CN109003672A - A kind of early stage of lung cancer detection classification integration apparatus and system based on deep learning - Google Patents

A kind of early stage of lung cancer detection classification integration apparatus and system based on deep learning Download PDF

Info

Publication number
CN109003672A
CN109003672A CN201810776622.1A CN201810776622A CN109003672A CN 109003672 A CN109003672 A CN 109003672A CN 201810776622 A CN201810776622 A CN 201810776622A CN 109003672 A CN109003672 A CN 109003672A
Authority
CN
China
Prior art keywords
lung
detection
lung cancer
tnet
cell carcinoma
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.)
Pending
Application number
CN201810776622.1A
Other languages
Chinese (zh)
Inventor
邹伟
高长龙
王燕妮
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Beijing Ruikebang Technology Co Ltd
Original Assignee
Beijing Ruikebang Technology Co Ltd
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 Beijing Ruikebang Technology Co Ltd filed Critical Beijing Ruikebang Technology Co Ltd
Priority to CN201810776622.1A priority Critical patent/CN109003672A/en
Publication of CN109003672A publication Critical patent/CN109003672A/en
Pending legal-status Critical Current

Links

Classifications

    • 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
    • G16H50/20—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics for computer-aided diagnosis, e.g. based on medical expert systems
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/0002—Inspection of images, e.g. flaw detection
    • G06T7/0012—Biomedical image inspection
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/10—Segmentation; Edge detection
    • 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
    • G16H50/70—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics for mining of medical data, e.g. analysing previous cases of other patients
    • 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
    • G06T2207/30—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
    • 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
    • G06T2207/30096—Tumor; Lesion

Landscapes

  • Engineering & Computer Science (AREA)
  • Health & Medical Sciences (AREA)
  • Medical Informatics (AREA)
  • Public Health (AREA)
  • Biomedical Technology (AREA)
  • Data Mining & Analysis (AREA)
  • General Health & Medical Sciences (AREA)
  • Epidemiology (AREA)
  • Pathology (AREA)
  • Databases & Information Systems (AREA)
  • Primary Health Care (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
  • Radiology & Medical Imaging (AREA)
  • Quality & Reliability (AREA)
  • Apparatus For Radiation Diagnosis (AREA)

Abstract

The invention discloses a kind of, and the early stage of lung cancer based on deep learning detects classification integration apparatus and system Cancer Early Test System；Equipment is stored including early stage of lung cancer detection touch control terminal, lung cancer detection server, lung cancer detection case；CETS is based on U-NET and devises TNet, for the early stage of lung cancer and Lung neoplasm detection classification；Patients with Lung CT images data are processed into the readable 3D lung image of CETS by data preprocessing module, are carried out semantic segmentation to lung image using TNet Lung neoplasm 3D detection sorter network, are exported Lung neoplasm region and probability by grade malignancy；Sorter network is detected by TNet lung cancer 3D to analyze the feature of Lung neoplasm, finally obtains the details examining reports such as lung cancer type, probability and Lung neoplasm type, form；CETS is firstly introduced light stream tracing algorithm in TNet, solves tubercle information lost in the shortage of data due to the lamellarity feature formation of CT images, the detection omission of conflicting and network, and auxiliary doctor carries out pulmonary cancer diagnosis, reduces the risk of wrong diagnosis and escape.

