CN109727227A - A kind of diagnosis of thyroid illness method based on SPECT image - Google Patents
A kind of diagnosis of thyroid illness method based on SPECT image Download PDFInfo
- Publication number
- CN109727227A CN109727227A CN201811402574.6A CN201811402574A CN109727227A CN 109727227 A CN109727227 A CN 109727227A CN 201811402574 A CN201811402574 A CN 201811402574A CN 109727227 A CN109727227 A CN 109727227A
- Authority
- CN
- China
- Prior art keywords
- weight
- thyroid
- network
- feature
- parameter
- 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
- 238000000034 method Methods 0.000 title claims abstract description 55
- 238000002603 single-photon emission computed tomography Methods 0.000 title claims abstract description 24
- 210000001685 thyroid gland Anatomy 0.000 title claims abstract description 17
- 238000003745 diagnosis Methods 0.000 title claims abstract description 16
- 238000013527 convolutional neural network Methods 0.000 claims abstract description 18
- 238000012549 training Methods 0.000 claims abstract description 11
- 208000024799 Thyroid disease Diseases 0.000 claims abstract description 9
- 208000021510 thyroid gland disease Diseases 0.000 claims abstract description 7
- 230000008569 process Effects 0.000 claims description 7
- 238000001514 detection method Methods 0.000 claims description 5
- 238000010801 machine learning Methods 0.000 claims description 5
- 230000009466 transformation Effects 0.000 claims description 2
- 238000013135 deep learning Methods 0.000 abstract description 6
- 201000010099 disease Diseases 0.000 description 8
- 208000037265 diseases, disorders, signs and symptoms Diseases 0.000 description 8
- 239000011159 matrix material Substances 0.000 description 7
- 206010020850 Hyperthyroidism Diseases 0.000 description 4
- 208000001204 Hashimoto Disease Diseases 0.000 description 3
- 208000030836 Hashimoto thyroiditis Diseases 0.000 description 3
- 206010061218 Inflammation Diseases 0.000 description 3
- 208000024770 Thyroid neoplasm Diseases 0.000 description 3
- 230000008901 benefit Effects 0.000 description 3
- 238000011156 evaluation Methods 0.000 description 3
- 230000006870 function Effects 0.000 description 3
- 208000003532 hypothyroidism Diseases 0.000 description 3
- 230000002989 hypothyroidism Effects 0.000 description 3
- 238000003384 imaging method Methods 0.000 description 3
- 230000004054 inflammatory process Effects 0.000 description 3
- 125000000325 methylidene group Chemical group [H]C([H])=* 0.000 description 3
- 238000002604 ultrasonography Methods 0.000 description 3
- 238000013528 artificial neural network Methods 0.000 description 2
- 238000004364 calculation method Methods 0.000 description 2
- 238000002405 diagnostic procedure Methods 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 238000002474 experimental method Methods 0.000 description 2
- 210000004907 gland Anatomy 0.000 description 2
- 230000009467 reduction Effects 0.000 description 2
- 239000005495 thyroid hormone Substances 0.000 description 2
- 229940036555 thyroid hormone Drugs 0.000 description 2
- 241000208340 Araliaceae Species 0.000 description 1
- 235000005035 Panax pseudoginseng ssp. pseudoginseng Nutrition 0.000 description 1
- 235000003140 Panax quinquefolius Nutrition 0.000 description 1
- 208000009453 Thyroid Nodule Diseases 0.000 description 1
- AUYYCJSJGJYCDS-LBPRGKRZSA-N Thyrolar Chemical class IC1=CC(C[C@H](N)C(O)=O)=CC(I)=C1OC1=CC=C(O)C(I)=C1 AUYYCJSJGJYCDS-LBPRGKRZSA-N 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000001574 biopsy Methods 0.000 description 1
- 238000009534 blood test Methods 0.000 description 1
- 238000004422 calculation algorithm Methods 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000003759 clinical diagnosis Methods 0.000 description 1
- 230000007812 deficiency Effects 0.000 description 1
- 230000001419 dependent effect Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000018109 developmental process Effects 0.000 description 1
- 238000002059 diagnostic imaging Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 235000013399 edible fruits Nutrition 0.000 description 1
- 230000002124 endocrine Effects 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 238000001914 filtration Methods 0.000 description 1
