CN110033042A - A kind of carcinoma of the rectum ring week incisxal edge MRI image automatic identifying method and system based on deep neural network - Google Patents
A kind of carcinoma of the rectum ring week incisxal edge MRI image automatic identifying method and system based on deep neural network Download PDFInfo
- Publication number
- CN110033042A CN110033042A CN201910298188.5A CN201910298188A CN110033042A CN 110033042 A CN110033042 A CN 110033042A CN 201910298188 A CN201910298188 A CN 201910298188A CN 110033042 A CN110033042 A CN 110033042A
- Authority
- CN
- China
- Prior art keywords
- network
- ring week
- area
- characteristic pattern
- 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
Links
Classifications
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/05—Detecting, measuring or recording for diagnosis by means of electric currents or magnetic fields; Measuring using microwaves or radio waves
- A61B5/055—Detecting, measuring or recording for diagnosis by means of electric currents or magnetic fields; Measuring using microwaves or radio waves involving electronic [EMR] or nuclear [NMR] magnetic resonance, e.g. magnetic resonance imaging
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/72—Signal processing specially adapted for physiological signals or for diagnostic purposes
- A61B5/7235—Details of waveform analysis
- A61B5/7264—Classification of physiological signals or data, e.g. using neural networks, statistical classifiers, expert systems or fuzzy systems
-
- 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
- G06F18/213—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods
-
- 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
-
- 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
- G06V10/25—Determination of region of interest [ROI] or a volume of interest [VOI]
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Artificial Intelligence (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- General Physics & Mathematics (AREA)
- Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
- Evolutionary Biology (AREA)
- Biophysics (AREA)
- Pathology (AREA)
- Biomedical Technology (AREA)
- Heart & Thoracic Surgery (AREA)
- Medical Informatics (AREA)
- Molecular Biology (AREA)
- Surgery (AREA)
- Animal Behavior & Ethology (AREA)
- General Health & Medical Sciences (AREA)
- Public Health (AREA)
- Veterinary Medicine (AREA)
- General Engineering & Computer Science (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Physiology (AREA)
- Radiology & Medical Imaging (AREA)
- Psychiatry (AREA)
- Signal Processing (AREA)
- High Energy & Nuclear Physics (AREA)
- Mathematical Physics (AREA)
- Fuzzy Systems (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
Abstract
The present invention relates to a kind of carcinoma of the rectum ring week incisxal edge MRI image automatic identifying method based on deep neural network, it include: FasterR-CNN network, the FasterR-CNN network includes feature extraction network, Area generation network and area-of-interest feature vector network；Image is input in trained FasterR-CNN network, firstly, generating convolution characteristic pattern using feature extraction network；Then, convolution characteristic pattern is screened using Area generation network, generates possible ring week positive margin region；Finally, convolution characteristic pattern and formation zone are returned and classified by area-of-interest feature vector network, the position in final output ring week positive margin region and probability.Method of the invention have the function of efficiently, high-accuracy, high stability, and to a certain degree can be to avoid error in judgement caused by doctor artificial origin.
Description
Technical field
The present invention relates to nerual network technique field, in particular to a kind of carcinoma of the rectum ring week based on deep neural network cuts
Edge MRI image automatic identifying method and system.
Background technique
The carcinoma of the rectum is one of most common malignant tumour, the 3rd and the 4th respectively away from cancer spectrum of disease incidence and case fatality rate
Position.The newest cancer statistical information in China, colorectal cancer estimates newly to send out every year and death is respectively 37.6 ten thousand and 5.0 ten thousand
Example, disease incidence and case fatality rate occupy the 5th.Although quoting the specifications diagnosis and treatment such as chemicotherapy and surgical technic, after rectum cancer treatment
Local relapse still between 3% to 30%, tumour involve ring week incisxal edge (CRM) be considered as predict local recurrence high wind
Danger is accused of, and the prognostic risk of " non-anatomical " is established in the 8th edition cancer staging system of american cancer joint committee (AJCC)
And outcome prediction assessment system, recommend using ring week incisxal edge as one of prognostic risk and diagnosis and treatment prediction index, and define its AJCC
Evidence-based medical grade is I grade.CRM by invade be local recurrence and low survival rate independent hazard factor, CRM pairs of Accurate Prediction
Therapeutic scheme is reasonably selected to be of great significance.
