较结果，得到所述候选结节区域的分类结果，其中，所述预存阀值根据对应三维卷积神经网络分类器中对应的ROC曲线确定。
可选地，所述通过预存的不同三维卷积神经网络分类器对应的各个预测模型对所述候选结节区域进行预测，得到所述候选结节区域对应的各个概率预测值步骤包括：
通过预存的不同三维卷积神经网络分类器对应的各个预测模型分别对所述候选结节区域进行下采样处理以及全连接处理，得到所述候选结节区域对应的各个概率预测值；
其中，所述对所述候选结节区域进下采样处理步骤包括对所述候选结节区域的卷积、激活、批标准化以及池化处理。
可选地，所述得到所述候选结节区域的目标概率预测值步骤之后包括：
输出所述目标概率预测值，并生成目标提示信息。
可选地，所述获取待检测电子计算机断层扫描CT图像，通过三维卷积神经像素分割网络对所述CT图像进行预测步骤之前包括：
线上多进程采集CT图像样本数据，将所述采集的CT图像样本数据作为第一样本数据，将所述第一样本数据作为输入数据以进行对应三维卷积神经像素分割网络以及三维卷积神经网络分类器的训练，并得到难分样本数据，其中，所述难分样本数据是第一样本数据中预测失败的数据；
对所述难分样本数据进行多次平移和水平翻转操作，以增大难分样本数据在所述第一样本数据中的比例，并将增大难分样本数据比例后的数据作为第二样本数据；
将所述第二样本数据作为输入数据以多次进行对应三维卷积神经像素分割网络以及三维卷积神经网络分类器的训练；
当训练后的所述三维卷积神经像素分割网络以及三维卷积神经网络分类器的预测准确率分别达到第一目标准确率、第二目标准确率时，执行获取待检测电子计算机断层扫描CT图像的步骤。
此外，为实现上述目的，本发明还提供一种可读存储介质，所述可读存储介质存储有一个或者一个以上程序，所述一个或者一个以上程序可被一个或者一个以上的处理器执行以用于：
获取待检测的电子计算机断层扫描CT图像，通过预存的三维卷积神经像素分割网络对所述CT图像进行像素分割处理，得到所述CT图像对应的概率图，对所述概率图进行连通域标记得到候选结节区域；
通过预存的不同三维卷积神经网络分类器对应的各个预测模型对所述候选结节区域进行预测，得到所述候选结节区域对应的各个概率预测值，融合处理所述各个概率预测值，得到所述候选结节区域的分类结果。
本发明通过获取待检测的电子计算机断层扫描CT图像，通过预存的三维卷积神经像素分割网络对所述CT图像进行像素分割处理，得到所述CT图像对应的概率图，对所述概率图进行连通域标记得到候选结节区域；通过预存的不同三维卷积神经网络分类器对应的各个预测模型对所述候选结节区域进行预测处理，得到所述候选结节区域对应的各个概率预测值，融合处理所述各个概率预测值，得到所述候选结节区域的分类结果。在本发明中，预存的三维卷积神经像素分割网络以及三维卷积神经网络分类器是已经训练完成的，具有一定准确度的处理网络，在第一阶段通过预存的三维卷积神经网络得到候选结节区域，在第二阶段，通过多模型融合，补偿不同模型的偶然性，得到候选结节区域的目标概率预测值，因而避免了由于肺结节的大小、形状以及密度等特征的不确定性造成肺结节检测精确度低的现象，解决了现有技术中基于CT图像自动化检测肺结节精确度欠佳的技术问题。
附图说明
图1为本发明CT图像肺结节检测方法第一实施例的流程示意图；
图2为本发明CT图像肺结节检测方法第二实施例的流程示意图；
图3是本发明实施例方法涉及的硬件运行环境的设备结构示意图；
图4是本发明CT图像肺结节检测方法的场景示意图。
本发明目的的实现、功能特点及优点将结合实施例，参照附图做进一步说明。
具体实施方式
应当理解，此处所描述的具体实施例仅仅用以解释本发明，并不用于限定本发明。
本发明提供一种CT图像肺结节检测方法，在本发明CT图像肺结节检测方法的第一实施例中，参照图1，所述CT图像肺结节检测方法包括：
获取待检测的电子计算机断层扫描CT图像，通过预存的三维卷积神经像素分割网络对所述CT图像进行像素分割处理，得到所述CT图像对应的概率图，对所述概率图进行连通域标记得到候选结节区域；通过预存的不同三维卷积神经网络分类器对应的各个预测模型对所述候选结节区域进行预测，得到所述候选结节区域对应的各个概率预测值，融合处理所述各个概率预测值，得到所述候选结节区域的分类结果。
具体步骤如下：
步骤S10，获取待检测的电子计算机断层扫描CT图像，通过预存的三维卷积神经像素分割网络对所述CT图像进行像素分割处理，得到所述CT图像对应的概率图，对所述概率图进行连通域标记得到候选结节区域；
在进行检测前，首先获取或者接收待检测的电子计算机断层扫描CT图像即所述CT图像，其中，所述CT图像是从医院通过CT采集肺结节得到的，所述CT图像中可以包含医生所标记的肺结节区域即结节标记区域，在本实施例中，在得到CT图像后，首先通过预存的三维卷积神经像素分割网络，对所述CT结节图像进行预测，以得到CT图像对应概率图，对该概率图进行连通域分析或者连通域标记得到候选结节区域。如图四所示，3D-Unet即三维卷积神经像素分割网络，通过阶段一即state1，得到候选结节区域（candidate screening），需要说明的是，三维卷积神经像素分割网络是已经训练完成的神经网络模型，该像素分割网络中最后一层卷积层使用的sigmoid函数激活，且该像素分割网络的训练使用dice loss以及focal loss损失函数，能够缓解正负样本的不平衡性。
具体地，参照图2，步骤S10包括：
步骤S11，对所述CT图像进行预设尺寸的区域分割处理，得到CT图像的子区域，通过预存的三维像素分割网络对所述子区域分别进行预设次数的下采样处理，对下采样处理后的所述子区域进行同样预设次数的上采样处理；
步骤S12，对所述下采样处理后以及上采样处理后分别得到的所述子区域进行桥接特征融合处理，得到与所述子区域尺寸相等的子概率图；