- 235000008434 ginseng Nutrition 0.000 description 1
- 238000010191 image analysis Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 230000003902 lesion Effects 0.000 description 1
- 230000003211 malignant effect Effects 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 238000007620 mathematical function Methods 0.000 description 1
- 230000004060 metabolic process Effects 0.000 description 1
- 239000000203 mixture Substances 0.000 description 1
- 238000003062 neural network model Methods 0.000 description 1
- 238000009206 nuclear medicine Methods 0.000 description 1
- 238000005192 partition Methods 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 102000004169 proteins and genes Human genes 0.000 description 1
- 108090000623 proteins and genes Proteins 0.000 description 1
- 238000009877 rendering Methods 0.000 description 1
- 238000012216 screening Methods 0.000 description 1
- 238000012706 support-vector machine Methods 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
- 206010043688 thyroid adenoma Diseases 0.000 description 1
- 201000002510 thyroid cancer Diseases 0.000 description 1
- 238000013526 transfer learning Methods 0.000 description 1
Abstract
The present invention provides a kind of diagnosis of thyroid illness method based on SPECT image, image classification is carried out using a kind of convolutional neural networks of improved DenseNet network structure, increase parameter influential for weight in the parallel link in Dense block, and the weight of the characteristic pattern of former each layer is made to carry out dynamic adjustment in training, so that network has greater flexibility, classification performance is improved.Embodiment shows that this method can obtain the performance better than other deep learning methods.The present invention can be widely applied to the diagnosis and other image classification problems of thyroid disease.
Description
Technical field
The present invention relates to a kind of diagnosis of thyroid illness methods based on SPECT image.
Background technique
Thyroid gland can produce thyroid hormone, play a crucial role in control human metabolism.First shape
Parathyrine (T4) and trilute (T3) are the two kinds of active thyroid hormones generated by thyroid gland, they have human body
Very big help, generation, body heat regulation and energy production and adjusting including protein.Thyroid disease is endocrine field
The second largest disease[1], serious thyroid disease may cause dead [1-3].
Clinically Diagnosis of Thyroid Diseases commonly because be known as very much, as clinical evaluation, blood test, imageological examination,
Biopsy etc..Wherein image method is a kind of very important thyroid gland diagnostic method, these images mainly include ultrasound, CT,
SPECT etc..Ultrasound is a kind of convenience, imaging method in real time, economic, is usually used in Clinical screening and judges the property of thyroid nodule
Matter.In recent years, the method based on convolutional neural networks also be used to identify Benign And Malignant Nodules of Thyroid Glands [4,5].CT images can be with
For finding the diseases such as thyroid adenoma and cancer, neural network method can be used for region thyreoidea regional partition and volumetric estimate
[6]。
SPECT is a kind of nuclear medicine using conventional gamma camera acquisition image data.SPECT imaging system is by pacifying
The conventional gamma camera composition of one or more on frame, detector can surround patient while collecting these images
Precisely and automatically rotate.It is substantially three-dimensional that the major advantage of SPECT imaging, which is the image generated,.Due to SPECT image
It is able to reflect thyroid function situation, is not changed in thyroid physical aspect, and when function generation obstacle, ultrasound
It can not just detected with CT images, therefore SPECT image can just find in time disease in disease early stage, identify difficult first
It plays an important role in shape gland disease.Clinical assistant diagnosis can reduce doctor's false diagnosis due to caused by the factors such as fatigue,
Carry out the auxiliary diagnosis work based on SPECT image, can be improved the accuracy of clinical diagnosis.