Description

A kind of early stage of lung cancer detection classification integration apparatus and system based on deep learning
Technical field
The invention belongs to medical image computer-aided diagnosis research field, it is more particularly to a kind of based on deep learning Early stage of lung cancer detection classification integration apparatus and system.
Background technique
Lung cancer is the highest malignant tumour of morbidity and mortality, and early symptom is unobvious, and medium and advanced lung cancer is extremely difficult It cures, therefore early stage of lung cancer detection is the main means for extending patient's life cycle, reducing the death rate；The early stage of lung cancer shows as lung more Tubercle, the more sizes of quantity are small, contrast is low, easily obscure with its hetero-organization；It is mainly carried out by the way of lung's CT examination at present Pulmonary nodule diagnosis, doctor need to judge its grade malignancy according to the size and form of Lung neoplasm in patient's CT images, and one For the CT picture number of a patients with lung cancer in hundred ranks, doctors'work burden is heavier, will receive doctor's warp in the accuracy of assessment It tests, the influence of the subjective factors such as degree of fatigue, personal mood, while each department expert doctor's maldistribution of the resources is even, takes biography System diagnosis and treatment mode, which is easy to appear, fails to pinpoint a disease in diagnosis and the case where mistaken diagnosis.
In recent years, as the staged of the raising of computer computation ability, data volume increases, depth learning technology is obtained Fast development is largely applied to medical field, and domestic and foreign scholars, research institution be in the calculating detected about the early stage of lung cancer A lot of research work is done in machine auxiliary system, major technique step includes data prediction and detects lung using convolutional network Tubercle, but focus is mainly placed on identification Lung neoplasm region and non-Lung neoplasm region by these methods, on the one hand not to lung knot The classification of section carries out careful differentiation, and the type of another aspect Lung neoplasm canceration, probability and Lung neoplasm type, size, form etc. Characteristic relation is close, and traditional detection convolutional network can only obtain the total probability that patient suffers from lung cancer, can not be to the specific type of lung cancer It distinguishes, goes to assess so generally requiring doctor, be not carried out the integration of early stage of lung cancer testing process.
In view of the deficiencies of the prior art, the invention proposes a kind of new early stage of lung cancer detection devices and system CETS (Cancer Early Test System): a kind of early stage of lung cancer detection classification based on deep learning integration apparatus and is System, which is capable of detecting when the type and probability of patient's early stage of lung cancer, while can be accurately identified the type of Lung neoplasm, ruler Very little, position and grade malignancy, doctor need to only dock CT equipment computer system and CETS device data input port, by Simple parameter setting operation is carried out on CETS touch control terminal, and the detailed detection report of patient's early stage of lung cancer and Lung neoplasm can be obtained It accuses.
Summary of the invention
CETS early stage of lung cancer detection service device system of the present invention is for the first time by light stream tracing algorithm and U-NET network knot Structure combines, and introduces biasing tubercle, develops a kind of 3D depth convolution net that can detect Lung neoplasm and lung cancer type automatically Network structure TNet, while providing quickly detection TNet-VE and accurate detection two kinds of detection models of TNet-VQ；Server simultaneously Detection system uses data enhancing, the alternately strategies such as training, sample balance in the training stage, so that CETS is in Lung neoplasm and lung In the detection of cancer, there is highly sensitive and precision, can precisely be identified micronodule, auxiliary doctor carries out lung cancer and examines It is disconnected, the false positive rate of pulmonary cancer diagnosis is reduced, early stage of lung cancer detection classification process integration is realized.
Early stage of lung cancer detection integration apparatus and system CETS of the present invention are mainly made of 3 parts: early stage of lung cancer inspection It surveys touch control terminal, lung cancer detection server and lung cancer detection case and stores equipment, primary operational process is as follows:
Step 1: corresponding model parameter is arranged according to actual needs, in early stage of lung cancer detection touch control terminal in doctor, and selection to be made Detection model (TNet-VE or TNet-VQ), CETS will read patient lungs' CT scan image automatically, be loaded onto early stage Lung cancer detection server；
Step 2: by patient lungs' CT scan image processing 3D lung image data readable at CETS；
Step 3: 3D lung image data are divided into the input TNet-VE or TNet-VQ Lung neoplasm 3D detection of 4D tensor data Sorter network, obtains the 4D tensor data comprising features such as Lung neoplasm type, position, size, grade malignancies, and merging data obtains Full lung Lung neoplasm testing result；
Step 4: TOP10 Lung neoplasm is picked out by grade malignancy from Lung neoplasm testing result, by TOP10 Lung neoplasm and bias junctions Section input TNet lung cancer 3D detects sorter network, obtains the testing result of patient's lung cancer type and probability, merges Lung neoplasm and lung Cancer detects classification results and generates examining report；
Step 5: doctor detects touch control terminal by the early stage of lung cancer and checks patient's examining report, chooses whether to print, while can be with Use patient's name and patient lungs' CT images as search condition, similar patients' case is retrieved in confirmed cases library, for doctor Raw comparative analysis, doctor confirm that testing result is errorless, CETS making a definite diagnosis into case storage equipment by automated back-up testing result Case library, and it is synchronized to PACS system, if testing result has differences with doctor's confirmed result, doctor can be by confirmed result The Model Self-Learning case library in case storage equipment is uploaded to, Model Self-Learning is used for, updates the detection of early stage of lung cancer server System.
Step 1 mainly includes setting model parameter, selection TNet-VE or TNet-VQ detection model and load CT images.
Step 2 specifically includes that image denoising, lung image are extracted and CT3D is standardized.
(1) image denoising operation: it is standard Heng Shi unit image by the CT pixel value video conversion of load, is filtered using two-dimentional intermediate value Wave algorithm removes salt-pepper noise, carries out binary conversion treatment to CT images using threshold method, is removed and checked using morphology opening operation The backgrounds such as bed, clothing.