步骤S13，对所述各个子概率图进行拼接还原得到所述CT图像对应的概率图；
步骤S14，对所述CT图像对应的概率图进行连通域标记，得到候选结节区域。
需要说明的是，图形处理器GPU的存储空间一般难以满足对整个CT图像的同时计算，因而需要对CT图像进行预设尺寸的区域分割处理，以满足GPU的计算能力，将分割处理后得到的图像作为CT图像的子区域，如在检测过程中，将CT图像分成128*128*128的小区域，该128*128*128的小区域即是CT图像的子区域。
在得到CT图像的子区域后，通过三维卷积神经像素分割网络对CT图像的子区域分别进行预测以得到各个子概率图，具体地，通过预存的三维像素分割网络对所述子区域分别进行预设次数的下采样处理，对下采样处理后的所述子区域进行同样预设次数的上采样处理，对所述下采样处理后以及上采样处理后分别得到的所述子区域进行桥接特征融合处理，得到与所述子区域尺寸相等的子概率图，以弥补下采样导致所述子区域的信息丢失，其中，由于通过同样次数的下采样处理与上采样处理，因而能够得到和原始CT图像的子区域形状相同的子概率图，桥接特征融合处理指的是在上采样和下采样阶段，对下采样以及上采样处理后的同尺寸的子区域之间加入桥接结构，对子区域的图像特征进行融合，因而能够避免可能存在的子区域信息的丢失。
进一步地，对所述各个子概率图进行拼接还原能够得到所述CT图像对应的概率图，通过准确获取概率图，能够为查找得到候选结节区域奠下基础。
其中，所述下采样处理包括对所述子区域的卷积、激活、批标准化以及池化处理，上采样处理包括对所述下采样处理后子区域的反卷积、激活、批标准化以及池化处理。
提取CT图像子区域的特征，其中，该特征包括图像纹理特征，对称特征等，在得到子概率图前，需要对该子区域进行预设次数的卷积、激活、批标准化以及池化处理，其中，卷积过程可以理解为：图像的一部分的统计特性与其他部分是一样的，即是在这一部分学习的特征也能用在相应另一部分上，因而将学习到的特征作为探测器，应用到这个图像的任意地方中去。即通过小范围图像所学习到的特征跟原本的大尺寸图像作卷积，并实现对图像上的任一像素位置获得一个不同特征的激活值，在得到激活值后，进行相应批标准化处理即校正处理，对应校正参数是预存的，在批标准化后，进行池化处理，在本实施例中，可以是最大池化处理即获取最大激活值，以提取局部特征，通过多个局部特征组合即可以得到下采样处理后的所述子区域。
对所述下采样处理后的所述子区域，进行同样预设次数的反卷积、激活、批标准化以及池化处理，得到与所述子区域尺寸相等的子概率图。
其中，反卷积处理即是卷积过程的逆向过程，在此不做具体说明。
在得到子概率图后，对所述各个子概率图进行拼接还原得到所述CT图像对应的概率图，需要说明的是，该各个子概率图是按照预设顺序拼接得到所述CT图像对应的概率图的，该预设顺序与像素分割CT图像中分割各个子区域的分割顺序映射关联。
步骤S20，通过预存的不同三维卷积神经网络分类器对应的各个预测模型对所述候选结节区域进行预测，得到所述候选结节区域对应的各个概率预测值，融合处理所述各个概率预测值，得到所述候选结节区域的分类结果。
需要说明的是，在本实施例中，训练有多个三维卷积神经网络分类器，该不同三维卷积神经网络分类器中的预测模型不同，具体地，可以是通过2个三维卷积神经网络分类器对应的2个预测模型，即采用2个预测模型融合方式对所述候选结节区域进行预测，得到所述候选结节区域对应的各个概率预测值，融合处理所述各个概率预测值，如图4中通过阶段二即state2，由于通过多个预测模型，因而能够消除模型之前的偶然性，提升检测精确度与准确度。在得到各个概率预测值后，通过该各个概率预测值，即可得到所述候选结节区域的分类结果。
具体地，所述通过预存的不同三维卷积神经网络分类器对应的各个预测模型对所述候选结节区域进行预测，得到所述候选结节区域对应的各个概率预测值步骤包括：
步骤S21，通过预存的不同三维卷积神经网络分类器对应的各个预测模型分别对所述候选结节区域进行下采样处理以及全连接处理，得到所述候选结节区域对应的各个概率预测值；
步骤S22，其中，所述对所述候选结节区域进下采样处理步骤包括对所述候选结节区域的卷积、激活、批标准化以及池化处理。
需要说明的是，不同三维卷积神经网络分类器分别对全部的候选结节区域进行预测，三维卷积神经网络分类器包括多个下采样层以及最后的全连接层，以实现对各个候选结节区域进行下采样处理以及全连接处理，其中，对各个候选结节区域进行下采样处理过程包括对所述候选结节区域的卷积、激活、批标准化以及池化处理，全连接处理即是将下采样处理后得到的各个节点连接，以综合处理各个节点对应的图像特征，以最后得到所述候选结节区域对应的各个概率预测值，其中，由于多个三维卷积神经网络分类器分别对所述候选结节区域进行预测，而每个候选结节区域对应每个三维卷积神经网络分类器得到一个概率预测值，因而，每个候选结节区域对应得到多个概率预测值。
其中，所述融合处理所述各个概率预测值，得到所述候选结节区域的分类结果步骤包括：
步骤S23，平均处理所述各个概率预测值，得到所述候选结节区域的目标概率预测值；
步骤S24，将所述目标概率预测值与预存阀值进行比较，并得到比较结果，基于该比较结果，得到所述候选结节区域的分类结果，其中，所述预存阀值根据三维卷积神经网络分类器中对应的ROC曲线确定。
在本实施例中，在得到每个候选结节区域对应得到多个概率预测值后，对该多个概率预测值进行平均处理，并将平均处理后的概率预测值作为候选结节区域的目标概率预测值，在得到目标概率预测值后，获取预存阀值，将目标概率预测值与预存阀值进行比较，得到比较结果，需要说明的是，预存阀值是可以调整的，具体地，预存阀值根据三维卷积神经网络分类器中不同模型整体的ROC曲线确定。在得到比较结果，基于该比较结果，即可得到所述候选结节区域的分类结果，该分类结果包括结节区域与非结节区域。