Magnetic resonance (MRI) is that the preferred image check method of noninvasively estimating is carried out to the carcinoma of the rectum, with it to mesorectum muscle
The high resolution of the subtle anatomical structures such as film, rectovesical septum fascia, it is considered to be judging CRM, whether there is or not the best inspections that cancer infiltrates
Checking method.3-tier architecture can be told on phased-array surface coil pairs high-resolution T 2WI: mucous membrane and submucosa (high RST) are consolidated
There are muscle layer (low signal) and mesorectum fat deposit (high RST)；And clearly illustrate line sample low signal wrapping intestines week fat
Mesorectal fascia specifies Mesorectal fascia and mesentery endolymph knot, thin vessels, lymphatic vessel and nervus vasculairs beam, with tomography
Anatomical slice has high consistency, mentions accurately to pick out lesion and measuring the distance between lesion and Mesorectal fascia
Having supplied may.Therefore, MRI is comprehensive cancer network (the National Comprehensive Cancer of US National
Network, NCCN) the preferred image check method for carcinoma of the rectum evaluation recommended of guide, instructs clinical to formulate rational therapy
Scheme and assessment to prognosis.And the diagnostic criteria that ring week incisxal edge (CRM) is involved on MRI is then referring to scholars such as Beets.Tan
Report, the shortest distance of tumour (including plantation cancerous node and metastatic lymph node) outer rim and Mesorectal fascia be less than 5mm with
It is highly relevant to be less than lmm for the shortest distance under microscope.MRI assesses CRM specificity and negative predictive value is 94%.China is clinical
2018 editions colorectal cancer guides of Society of Oncology are recommended to use pelvic cavity High-resolution MRI and judge primary tumo(u)r, metastatic in mesorectum
Lymph node, cancerous node, the relationship gap < of rectal wall outer vascular invasion and Mesorectal fascia (MRF), adjacent organs and structure
1mm is diagnosed as the CRM positive.
Currently, image doctor cuts carcinoma of the rectum ring week according to the above diagnostic criteria and high-resolution magnetic resonance imaging feature
Whether edge is diagnosed by infringement.The shape in conjunction with tumour, lymphnode metastatic and cancerous node, boundary and signal strength is needed
Judge that cancerous swelling infiltrates, and correspond in horizontal position, Coronal and sagittal image, comprehensively considers, comparison determines CRM sun repeatedly
Property.Image doctor grasps disease according to reading report and corresponding imaged image according to the above diagnosis generation reading report, clinician
Feelings formulate therapeutic scheme or judge situation of change after oncotherapy.
In face of a large amount of at present, multidimensional and multi-parameter images, iconography doctor integrates various factors in a short time, makes
Correctly, relative difficulty is timely diagnosed；Not being true to type, CRM is invaded in image, and different images diagnosis can be inconsistent, it is difficult to visitor
Provide specific probability numbers with seeing.
Summary of the invention
The present invention proposes that a kind of carcinoma of the rectum ring week based on deep neural network and is at incisxal edge MRI image automatic identifying method
System, solves the problems, such as to need in the prior art manually to identify carcinoma of the rectum ring week incisxal edge MRI image.
The technical scheme of the present invention is realized as follows:
A kind of carcinoma of the rectum ring week incisxal edge MRI image automatic identifying method based on deep neural network, comprising: FasterR-
CNN network, the FasterR-CNN network include feature extraction network, Area generation network and area-of-interest feature vector
Network；
Image is input in trained FasterR-CNN network, firstly, generating convolution using feature extraction network
Characteristic pattern；Then, convolution characteristic pattern is screened using Area generation network, generates possible ring week positive margin region；
Finally, convolution characteristic pattern and formation zone are returned and classified by area-of-interest feature vector network, final output ring
The position in all positive margin regions and probability.
Optionally, the feature extraction network is VGG structure or ZF structure.
Optionally, the Area generation network adds one layer of convolutional layer after the last layer of the feature extraction network,
All candidate frames undetermined are differentiated on the convolution characteristic pattern extracted, generating on convolution characteristic pattern to be ring week
The region of positive margin.