(2) lung image extraction operation: using algorithm of region growing selected seed point, carries out 8 connected region growths, extracts most Big connected component, then by lung segmentation at pulmo two parts, carries out algorithm of convex hull and swollen as initial lung contours respectively It is swollen that lung is repaired to obtain lung's exposure mask, original lung Heng Shi unit image data is then converted to BMP format, uses lung Portion's exposure mask overlaps multiplication and obtains lung image data.
(3) CT3D normalizing operation: since the CT images data of Different hospital have differences on slice size, thickness, in order to Coronal-plane, sagittal plane and cross-sectional slices thickness are adjusted using Lanczos interpolation algorithm to image resampling convenient for processing For 1mm*1mm*1mm, 2D image data is unified conversion processing into the 3D image data of 1024*1024*512.
Step 3 specifically includes that Image Segmentation, Lung neoplasm detection mark, image merge, full lung light stream tracking is corrected.
(1) Image Segmentation operates: in order to solve GPU low memory problem, lung's 3D image data of input being cut into unit Having a size of 256*256*256*1, and carry the 4D tensor set of location information.
(2) Lung neoplasm detect labeling operation: by 4D tensor set input TNet Lung neoplasm 3D detect sorter network, by convolution, Residual noise reduction, deconvolution, data fusion, local light stream tracking amendment and RPN layers of processing, obtain the tensor information of 4D, including shadow As data, anchor point information, Lung neoplasm type, three-dimensional coordinate and window diameter, then by 4D tensor information input fully-connected network, It is divided into 7 classes by grade malignancy by Lung neoplasm: high-risk solid nodules, middle danger solid nodules, low danger solid nodules；High-risk part-solid Tubercle, middle danger part-solid tubercle；Then middle danger ground-glass opacity tubercle, low danger ground-glass opacity tubercle pass through global optical flow Tracking amendment is merged into Lung neoplasm detection 3D image.
(3) TNet Lung neoplasm 3D detects sorter network and is based on U-NET, merges in convolutional layer, image draw at two node of layer for the first time Enter light stream tracing algorithm, for calculating the relevance between same CT images difference channel, between lung's difference CT images, solves Lost tubercle letter is omitted in the shortage of data that is formed due to the lamellarity feature of CT images, the detection of conflicting and network Breath, improves the precision of identification；And classification is carried out to Lung neoplasm using full articulamentum, while Lung neoplasm 3D detects sorter network There are two network structure is available for model: TNet-VE accurately detects sorter network and TNet-VQ quickly detects sorter network.
(4) TNet-VE Lung neoplasm 3D accurately detects sorter network: by 4 3D convolutional layers, 6 3D residual blocks, 2 3D deconvolution Layer, 2 fused layers, 2 light stream tracking layer, 1 RPN output layer and 1 full articulamentum composition, wherein preceding 2 convolution kernels are 2* 2*2*24, latter two convolution kernel are 1*1*1* (64,15), and residual block is by 3 residual unit 3* (3*3*3 convolution+BN+RELU+3* 3*3 convolution+BN+RELU)+2*2 maximum pondization composition, 3D deconvolution core size is 2*2*2, and is introduced in fused layer Location information, full connection includes two layers of hidden layer and 7 output units, is classified using sigmoid function.
(5) TNet-VQ Lung neoplasm 3D quickly detects sorter network: by 8 3D convolutional layers, 2 3D residual blocks, 2 3D deconvolution Layer, 2 fused layers, 2 light stream tracking layer, 1 RPN output layer and 1 full articulamentum composition, wherein preceding 2 convolution kernels are 2* 2*2*24,3-6 convolution unit are made of convolution+RELU+2*2 maximum pond, and the 3rd convolution kernel is 3*3*3*32,4-6 A convolution kernel is 3*3*3*64, and most latter two convolution kernel is 1*1*1* (64,15), and residual block is by 3 residual unit 3* (3*3*3 Convolution+BN+RELU+3*3*3 convolution+BN+RELU)+2*2 maximum pondization composition, 3D deconvolution core size is 2*2*2, and Location information is introduced in fused layer, full connection includes two layers of hidden layer and 7 output units, using sigmoid function into Row classification.
(6) loss function increases data enhancing and L1 regularization operation on the basis of intersecting entropy function:
L=Lcls+pLreg
LclsTo intersect entropy loss item, LregItem is lost to return:
S indicates L1 norm, GkIndicate frame central point three-dimensional coordinate and side length, A indicates true Lung neoplasm and three-dimensional coordinate and side length.
(7) image union operation: full articulamentum is classified and exports the 4D tensor image data merging of output.
(8) amendment operation is tracked in full lung light stream: being used light stream tracing algorithm, is corrected due to the lamellarity feature formation of CT images Lost tubercle information is omitted in shortage of data, the detection of conflicting and network, is generated and is had Lung neoplasm type, size and evil The complete image of the information such as property degree；
Step 4 specifically includes that Lung neoplasm core feature extracts, Lung neoplasm canceration probability calculation, lung cancer detection classification.
(1) Lung neoplasm core feature extraction operation: TOP10 lung is extracted from step 3 Lung neoplasm testing result by grade malignancy The included biasing tubercle of TOP10 Lung neoplasm and model is inputted TNet Lung neoplasm 3D detection sorter network together, obtains lung by tubercle Then tubercle 4D tensor data extract data center's point 2*2 data information, and carry out maximum pondization operation, finally obtain 11 A Lung neoplasm core 128D characteristic.Tubercle is biased in the module as error correction item, can prevent Lung neoplasm from detecting net Network omits Malignant Nodules information, and data are obtained by Model Self-Learning in training process.
(2) Lung neoplasm canceration probability calculation operates: Lung neoplasm canceration probability mainly passes through perceptron model and calculates realization, by 11 Lung neoplasm core 128D characteristic inputs perceptron, and the canceration type and probability of 11 Lung neoplasms, perceptron mould can be obtained Type includes 64 hidden units and output unit (using sigmoid activation primitive).