需要说明的是，在得到目标概率预测值后，可以对目标概率预测值进行输出处理，并生成目标提示信息，该目标提示信息对应提示方式并不做具体限定。
本发明通过获取待检测的电子计算机断层扫描CT图像，通过预存的三维卷积神经像素分割网络对所述CT图像进行像素分割处理，得到所述CT图像对应的概率图，对所述概率图进行连通域标记得到候选结节区域；通过预存的不同三维卷积神经网络分类器对应的各个预测模型对所述候选结节区域进行预测处理，得到所述候选结节区域对应的各个概率预测值，融合处理所述各个概率预测值，得到所述候选结节区域的分类结果。在本发明中，三维卷积神经像素分割网络以及三维卷积神经网络分类器是已经训练完成的，具有一定准确度的处理网络，在第一阶段通过三维卷积神经网络得到候选结节区域，在第二阶段，通过多模型融合，补偿不同模型的偶然性，得到候选结节区域的目标概率预测值，因而避免了由于肺结节的大小、形状以及密度等特征的不确定性造成肺结节检测精确度低的现象，解决了现有检测方法难以满足市场对肺结节检测精确度的需求的技术问题。
进一步地，本发明提供CT结节检测方法的另一实施例，在该实施例中，所述获取待检测电子计算机断层扫描CT图像，通过三维卷积神经像素分割网络对所述CT图像进行预测步骤之前包括：
步骤S01，线上多进程采集CT图像样本数据，将所述采集的CT图像样本数据作为第一样本数据，将所述第一样本数据作为输入数据以进行对应三维卷积神经像素分割网络以及三维卷积神经网络分类器的训练，并得到难分样本数据，其中，所述难分样本数据是第一样本数据中预测失败的数据；
步骤S02，对所述难分样本数据进行多次平移和水平翻转操作，以增大难分样本数据在所述第一样本数据中的比例，并将增大难分样本数据比例后的数据作为第二样本数据；
步骤S03，将所述第二样本数据作为输入数据以多次进行对应三维卷积神经像素分割网络以及三维卷积神经网络分类器的训练；
步骤S04，当训练后的所述三维卷积神经像素分割网络以及三维卷积神经网络分类器的预测准确率分别达到第一目标准确率、第二目标准确率时，执行获取待检测电子计算机断层扫描CT图像的步骤。
在本实施例中，在三维卷积神经像素分割网络以及三维卷积神经网络分类器训练过程中，不断进行线上多进程采集CT图像样本数据即第一样本数据，如图4所示，在第一样本数据作为输入数据进行对应三维卷积神经像素分割网络以及三维卷积神经网络分类器的训练后，通过平移和水平翻转等操作实现难分样本的数据增强，以实现第二样本数据中难分样本数据比例增大，因而，在将所述第二样本数据作为输入数据再次进行相应三维卷积神经网络模型的训练过程中，能够提高所述维卷积神经网络模型的敏感率。
参照图3，图3是本发明实施例方案涉及的硬件运行环境的设备结构示意图。
本发明实施例CT图像肺结节检测设备可以是PC，便携计算机，也可以是移动终端等终端设备。
如图3所示，该CT图像肺结节检测设备可以包括：处理器1001，例如CPU，存储器1005，通信总线1002。其中，通信总线1002用于实现处理器1001和存储器1005之间的连接通信。存储器1005可以是高速RAM存储器，也可以是稳定的存储器（non-volatile memory），例如磁盘存储器。存储器1005可选的还可以是独立于前述处理器1001的存储设备。
可选地，该CT图像肺结节检测设备还可以包括用户接口、网络接口、摄像头、RF（Radio Frequency，射频）电路，传感器、音频电路、WiFi模块等等。用户接口可以包括显示屏（Display）、输入单元比如键盘（Keyboard），可选用户接口还可以包括标准的有线接口、无线接口。网络接口可选的可以包括标准的有线接口、无线接口（如WI-FI接口）。
本领域技术人员可以理解，图3中示出的CT图像肺结节检测设备结构并不构成对CT图像肺结节检测设备的限定，可以包括比图示更多或更少的部件，或者组合某些部件，或者不同的部件布置。
如图3所示，作为一种计算机存储介质的存储器1005中可以包括操作系统、网络通信模块以及CT图像肺结节检测程序。操作系统是管理和控制CT图像肺结节检测设备硬件和软件资源的程序，支持CT图像肺结节检测程序以及其它软件和/或程序的运行。网络通信模块用于实现存储器1005内部各组件之间的通信，以及与CT图像肺结节检测设备中其它硬件和软件之间通信。
在图3所示的CT图像肺结节检测设备中，处理器1001用于执行存储器1005中存储的CT图像肺结节检测程序，实现上述任一项所述的CT图像肺结节检测方法的步骤。
本发明CT图像肺结节检测设备具体实施方式与上述CT图像肺结节检测方法各实施例基本相同，在此不再赘述。
本发明还提供一种CT图像肺结节检测装置，所述CT图像肺结节检测装置包括：
获取模块，用于获取待检测的电子计算机断层扫描CT图像，通过预存的三维卷积神经像素分割网络对所述CT图像进行像素分割处理，得到所述CT图像对应的概率图，对所述概率图进行连通域标记得到候选结节区域；
模型融合模块，用于通过预存的不同三维卷积神经网络分类器对应的各个预测模型对所述候选结节区域进行预测，得到所述候选结节区域对应的各个概率预测值，融合处理所述各个概率预测值，得到所述候选结节区域的分类结果。
本发明CT图像肺结节检测装置具体实施方式与上述CT图像肺结节检测方法各实施例基本相同，在此不再赘述。
本发明提供了一种可读存储介质，所述可读存储介质存储有一个或者一个以上程序，所述一个或者一个以上程序还可被一个或者一个以上的处理器执行以用于实现上述任一项所述的CT图像肺结节检测方法的步骤。
本发明可读存储介质具体实施方式与上述CT图像肺结节检测方法各实施例基本相同，在此不再赘述。
以上仅为本发明的优选实施例，并非因此限制本发明的专利范围，凡是利用本发明说明书及附图内容所作的等效结构或等效流程变换，或直接或间接运用在其他相关的技术领域，均同理包括在本发明的专利处理范围内。