Machine learning is a kind of important aided diagnosis method, has been largely used to the detection and diagnosis based on medical image
In.Supervised learning is a kind of important machine learning method, is learnt by the training sample and corresponding disease label of image
Mathematical function, and any class disease [7-8] judged until the lesion in image or belong to.Main supervised learning algorithm includes
Neural network, support vector machine and deep learning method.Deep learning is a kind of preferable machine learning method of application effect, with
The development of graphics processing unit (GPU), deep learning breakthrough performance is obtained in various medical applications.Convolutional Neural
Network (CNN) is a kind of widely applied deep learning method [9-18] in medical image analysis field.CNN is with 2D or 3D rendering
As input, there are multilayered structure, including pond layer, convolutional layer, RELU layers and full articulamentum etc., there is local sensing, weight
The characteristics of shared and more convolution kernels, to significantly reduce the calculation amount of the quantity of parameters in neural network model.
Classify although existing CNN method is used directly for SPECT image to thyroid disease, realizes thyroid gland
The diagnosis of disease, but these existing methods have that accuracy is low, performance is bad.The present invention is directed to this problem, mentions
A kind of diagnosis of thyroid illness method based on SPECT image out.
Summary of the invention
It is applied to the problem that accuracy is low, performance is bad existing for SPECT image, the present invention for existing CNN method
A kind of diagnosis of thyroid illness method based on SPECT image is proposed, this method uses a kind of improved CNN network structure,
Trainable weight parameter is added in parallel link by the structure, allows the network to the ginseng for learning weight during the training period
Number overcomes the problems, such as that parallel link existing for original network will lead to information redundancy and reduce network performance, to improve knowledge
The accuracy of other method.
DenseNet is a kind of widely applied CNN network structure [18].Its main feature is that alleviating ladder by intensively connection
The problem of degree disappears reinforces feature propagation, reduces parameter amount.In network each layer of input be all the output of all layers of front and
Collection, and the characteristic pattern that this layer is learnt can also be directly transmitted to be used as input for all layers behind.DenseNet utilizes every layer of reduction
Calculation amount and feature multiplexing improve the efficiency of network, by all layers after allowing l layers of input to directly influence,
At l layers, output and input relationship have
yl=Fl([x0,x1,...,xl-1],{Wl}), (1)
Wherein, l indicates the current number of plies, ylIt is the output of this layer;[x0,x1,...,xl-1] be 0,1 ..., in l-1 layers
The characteristic pattern (feature map) of generation, merges (concatenation) with the dimension in channel;FlIndicate non-linear change
It changes, including BN, the latticed forms such as convolution of ReLU and 3x3, WlIndicate FlParameter.
The present invention improves network structure on the basis of DenseNet.Parallel link in DenseNet is with identical
Weight connect all pervious features, but not all pervious feature is all useful, therefore this connection will lead to
Information redundancy and reduction network performance.
The present invention provides a kind of diagnosis of thyroid illness method based on SPECT image, by using convolutional neural networks
Machine learning method classify to SPECT image, achieve the purpose that detection or identification thyroid disease, aforementioned convolution mind
DenseNet network structure or improved DenseNet network structure are used through network, it is characterised in that:
First, aforementioned convolutional neural networks connection has the feature that in each intensive connection dense block module
Trainable weight parameter is added in each parallel link by the inside, and initialization value is set as 1, to will not influence training
Preceding weight, in the forward propagation process, by each layer of feature in network and corresponding multiplied by weight, obtain in this way
The l layers of relationship output and input are
yl=Fl([x0·kl,0,x1·kl,1,...,xl-1·kl,l-1],{Wl}), (2)
Wherein, l indicates the current number of plies, ylIt is the output of this layer;[x0,x1,...,xl-1] be 0,1 ..., in l-1 layers
The characteristic pattern of generation;FlIndicate nonlinear transformation, latticed form including but not limited to below: the convolution of BN, ReLU and 3x3;Wl
Indicate FlParameter;kl,0,kl,1,...,kl,l-1It refers to working as x0,x1,...,xl-1X is determined when being connected to l layers0,x1,...,xl-1
Weight parameter;
Second, it is aforementioned that aforementioned convolutional neural networks have the feature that network learns during the training period in learning process
The parameter of weight, in back-propagation process, the value of aforementioned weight parameter indicates the influence degree of individual features figure, when this is corresponding
Characteristic pattern in classification task comprising more useful information or when playing main, join by the corresponding weight of the individual features figure
Number will be relatively large;And working as this feature figure includes less useful information in classification task, perhaps plays a secondary role or does not rise
When effect, which will be relatively small.