Optionally, the Area generation network carries out sliding window feature extraction on newly added convolutional layer.
Optionally, the Area generation network on newly added convolutional layer carry out sliding window feature extraction the step of,
It include: that original image is mapped back from newly added convolutional layer according to one-to-one point by spatial pyramid first, in characteristic pattern
Carry out positive and negative label differentiation on each position, estimate each candidate frame undetermined be target or be not target probability.
Optionally, using k scale and k length-width ratio, k × k anchor point is generated in each sliding position；For each anchor point
Distribute a two-value class label.
Optionally, the overlapping for giving actual boundary frame has highest friendship and the anchor point than IoU, alternatively, with actual boundary frame
Overlapping be more than 0.7IoU anchor point distribute a positive label；For all real border frames, to IoU lower than 0.3 it is non-just
The anchor point in face distributes a negative label.
Optionally, after generation may be for the region of ring week positive margin on convolution characteristic pattern, further includes: utilize non-pole
Big value suppressing method merges adjacent domain.
Optionally, the area-of-interest feature vector network and the Area generation network share convolution characteristic pattern, warp
The area-of-interest pond layer and the full articulamentum of subsequent two sons for crossing area-of-interest feature vector network, obtain candidate frame
Coordinate and classification probability score.
Optionally, in the training process of the FasterR-CNN network, by Area generation network and region of interest characteristic of field
Vector network carries out alternative two stage training, and parameter is constantly finely tuned in iteration, then candidate by boundingbox regression calibrations
Frame position.
The invention also provides a kind of carcinoma of the rectum ring week incisxal edge MRI image automatic identification system based on deep neural network
System, comprising: FasterR-CNN network, the FasterR-CNN network include feature extraction network, Area generation network and sense
Interest provincial characteristics vector network；
Feature extraction network is for generating convolution characteristic pattern；Area generation network screens the convolution characteristic pattern,
Generate possible ring week positive margin region；Area-of-interest feature vector network returns convolution characteristic pattern and formation zone
Return and classifies, the position in output ring week positive margin region and probability.
Optionally, the Area generation network adds one layer of convolutional layer after the last layer of the feature extraction network,
All candidate frames undetermined are differentiated on the convolution characteristic pattern extracted, generating on convolution characteristic pattern to be ring week
The region of positive margin.
Optionally, the area-of-interest feature vector network and the Area generation network share convolution characteristic pattern, warp
The area-of-interest pond layer and the full articulamentum of subsequent two sons for crossing area-of-interest feature vector network obtain prediction side
The coordinate of frame and the probability score of classification.
The beneficial effects of the present invention are:
(1) have the function of efficient, high-accuracy, high stability；
(2) to a certain degree can be to avoid error in judgement caused by doctor artificial origin
Detailed description of the invention
In order to more clearly explain the embodiment of the invention or the technical proposal in the existing technology, to embodiment or will show below
There is attached drawing needed in technical description to be briefly described, it should be apparent that, the accompanying drawings in the following description is only this
Some embodiments of invention for those of ordinary skill in the art without creative efforts, can be with
It obtains other drawings based on these drawings.
Fig. 1 is a kind of carcinoma of the rectum ring week incisxal edge MRI image automatic identifying method based on deep neural network of the present invention
One optional implementation flow chart；
Fig. 2 is the P-R curve synoptic diagram of FasterR-CNN network；
Fig. 3 is the ROC curve diagram of FasterR-CNN network；
Fig. 4 is a kind of carcinoma of the rectum ring week incisxal edge MRI image automatic recognition system based on deep neural network of the present invention
One optional implementation structural schematic diagram.
Specific embodiment
Following will be combined with the drawings in the embodiments of the present invention, and technical solution in the embodiment of the present invention carries out clear, complete
Site preparation description, it is clear that described embodiments are only a part of the embodiments of the present invention, instead of all the embodiments.It is based on
Embodiment in the present invention, it is obtained by those of ordinary skill in the art without making creative efforts every other
Embodiment shall fall within the protection scope of the present invention.