(3) lung cancer detection sort operation: the canceration type of 11 Lung neoplasms and probability are inputted into full articulamentum, full articulamentum includes Two layers of hidden layer and 15 output units, are calculated the type and probability that patient suffers from lung cancer, which presses base for lung cancer This type and mixed type are subdivided into 15 classes: basic class: adenocarcinoma of lung, squamous cell carcinoma, large cell carcinoma, small cell carcinoma；Mix II type: lung Gland cancer+squamous cell carcinoma, adenocarcinoma of lung+large cell carcinoma, adenocarcinoma of lung+small cell carcinoma, squamous cell carcinoma+large cell carcinoma, scaly epithelium Cancer+small cell carcinoma, large cell carcinoma+small cell carcinoma；Mix type III: adenocarcinoma of lung+squamous cell carcinoma+large cell carcinoma, adenocarcinoma of lung+squama Columnar epithelium cancer+small cell carcinoma, adenocarcinoma of lung+large cell carcinoma+small cell carcinoma, squamous cell carcinoma+large cell carcinoma+small cell carcinoma；Mixing IV type: adenocarcinoma of lung+squamous cell carcinoma+large cell carcinoma+small cell carcinoma.
Step 5, which specifically includes that, merges the testing result of step 3 and step 4, generates patient's early stage of lung cancer examining report, And it is shown in touch control terminal.
Compared to other existing early stage of lung cancer detection methods, the invention has the following advantages that
(1) CETS early stage of lung cancer detection service device has used TNet to detect network, and TNet is firstly introduced on the basis of U-NET Light stream tracing algorithm solves the shortage of data due to the lamellarity feature formation of CT images, the detection of conflicting and network Lost tubercle information is omitted, and is able to carry out the detection classification of Lung neoplasm and lung cancer, wrong diagnosis and escape risk is reduced, mentions The accuracy of high early stage of lung cancer diagnosis.
(2) CETS realizes the integration at CT images end to lung cancer detection end process, while providing TNet-VE and TNet-VQ Two kinds of computation models, after doctor chooses computation model according to actual needs, CETS loads CT images data automatically and is detected, It can quickly obtain comprising the details detection report such as Lung neoplasm type, position, size, grade malignancy and lung cancer type, probability It accuses.
(3) CETS early stage of lung cancer detection service device system uses data enhancing, Lung neoplasm detection and lung cancer inspection in the training stage Test cross is for strategies such as training, sample balances, so that CETS in the size and Morphology observation of Lung neoplasm, has highly sensitive and essence Accuracy.
(4) other than it can use patient's name and retrieve, CETS touch control terminal post-processing module is additionally provided to scheme doctor The function of figure is searched, doctor can be used patient CT diagnosis imaging as search condition, retrieve history similar case, be convenient for doctor Comparative analysis.
Detailed description of the invention
Fig. 1 CETS early stage of lung cancer detects integration apparatus
Fig. 2 CETS early stage of lung cancer detection service device flow chart
Fig. 3 CETS early stage of lung cancer detection service device preprocessing module
Fig. 4 standard Heng Shi unit CT images
Fig. 5 binary conversion treatment
Fig. 6 convex closure and expansion are repaired
Fig. 7 lung exposure mask
Fig. 8 lung image
Fig. 9 CETS early stage of lung cancer detection service device Lung neoplasm accurately detects sorter network (TNet-VE)
Figure 10 Lung neoplasm testing result
Figure 11 CETS early stage of lung cancer detection service device lung cancer detection sorter network
Figure 12 CETS early stage of lung cancer detection service device Lung neoplasm quickly detects sorter network (TNet-VQ)
Specific embodiment:
The present invention is described in detail with specific embodiment with reference to the accompanying drawing.
As shown in Figure 1, early stage of lung cancer detection device of the present invention and system CETS are whole by early stage of lung cancer detection touch-control Hold (100), early stage of lung cancer detection service device (200) and early stage of lung cancer detection case storage equipment (300) composition.
The early stage of lung cancer detects touch control terminal (100), includes mainly model parameter setup module and post-processing module, is used for Parameter setting, detection model selection and testing result before CETS detection are checked, print and are retrieved.
As shown in Fig. 2, early stage of lung cancer detection service device (200) mainly includes preprocessing module and lung cancer detection depth model (Lung neoplasm 3D detects sorter network+lung cancer 3D and detects sorter network)；Patient 3D pulmonary data can be divided by CETS automatically 256*256*256*1 data block detects knuckle areas using Lung neoplasm 3D detection sorter network and classifies to it, then TOP10 Lung neoplasm and biasing tubercle input lung cancer 3D are detected into sorter network, obtain the detection classification knot of lung cancer classification and probability Fruit generates the early stage of lung cancer and detects classification report, and shows in system touch control terminal.
Early stage of lung cancer detection case storage equipment (300) connects with the PACS system of hospital: mainly including confirmed cases library With Model Self-Learning case library, doctor confirms that testing result is errorless, and automated back-up testing result to case is stored equipment by CETS Confirmed cases library, and it is synchronized to PACS system, if testing result has differences with doctor's confirmed result, doctor can will be made a definite diagnosis As a result case storage device model self study case library is uploaded to, Model Self-Learning is used for, updates server detection system.
It mainly includes 101~102 that the CETS early stage of lung cancer, which detects touch control terminal,.
Parameter setting module 101: doctor uses accurate detection model by parameter setting module setting model parameter, selection TNet-VE or quick detection model TNet-VQ are detected, and CETS loads patient lungs' CT images to the early stage of lung cancer automatically and detects Server is illustrated by taking accurate detection model TNet-VE as an example in this embodiment.
After the completion of post-processing module 102:CETS detection, in post-processing module, doctor can check lung in touch control terminal Cancer examining report, and can be printed, while patient's name and patient lungs' CT diagnostic image can be used as searching bar Part retrieves similar patients' case in the confirmed cases library of storage equipment, is convenient for doctor's comparative analysis.
As shown in figure 3, CETS early stage of lung cancer detection service device preprocessing module includes 201~203.
Image denoises unit 201: in order to eliminate otherness existing for the pixel value of Different hospital CT scan image, unification will CT pixel value video conversion standard Heng Shi unit image, as shown in figure 4, and being made an uproar using the two dimension median filter algorithm removal spiced salt Sound is carried out binary conversion treatment to CT images as threshold value using -700, checked as shown in figure 5, being removed using morphology opening operation Bed.