Above-mentioned improvement has the benefit that the feature weight due to each layer is no longer fixed, net in connection
Network has greater flexibility, and has the ability for filtering invalid feature;Meanwhile the pond layer in network is replaced by extension
Convolutional layer, to save the useful information of feature as much as possible.
The present invention utilizes the improved network structure and learning process of DenseNet, feature summation can be overcome to weaken scarce
Point obtains more accurate testing result.With reference to the accompanying drawing, specific implementation example and its advantages are made further
It is bright.
Detailed description of the invention
The improved network structure of Fig. 1
The specific network information of Fig. 2 embodiment
The mean accuracy curve of Fig. 3 difference the number of iterations
The confusion matrix of Fig. 4 distinct methods
Specific embodiment
With reference to the accompanying drawing, to the specific of the diagnostic method of the thyroid disease provided by the invention based on SPECT image
Embodiment is described as follows:
Fig. 1 gives improved network structure, and increase parallel link in intensively connection dense block module can
Training weight parameter.Fig. 2 gives a kind of specific network information of embodiment, and realization of the invention including but not limited to should
The network information.
Web vector graphic deep learning frame PyTorch in the specific embodiment of the invention is realized, is arranged according to Fig. 2 and is connected
Each layer of network, according to the mode of learning of Fig. 1 and formula (2) setting network.The DenseNet121 provided frame is loaded,
Network is trained in advance using transfer learning method ImageNet, and then it is finely adjusted with SPECT image data set.It uses
SGD trains network, momentum 0.9.Each include 5 images in small batches, and each image size is 255 × 255.Initial study
Rate is set as 0.001, and the learning rate of every 5 cycles of training is originally 1/10, and loss function is set as intersecting entropy loss.
The performance of the method for the present invention for further evaluation has carried out identical 10 experiments in the present embodiment, and to knot
Fruit is averaged.The present invention is equipped with two NVIDIA Geforce 1080Ti GPU, an Intel Xeon E5- at one
The work station of 2620 CPU carries out experiment.
Four kinds of common thyroid diseases of SPECT diagnostic imaging that the data set of this embodiment uses, including thyroid gland
Hyperfunction, hypothyroidism, methylene inflammation and Hashimoto's disease.In the SPECT data set, there are 800 width normal thyroids
Image, the image of 650 width hyperthyroidism, the image of 200 width hypothyroidism, the image of 650 width methylene inflammation, and
The image of 650 width bridge this hyperthyroidism.
Illustrate the beneficial effect that embodiment of the invention provides below.The present invention compares more on the same data set
The performance of a network, including DeaveNet121, ReNet101, VGG19 and EnEntuv3;Multiple indexs are also compared, including are divided
Class precision, precision, recall rate, F1 score and confusion matrix.1 be the results are shown in Table to table 6.
