CN111598875A - Method, system and device for building thyroid nodule automatic detection model - Google Patents
Method, system and device for building thyroid nodule automatic detection model Download PDFInfo
- Publication number
- CN111598875A CN111598875A CN202010417220.XA CN202010417220A CN111598875A CN 111598875 A CN111598875 A CN 111598875A CN 202010417220 A CN202010417220 A CN 202010417220A CN 111598875 A CN111598875 A CN 111598875A
- Authority
- CN
- China
- Prior art keywords
- thyroid
- thyroid nodule
- detection model
- training
- image
- 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
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
- G06T7/0012—Biomedical image inspection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10132—Ultrasound image
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Medical Informatics (AREA)
- Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
- Radiology & Medical Imaging (AREA)
- Quality & Reliability (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a method, a system and a device for building an automatic thyroid nodule detection model based on a convolutional neural network, wherein the method comprises the following steps: denoising thyroid ultrasound image data to obtain a thyroid ultrasound image training data set; training a thyroid nodule detection model using a Yolov3 network based on a training dataset; training a thyroid nodule benign and malignant recognition model by using a Resnet network based on a training data set; and fusing the thyroid nodule detection model and the thyroid nodule benign and malignant identification model to generate the thyroid nodule automatic detection model.
Description
Technical Field
The invention relates to the technical field of artificial intelligence, in particular to a method, a system and a device for building an automatic thyroid nodule detection model based on a convolutional neural network.
Background
In the past 20 years, the detection of thyroid nodules is increasing, and as most of the thyroid nodules are, accurate judgment of thyroid nodules and nodular character is crucial to patients, so that on one hand, the medical cost of the patients in detection can be greatly reduced, and more expensive pain measures such as puncture and biopsy are avoided; more importantly, the precise diagnosis of the nodule and its malignancy and malignancy is critical to the treatment of the patient. Ultrasound is a common means for thyroid nodule detection, and radiologists summarize ultrasound features for detecting malignant tumors, which include hypoechogenicity, no halo, microcalcification, solidity, intra-nodule blood flow, etc. Based on these features, international general thyroid imaging reporting and data system (TI-RADS) standards were developed that strictly classify thyroid nodules and malignancies for reference by radiologists. TI-RADS classifies thyroid nodules into six classes 2, 3, 4a, 4b, 4c, and 5, meaning no nodules, possibly benign nodules, one suspect feature, two suspect features, three or more suspect features, and possibly malignant nodules, respectively. The TI-RADS standard is then evaluated as a current standard for ultrasonically diagnosing thyroid malignant nodules, the time is consumed, the effect is not good, the detection accuracy is often related to personal experience of doctors, and the change of thyroid nodule echo patterns limits the judgment capability of radiologists.
On the other hand, since the ultrasound features contained in the ultrasound image can be digitally processed, such automatic detection using a machine learning method is natural, and thus an automatic or semi-automatic classification system based on image features will be possible. Image features are first extracted using machine learning, and many manual thyroid ultrasound image features using different methods have been studied extensively in recent years. And using the extracted features, executing a supervision classification task through an existing machine learning classifier, such as a support vector machine in common use, so as to perform a thyroid nodule automatic detection task. However, when the machine learning method is used for extracting features, a large amount of manpower and experienced experts are needed, time and labor are consumed, and the extracted features are not good in effect.
With the revival of deep learning, the automatic feature extraction by using a Convolutional Neural Network (CNN) becomes a general method for processing tasks such as automatic image recognition, detection, segmentation and the like. The CNN extraction features have two advantages, one is that the CNN extraction features are obtained through network automatic learning, manual extraction is not needed, and the CNN extraction features are simple and easy to implement, so that the CNN extraction features are more effective than manual extraction by a machine learning method; secondly, the features extracted by the CNN are more robust, can adapt to image changes caused by shape changes and the like due to camera lenses, different illumination conditions, different postures, partial occlusion, horizontal and vertical movement, and obtain better effects. Therefore, the research of using CNN to carry out the automatic identification work of thyroid nodules is published in succession, and a very good effect is obtained.
In view of the current state of research, although research on benign and malignant thyroid nodules has been carried out, in actual work, the thyroid nodules need to be located first, and then the benign and malignant thyroid nodules can be identified, and the work in this respect is lacked, so that further research is needed for automatic detection of thyroid nodules.
Disclosure of Invention
The invention aims to provide a method, a system and a device for building an automatic thyroid nodule detection model based on a convolutional neural network, and aims to solve the problems in the prior art.