Claims (19)

  1. 一种CT图像肺结节检测方法，其特征在于，所述CT图像肺结节检测方法包括：
    获取待检测的电子计算机断层扫描CT图像，通过预存的三维卷积神经像素分割网络对所述CT图像进行像素分割处理，得到所述CT图像对应的概率图，对所述概率图进行连通域标记得到候选结节区域；
    通过预存的不同三维卷积神经网络分类器对应的各个预测模型对所述候选结节区域进行预测，得到所述候选结节区域对应的各个概率预测值，融合处理所述各个概率预测值，得到所述候选结节区域的分类结果。
  2. 如权利要求1所述的CT图像肺结节检测方法，其特征在于，
    所述通过预存的三维卷积神经像素分割网络对所述CT图像进行像素分割处理，得到所述CT图像对应的概率图，对所述概率图进行连通域标记得到候选结节区域步骤包括：
    对所述CT图像进行预设尺寸的区域分割处理，得到CT图像的子区域，通过预存的三维像素分割网络对所述子区域分别进行预设次数的下采样处理，对下采样处理后的所述子区域进行同样预设次数的上采样处理；
    对所述下采样处理后以及上采样处理后分别得到的所述子区域进行桥接特征融合处理，得到与所述子区域尺寸相等的子概率图；
    对所述各个子概率图进行拼接还原得到所述CT图像对应的概率图；
    对所述CT图像对应的概率图进行连通域标记，得到候选结节区域。
  3. 如权利要求2所述的CT图像肺结节检测方法，其特征在于，所述融合处理所述各个概率预测值，得到所述候选结节区域的分类结果步骤包括：
    平均处理所述各个概率预测值，得到所述候选结节区域的目标概率预测值；
    将所述目标概率预测值与预存阀值进行比较，并得到比较结果，基于该比较结果，得到所述候选结节区域的分类结果，其中，所述预存阀值根据对应三维卷积神经网络分类器中对应的ROC曲线确定。
  4. 如权利要求2所述的CT图像肺结节检测方法，其特征在于，所述通过预存的不同三维卷积神经网络分类器对应的各个预测模型对所述候选结节区域进行预测，得到所述候选结节区域对应的各个概率预测值步骤包括：
    通过预存的不同三维卷积神经网络分类器对应的各个预测模型分别对所述候选结节区域进行下采样处理以及全连接处理，得到所述候选结节区域对应的各个概率预测值；
    其中，所述对所述候选结节区域进下采样处理步骤包括对所述候选结节区域的卷积、激活、批标准化以及池化处理。
  5. 如权利要求2所述的CT图像肺结节检测方法，其特征在于，所述获取待检测电子计算机断层扫描CT图像，通过三维卷积神经像素分割网络对所述CT图像进行预测步骤之前包括：
    线上多进程采集CT图像样本数据，将所述采集的CT图像样本数据作为第一样本数据，将所述第一样本数据作为输入数据以进行对应三维卷积神经像素分割网络以及三维卷积神经网络分类器的训练，并得到难分样本数据，其中，所述难分样本数据是第一样本数据中预测失败的数据；
    对所述难分样本数据进行多次平移和水平翻转操作，以增大难分样本数据在所述第一样本数据中的比例，并将增大难分样本数据比例后的数据作为第二样本数据；
    将所述第二样本数据作为输入数据以多次进行对应三维卷积神经像素分割网络以及三维卷积神经网络分类器的训练；
    当训练后的所述三维卷积神经像素分割网络以及三维卷积神经网络分类器的预测准确率分别达到第一目标准确率、第二目标准确率时，执行获取待检测电子计算机断层扫描CT图像的步骤。
  6. 如权利要求2所述的CT图像肺结节检测方法，其特征在于，所述下采样处理包括对所述子区域的卷积、激活、批标准化以及池化处理，上采样处理包括对所述下采样处理后子区域的反卷积、激活、批标准化以及池化处理。
  7. 如权利要求6所述的CT图像肺结节检测方法，其特征在于，所述融合处理所述各个概率预测值，得到所述候选结节区域的分类结果步骤包括：
    平均处理所述各个概率预测值，得到所述候选结节区域的目标概率预测值；
    将所述目标概率预测值与预存阀值进行比较，并得到比较结果，基于该比较结果，得到所述候选结节区域的分类结果，其中，所述预存阀值根据对应三维卷积神经网络分类器中对应的ROC曲线确定。
  8. 如权利要求6所述的CT图像肺结节检测方法，其特征在于，所述通过预存的不同三维卷积神经网络分类器对应的各个预测模型对所述候选结节区域进行预测，得到所述候选结节区域对应的各个概率预测值步骤包括：
    通过预存的不同三维卷积神经网络分类器对应的各个预测模型分别对所述候选结节区域进行下采样处理以及全连接处理，得到所述候选结节区域对应的各个概率预测值；
    其中，所述对所述候选结节区域进下采样处理步骤包括对所述候选结节区域的卷积、激活、批标准化以及池化处理。
  9. 如权利要求6所述的CT图像肺结节检测方法，其特征在于，所述获取待检测电子计算机断层扫描CT图像，通过三维卷积神经像素分割网络对所述CT图像进行预测步骤之前包括：
    线上多进程采集CT图像样本数据，将所述采集的CT图像样本数据作为第一样本数据，将所述第一样本数据作为输入数据以进行对应三维卷积神经像素分割网络以及三维卷积神经网络分类器的训练，并得到难分样本数据，其中，所述难分样本数据是第一样本数据中预测失败的数据；
    对所述难分样本数据进行多次平移和水平翻转操作，以增大难分样本数据在所述第一样本数据中的比例，并将增大难分样本数据比例后的数据作为第二样本数据；
    将所述第二样本数据作为输入数据以多次进行对应三维卷积神经像素分割网络以及三维卷积神经网络分类器的训练；
    当训练后的所述三维卷积神经像素分割网络以及三维卷积神经网络分类器的预测准确率分别达到第一目标准确率、第二目标准确率时，执行获取待检测电子计算机断层扫描CT图像的步骤。
  10. 如权利要求1所述的CT图像肺结节检测方法，其特征在于，所述融合处理所述各个概率预测值，得到所述候选结节区域的分类结果步骤包括：
    平均处理所述各个概率预测值，得到所述候选结节区域的目标概率预测值；
    将所述目标概率预测值与预存阀值进行比较，并得到比较结果，基于该比较结果，得到所述候选结节区域的分类结果，其中，所述预存阀值根据对应三维卷积神经网络分类器中对应的ROC曲线确定。