1 hyperthyroidism classification indicators of table
Precision | Recall rate | F1 score | |
The method of the present invention | 1.00 | 1.00 | 1.00 |
DenseNet121 | 1.00 | 1.00 | 1.00 |
ResNet101 | 1.00 | 0.99 | 0.99 |
InceptionV3 | 1.00 | 1.00 | 1.00 |
VGG19 | 1.00 | 1.00 | 1.00 |
The classification indicators of 2 hypothyroidism of table
Precision | Recall rate | F1-score | |
The method of the present invention | 0.93 | 0.94 | 0.93 |
DenseNet121 | 0.92 | 0.87 | 0.89 |
ResNet101 | 0.92 | 0.89 | 0.90 |
InceptionV3 | 0.91 | 0.87 | 0.89 |
VGG19 | 0.93 | 0.84 | 0.88 |
3 methylene inflammation classification indicators of table
Precision | Recall rate | F1-score | |
The method of the present invention | 0.95 | 0.91 | 0.92 |
DenseNet121 | 0.88 | 0.92 | 0.90 |
ResNet101 | 0.89 | 0.93 | 0.91 |
InceptionV3 | 0.87 | 0.91 | 0.89 |
VGG19 | 0.85 | 0.93 | 0.89 |
4 Hashimoto thyroiditis classification indicators of table
Precision | Recall rate | F1-score | |
The method of the present invention | 1.00 | 1.00 | 1.00 |
DenseNet121 | 1.00 | 1.00 | 1.00 |
ResNet101 | 0.99 | 0.99 | 0.99 |
InceptionV3 | 1.00 | 0.99 | 0.99 |
VGG19 | 1.00 | 0.99 | 1.00 |
The normal classification indicators of table 5
Precision | Recall rate | F1-score | |
The method of the present invention | 1.00 | 1.00 | 1.00 |
DenseNet121 | 1.00 | 1.00 | 1.00 |
ResNet101 | 0.99 | 1.00 | 1.00 |
InceptionV3 | 0.99 | 1.00 | 0.99 |
VGG19 | 0.99 | 1.00 | 1.00 |
Table 6 is averaged classification indicators
Average Precision | Average Recall rate | Average F1-score | |
The method of the present invention | 0.98 | 0.97 | 0.97 |
DenseNet121 | 0.97 | 0.96 | 0.96 |
ResNet101 | 0.97 | 0.96 | 0.97 |
InceptionV3 | 0.96 | 0.96 | 0.96 |
VGG19 | 0.96 | 0.96 | 0.96 |
The experimental results showed that this method has higher detection and identification accuracy than other convolutional neural networks methods,
Classification performance is also superior to other methods.For hyperthyroidism, Hashimoto thyroiditis and normal is proposed by the present invention
Method and other most methods can obtain good accuracy.For other two classes, due to training sample or sample distribution
Deficiency, all methods have some classification errors.However be compared with other methods, network proposed by the present invention is to SPECT first
Shape gland image has stronger ability in feature extraction, and can make full use of all potential informations in image.Even if therefore two classes
Image has less training data or unreasonable sample distribution, method classification error in all methods proposed by the present invention
Rate is still minimum.
The comparison of the mean accuracy of the different the number of iterations of distinct methods is given in Fig. 3, and method of the invention can be
Optimal precision is obtained under different the number of iterations.This means that the network proposed has excellent general classification performance, without
Dependent on the number of iterations.
Confusion matrix, also known as error matrix are to indicate to carry out classification method in the matrix form in computer learning field
A kind of method of evaluation.The concrete class that each list diagram picture of confusion matrix is classified into, the sum of each column indicate real
Border is classified as the amount of images of classification, and every a line indicates the real property classification of image, and the sum of every a line indicates the category
Image instance quantity.Fig. 4 gives the confusion matrix of distinct methods.Classification error proposed by the present invention is minimum, it is seen that
Its method has the detection accuracy better than other convolutional neural networks methods.
Bibliography
[1]Z.Parry,R.Macnab.Thyroid disease and thyroid surgery.Anaesthesia&
Intensive Care Medicine,18(10):488-495,2017.