The invention provides a method for constructing an automatic thyroid nodule detection model based on a convolutional neural network, which comprises the following steps:
denoising thyroid ultrasound image data to obtain a thyroid ultrasound image training data set;
training a thyroid nodule detection model using a Yolov3 network based on a training dataset;
training a thyroid nodule benign and malignant recognition model by using a Resnet network based on a training data set;
and fusing the thyroid nodule detection model and the thyroid nodule benign and malignant identification model to generate the thyroid nodule automatic detection model.
The invention provides a thyroid nodule automatic detection model construction system based on a convolutional neural network, which comprises the following steps:
the noise reduction module is used for reducing noise of the thyroid ultrasound image data to obtain a thyroid ultrasound image training data set;
a first training module for training a thyroid nodule detection model using a Yolov3 network based on a training data set;
the second training module is used for training a thyroid nodule benign and malignant recognition model by using a Resnet network based on a training data set;
and the fusion module is used for fusing the thyroid nodule detection model with the thyroid nodule benign and malignant identification model to generate a thyroid nodule automatic detection model.
The embodiment of the invention also provides a device for constructing the automatic thyroid nodule detection model based on the convolutional neural network, which comprises the following components: the automatic thyroid nodule detection model building method based on the convolutional neural network comprises a memory, a processor and a computer program which is stored on the memory and can run on the processor, wherein the computer program is executed by the processor to realize the steps of the automatic thyroid nodule detection model building method based on the convolutional neural network.
The embodiment of the invention also provides a computer readable storage medium, wherein an implementation program for information transmission is stored on the computer readable storage medium, and when the program is executed by a processor, the steps of the thyroid nodule automatic detection model construction method based on the convolutional neural network are realized.
By adopting the embodiment of the invention, the automatic identification and detection technology of the deep learning image is applied, and the latest detection and identification framework is used to complete the automatic detection task of the thyroid nodule and identify the nodule, so that the detection work can be completed at the early stage of thyroid cancer, and thus expensive work such as puncture and the like is not used to help patients to screen, and doctors are assisted to complete the detection and benign and malignant screening of the thyroid nodule.
The foregoing description is only an overview of the technical solutions of the present invention, and the embodiments of the present invention are described below in order to make the technical means of the present invention more clearly understood and to make the above and other objects, features, and advantages of the present invention more clearly understandable.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described below, and it is obvious that the drawings in the following description are some embodiments of the present invention, and other drawings can be obtained by those skilled in the art without creative efforts.
FIG. 1 is a flow chart of a method for building an automatic thyroid nodule detection model based on a convolutional neural network according to an embodiment of the present invention;
FIG. 2 is a schematic diagram of a thyroid nodule automatic detection model construction method according to an embodiment of the present invention;
FIG. 3 is a schematic diagram of a Yolov3 network according to an embodiment of the present invention;
FIG. 4 is a schematic diagram of a convolutional neural network-based thyroid nodule automatic detection model construction system according to an embodiment of the present invention;
fig. 5 is a schematic diagram of a convolutional neural network-based thyroid nodule automatic detection model construction device according to an embodiment of the present invention.
Detailed Description
The embodiment of the invention provides a method for constructing an automatic thyroid detection and identification model based on a convolutional neural network, which comprises the following steps: first, the data set is denoised. The original thyroid ultrasound image has information such as instruments and dates, and has a large influence on the identification effect, so that image morphological content noise reduction processing needs to be performed on the data. Second, a thyroid nodule detection model is constructed. The invention uses the latest Yolov3 network architecture to carry out model building, and the Yolov3 has the best effect on data sets such as Coco, has much higher speed than classical Faster-Rcnn and has little difference in precision, and is widely used on many detection tasks. Third, the thyroid nodule benign and malignant auto-recognition task. The invention uses the latest ResNet network architecture to build the model, and ResNet has the best effect on ImageNet data sets and is widely used in various image recognition and detection tasks. Fourth, difficult excavation. In training the thyroid nodule automatic identification model Yolov3, some samples are difficult to learn by a network, so learning of the samples by adding extra weight is required, and a method of hard mining is used. Fifthly, the thyroid nodule automatic detection framework is fused with the nodule benign and malignant automatic identification framework. In order to fit practical application, an end-to-end mode for automatically identifying thyroid nodule and detecting benign and malignant identification needs to be formed, and the thyroid nodule and the benign and malignant identification need to be fused together.