  11. 如权利要求10所述的CT图像肺结节检测方法，其特征在于，所述通过预存的不同三维卷积神经网络分类器对应的各个预测模型对所述候选结节区域进行预测，得到所述候选结节区域对应的各个概率预测值步骤包括：
    通过预存的不同三维卷积神经网络分类器对应的各个预测模型分别对所述候选结节区域进行下采样处理以及全连接处理，得到所述候选结节区域对应的各个概率预测值；
    其中，所述对所述候选结节区域进下采样处理步骤包括对所述候选结节区域的卷积、激活、批标准化以及池化处理。
  12. 如权利要求10所述的CT图像肺结节检测方法，其特征在于，所述获取待检测电子计算机断层扫描CT图像，通过三维卷积神经像素分割网络对所述CT图像进行预测步骤之前包括：
    线上多进程采集CT图像样本数据，将所述采集的CT图像样本数据作为第一样本数据，将所述第一样本数据作为输入数据以进行对应三维卷积神经像素分割网络以及三维卷积神经网络分类器的训练，并得到难分样本数据，其中，所述难分样本数据是第一样本数据中预测失败的数据；
    对所述难分样本数据进行多次平移和水平翻转操作，以增大难分样本数据在所述第一样本数据中的比例，并将增大难分样本数据比例后的数据作为第二样本数据；
    将所述第二样本数据作为输入数据以多次进行对应三维卷积神经像素分割网络以及三维卷积神经网络分类器的训练；
    当训练后的所述三维卷积神经像素分割网络以及三维卷积神经网络分类器的预测准确率分别达到第一目标准确率、第二目标准确率时，执行获取待检测电子计算机断层扫描CT图像的步骤。
  13. 如权利要求1所述的CT图像肺结节检测方法，其特征在于，所述通过预存的不同三维卷积神经网络分类器对应的各个预测模型对所述候选结节区域进行预测，得到所述候选结节区域对应的各个概率预测值步骤包括：
    通过预存的不同三维卷积神经网络分类器对应的各个预测模型分别对所述候选结节区域进行下采样处理以及全连接处理，得到所述候选结节区域对应的各个概率预测值；
    其中，所述对所述候选结节区域进下采样处理步骤包括对所述候选结节区域的卷积、激活、批标准化以及池化处理。
  14. 如权利要求13所述的CT图像肺结节检测方法，其特征在于，所述获取待检测电子计算机断层扫描CT图像，通过三维卷积神经像素分割网络对所述CT图像进行预测步骤之前包括：
    线上多进程采集CT图像样本数据，将所述采集的CT图像样本数据作为第一样本数据，将所述第一样本数据作为输入数据以进行对应三维卷积神经像素分割网络以及三维卷积神经网络分类器的训练，并得到难分样本数据，其中，所述难分样本数据是第一样本数据中预测失败的数据；
    对所述难分样本数据进行多次平移和水平翻转操作，以增大难分样本数据在所述第一样本数据中的比例，并将增大难分样本数据比例后的数据作为第二样本数据；
    将所述第二样本数据作为输入数据以多次进行对应三维卷积神经像素分割网络以及三维卷积神经网络分类器的训练；
    当训练后的所述三维卷积神经像素分割网络以及三维卷积神经网络分类器的预测准确率分别达到第一目标准确率、第二目标准确率时，执行获取待检测电子计算机断层扫描CT图像的步骤。
  15. 如权利要求1所述的CT图像肺结节检测方法，其特征在于，所述得到所述候选结节区域的目标概率预测值步骤之后包括：
    输出所述目标概率预测值，并生成目标提示信息。
  16. 如权利要求15所述的CT图像肺结节检测方法，其特征在于，所述获取待检测电子计算机断层扫描CT图像，通过三维卷积神经像素分割网络对所述CT图像进行预测步骤之前包括：
    线上多进程采集CT图像样本数据，将所述采集的CT图像样本数据作为第一样本数据，将所述第一样本数据作为输入数据以进行对应三维卷积神经像素分割网络以及三维卷积神经网络分类器的训练，并得到难分样本数据，其中，所述难分样本数据是第一样本数据中预测失败的数据；
    对所述难分样本数据进行多次平移和水平翻转操作，以增大难分样本数据在所述第一样本数据中的比例，并将增大难分样本数据比例后的数据作为第二样本数据；
    将所述第二样本数据作为输入数据以多次进行对应三维卷积神经像素分割网络以及三维卷积神经网络分类器的训练；
    当训练后的所述三维卷积神经像素分割网络以及三维卷积神经网络分类器的预测准确率分别达到第一目标准确率、第二目标准确率时，执行获取待检测电子计算机断层扫描CT图像的步骤。
  17. 一种CT图像肺结节检测装置，其特征在于，所述CT图像肺结节检测装置包括：
    获取模块，用于获取待检测的电子计算机断层扫描CT图像，通过预存的三维卷积神经像素分割网络对所述CT图像进行像素分割处理，得到所述CT图像对应的概率图，对所述概率图进行连通域标记得到候选结节区域；
    模型融合模块，用于通过预存的不同三维卷积神经网络分类器对应的各个预测模型对所述候选结节区域进行预测，得到所述候选结节区域对应的各个概率预测值，融合处理所述各个概率预测值，得到所述候选结节区域的分类结果。
  18. 一种CT图像肺结节检测设备，其特征在于，所述CT图像肺结节检测设备包括：存储器、处理器，通信总线以及存储在所述存储器上的CT图像肺结节检测程序，
    所述通信总线用于实现处理器与存储器间的通信连接；
    所述处理器用于执行所述CT图像肺结节检测程序，以实现如权利要求1所述的CT图像肺结节检测方法的步骤。
  19. 一种可读存储介质，其特征在于，所述可读存储介质上存储有CT图像肺结节检测程序，所述CT图像肺结节检测程序被处理器执行时实现如权利要求1所述的CT图像肺结节检测方法的步骤。
PCT/CN2018/104189 2018-06-28 2018-09-05 Ct图像肺结节检测方法、设备及可读存储介质 WO2020000643A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201810695624.8 2018-06-28
CN201810695624.8A CN109003260B (zh) 2018-06-28 2018-06-28 Ct图像肺结节检测方法、装置、设备及可读存储介质