Lung image extraction unit 202: algorithm of region growing selected seed point is used, and carries out 8 connected region growths, is picked Except in every binaryzation CT images be less than 20mm2It is greater than 0.9 connected component with eccentricity, by all binaryzation CT images data It is combined into a 3D image data, retains volume CT images number corresponding to the connected component of 0.6-8L in 3D image data According to rejecting extracts largest connected component, as initial lung contours, then by lung segmentation not in the CT images in this region At pulmo two parts, algorithm of convex hull is carried out respectively and lung is repaired in expansion, as shown in Figure 6；To include near lung wall and Tubercle finally obtains lung's exposure mask, as shown in Figure 7；Then the Heng Shi unit image of original CT images is cut, removes HU Value converts it into BMP format image not in the data in [- 1100 ,+500] section, heavy with it using obtained lung's exposure mask figure Folded multiplication obtains lung image, the numerical value in region and bone except lung is disposed as 170, as shown in Figure 8.
CT3D Standardisation Cell 203: the lung image data extracted are cut, and unified back gauge is 20 voxels, Then coronal-plane, sagittal plane and cross-sectional slices thickness are adjusted to image resampling by 1mm* using Lanczos interpolation algorithm 2D image is uniformly converted to the 3D image data of 1024*1024*512 by 1mm*1mm.
As shown in figure 9, CETS early stage of lung cancer detection service device Lung neoplasm detection categorization module includes 204~210.
Image Segmentation unit 204: in order to solve GPU low memory problem, lung's 3D image data of input is cut into Unit size is 256*256*256*1, and carries the 4D tensor set of location information.
Image, semantic cutting unit 205: accurately detecting sorter network for 4D tensor information input Lung neoplasm TNet-VE, warp It crosses T1 convolutional layer to handle to obtain 256*256*24 image data, handles to obtain 256*256*24 image data by T2 convolutional layer, It handles to obtain 128*128*32 image data by T3 residual error layer, handles to obtain 64*64*64 image data by T4 residual error layer, It handles to obtain 32*32*64 image data by T5 residual error layer, handles to obtain 16*16*64 image data by T6 residual error layer, pass through It crosses T7 warp lamination and obtains 32*32*64 image data, 32*32*64 image data, the T5 residual error layer that T7 warp lamination is obtained Obtained 32*32*64 data fusion is handled, successively obtains 64*64*64 by T8 fused layer, T9 residual error layer, T10 warp lamination Image data, the 64*64*64 image data that 64*64*64 image data that T10 is obtained, T4 residual error layer are handled and position Information data 64*64*3 is merged, and is obtained 64*64*128 by T11 fused layer, T12 residual error layer, is obtained by T13 convolutional layer To 64*64*64 image data, 64*64*15 image data is obtained by T14 convolutional layer.
Amending unit 206 is tracked in local light stream: the 64*64*15 image data that T14 convolutional layer is obtained inputs local light stream Tracking layer, the tubercle information that amendment detection network may be omitted, obtains 64*64*15 image data.
RPN unit 207: light stream tracking result is exported into the tensor for 4D, in addition to image data further includes anchor point information, lung Tubercle class probability (sigmoid function), three-dimensional coordinate and window diameter.
Lung neoplasm taxon 208: the type of Lung neoplasm, ruler are calculated according to obtained Lung neoplasm 4D tensor information The information such as very little, position and grade malignancy obtain Lung neoplasm detection classification results, as shown in Figure 10.
Image combining unit 209: according to the 4D tensor information of full articulamentum classification output by data merged block.
Full lung light stream tracing unit 210: light stream tracing algorithm is used, according to the relevance between CT images, to Lung neoplasm Position form on each frame CT is tracked positioning, corrects the shortage of data due to the lamellarity feature formation of CT images, phase Lost tubercle information is omitted in mutual contradiction and network detection, is ultimately produced with Lung neoplasm type, size and grade malignancy Etc. information complete image.
As shown in figure 11, CETS early stage of lung cancer detection service device lung cancer detection categorization module 211~213.
Lung neoplasm core feature extraction unit 211: automatically extracting out TOP10 Lung neoplasm data, and loads biasing tubercle and make For error correction, above-mentioned TOP10 Lung neoplasm data and bias junctions joint number are detected into sorter network according to input TNet Lung neoplasm again, It obtains then extracting data center 2*2 pixel data information, and carry out by Lung neoplasm taxon treated image data Maximum pondization operation obtains 128D characteristic.
Lung neoplasm canceration probability calculation unit 212: 11 Lung neoplasm core 128D features are passed through into hidden layer and output layer Processing, obtains the canceration type and probability of 11 Lung neoplasms.
Lung cancer detection taxon 213: according to the canceration type and probability of 11 Lung neoplasms, patient's lung cancer point is calculated The probability distribution of class obtains lung cancer detection classification results.
It includes 301~302 that the early stage of lung cancer, which detects case storage equipment,.
Confirmed cases library module 301, system automatically merge Lung neoplasm and lung cancer classification and Detection result, generate final detection Report display, doctor confirm that testing result is errorless, and automated back-up testing result to confirmed cases library, and is synchronized to PACS by system System.
Model Self-Learning case library module 302 is cured if doctor's last diagnostic report and CETS testing result have differences Life can will make a definite diagnosis report by system touch control terminal and be uploaded to storage device model self study case module, and default setting works as case When example is more than 50 (doctor can change this parameter by touch control terminal parameter module), model will carry out self study, update CETS early detection server system, keeps testing result more accurate.
Quick detection model TNet-VQ specific embodiment process is substantially identical as accurate detection model TNet-VE, mainly Difference is the+2*2 maximum pond residual error layer 3* in TNet-VE (3*3*3 convolution+BN+RELU+3*3*3 convolution+BN+RELU) Topology update is convolutional layer (3*3*3 convolution+RELU+2*2 maximum pond), to improve calculating speed, but can be sacrificed certain Accuracy is detected, as shown in figure 12.
This example is only exemplary of the invention, and the embodiment of CETS equipment and system is not limited thereto Example can modify and optimize to equipment of the present invention and system completely for researcher in this field, this A little essences are identical with the content of present invention, and within the different change of the form of expression all belongs to the scope of protection of the present invention.