The technical solutions of the present invention will be described clearly and completely with reference to the following embodiments, and it should be understood that the described embodiments are some, but not all, embodiments of the present invention. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
In the description of the present invention, it is to be understood that the terms "center", "longitudinal", "lateral", "length", "width", "thickness", "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", "clockwise", "counterclockwise", and the like, indicate orientations and positional relationships based on those shown in the drawings, and are used only for convenience of description and simplicity of description, and do not indicate or imply that the device or element being referred to must have a particular orientation, be constructed and operated in a particular orientation, and thus, should not be considered as limiting the present invention.
Furthermore, the terms "first", "second" and "first" are used for descriptive purposes only and are not to be construed as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Thus, features defined as "first", "second", may explicitly or implicitly include one or more of the described features. In the description of the present invention, "a plurality" means two or more unless specifically defined otherwise. Furthermore, the terms "mounted," "connected," and "connected" are to be construed broadly and may, for example, be fixedly connected, detachably connected, or integrally connected; can be mechanically or electrically connected; they may be connected directly or indirectly through intervening media, or they may be interconnected between two elements. The specific meanings of the above terms in the present invention can be understood in specific cases to those skilled in the art.
Method embodiment
According to an embodiment of the present invention, a method for building an automatic thyroid nodule detection model based on a convolutional neural network is provided, fig. 1 is a flowchart of the method for building an automatic thyroid nodule detection model based on a convolutional neural network according to an embodiment of the present invention, and fig. 2 is a schematic diagram of the method for building an automatic thyroid nodule detection model according to an embodiment of the present invention, as shown in fig. 1 and 2, the method for building an automatic thyroid nodule detection model based on a convolutional neural network according to an embodiment of the present invention specifically includes:
101, denoising thyroid ultrasound image data to obtain a thyroid ultrasound image training data set; step 101 specifically includes: carrying out graying processing on the thyroid ultrasound image to obtain a binarized image; and (3) performing image opening operation on the basis of the binary image, namely corroding the image and then expanding the image to finish the noise reduction of the thyroid ultrasound image data.
Specifically, the quality of an image preprocessing algorithm directly relates to the effect of subsequent image processing, such as image segmentation, target recognition, edge extraction, and the like, and in order to obtain a high-quality digital image, noise reduction processing is often required to be performed on the image, so that the integrity of original information is maintained as much as possible, and useless information in a signal can be removed. Although various image noise reduction algorithms are added as if the image noise reduction algorithms are new in the spring of the rain, many methods have a general disadvantage that the details or the edge information of the image is often lost while the noise is reduced. The thyroid ultrasound images have a significant portion of noise information such as instrumentation, time, thumbnail images, and others that are noisy, and therefore it is necessary to remove these noises before training the test. Firstly, carrying out gray processing on an image, wherein the threshold value is 0, and obtaining a binarized image; and performing image opening operation on the basis of the binary image. The image opening operation and closing operation are related to the expansion and corrosion operation, and are composed of the combination of the expansion operation and the corrosion operation and the set operation. The opening operation is to corrode the image and then expand the image, so that noise can be removed, and original image information is kept unchanged.
102, training a thyroid nodule detection model by using a Yolov3 network based on a training data set;
fig. 3 is a schematic diagram of the Yolov3 network according to the embodiment of the present invention, as shown in fig. 3, Yolov3 is one of the most popular detection frameworks at present, and as the name suggests, there are 2 versions before Yolov3, where the fundamental improvement of Yolov3 is: the network structure is adjusted, multi-scale features are used for object detection, and object classification Logistic replaces softmax. The new network structure is Darknet-53, the idea of Resnet is borrowed, residual modules are added into the network, and therefore the gradient problem of the deep-level network is solved, each residual module is composed of two convolution layers and one short connection, no pooling layer and a full connection layer are arranged in the whole v3 structure, the down-sampling of the network is achieved by setting the stride of convolution to be 2, and the size of an image is reduced to half after passing through the convolution layer. The multi-scale detection is used for predicting a plurality of scales, and the specific form is realized by performing upsampling and splicing operations on certain final layers of network prediction, so that the characteristics on different scales can be learned. When predicting the object type, softmax is not used, and the output of logistic is used for prediction instead. This enables multi-tagged objects to be supported. When the Yolov3 is used for training a thyroid detection model, only thyroid nodules are detected, and good and malignant identification is not carried out on the nodules, so that the model only focuses on the detection of the nodules, and the nodule detection rate is improved.