The invention discloses a kind of carcinoma of the rectum ring week incisxal edge MRI image automatic identifying method based on deep neural network,
It is final to identify using convolutional neural networks deep learning FasterR-CNN technology to carcinoma of the rectum ring week positive margin automatic identification
The coordinate and probability score of target area out.Doctor can be made after comprehensive analysis according to the area identification and probability score of acquisition
Determine diagnosis and treatment scheme and assessment prognosis, substantially increases the working efficiency of doctor.
Fig. 1 shows the carcinoma of the rectum ring week based on deep neural network one of incisxal edge MRI image automatic identifying method can
Select embodiment.
In the alternative embodiment, a kind of carcinoma of the rectum ring week incisxal edge MRI image automatic identification side based on deep neural network
Method, including FasterR-CNN network, the FasterR-CNN network include feature extraction network, Area generation network (RPN)
With area-of-interest feature vector network, image is input in trained FasterR-CNN network, firstly, using special
Sign is extracted network generation convolution characteristic pattern and is then screened using Area generation network (RPN) to convolution characteristic pattern, generates
Possible ring week positive margin region finally carries out convolution characteristic pattern and formation zone by area-of-interest feature vector network
Return (target detection) and classification, the position in final output ring week positive margin region and probability.
Wherein, feature extraction network is existing network structure, such as VGG structure or ZF structure.Area generation network is
One layer of convolutional layer is added after the last layer of feature extraction network, to all times undetermined on the convolution characteristic pattern extracted
Frame is selected to be differentiated, generating on convolution characteristic pattern to be the region of ring week positive margin.Area-of-interest feature vector net
Network and Area generation network share convolution characteristic pattern, by the area-of-interest pond layer of area-of-interest feature vector network with
And the full articulamentum of subsequent two sons, the coordinate of candidate frame and the probability score of classification can be obtained.
Using the alternative embodiment, RPN network carries out sliding window feature extraction on newly added convolutional layer, overcomes
The problem of sliding window extracts time-consuming and storage redundancy is being carried out in original image.
Optionally, the RPN network carries out the step of sliding window feature extraction, comprising: passes through spatial pyramid first
Original image is mapped back from newly added convolutional layer according to one-to-one point, designs k different scale in each position of characteristic pattern
Upper to carry out positive and negative label differentiation, convolutional layer has 4k output, encodes the coordinate of k bounding box, and classification layer exports 2k score,
Estimate each candidate frame undetermined be target or be not target probability.Optionally, using k scale and k length-width ratio, every
A sliding position generates k × k anchor point.Under default situations, using 3 scales and 3 length-width ratios, generated in each sliding position
9 anchor points.Two-value class label (be target or be not target) is distributed for each anchor point in candidate region in order to obtain.It gives
The following two kinds anchor point distributes a positive label (being target): (1) overlapping with actual boundary frame that there is highest to hand over and than (IoU)
Anchor point, alternatively, (2) it is Chong Die with actual boundary frame be more than 0.7IoU anchor point.For all real border frames, if one
The IoU of anchor point is lower than 0.3, then distributes a negative label (not being target) to non-frontal anchor point.
Optionally, after generation may be for the region of ring week positive margin on convolution characteristic pattern, the method also includes:
Merge adjacent domain using non-maxima suppression method, reduce candidate region, is subsequent recurrence (i.e. target detection) and classification
Reduce largely unnecessary compute repeatedly.
Optionally, in training process, Area generation network and area-of-interest feature vector network are subjected to two stages friendship
For training, parameter is constantly finely tuned in iteration, then by bounding box regression calibrations candidate frame position, is finally obtained optimal
The result of change.
A specific embodiment of FasterR-CNN network training process is given below.
In the training process, use and be identified as 1020 images of the CRM positive in 192 patients as training set data,
The good VGG16 with 13 convolutional layers and 3 full articulamentums of pre-training is mentioned as the feature of initialization in ImageNet
Network is taken, all weights are endowed that meet deviation zero average in Area generation network and area-of-interest feature vector network
The random number of Gaussian Profile.Training process uses two stages, and two stages respectively include 80000 candidate regions RPN
(preceding 60000 learning rates are 0.0001, and rear 20000 learning rates are 0.00001) with 40000 times based on candidate for training
(preceding 30000 learning rates are 0.0001, and rear 10000 learning rates are for the classification and recurrence of the feature vector in region
0.00001)；Amount of exercise (momentum) is 0.9, weighted delay 0.0005；The anchor scale of Area generation network is set as
1282,2562,5122, anchor ratio is set as 0.5,1,2；SGD (Stochastic Gradient is utilized in the training process
Descent) method provides the data of end-to-end backpropagation, and adjustment weighting even depth learning network parameter reduces loss function
Value, makes network convergence.