Claims (4)

1. a kind of early stage of lung cancer detection classification integration apparatus and system (CETS) based on deep learning, which is characterized in that packet It includes with lower module:
(1) early stage of lung cancer detects touch control terminal: mainly including parameter setting module and post-processing module；Doctor passes through CETS touch-control Terminal parameter setting module setting model parameter, selection are examined using accurate model TNet-VE or accelerated model TNet-VQ It surveys；CETS loads patient lungs' CT images to early stage of lung cancer detection service device, after the completion of detection, after CETS touch control terminal automatically In processing module, doctor can check lung cancer and Lung neoplasm detection classification report, and can be printed, while trouble can be used Person's name and patient lungs CT diagnosis imaging retrieve similar patients' case, just as search condition in CETS confirmed cases library Yu doctor's comparative analysis；
(2) early stage of lung cancer detection service device: mainly including preprocessing module and lung cancer detection depth model, and depth model uses TNet detects sorter network, which improves on the basis of U-NET, is firstly introduced light stream tracing algorithm, solve due to Lost tubercle information is omitted in the shortage of data of the lamellarity feature formation of CT images, the detection of conflicting and network, and Accurate detection TNet-VE and quickly detection two kinds of models of TNet-VQ are provided；Preprocessing module automatically will be at patient lungs' CT images The readable three-dimensional lung image of CETS is managed into, detects tubercle using TNet-VE or TNet-VQ Lung neoplasm 3D detection sorter network Region, and classify to tubercle, while TOP10 Lung neoplasm and biasing tubercle input TNet or TNet-VQ lung cancer will be extracted 3D detects sorter network, obtains lung cancer detection as a result, merging Lung neoplasm and lung cancer detection result generation examining report；
(3) early stage of lung cancer detection case stores equipment: connecting with Hospital PACS, mainly includes confirmed cases library module and mould Type self study case library module；After the completion of detection, doctor confirms that testing result is errorless, and system is by automated back-up testing result to really It diagnoses a disease a library, and is synchronized to PACS system, if testing result has differences with doctor's confirmed result, doctor can will make a definite diagnosis knot Fruit uploads to Model Self-Learning case library, is used for Model Self-Learning, updates server detection system.
2. a kind of early stage of lung cancer detection classification integration apparatus and system based on deep learning according to claim 1 (CETS), which is characterized in that the detection touch control terminal of the early stage of lung cancer described in the module (1) specifically includes:
(1) parameter setting module: model parameter is arranged by parameter setting module in doctor, and selection uses accurate detection model TNet-VE or quick detection model TNet-VQ are detected, and CETS loads patient lungs' CT images to the early stage of lung cancer automatically and detects Server；
(2) post-processing module: after the completion of server detection, in terminal post-processing module, doctor can check lung cancer and lung knot Section detection classification report, and can be printed, while patient's name and patient lungs' CT diagnosis imaging can be used as searching Rope condition retrieves similar patients' case in CETS confirmed cases library, is convenient for doctor's comparative analysis.
3. a kind of early stage of lung cancer detection classification integration apparatus and system based on deep learning according to claim 1 (CETS), which is characterized in that early stage of lung cancer detection service implement body described in the module (2) includes:
(1) TNet-VE Lung neoplasm 3D accurately detects disaggregated model: by 4 3D convolutional layers, 6 3D residual blocks, 2 3D deconvolution Layer, 2 fused layers, 2 light stream tracking layer, 1 RPN layers and 1 full articulamentum compositions, wherein preceding 2 convolution kernels are 2*2*2* 24, latter two convolution kernel is 1*1*1* (64,15), and residual block is by 3 residual unit 3* (3*3*3 convolution+BN+RELU+3*3*3 Convolution+BN+RELU)+2*2 maximum pondization composition, 3D deconvolution core size is 2*2*2, and position is introduced in fused layer Confidence breath, full connection includes two hidden layers and 7 output units, is classified using sigmoid function；
(2) TNet-VQ Lung neoplasm 3D quickly detects disaggregated model: by 8 3D convolutional layers, 2 3D residual blocks, 2 3D deconvolution Layer, 2 fused layers, 2 light stream tracking layer, 1 RPN layers and 1 full articulamentum compositions, wherein preceding 2 convolution kernels are 2*2*2* 24,3-6 convolution unit is made of convolution+RELU+2*2 maximum pond, and the 3rd convolution kernel is 3*3*3*32, the 4-6 convolution Core is 3*3*3*64, and most latter two convolution kernel is 1*1*1* (64,15), and residual block is by 3 residual unit 3* (3*3*3 convolution+BN + RELU+3*3*3 convolution+BN+RELU)+2*2 maximum pondization composition, 3D deconvolution core size is 2*2*2, and in fused layer In introduce location information, full connection includes two hidden layers and 7 output units, is classified using sigmoid function；
(3) image denoises unit: CT pixel value video conversion standard Heng Shi unit image is gone using two dimension median filter algorithm Except salt-pepper noise, binary conversion treatment is carried out to CT images using threshold method, and use morphology opening operation removal examination couch and clothing The backgrounds such as object；
(4) lung image extraction unit: using algorithm of region growing selected seed point, carries out 8 connected region growths, obtains initial Lung outlines carry out algorithm of convex hull respectively and expansion are repaired to obtain to lung then by lung segmentation at pulmo two parts Lung's original CT Heng Shi unit image is converted to BMP format image, is overlapped using lung's exposure mask mutually multiplied by lung's exposure mask To lung image data；
(5) CT3D Standardisation Cell: using Lanczos interpolation algorithm to picture resampling, by coronal-plane, sagittal plane and cross section Slice thickness is adjusted to 1mm*1mm*1mm, and 2D image is uniformly converted to the 3D image data of 1024*1024*512；
(6) in order to solve GPU low memory problem, lung's 3D image data of input Image Segmentation unit: is cut into unit The 4D tensor set of location information is carried having a size of 256*256*256*1；
(7) image, semantic cutting unit: 4D tensor set input TNet Lung neoplasm 3D is detected into sorter network, by convolution, residual Poor processing, deconvolution, data aggregation layer processing, obtain image, semantic segmentation result；
(8) amending unit is tracked in local light stream: being used light stream tracing algorithm, is carried out location tracking to Lung neoplasm, correct TNet net The Malignant Nodules information that network or data are omitted, improves the accuracy of identification；
(9) RPN unit: light stream tracking result is exported into the tensor for 4D, in addition to image data further includes anchor point information, Lung neoplasm Class probability (sigmoid function), three-dimensional coordinate and window diameter；
(10) Lung neoplasm taxon: being classified using full articulamentum and sigmoid function, according to obtained Lung neoplasm 4D It measures information and Lung neoplasm is divided into 7 classes by grade malignancy and size: high-risk solid nodules, middle danger solid nodules, low danger solid nodules； High-risk part-solid tubercle, middle danger part-solid tubercle；Middle danger ground-glass opacity tubercle, low danger ground-glass opacity tubercle；
(11) lung 3D image image combining unit: is helped into the 4D tensor information merging of full articulamentum classification output；
(12) amending unit is tracked in full lung light stream: using light stream tracing algorithm, amendment is formed due to the lamellarity feature of CT images Shortage of data, the detection of conflicting and network omit lost tubercle information, generate with Lung neoplasm type, size and The complete image of the information such as grade malignancy；
(13) loss function increases data enhancing and L1 regularization operation on the basis of intersecting entropy function, is finally lost Function are as follows:
L=Lcls+pLreg
LclsTo intersect entropy loss item, LregItem is lost to return:
S indicates L1 norm, GkIndicate frame central point three-dimensional coordinate and side length, A indicates true tubercle and three-dimensional coordinate and side length；
(14) Lung neoplasm core feature extraction unit: extracting TOP10 Lung neoplasm data, but detects network or CT images sometimes Some Malignant Nodules information can be omitted, therefore we introduce a biasing tubercle as error correction, and in the training stage pair Biasing tubercle is trained, and obtains bias junctions joint number evidence by the self study of model, and above-mentioned 11 Lung neoplasms are inputted TNet net Network obtains then extracting its center 2*2 data information, and carry out most by Lung neoplasm taxon treated 4D tensor data Great Chiization obtains 128D feature；
(15) Lung neoplasm canceration probability calculation unit: the double-deck perceptron layer contains with 64 hidden units and output unit, defeated Out layer activation primitive use sigmoid function, according to Lung neoplasm 128D characteristic calculate the canceration of the tubercle type and Probability；
(16) lung cancer detection taxon: the type and probability that patient suffers from lung cancer are calculated using full articulamentum, full connection includes two Layer hidden layer and 15 output units, are subdivided into 15 classes: basic model by basic model and mixed type for lung cancer: in adenocarcinoma of lung, squamous Skin cancer, large cell carcinoma, small cell carcinoma；Mixing II type: adenocarcinoma of lung+squamous cell carcinoma, adenocarcinoma of lung+large cell carcinoma, adenocarcinoma of lung+small thin Born of the same parents' cancer, squamous cell carcinoma+large cell carcinoma, squamous cell carcinoma+small cell carcinoma, large cell carcinoma+small cell carcinoma；Mix type III: lung gland Cancer+squamous cell carcinoma+large cell carcinoma, adenocarcinoma of lung+squamous cell carcinoma+small cell carcinoma, adenocarcinoma of lung+large cell carcinoma+small cell carcinoma, squama Columnar epithelium cancer+large cell carcinoma+small cell carcinoma；Mix IV type: adenocarcinoma of lung+squamous cell carcinoma+large cell carcinoma+small cell carcinoma.
4. a kind of early stage of lung cancer detection classification integration apparatus and system based on deep learning according to claim 1, It is characterized in that, the detection case storage equipment of the early stage of lung cancer described in the module (3) specifically includes:
(1) confirmed cases library module: storage patient makes a definite diagnosis report, and is synchronized to Hospital PACS in time；
(2) Model Self-Learning case library module: storage examining report with make a definite diagnosis the case that has differences of report, for model from Study updates early stage of lung cancer detection service device.
CN201810776622.1A 2018-07-16 2018-07-16 A kind of early stage of lung cancer detection classification integration apparatus and system based on deep learning Pending CN109003672A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201810776622.1A CN109003672A (en) 2018-07-16 2018-07-16 A kind of early stage of lung cancer detection classification integration apparatus and system based on deep learning