Step 102 specifically includes:
mining a training data set by adopting a Focal loss mode, reducing the weight of a simple negative sample in training, and training a thyroid nodule detection model by using a Yolov3 network based on the training data after the Focal loss mining. Specifically, the method comprises the following steps:
samples are encountered in a training detection model framework and are difficult to train, positive samples which are easy to predict by a network in negative samples are often called difficult negative samples, and the training of the difficult negative samples is greatly helpful for improving the classification performance of the network. The method can effectively select the difficult samples by using an initial sample set (to train a network and then to predict the remaining negative samples in the negative sample set by using the trained network, wherein the negative samples with the highest score, namely the negative samples which are most easily judged as positive samples, are selected as the difficult samples, but the efficiency is low, and the training is laborious and time-consuming, so that the method adopts Focal loss to mine, wherein the Focal loss is mainly used for solving the problem of serious imbalance of the proportion of the positive samples and the negative samples in target detection, as shown in formula 1, the loss function reduces the weight of a large number of simple negative samples in the training.
focalloss＝-y′alog(y′)-(1-y′n) log (y') equation 1;
wherein a is a hyper-parameter greater than 0, in order to reduce the loss of easily classifiable samples, so that more attention is paid to difficult, misclassified samples; and y' is the network output result.
Therefore, the network training can pay more attention to the difficult samples, and meanwhile, the samples do not need to be manually selected, so that the network effect is improved.
103, training a thyroid nodule benign and malignant identification model by using a Resnet network based on a training data set; specifically, the method comprises the following steps:
the thyroid recognition model uses a classical Resnet model, the problems of consumption of computing resources, easiness in overfitting of the model, disappearance of gradients and the like are generally accompanied by the increase of the number of network layers in the deep science, and the problems are solved through the technical means such as GPU clustering, Batch Normalization, Dropout and the like. From the perspective of information theory, in the forward transmission process, as the layer number increases, the image information contained in the feature map decreases layer by layer, and Resnet uses a residual structure to do identity mapping to solve the problem perfectly, so that excellent effects are obtained in each visual task, and the Resnet is used as an alphago basic network. The present invention therefore uses Resnet to train thyroid identification models to automatically identify the malignancy and benign of the nodules.
And step 104, fusing the thyroid nodule detection model and the thyroid nodule benign and malignant identification model to generate a thyroid nodule automatic detection model.
Specifically, the thyroid nodule detection model provided by the embodiment of the invention is only responsible for detecting nodules, so that the nodule detection rate can be improved; the thyroid nodule benign and malignant identification model is only responsible for identifying the nodule benign and malignant, so that the benign and malignant detection accuracy can be improved, but the thyroid nodule benign and malignant identification model and the thyroid nodule benign and malignant identification model are separated, so that the thyroid nodule benign and malignant identification model and the thyroid nodule malignant identification model are inconvenient to use, and therefore the thyroid nodule benign and malignant identification model and the thyroid nodule malignant identification model need to. Firstly, a thyroid nodule detection model is used for detecting whether nodules exist in a test image, and if not, the thyroid nodule detection model is not sent into a thyroid nodule benign and malignant identification model for benign and malignant identification; if the confidence coefficient is higher than 0.7, the thyroid nodule benign and malignant identification model is sent to carry out benign and malignant identification. This allows nodule detection nodule identification to be accomplished in an end-to-end fashion.
In the embodiment of the present invention, the detection of the model is further required, and the specific operations include:
inputting thyroid ultrasonic image test data into a thyroid nodule detection model, if the output result of the thyroid nodule detection model is that a thyroid nodule exists, inputting a thyroid ultrasonic image with the confidence coefficient higher than a preset value into a thyroid nodule benign and malignant identification model, and outputting a final thyroid nodule benign and malignant result; if the output result of the thyroid nodule detection model is that no thyroid nodule exists, directly outputting the result; and detecting the final output result of the thyroid nodule automatic detection model, so as to detect the thyroid nodule automatic detection model, and finally obtaining the trained thyroid nodule automatic detection model if the thyroid nodule automatic detection model passes the detection.
Finally, the thyroid ultrasonic image can be automatically detected through the trained thyroid nodule automatic detection model.