[2]C-Y.Chang,S-J.Chen,M-F.Tsai.Application of support-vector-machine-
based method for feature selection and classification of thyroid nodules in
ultrasound images.Pattern Recognition,43(10):3494-3506,2010.
[3]U.Raghavendra,A.Gudigar,M.Maithri,A.Gertych,et al.Optimized multi-
level elongated quinary patterns for the assessment of thyroid nodules in
ultrasound images.Computers in Biology and Medicine,95(1):55-62,2018.
[4]J.Chi,E.Walia,P.Babyn,J.Wang,G.Groot,M.Eramian.Thyroid nodule
classification in ultrasound images by fine-tuning deep convolutional neural
network.Journal of Digital Imaging,30(4):477-486,2017.
[5]J.Ma,F.Wu,T.Jiang,J.Zhu,D.Kong.Cascade convolutional neural
networks for automatic detection of thyroid nodules in ultrasound
images.Medical Physics,44(5):1678-1691,2017.
[6]C.Chang,Y.Hong,P.Chung,C.Tseng.A neural network for thyroid
segmentation and volume estimation in CT images.IEEE Computational
Intelligence Magazine,6(4):43-55,2011.
[7]Y.LeCun,Y.Bengio,G.Hinton.Deep learning.Nature,521(7553):436-444,
2015.
[8]J.Schmidhuber.Deep learning in neural networks:An overview.Neural
Networks,61(1):85-117,2015.
[9]J.Ker,L.Wang,J.Rao,T.Lim.Deep learning applications in medical
image analysis.IEEE Access,6(1):9375-9389,2018.
[10]D.Shen,G.Wu,H.Suk.Deep learning in medical image analysis.Annual
Review of Biomedical Engineering,19(1):221-248,2017.
[11]G.Litjens,T.Kooi,B.Bejnordi,A.Setio,et al.A survey on deep
learning in medical image analysis.Medical Image Analysis,42(1):60-88,2017.
[12]G Evgin,G Numan.Deep learning in medical image analysis:Recent
advances and future trends.In:Proceedings of the international conferences on
computer graphics,visualization,computer vision and image processing
(CGVCVIP),Lisbon,Portugal,pp.305-310,2017.
[13]K.He,X.Zhang,S.Ren,J.Sun,Deep residual learning for image
recognition.In Proceedings of IEEE conference on computer vision and pattern
recognition(CVPR),pp.770-778,2016.
[14]N.Srivastava,G.Hinton,A.Krizhevsky,et al.Dropout:a simple way to
prevent neural networks from overfitting.Journal of Machine Learning
Research,15(1):1929-1958,2014.
[15]S.Ioffe,C.Szegedy.Batch normalization:accelerating deep network
training by reducing internal covariate shift.In Proceedings of international
conference on machine learning(JMLR),pp.448-456,2015.
[16]S.Hoochang,H.Roth,M.Gao,et al.Deep convolutional neural networks
for computer-aided detection:CNN architectures,dataset characteristics and
transfer learning.IEEE Transactions on Medical Imaging,35(5):1285-1298,2016.
[17]H.Zhang.M.Cisse,Y.Dauphin,et al.Mixup:Beyond empirical risk
minimization.In Proceedings of international conference on learning
representations(ICML),2018.
[18]G.Huang,Z.Liu Z,L.Maaten,K.Weinberger.Densely connected
convolutional networks.In Proceedings of IEEE conference on computer vision
and pattern recognition(CVPR),pp.2261-2269,2017.