In the training process, FasterR-CNN network will not only identify marked image, it is also necessary to learn same patient and connect
Continuous normal picture is used to training and improves discrimination memory capability.Therefore, establishing 10980 includes rectum ring week incisxal edge sun
Property High-resolution MRI T2WI image database, the database cover in clinic diagnosis be used to determine occur lesion original
It swells tumor, the outer vascular invasion image of lymphnode metastatic, cancerous node and rectal wall in mesorectum, for FasterR-CNN network
Learn and remembers with normal anatomical structures, the difference of inflammatory reaction hyperplasia region.
Using 200 images in database as test set, input completes the FasterR-CNN network of training, has recorded essence
True rate (Precision) and recall rate (Recall) are simultaneously depicted as PR curve, as shown in Fig. 2, obtaining area under a curve and being
0.3670, i.e. AP=0.3670, the value show that, for carcinoma of the rectum ring week positive margin image, FasterR-CNN network carries out
Effective training.
All marked regions of test set are subjected to true positives/false positive classification, obtain the kidney-Yang under different probability threshold value
Property rate (TP) and false positive rate (FP), be depicted as ROC curve such as Fig. 3, the area under ROC curve, i.e. AUC value be calculated, should
Value reflects the accuracy of test set data markers.It is analyzed with ROC curve figure shown in Fig. 3, is calculated using trapezoidal method
Area under ROC curve, AUC=0.9534, it is possible thereby to judge that the diagnosis capability for completing the FasterR-CNN network of training is better than
Iconography expert.
In this test, FasterR-CNN network is 0.2 second to the automatic identification time of an image, with a disease
The image averaging quantity 50 of people is opened to calculate, and the automatic Diagnostic Time of artificial intelligence is 10 seconds, relative to examining for iconography expert
Disconnected (600 seconds) time has obviously advantage.
As it can be seen that the carcinoma of the rectum ring week incisxal edge MRI image automatic identifying method of the invention based on deep neural network has
Efficiently, the effect of high-accuracy, high stability, and to a certain degree can be to avoid error in judgement caused by doctor artificial origin.
Fig. 4 shows the carcinoma of the rectum ring week based on deep neural network one of incisxal edge MRI image automatic recognition system can
Select embodiment.
In the alternative embodiment, a kind of carcinoma of the rectum ring week incisxal edge MRI image automatic identification system based on deep neural network
System, including FasterR-CNN network, the FasterR-CNN network include feature extraction network, Area generation network (RPN)
With area-of-interest feature vector network, image is input in trained FasterR-CNN network, firstly, using special
Sign is extracted network generation convolution characteristic pattern and is then screened using Area generation network (RPN) to convolution characteristic pattern, generates
Possible ring week positive margin region finally carries out convolution characteristic pattern and formation zone by area-of-interest feature vector network
Return (target detection) and classification, the position in final output ring week positive margin region and probability.
Wherein, feature extraction network is existing network structure, such as VGG structure or ZF structure.Area generation network is
One layer of convolutional layer is added after the last layer of feature extraction network, to all times undetermined on the convolution characteristic pattern extracted
Frame is selected to be differentiated, generating on convolution characteristic pattern to be the region of ring week positive margin.Area-of-interest feature vector net
Network and Area generation network share convolution characteristic pattern, by the area-of-interest pond layer of area-of-interest feature vector network with
And the full articulamentum of subsequent two sons, the coordinate of candidate frame and the probability score of classification can be obtained.
Using the alternative embodiment, RPN network carries out sliding window feature extraction on newly added convolutional layer, overcomes
The problem of sliding window extracts time-consuming and storage redundancy is being carried out in original image.