Publications (1)

Publication Number Publication Date
WO2020000643A1 true WO2020000643A1 (zh) 2020-01-02

Family

ID=64602134

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2018/104189 WO2020000643A1 (zh) 2018-06-28 2018-09-05 Ct图像肺结节检测方法、设备及可读存储介质

Country Status (3)

Country Link
US (1) US10909682B2 (zh)
CN (1) CN109003260B (zh)
WO (1) WO2020000643A1 (zh)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112966609A (zh) * 2021-03-05 2021-06-15 北京百度网讯科技有限公司 目标检测方法及装置

Families Citing this family (72)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111402191B (zh) * 2018-12-28 2023-04-18 阿里巴巴集团控股有限公司 一种目标检测方法、装置、计算设备及介质
CN109961423A (zh) * 2019-02-15 2019-07-02 平安科技（深圳）有限公司 一种基于分类模型的肺结节检测方法、服务器及存储介质
CN109978004B (zh) * 2019-02-21 2024-03-29 平安科技（深圳）有限公司 图像识别方法及相关设备
CN109961446B (zh) * 2019-03-27 2021-06-01 深圳视见医疗科技有限公司 Ct/mr三维图像分割处理方法、装置、设备及介质
CN110141776A (zh) * 2019-04-02 2019-08-20 成都真实维度科技有限公司 瘤体内放射性粒子源线性填充方法
CN110136096A (zh) * 2019-04-02 2019-08-16 成都真实维度科技有限公司 一种基于断层扫描图数据集进行病变区域分割的方法
CN110097564B (zh) * 2019-04-04 2023-06-16 平安科技（深圳）有限公司 基于多模型融合的图像标注方法、装置、计算机设备及存储介质
CN110148192B (zh) * 2019-04-18 2023-05-30 上海联影智能医疗科技有限公司 医学影像成像方法、装置、计算机设备和存储介质
CN110084810B (zh) * 2019-05-06 2021-11-05 成都医云科技有限公司 一种肺结节图像检测方法、模型训练方法、装置及存储介质
CN110223300A (zh) * 2019-06-13 2019-09-10 北京理工大学 Ct图像腹部多器官分割方法及装置
CN110600105B (zh) * 2019-08-27 2022-02-01 武汉科技大学 一种ct影像数据处理方法、装置及存储介质
CN110689112A (zh) * 2019-09-02 2020-01-14 杭州深睿博联科技有限公司 数据处理的方法及装置
CN110634144B (zh) * 2019-09-23 2022-08-02 武汉联影医疗科技有限公司 一种卵圆孔定位方法、装置和存储介质
CN110766659A (zh) * 2019-09-24 2020-02-07 西人马帝言(北京)科技有限公司 医学图像识别方法、装置、设备和介质
CN110866925B (zh) * 2019-10-18 2023-05-26 拜耳股份有限公司 一种用于图像分割的方法和装置
CN110807788B (zh) * 2019-10-21 2023-07-21 腾讯科技（深圳）有限公司 医学图像处理方法、装置、电子设备及计算机存储介质
CN110728675A (zh) * 2019-10-22 2020-01-24 慧影医疗科技（北京）有限公司 肺结节分析装置、模型训练方法、装置及分析设备
CN110992312B (zh) * 2019-11-15 2024-02-27 上海联影智能医疗科技有限公司 医学图像处理方法、装置、存储介质及计算机设备
CN110991535A (zh) * 2019-12-04 2020-04-10 中山大学 一种基于多类型医学数据的pCR预测方法
US11684333B2 (en) 2020-01-03 2023-06-27 National Taiwan University Medical image analyzing system and method thereof
TWI767439B (zh) * 2020-01-03 2022-06-11 國立臺灣大學 醫療影像分析系統及其方法
CN111292339B (zh) * 2020-01-22 2023-01-10 北京航空航天大学 基于w型网络结构的临床颞骨ct多结构提取方法及装置
CN111311578B (zh) * 2020-02-17 2024-05-03 腾讯科技（深圳）有限公司 基于人工智能的对象分类方法以及装置、医学影像设备
US11113577B1 (en) * 2020-02-27 2021-09-07 GE Precision Healthcare LLC Systems and methods for detecting laterality of a medical image
CN111402219A (zh) * 2020-03-11 2020-07-10 北京深睿博联科技有限责任公司 陈旧性脑梗死检测方法和装置
CN111445478B (zh) * 2020-03-18 2023-09-08 吉林大学 一种用于cta图像的颅内动脉瘤区域自动检测系统和检测方法
CN111445449B (zh) * 2020-03-19 2024-03-01 上海联影智能医疗科技有限公司 感兴趣区域的分类方法、装置、计算机设备和存储介质
CN111583179B (zh) * 2020-04-02 2022-09-16 天津大学 一种基于浮动切割的肺结节深度学习分类方法
CN111462101A (zh) * 2020-04-07 2020-07-28 广州柏视医疗科技有限公司 基于新型冠状病毒肺炎ct检测的分期设备及其使用方法
CN111583188B (zh) * 2020-04-15 2023-12-26 武汉联影智融医疗科技有限公司 手术导航标记点定位方法、存储介质及计算机设备
CN111553892B (zh) * 2020-04-23 2021-11-05 北京小白世纪网络科技有限公司 基于深度学习的肺结节分割计算方法、装置及系统
CN111626257B (zh) * 2020-06-03 2023-06-06 兰波(苏州)智能科技有限公司 硅藻检测识别用多概率组合预测后处理方法、系统及存储介质