Claims (1)
1. a kind of diagnosis of thyroid illness method based on SPECT image, by using the machine learning side of convolutional neural networks
Method classifies to SPECT image, achievees the purpose that detection or identification thyroid disease, aforementioned convolutional neural networks use
DenseNet network structure or improved DenseNet network structure, it is characterised in that:
First, aforementioned convolutional neural networks connection has the feature that in each intensive connection dense block module
Trainable weight parameter is added in each parallel link by face, and initialization value is set as 1, thus before will not influence training
Weight, in the forward propagation process, by each layer of feature in network and corresponding multiplied by weight, obtain in this way in l
The relationship that outputs and inputs of layer be
yl=Fl([x0·kl,0,x1·kl,1,...,xl-1·kl,l-1],{Wl}), (2)
Wherein, l indicates the current number of plies, ylIt is the output of this layer;[x0,x1,...,xl-1] be 0,1 ..., generated in l-1 layers
Characteristic pattern;FlIndicate nonlinear transformation, latticed form including but not limited to below: the convolution of BN, ReLU and 3x3;WlIt indicates
FlParameter;kl,0,kl,1,...,kl,l-1It refers to working as x0,x1,...,xl-1X is determined when being connected to l layers0,x1,...,xl-1Power
The parameter of weight;
Second, aforementioned convolutional neural networks have the feature that network learns aforementioned weight during the training period in learning process
Parameter, in back-propagation process, the value of aforementioned weight parameter indicates the influence degree of individual features figure, when the individual features
For figure in classification task comprising more useful information or when playing main, the corresponding weight parameter of the individual features figure will
It is relatively large;And working as this feature figure includes less useful information in classification task, perhaps plays a secondary role or does not work
When, which will be relatively small.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201811402574.6A CN109727227A (en) | 2018-11-23 | 2018-11-23 | A kind of diagnosis of thyroid illness method based on SPECT image |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201811402574.6A CN109727227A (en) | 2018-11-23 | 2018-11-23 | A kind of diagnosis of thyroid illness method based on SPECT image |
Publications (1)
Publication Number | Publication Date |
---|---|
CN109727227A true CN109727227A (en) | 2019-05-07 |
Family
ID=66295106
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201811402574.6A Pending CN109727227A (en) | 2018-11-23 | 2018-11-23 | A kind of diagnosis of thyroid illness method based on SPECT image |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109727227A (en) |
Cited By (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111340906A (en) * | 2020-02-25 | 2020-06-26 | 哈尔滨工业大学(深圳)(哈尔滨工业大学深圳科技创新研究院) | SPECT (single photon emission computed tomography) tomographic image reconstruction method, device and equipment combining ART (ART) and UNet algorithm |
CN112070755A (en) * | 2020-09-14 | 2020-12-11 | 内江师范学院 | New coronary pneumonia image identification method based on combination of deep learning and transfer learning |
CN112070089A (en) * | 2020-09-23 | 2020-12-11 | 西安交通大学医学院第二附属医院 | Ultrasonic image-based intelligent diagnosis method and system for diffuse thyroid diseases |
CN114926486A (en) * | 2022-05-12 | 2022-08-19 | 哈尔滨工业大学人工智能研究院有限公司 | Thyroid ultrasound image intelligent segmentation method based on multi-level improvement |
Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107680082A (en) * | 2017-09-11 | 2018-02-09 | 宁夏医科大学 | Lung tumor identification method based on depth convolutional neural networks and global characteristics |
CN108171232A (en) * | 2017-11-15 | 2018-06-15 | 中山大学 | The sorting technique of bacillary and viral children Streptococcus based on deep learning algorithm |
-
2018
- 2018-11-23 CN CN201811402574.6A patent/CN109727227A/en active Pending
Patent Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107680082A (en) * | 2017-09-11 | 2018-02-09 | 宁夏医科大学 | Lung tumor identification method based on depth convolutional neural networks and global characteristics |