Optionally, the RPN network carries out sliding window feature extraction, comprising: first by spatial pyramid according to one
Corresponding point maps back original image from newly added convolutional layer, and k different scale of design carries out on each position of characteristic pattern
Positive and negative label differentiates that convolutional layer has 4k output, encodes the coordinate of k bounding box, and classification layer exports 2k score, and estimation is every
A candidate frame undetermined be target or be not target probability.Optionally, using k scale and k length-width ratio, in each sliding
Position generates k × k anchor point.Under default situations, using 3 scales and 3 length-width ratios, 9 anchors are generated in each sliding position
Point.Two-value class label (be target or be not target) is distributed for each anchor point in candidate region in order to obtain.To following two
Kind of anchor point distributes a positive label (being target): (1) it is with actual boundary frame overlapping with highest friendship and than the anchor point of (IoU),
Alternatively, (2) it is Chong Die with actual boundary frame be more than 0.7IoU anchor point.For all real border frames, if an anchor point
IoU be lower than 0.3, then distribute a negative label (not being target) to non-frontal anchor point.
Optionally, after generation may be for the region of ring week positive margin on convolution characteristic pattern, the system also includes:
Merge adjacent domain using non-maxima suppression method, reduce candidate region, is subsequent recurrence (i.e. target detection) and classification
Reduce largely unnecessary compute repeatedly.
Optionally, in FasterR-CNN network training process, by Area generation network and area-of-interest feature vector
Network carries out alternative two stage training, parameter is constantly finely tuned in iteration, then pass through bounding box regression calibrations candidate frame
Position finally obtains the result of optimization.
The foregoing is merely illustrative of the preferred embodiments of the present invention, is not intended to limit the invention, all in essence of the invention
Within mind and principle, any modification, equivalent replacement, improvement and so on be should all be included in the protection scope of the present invention.
Claims (10)
1. a kind of carcinoma of the rectum ring week incisxal edge MRI image automatic identifying method based on deep neural network, which is characterized in that packet
Include: FasterR-CNN network, the FasterR-CNN network include feature extraction network, Area generation network and region of interest
Characteristic of field vector network；
Image is input in trained FasterR-CNN network, firstly, generating convolution feature using feature extraction network
Figure；Then, convolution characteristic pattern is screened using Area generation network, generates possible ring week positive margin region；Finally,
Convolution characteristic pattern and formation zone are returned and are classified by area-of-interest feature vector network, final output ring week incisxal edge
The position of positive region and probability.
2. a kind of carcinoma of the rectum ring week incisxal edge MRI image automatic identification side based on deep neural network as described in claim 1
Method, which is characterized in that the feature extraction network is VGG structure or ZF structure.
3. a kind of carcinoma of the rectum ring week incisxal edge MRI image automatic identification side based on deep neural network as described in claim 1
Method, which is characterized in that the Area generation network adds one layer of convolutional layer after the last layer of the feature extraction network,
All candidate frames undetermined are differentiated on the convolution characteristic pattern extracted, generating on convolution characteristic pattern may be to cut in ring week
The region of the edge positive.
4. a kind of carcinoma of the rectum ring week incisxal edge MRI image automatic identification side based on deep neural network as claimed in claim 3
Method, which is characterized in that the Area generation network carries out sliding window feature extraction on newly added convolutional layer.
5. a kind of carcinoma of the rectum ring week incisxal edge MRI image automatic identification side based on deep neural network as claimed in claim 4
Method, which is characterized in that the Area generation network is the step of carrying out sliding window feature extraction on newly added convolutional layer, packet
It includes: original image being mapped back from newly added convolutional layer according to one-to-one point by spatial pyramid first, in the every of characteristic pattern
Carry out positive and negative label differentiation on a position, estimate each candidate frame undetermined be target or be not target probability.
6. a kind of carcinoma of the rectum ring week incisxal edge MRI image automatic identification side based on deep neural network as described in claim 1
Method, which is characterized in that the area-of-interest feature vector network and the Area generation network share convolution characteristic pattern pass through
The area-of-interest pond layer of area-of-interest feature vector network and the full articulamentum of subsequent two sons, obtain candidate frame
The probability score of coordinate and classification.
7. a kind of carcinoma of the rectum ring week incisxal edge MRI image automatic identification side based on deep neural network as described in claim 1
Method, which is characterized in that in the FasterR-CNN network training process, by Area generation network and region of interest characteristic of field to
It measures network and carries out alternative two stage training, parameter is constantly finely tuned in iteration, then is candidate by bounding box regression calibrations
Frame position.