CN111652217B (zh) * 2020-06-03 2022-05-03 北京易真学思教育科技有限公司 文本检测方法、装置、电子设备及计算机存储介质
CN111784646A (zh) * 2020-06-18 2020-10-16 北京智像科技有限公司 图像处理装置及电子设备
CN112001218B (zh) * 2020-06-19 2023-05-30 中国科学院计算技术研究所 一种基于卷积神经网络的三维颗粒类别检测方法及系统
CN111815579B (zh) * 2020-06-24 2024-04-26 浙江大华技术股份有限公司 一种图像变化检测方法、装置及计算机可读存储介质
CN111862001B (zh) * 2020-06-28 2023-11-28 微医云(杭州)控股有限公司 Ct影像的半自动标注方法及装置、电子设备、存储介质
CN111767952B (zh) * 2020-06-30 2024-03-29 重庆大学 一种可解释的肺结节良恶性分类方法
CN111915568B (zh) * 2020-07-08 2023-07-25 深圳大学 图像篡改定位模型的生成方法、图像篡改定位方法及设备
CN111798437A (zh) * 2020-07-09 2020-10-20 兴义民族师范学院 一种基于ct影像的新型冠状病毒肺炎ai快速诊断方法
CN112053319B (zh) * 2020-07-22 2022-12-02 清华大学 图像处理的方法和装置
CN111915698A (zh) * 2020-08-21 2020-11-10 南方科技大学 一种脉管浸润检测方法、装置、计算机设备及存储介质
CN112085714B (zh) * 2020-08-31 2023-12-15 广州视源电子科技股份有限公司 一种肺结节检测方法、模型训练方法、装置、设备及介质
CN112102266B (zh) * 2020-08-31 2022-09-20 清华大学 基于注意力机制的脑梗死医学影像分类模型的训练方法
CN111932482B (zh) * 2020-09-25 2021-05-18 平安科技（深圳）有限公司 图像中目标物的检测方法、装置、电子设备及存储介质
CN112102311B (zh) * 2020-09-27 2023-07-18 平安科技（深圳）有限公司 甲状腺结节图像的处理方法、装置和计算机设备
CN112348774A (zh) * 2020-09-29 2021-02-09 深圳市罗湖区人民医院 一种适用于膀胱癌的ct影像分割方法、终端以及存储介质
CN112102326B (zh) * 2020-10-26 2023-11-07 北京航星机器制造有限公司 一种安检ct图像目标物的提取和分割方法
CN112419306B (zh) * 2020-12-11 2024-03-15 长春工业大学 一种基于nas-fpn的肺结节检测方法
CN112669312A (zh) * 2021-01-12 2021-04-16 中国计量大学 一种基于深度特征对称融合的胸片肺炎检测方法及系统
CN112785565B (zh) * 2021-01-15 2024-01-05 上海商汤智能科技有限公司 目标检测方法及装置、电子设备和存储介质
CN113017670A (zh) * 2021-03-01 2021-06-25 北京小白世纪网络科技有限公司 一种基于3D UNet的纵隔肿物识别方法、装置及存储介质
CN112927210A (zh) * 2021-03-08 2021-06-08 常州市第一人民医院 一种能够定量分析肾表面结节的量化方法
CN113077419A (zh) * 2021-03-19 2021-07-06 零氪智慧医疗科技(天津)有限公司 用于髋关节ct影像识别的信息处理方法及装置
CN112967279A (zh) * 2021-04-02 2021-06-15 慧影医疗科技（北京）有限公司 一种检测肺结节的方法、装置、存储介质和电子设备
CN112949767B (zh) * 2021-04-07 2023-08-11 北京百度网讯科技有限公司 样本图像增量、图像检测模型训练及图像检测方法
CN113269672B (zh) * 2021-04-14 2023-02-24 佛山科学技术学院 一种超分辨率的细胞图像构建方法和系统
CN113240666B (zh) * 2021-06-04 2024-04-16 科大讯飞股份有限公司 一种医学影像预处理方法、装置、设备及存储介质
CN113408595B (zh) * 2021-06-09 2022-12-13 北京小白世纪网络科技有限公司 病理图像处理的方法、装置、电子设备及可读存储介质
CN113506289B (zh) * 2021-07-28 2024-03-29 中山仰视科技有限公司 一种利用双流网络进行肺结节假阳性分类的方法
CN113470026B (zh) * 2021-09-02 2021-11-05 北京字节跳动网络技术有限公司 息肉识别方法、装置、介质及设备
CN113947613B (zh) * 2021-12-21 2022-03-29 腾讯科技（深圳）有限公司 目标区域检测方法、装置、设备及存储介质
CN114387227A (zh) * 2021-12-23 2022-04-22 沈阳东软智能医疗科技研究院有限公司 结节类型预测方法、装置、存储介质及电子设备
CN114004425B (zh) * 2021-12-29 2022-06-07 北京京东振世信息技术有限公司 物品流转信息预测模型生成方法、信息生成方法、装置
CN114677383B (zh) * 2022-03-03 2024-03-15 西北工业大学 基于多任务学习的肺结节检测分割方法
CN114820535B (zh) * 2022-05-05 2023-09-12 深圳市铱硙医疗科技有限公司 动脉瘤的图像检测方法、装置、计算机设备及存储介质
CN114757943B (zh) * 2022-06-10 2022-09-27 肺诊网(苏州)网络科技有限公司 一种数字影像人工智能分析方法和系统
CN115272206B (zh) * 2022-07-18 2023-07-04 深圳市医未医疗科技有限公司 医学图像处理方法、装置、计算机设备及存储介质
CN116310513B (zh) * 2023-02-14 2023-12-05 成都泰莱生物科技有限公司 基于肺部CT与5hmC标志物融合的肺结节分类方法及产品
CN116228745A (zh) * 2023-04-24 2023-06-06 徐州医科大学 一种用于肺结节良恶性分类的深度卷积网络构建方法
CN117237359B (zh) * 2023-11-15 2024-02-20 天津市恒一机电科技有限公司 传送带撕裂检测方法、装置、存储介质和电子设备
CN117830302A (zh) * 2024-03-04 2024-04-05 瀚依科技(杭州)有限公司 肺段分割的优化方法及装置、电子设备和存储介质