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201810776622.1A CN109003672A (en) 2018-07-16 2018-07-16 A kind of early stage of lung cancer detection classification integration apparatus and system based on deep learning

Publications (1)

Publication Number Publication Date
CN109003672A true CN109003672A (en) 2018-12-14

Family

ID=64598374

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201810776622.1A Pending CN109003672A (en) 2018-07-16 2018-07-16 A kind of early stage of lung cancer detection classification integration apparatus and system based on deep learning

Country Status (1)

Country Link
CN (1) CN109003672A (en)

Cited By (31)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109785311A (en) * 2019-01-14 2019-05-21 深圳和而泰数据资源与云技术有限公司 A kind of methods for the diagnosis of diseases and relevant device
CN109886243A (en) * 2019-03-01 2019-06-14 腾讯科技（深圳）有限公司 Image processing method, device, storage medium, equipment and system
CN109886179A (en) * 2019-02-18 2019-06-14 深圳视见医疗科技有限公司 The image partition method and system of cervical cell smear based on Mask-RCNN
CN109903846A (en) * 2019-02-21 2019-06-18 吉林禾熙科技开发有限公司 Screening lung cancer system and method based on medical big data
CN110097567A (en) * 2019-04-18 2019-08-06 天津大学 In conjunction with the three-dimensional brain tumor image partition method of improved FCNN and level set
CN110246578A (en) * 2019-06-12 2019-09-17 吉林大学第一医院 A kind of lung cancer sample rapid detection method
CN110310281A (en) * 2019-07-10 2019-10-08 重庆邮电大学 Lung neoplasm detection and dividing method in a kind of Virtual Medical based on Mask-RCNN deep learning
CN110310270A (en) * 2019-06-29 2019-10-08 东南大学苏州医疗器械研究院 Tumour microballoon profile testing method and device based on U-net network model
CN110473208A (en) * 2019-07-31 2019-11-19 上海联影智能医疗科技有限公司 Structural form method of discrimination, computer equipment and storage medium
CN110634132A (en) * 2019-08-30 2019-12-31 浙江大学 Method for automatically generating tuberculosis quantitative diagnosis report based on deep learning 3D CT image
TWI681406B (en) * 2018-12-20 2020-01-01 中國醫藥大學附設醫院 Deep learning of tumor image-aided prediction of prognosis of patients with uterine cervical cancer system, method and computer program product thereof
CN110705570A (en) * 2019-09-20 2020-01-17 桂林电子科技大学 Image feature identification method
CN110782441A (en) * 2019-10-22 2020-02-11 浙江大学 DR image pulmonary tuberculosis intelligent segmentation and detection method based on deep learning
CN110797109A (en) * 2019-10-15 2020-02-14 山东朗润医疗系统有限公司 Automatic classification system for magnetic resonance images
CN111354442A (en) * 2018-12-20 2020-06-30 中国医药大学附设医院 Tumor image deep learning assisted cervical cancer patient prognosis prediction system and method
CN111489833A (en) * 2019-12-19 2020-08-04 上海杏脉信息科技有限公司 Lung cell pathology rapid on-site evaluation system and method and computer readable storage medium
CN111653356A (en) * 2020-04-20 2020-09-11 浙江大学 New coronary pneumonia screening method and new coronary pneumonia screening system based on deep learning
CN111666972A (en) * 2020-04-28 2020-09-15 清华大学 Liver case image classification method and system based on deep neural network
CN112365973A (en) * 2020-11-02 2021-02-12 太原理工大学 Pulmonary nodule auxiliary diagnosis system based on countermeasure network and fast R-CNN
CN112381773A (en) * 2020-11-05 2021-02-19 东风柳州汽车有限公司 Key cross section data analysis method, device, equipment and storage medium
CN112419262A (en) * 2020-11-19 2021-02-26 上海联影智能医疗科技有限公司 Bone marrow smear microscopic image analysis method and device, electronic equipment and storage medium
CN113344887A (en) * 2021-06-16 2021-09-03 南通大学 Interstitial pneumonia assessment method based on deep learning and fuzzy logic
CN113506289A (en) * 2021-07-28 2021-10-15 中山仰视科技有限公司 Method for classifying false positive of pulmonary nodule by using double-flow network
CN113658657A (en) * 2021-08-24 2021-11-16 中山仰视科技有限公司 Lung nodule detection method and device combining depth network and optical flow mechanism
CN113889261A (en) * 2021-09-23 2022-01-04 之江实验室 Pathological feature assistance-based PET/CT automatic lung cancer diagnosis classification model training method
CN113989277A (en) * 2021-12-24 2022-01-28 青岛美迪康数字工程有限公司 Imaging method and device for medical radiation diagnosis and treatment examination
CN115346658A (en) * 2022-07-15 2022-11-15 一选(浙江)医疗科技有限公司 Early intelligent auxiliary lung cancer detection system and method based on big data technology
CN115439686A (en) * 2022-08-30 2022-12-06 一选(浙江)医疗科技有限公司 Method and system for detecting attention object based on scanned image
CN116934757A (en) * 2023-09-18 2023-10-24 电子科技大学 Method, equipment and storage medium for lung nodule false positive pruning
CN117079093A (en) * 2023-10-18 2023-11-17 澄影科技(北京)有限公司 Method, device and equipment for predicting abnormal symptoms based on multi-mode image data
CN118262901A (en) * 2024-04-07 2024-06-28 中国人民解放军总医院第六医学中心 Deep learning-based lung cancer type prediction system

Citations (4)

* Cited by examiner, † Cited by third party
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
CN106780460A (en) * 2016-12-13 2017-05-31 杭州健培科技有限公司 A kind of Lung neoplasm automatic checkout system for chest CT image
CN106909778A (en) * 2017-02-09 2017-06-30 北京市计算中心 A kind of Multimodal medical image recognition methods and device based on deep learning
CN108062525A (en) * 2017-12-14 2018-05-22 中国科学技术大学 A kind of deep learning hand detection method based on hand region prediction

Patent Citations (4)

* Cited by examiner, † Cited by third party
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
CN106780460A (en) * 2016-12-13 2017-05-31 杭州健培科技有限公司 A kind of Lung neoplasm automatic checkout system for chest CT image
CN106909778A (en) * 2017-02-09 2017-06-30 北京市计算中心 A kind of Multimodal medical image recognition methods and device based on deep learning
CN108062525A (en) * 2017-12-14 2018-05-22 中国科学技术大学 A kind of deep learning hand detection method based on hand region prediction

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
YUNPENG CHEN 等: "Sharing residual units through collective tensor factorization to improve deep neural networks", 《HTTP://ARXIV.ORG/PDF/1703.02180.PDF》 *