8. a kind of carcinoma of the rectum ring week incisxal edge MRI image automatic recognition system based on deep neural network, which is characterized in that packet
Include: FasterR-CNN network, the FasterR-CNN network include feature extraction network, Area generation network and region of interest
Characteristic of field vector network；
Feature extraction network is for generating convolution characteristic pattern；
Area generation network screens the convolution characteristic pattern, generates possible ring week positive margin region；
Area-of-interest feature vector network is returned and is classified to convolution characteristic pattern and formation zone, output ring week incisxal edge sun
The position in property region and probability.
9. a kind of carcinoma of the rectum ring week incisxal edge MRI image automatic identification system based on deep neural network as claimed in claim 8
System, which is characterized in that the Area generation network adds one layer of convolutional layer after the last layer of the feature extraction network,
All candidate frames undetermined are differentiated on the convolution characteristic pattern extracted, generating on convolution characteristic pattern may be to cut in ring week
The region of the edge positive.
10. a kind of carcinoma of the rectum ring week incisxal edge MRI image automatic identification system based on deep neural network as claimed in claim 8
System, which is characterized in that the area-of-interest feature vector network and the Area generation network share convolution characteristic pattern pass through
The area-of-interest pond layer of area-of-interest feature vector network and the full articulamentum of subsequent two sons obtain prediction frame
Coordinate and classification probability score.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910298188.5A CN110033042A (en) | 2019-04-15 | 2019-04-15 | A kind of carcinoma of the rectum ring week incisxal edge MRI image automatic identifying method and system based on deep neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910298188.5A CN110033042A (en) | 2019-04-15 | 2019-04-15 | A kind of carcinoma of the rectum ring week incisxal edge MRI image automatic identifying method and system based on deep neural network |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN110033042A true CN110033042A (en) | 2019-07-19 |
Family
ID=67238273
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910298188.5A Pending CN110033042A (en) | 2019-04-15 | 2019-04-15 | A kind of carcinoma of the rectum ring week incisxal edge MRI image automatic identifying method and system based on deep neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110033042A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110991535A (en) * | 2019-12-04 | 2020-04-10 | 中山大学 | pCR prediction method based on multi-type medical data |
| CN111127400A (en) * | 2019-11-29 | 2020-05-08 | 深圳蓝韵医学影像有限公司 | Method and device for detecting breast lesions |
Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101449291A (en) * | 2006-03-24 | 2009-06-03 | Im3D有限公司 | Process and system for automatically recognising preneoplastic abnormalities in anatomical structures, and corresponding computer program |
| US20170053398A1 (en) * | 2015-08-19 | 2017-02-23 | Colorado Seminary, Owner and Operator of University of Denver | Methods and Systems for Human Tissue Analysis using Shearlet Transforms |
| CN107451615A (en) * | 2017-08-01 | 2017-12-08 | 广东工业大学 | Thyroid papillary carcinoma Ultrasound Image Recognition Method and system based on Faster RCNN |
| CN108550133A (en) * | 2018-03-02 | 2018-09-18 | 浙江工业大学 | A kind of cancer cell detection method based on Faster R-CNN |
| CN108717554A (en) * | 2018-05-22 | 2018-10-30 | 复旦大学附属肿瘤医院 | A kind of thyroid tumors histopathologic slide image classification method and its device |
| CN109118485A (en) * | 2018-08-13 | 2019-01-01 | 复旦大学 | Digestive endoscope image classification based on multitask neural network cancer detection system early |
| CN109242844A (en) * | 2018-09-04 | 2019-01-18 | 青岛大学附属医院 | Pancreatic tumour automatic recognition system based on deep learning, computer equipment, storage medium |
| CN109584218A (en) * | 2018-11-15 | 2019-04-05 | 首都医科大学附属北京友谊医院 | A kind of construction method of gastric cancer image recognition model and its application |
-
2019
- 2019-04-15 CN CN201910298188.5A patent/CN110033042A/en active Pending