CN108171232A (en) * | 2017-11-15 | 2018-06-15 | 中山大学 | The sorting technique of bacillary and viral children Streptococcus based on deep learning algorithm |
Non-Patent Citations (1)
Title |
---|
LIYONG MA: "Diagnosis of Thyroid Diseases Using SPECT Images Based on Convolutional Neural Network" * |
Cited By (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111340906A (en) * | 2020-02-25 | 2020-06-26 | 哈尔滨工业大学(深圳)(哈尔滨工业大学深圳科技创新研究院) | SPECT (single photon emission computed tomography) tomographic image reconstruction method, device and equipment combining ART (ART) and UNet algorithm |
CN112070755A (en) * | 2020-09-14 | 2020-12-11 | 内江师范学院 | New coronary pneumonia image identification method based on combination of deep learning and transfer learning |
CN112070089A (en) * | 2020-09-23 | 2020-12-11 | 西安交通大学医学院第二附属医院 | Ultrasonic image-based intelligent diagnosis method and system for diffuse thyroid diseases |
CN114926486A (en) * | 2022-05-12 | 2022-08-19 | 哈尔滨工业大学人工智能研究院有限公司 | Thyroid ultrasound image intelligent segmentation method based on multi-level improvement |
CN114926486B (en) * | 2022-05-12 | 2023-02-07 | 哈尔滨工业大学人工智能研究院有限公司 | Thyroid ultrasound image intelligent segmentation method based on multi-level improvement |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Li et al. | Benign and malignant classification of mammogram images based on deep learning | |
Al-Antari et al. | Evaluation of deep learning detection and classification towards computer-aided diagnosis of breast lesions in digital X-ray mammograms | |
Rouhi et al. | Benign and malignant breast tumors classification based on region growing and CNN segmentation | |
CN109727227A (en) | A kind of diagnosis of thyroid illness method based on SPECT image | |
Hu et al. | AS-Net: Attention Synergy Network for skin lesion segmentation | |
Le et al. | Liver tumor segmentation from MR images using 3D fast marching algorithm and single hidden layer feedforward neural network | |
Xu et al. | Convolutional-neural-network-based approach for segmentation of apical four-chamber view from fetal echocardiography | |
Zheng et al. | MDCC-Net: multiscale double-channel convolution U-Net framework for colorectal tumor segmentation | |
Ma et al. | Diagnosis of thyroid diseases using SPECT images based on convolutional neural network | |
Weimin et al. | Enhancing Liver Segmentation: A Deep Learning Approach with EAS Feature Extraction and Multi-Scale Fusion | |
Zhan et al. | CFNet: A medical image segmentation method using the multi-view attention mechanism and adaptive fusion strategy | |
Aslam et al. | Liver-tumor detection using CNN ResUNet | |
Ru et al. | Attention guided neural ODE network for breast tumor segmentation in medical images | |
Yadav et al. | Deep learning-based CAD system design for thyroid tumor characterization using ultrasound images | |
Zhi et al. | Deep neural network pulmonary nodule segmentation methods for CT images: Literature review and experimental comparisons | |
Zheng et al. | Segmentation of thyroid glands and nodules in ultrasound images using the improved U-Net architecture | |
Kwong et al. | A survey on deep learning approaches for breast cancer diagnosis | |
Tao et al. | Tooth CT Image Segmentation Method Based on the U-Net Network and Attention Module. | |
Wang et al. | Triplanar convolutional neural network for automatic liver and tumor image segmentation | |
Xiao et al. | PET and CT image fusion of lung cancer with siamese pyramid fusion network | |
Wang et al. | 2.75 D: Boosting learning by representing 3D Medical imaging to 2D features for small data | |
Ma et al. | AMSeg: A Novel Adversarial Architecture based Multi-scale Fusion Framework for Thyroid Nodule Segmentation | |
Raina et al. | Slim u-net: Efficient anatomical feature preserving u-net architecture for ultrasound image segmentation | |
Deng et al. | BE-FNet: 3D bounding box estimation feature pyramid network for accurate and efficient maxillary sinus segmentation | |
Zhong et al. | Mix Transformer depth-wise separable convolution UNet for breast mass segmentation in mammographic |
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 |
Application publication date: 20190507 |
|
WD01 | Invention patent application deemed withdrawn after publication |