Cited By (47)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US10896502B2 (en) 2018-12-20 2021-01-19 China Medical University Hospital Prediction system, method and computer program product thereof
CN111354442A (en) * 2018-12-20 2020-06-30 中国医药大学附设医院 Tumor image deep learning assisted cervical cancer patient prognosis prediction system and method
TWI681406B (en) * 2018-12-20 2020-01-01 中國醫藥大學附設醫院 Deep learning of tumor image-aided prediction of prognosis of patients with uterine cervical cancer system, method and computer program product thereof
CN109785311B (en) * 2019-01-14 2021-06-04 深圳和而泰数据资源与云技术有限公司 Disease diagnosis device, electronic equipment and storage medium
CN109785311A (en) * 2019-01-14 2019-05-21 深圳和而泰数据资源与云技术有限公司 A kind of methods for the diagnosis of diseases and relevant device
CN109886179A (en) * 2019-02-18 2019-06-14 深圳视见医疗科技有限公司 The image partition method and system of cervical cell smear based on Mask-RCNN
CN109886179B (en) * 2019-02-18 2023-03-28 深圳视见医疗科技有限公司 Image segmentation method and system of cervical cell smear based on Mask-RCNN
CN109903846A (en) * 2019-02-21 2019-06-18 吉林禾熙科技开发有限公司 Screening lung cancer system and method based on medical big data
CN110458127A (en) * 2019-03-01 2019-11-15 腾讯医疗健康(深圳)有限公司 Image processing method, device, equipment and system
CN110458127B (en) * 2019-03-01 2021-02-26 腾讯医疗健康(深圳)有限公司 Image processing method, device, equipment and system
CN109886243B (en) * 2019-03-01 2021-03-26 腾讯医疗健康(深圳)有限公司 Image processing method, device, storage medium, equipment and system
CN109886243A (en) * 2019-03-01 2019-06-14 腾讯科技（深圳）有限公司 Image processing method, device, storage medium, equipment and system
CN110097567A (en) * 2019-04-18 2019-08-06 天津大学 In conjunction with the three-dimensional brain tumor image partition method of improved FCNN and level set
CN110246578A (en) * 2019-06-12 2019-09-17 吉林大学第一医院 A kind of lung cancer sample rapid detection method
CN110310270B (en) * 2019-06-29 2020-11-20 东南大学苏州医疗器械研究院 Tumor microsphere contour detection method and device based on U-net network model
CN110310270A (en) * 2019-06-29 2019-10-08 东南大学苏州医疗器械研究院 Tumour microballoon profile testing method and device based on U-net network model
CN110310281B (en) * 2019-07-10 2023-03-03 重庆邮电大学 Mask-RCNN deep learning-based pulmonary nodule detection and segmentation method in virtual medical treatment
CN110310281A (en) * 2019-07-10 2019-10-08 重庆邮电大学 Lung neoplasm detection and dividing method in a kind of Virtual Medical based on Mask-RCNN deep learning
CN110473208A (en) * 2019-07-31 2019-11-19 上海联影智能医疗科技有限公司 Structural form method of discrimination, computer equipment and storage medium
CN110634132A (en) * 2019-08-30 2019-12-31 浙江大学 Method for automatically generating tuberculosis quantitative diagnosis report based on deep learning 3D CT image
CN110705570B (en) * 2019-09-20 2022-06-21 桂林电子科技大学 Image feature identification method
CN110705570A (en) * 2019-09-20 2020-01-17 桂林电子科技大学 Image feature identification method
CN110797109A (en) * 2019-10-15 2020-02-14 山东朗润医疗系统有限公司 Automatic classification system for magnetic resonance images
CN110782441A (en) * 2019-10-22 2020-02-11 浙江大学 DR image pulmonary tuberculosis intelligent segmentation and detection method based on deep learning
CN111489833A (en) * 2019-12-19 2020-08-04 上海杏脉信息科技有限公司 Lung cell pathology rapid on-site evaluation system and method and computer readable storage medium
CN111653356A (en) * 2020-04-20 2020-09-11 浙江大学 New coronary pneumonia screening method and new coronary pneumonia screening system based on deep learning
CN111666972A (en) * 2020-04-28 2020-09-15 清华大学 Liver case image classification method and system based on deep neural network
CN112365973A (en) * 2020-11-02 2021-02-12 太原理工大学 Pulmonary nodule auxiliary diagnosis system based on countermeasure network and fast R-CNN
CN112365973B (en) * 2020-11-02 2022-04-19 太原理工大学 Pulmonary nodule auxiliary diagnosis system based on countermeasure network and fast R-CNN
CN112381773A (en) * 2020-11-05 2021-02-19 东风柳州汽车有限公司 Key cross section data analysis method, device, equipment and storage medium
CN112381773B (en) * 2020-11-05 2023-04-18 东风柳州汽车有限公司 Key cross section data analysis method, device, equipment and storage medium
CN112419262A (en) * 2020-11-19 2021-02-26 上海联影智能医疗科技有限公司 Bone marrow smear microscopic image analysis method and device, electronic equipment and storage medium
CN113344887A (en) * 2021-06-16 2021-09-03 南通大学 Interstitial pneumonia assessment method based on deep learning and fuzzy logic
CN113506289A (en) * 2021-07-28 2021-10-15 中山仰视科技有限公司 Method for classifying false positive of pulmonary nodule by using double-flow network
CN113506289B (en) * 2021-07-28 2024-03-29 中山仰视科技有限公司 Method for classifying false positives of lung nodules by using double-flow network
CN113658657A (en) * 2021-08-24 2021-11-16 中山仰视科技有限公司 Lung nodule detection method and device combining depth network and optical flow mechanism
CN113889261B (en) * 2021-09-23 2022-06-10 之江实验室 Pathological feature assistance-based PET/CT automatic lung cancer diagnosis classification model training method
CN113889261A (en) * 2021-09-23 2022-01-04 之江实验室 Pathological feature assistance-based PET/CT automatic lung cancer diagnosis classification model training method
CN113989277A (en) * 2021-12-24 2022-01-28 青岛美迪康数字工程有限公司 Imaging method and device for medical radiation diagnosis and treatment examination
CN115346658A (en) * 2022-07-15 2022-11-15 一选(浙江)医疗科技有限公司 Early intelligent auxiliary lung cancer detection system and method based on big data technology
CN115439686B (en) * 2022-08-30 2024-01-09 一选(浙江)医疗科技有限公司 Method and system for detecting object of interest based on scanned image
CN115439686A (en) * 2022-08-30 2022-12-06 一选(浙江)医疗科技有限公司 Method and system for detecting attention object based on scanned image
CN116934757A (en) * 2023-09-18 2023-10-24 电子科技大学 Method, equipment and storage medium for lung nodule false positive pruning
CN116934757B (en) * 2023-09-18 2023-11-21 电子科技大学 Method, equipment and storage medium for lung nodule false positive pruning
CN117079093B (en) * 2023-10-18 2023-12-12 澄影科技(北京)有限公司 Method, device and equipment for predicting abnormal symptoms based on multi-mode image data
CN117079093A (en) * 2023-10-18 2023-11-17 澄影科技(北京)有限公司 Method, device and equipment for predicting abnormal symptoms based on multi-mode image data
CN118262901A (en) * 2024-04-07 2024-06-28 中国人民解放军总医院第六医学中心 Deep learning-based lung cancer type prediction system

Similar Documents

Publication Publication Date Title
CN109003672A (en) A kind of early stage of lung cancer detection classification integration apparatus and system based on deep learning
NL2019410B1 (en) Computer-aided diagnostics using deep neural networks
CN112101451B (en) Breast cancer tissue pathological type classification method based on generation of antagonism network screening image block
Liu Stbi-yolo: A real-time object detection method for lung nodule recognition
CN109272510B (en) Method for segmenting tubular structure in three-dimensional medical image
Chaki et al. A deep learning based four-fold approach to classify brain MRI: BTSCNet
CN110084237A (en) Detection model construction method, detection method and the device of Lung neoplasm
Liu et al. A CADe system for nodule detection in thoracic CT images based on artificial neural network
Naresh et al. Early detection of lung cancer using neural network techniques
US10706534B2 (en) Method and apparatus for classifying a data point in imaging data
CN111311536A (en) Medical image data
Aslantas et al. CADBOSS: A computer-aided diagnosis system for whole-body bone scintigraphy scans
Ribeiro et al. Handling inter-annotator agreement for automated skin lesion segmentation
Nayan et al. A deep learning approach for brain tumor detection using magnetic resonance imaging
CN111402231B (en) Automatic evaluation system and method for lung CT image quality
CN115115570A (en) Medical image analysis method and apparatus, computer device, and storage medium
Aslam et al. Liver-tumor detection using CNN ResUNet
Abd Al-Ameer et al. Lung cancer detection using image processing and deep learning
Tiwari et al. Optimized ensemble of hybrid rnn-gan models for accurate and automated lung tumour detection from ct images
Hu et al. A multi-instance networks with multiple views for classification of mammograms
Mandour An Exhaustive Review of Neutrosophic Logic in Addressing Image Processing Issues
Üzen Convmixer-based encoder and classification-based decoder architecture for breast lesion segmentation in ultrasound images
Borah et al. Performance Analysis of Breast Cancer Classification from Mammogram Images Using Vision Transformer
Malik et al. Lung cancer detection at initial stage by using image processing and classification techniques
Sivasankaran et al. Lung Cancer Detection Using Image Processing Technique Through Deep Learning Algorithm.

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
WD01 Invention patent application deemed withdrawn after publication
WD01 Invention patent application deemed withdrawn after publication

Application publication date: 20181214