Patent Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101449291A (en) * | 2006-03-24 | 2009-06-03 | Im3D有限公司 | Process and system for automatically recognising preneoplastic abnormalities in anatomical structures, and corresponding computer program |
| US20170053398A1 (en) * | 2015-08-19 | 2017-02-23 | Colorado Seminary, Owner and Operator of University of Denver | Methods and Systems for Human Tissue Analysis using Shearlet Transforms |
| CN107451615A (en) * | 2017-08-01 | 2017-12-08 | 广东工业大学 | Thyroid papillary carcinoma Ultrasound Image Recognition Method and system based on Faster RCNN |
| CN108550133A (en) * | 2018-03-02 | 2018-09-18 | 浙江工业大学 | A kind of cancer cell detection method based on Faster R-CNN |
| CN108717554A (en) * | 2018-05-22 | 2018-10-30 | 复旦大学附属肿瘤医院 | A kind of thyroid tumors histopathologic slide image classification method and its device |
| CN109118485A (en) * | 2018-08-13 | 2019-01-01 | 复旦大学 | Digestive endoscope image classification based on multitask neural network cancer detection system early |
| CN109242844A (en) * | 2018-09-04 | 2019-01-18 | 青岛大学附属医院 | Pancreatic tumour automatic recognition system based on deep learning, computer equipment, storage medium |
| CN109584218A (en) * | 2018-11-15 | 2019-04-05 | 首都医科大学附属北京友谊医院 | A kind of construction method of gastric cancer image recognition model and its application |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111127400A (en) * | 2019-11-29 | 2020-05-08 | 深圳蓝韵医学影像有限公司 | Method and device for detecting breast lesions |
| CN110991535A (en) * | 2019-12-04 | 2020-04-10 | 中山大学 | pCR prediction method based on multi-type medical data |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111539930B (en) | Dynamic ultrasonic breast nodule real-time segmentation and identification method based on deep learning | |
| CN109242844A (en) | Pancreatic tumour automatic recognition system based on deep learning, computer equipment, storage medium | |
| ES2914387T3 (en) | immediate study | |
| US11051790B2 (en) | System comprising indicator features in high-resolution micro-ultrasound images | |
| CN109447065A (en) | A kind of method and device of breast image identification | |
| US8116546B2 (en) | Lung cancer diagnosis using magnetic resonance imaging data obtained at three time points | |
| CN108520518A (en) | A kind of thyroid tumors Ultrasound Image Recognition Method and its device | |
| CN108268870A (en) | Multi-scale feature fusion ultrasonoscopy semantic segmentation method based on confrontation study | |
| Zhang et al. | Intelligent scanning: Automated standard plane selection and biometric measurement of early gestational sac in routine ultrasound examination | |
| CN108364293A (en) | A kind of on-line training thyroid tumors Ultrasound Image Recognition Method and its device | |
| CN109512464A (en) | A kind of disorder in screening and diagnostic system | |
| KR20200080626A (en) | Method for providing information of lesion diagnosis and device for providing information of lesion diagnosis using the same | |
| CN110197713A (en) | A kind of processing method of medical imaging, device, equipment and medium | |
| Rodriguez et al. | Optical fish trajectory measurement in fishways through computer vision and artificial neural networks | |
| CN109363698A (en) | A kind of method and device of breast image sign identification | |
| US20100183210A1 (en) | Computer-assisted analysis of colonic polyps by morphology in medical images | |
| KR20190119731A (en) | Method for predicting cancer suspicious area and device for predicting cancer suspicious area using the same | |
| US20200143532A1 (en) | Method and apparatus for analyzing abdominal disease based on medical image | |
| CN108765427A (en) | A kind of prostate image partition method | |
| CN104217437A (en) | Lesion area partitioning method for prostate KVCT image | |
| CN110033042A (en) | A kind of carcinoma of the rectum ring week incisxal edge MRI image automatic identifying method and system based on deep neural network | |
| CN115661107A (en) | Image analysis method, system and equipment based on bladder cancer risk stratification | |
| CN113633315A (en) | Ultrasonic image processing method using ultrasonic image processing system | |
| Kim et al. | Deep convolutional neural network for classification of thyroid nodules on ultrasound: Comparison of the diagnostic performance with that of radiologists | |
| CN112489788A (en) | Multi-modal image analysis method and system for cancer diagnosis |
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 |