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN105976367A (zh) * 2016-04-29 2016-09-28 上海联影医疗科技有限公司 图像分割方法、肺结节检测方法及其计算机辅助检测系统
CN106780460A (zh) * 2016-12-13 2017-05-31 杭州健培科技有限公司 一种用于胸部ct影像的肺结节自动检测系统
CN106940816A (zh) * 2017-03-22 2017-07-11 杭州健培科技有限公司 基于3d全连接卷积神经网络的ct图像肺结节检测系统
CN107480677A (zh) * 2017-08-07 2017-12-15 北京深睿博联科技有限责任公司 一种识别三维ct图像中感兴趣区域的方法及装置
US20180089840A1 (en) * 2016-09-26 2018-03-29 Siemens Medical Solutions Usa, Inc. Automatic Medical Image Retrieval
CN108010021A (zh) * 2017-11-30 2018-05-08 上海联影医疗科技有限公司 一种医学图像处理系统和方法

Family Cites Families (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20090092299A1 (en) * 2007-10-03 2009-04-09 Siemens Medical Solutions Usa, Inc. System and Method for Joint Classification Using Feature Space Cluster Labels
CN106650830A (zh) * 2017-01-06 2017-05-10 西北工业大学 基于深层模型和浅层模型决策融合的肺结节ct图像自动分类方法
US10692211B2 (en) * 2017-06-20 2020-06-23 Case Western Reserve University Intra-perinodular textural transition (IPRIS): a three dimenisonal (3D) descriptor for nodule diagnosis on lung computed tomography (CT) images
CN107590797B (zh) * 2017-07-26 2020-10-30 浙江工业大学 一种基于三维残差神经网络的ct影像肺结节检测方法
US11341631B2 (en) * 2017-08-09 2022-05-24 Shenzhen Keya Medical Technology Corporation System and method for automatically detecting a physiological condition from a medical image of a patient
CN108257128B (zh) * 2018-01-30 2020-09-18 浙江大学 一种基于3d卷积神经网络的肺结节检测装置的建立方法

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN105976367A (zh) * 2016-04-29 2016-09-28 上海联影医疗科技有限公司 图像分割方法、肺结节检测方法及其计算机辅助检测系统
US20180089840A1 (en) * 2016-09-26 2018-03-29 Siemens Medical Solutions Usa, Inc. Automatic Medical Image Retrieval
CN106780460A (zh) * 2016-12-13 2017-05-31 杭州健培科技有限公司 一种用于胸部ct影像的肺结节自动检测系统
CN106940816A (zh) * 2017-03-22 2017-07-11 杭州健培科技有限公司 基于3d全连接卷积神经网络的ct图像肺结节检测系统
CN107480677A (zh) * 2017-08-07 2017-12-15 北京深睿博联科技有限责任公司 一种识别三维ct图像中感兴趣区域的方法及装置
CN108010021A (zh) * 2017-11-30 2018-05-08 上海联影医疗科技有限公司 一种医学图像处理系统和方法

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112966609A (zh) * 2021-03-05 2021-06-15 北京百度网讯科技有限公司 目标检测方法及装置
CN112966609B (zh) * 2021-03-05 2023-08-11 北京百度网讯科技有限公司 目标检测方法及装置

Also Published As

Publication number Publication date
US20200005460A1 (en) 2020-01-02
CN109003260B (zh) 2021-02-09
US10909682B2 (en) 2021-02-02
CN109003260A (zh) 2018-12-14

Similar Documents

Publication Publication Date Title
WO2020000643A1 (zh) Ct图像肺结节检测方法、设备及可读存储介质
WO2021012508A1 (zh) Ai影像识别方法、装置、设备及存储介质
WO2018135881A1 (en) Vision intelligence management for electronic devices
WO2017010695A1 (en) Three dimensional content generating apparatus and three dimensional content generating method thereof
WO2016171341A1 (ko) 클라우드 기반 병리 분석 시스템 및 방법
TWI748242B (zh) 掃描晶圓的系統及方法
WO2019164232A1 (ko) 전자 장치, 이의 영상 처리 방법 및 컴퓨터 판독가능 기록 매체
JP7146857B2 (ja) ディープラーニングに基づきウェーハ欠陥画像を用いてウェーハの欠陥を分類するための方法およびシステム
WO2019050360A1 (en) ELECTRONIC DEVICE AND METHOD FOR AUTOMATICALLY SEGMENTING TO BE HUMAN IN AN IMAGE
WO2013015528A1 (en) Apparatus, method, and medium detecting object pose
WO2020253048A1 (zh) 基于深度学习的影像识别方法、装置、设备及存储介质
WO2021107610A1 (en) Method and system for generating a tri-map for image matting
WO2016027983A1 (en) Method and electronic device for classifying contents
WO2020139009A1 (ko) 뇌혈관 질환 학습 장치, 뇌혈관 질환 검출 장치, 뇌혈관 질환 학습 방법 및 뇌혈관 질환 검출 방법
WO2020076133A1 (ko) 암 영역 검출의 유효성 평가 장치
Althubiti et al. Circuit manufacturing defect detection using VGG16 convolutional neural networks
WO2020019671A1 (zh) 一种乳腺肿块检测与分类系统、计算机可读存储介质
WO2020017875A1 (en) Electronic apparatus, method for processing image and computer-readable recording medium
WO2020207038A1 (zh) 基于人脸识别的人数统计方法、装置、设备及存储介质
WO2019009664A1 (en) APPARATUS FOR OPTIMIZING THE INSPECTION OF THE OUTSIDE OF A TARGET OBJECT AND ASSOCIATED METHOD
WO2020258672A1 (zh) 网络访问的异常检测方法和装置
WO2020114184A1 (zh) 联合建模方法、装置、设备以及计算机可读存储介质
WO2020080734A1 (ko) 얼굴 인식 방법 및 얼굴 인식 장치
WO2017099292A1 (ko) 물체-행동 관계 모델에 기반한 행동 인식 방법 및 그 장치
JP4074624B2 (ja) パターン検査方法

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 18924784

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 18924784

Country of ref document: EP

Kind code of ref document: A1