As can be seen from the above description, the data is denoised and boosted before the thyroid identification model is trained. And (4) constructing a thyroid detection model by using Yolov3 to automatically detect the nodule on the thyroid ultrasound image. The most popular ResNet is used for constructing an automatic thyroid identification model for thyroid benign and malignant identification. And 4, carrying out difficult excavation on the thyroid ultrasound image detection frame by using Focal loss, and improving the detection effect. And (3) using a thyroid nodule fusion automatic detection frame and a nodule benign and malignant automatic identification frame to finish end-to-end detection and identification.
System embodiment
According to an embodiment of the present invention, a thyroid nodule automatic detection model building system based on a convolutional neural network is provided, fig. 4 is a schematic diagram of a thyroid nodule automatic detection model building system based on a convolutional neural network according to an embodiment of the present invention, and as shown in fig. 4, the thyroid nodule automatic detection model building system based on a convolutional neural network according to an embodiment of the present invention specifically includes:
the noise reduction module 40 is configured to reduce noise of the thyroid ultrasound image data to obtain a thyroid ultrasound image training data set; the noise reduction module 40 is specifically configured to:
carrying out graying processing on the thyroid ultrasound image to obtain a binarized image;
and (3) performing image opening operation on the basis of the binary image, namely corroding the image and then expanding the image to finish the noise reduction of the thyroid ultrasound image data.
A first training module 42, configured to train a thyroid nodule detection model using a Yolov3 network based on a training data set; the first training module 42 is specifically configured to:
mining a training data set by adopting a Focal loss mode, reducing the weight of a simple negative sample in training, and training a thyroid nodule detection model by using a Yolov3 network based on the training data after the Focal loss mining.
A second training module 44, configured to train a thyroid nodule benign and malignant recognition model using a Resnet network based on the training data set;
and the fusion module 46 is used for fusing the thyroid nodule detection model and the thyroid nodule benign and malignant identification model to generate a thyroid nodule automatic detection model.
In an embodiment of the present invention, the system further includes:
the test module is used for inputting the thyroid ultrasonic image test data into the thyroid nodule detection model, if the output result of the thyroid nodule detection model is that thyroid nodules exist, the thyroid ultrasonic image with the confidence coefficient higher than the preset value is input into the thyroid nodule benign and malignant identification model, and the final thyroid nodule benign and malignant result is output; if the output result of the thyroid nodule detection model is that no thyroid nodule exists, directly outputting the result; detecting the final output result of the automatic thyroid nodule detection model so as to detect the automatic thyroid nodule detection model, and finally obtaining a trained automatic thyroid nodule detection model if the detection is passed;
and the detection module is used for automatically detecting the thyroid ultrasonic image through the trained thyroid nodule automatic detection model.
The embodiment of the present invention is a system embodiment corresponding to the above method embodiment, and specific operations of each module may be understood by referring to the description in the above method embodiment, which is not described herein again.
Apparatus embodiment one
The embodiment of the invention provides a convolutional neural network-based thyroid nodule automatic detection model construction device, as shown in fig. 5, comprising: a memory 50, a processor 52 and a computer program stored on the memory 50 and executable on the processor 52, which computer program, when executed by the processor 52, carries out the following method steps:
101, denoising thyroid ultrasound image data to obtain a thyroid ultrasound image training data set; step 101 specifically includes: carrying out graying processing on the thyroid ultrasound image to obtain a binarized image; and (3) performing image opening operation on the basis of the binary image, namely corroding the image and then expanding the image to finish the noise reduction of the thyroid ultrasound image data.
Specifically, the quality of an image preprocessing algorithm directly relates to the effect of subsequent image processing, such as image segmentation, target recognition, edge extraction, and the like, and in order to obtain a high-quality digital image, noise reduction processing is often required to be performed on the image, so that the integrity of original information is maintained as much as possible, and useless information in a signal can be removed. Although various image noise reduction algorithms are added as if the image noise reduction algorithms are new in the spring of the rain, many methods have a general disadvantage that the details or the edge information of the image is often lost while the noise is reduced. The thyroid ultrasound images have a significant portion of noise information such as instrumentation, time, thumbnail images, and others that are noisy, and therefore it is necessary to remove these noises before training the test. Firstly, carrying out gray processing on an image, wherein the threshold value is 0, and obtaining a binarized image; and performing image opening operation on the basis of the binary image. The image opening operation and closing operation are related to the expansion and corrosion operation, and are composed of the combination of the expansion operation and the corrosion operation and the set operation. The opening operation is to corrode the image and then expand the image, so that noise can be removed, and original image information is kept unchanged.
102, training a thyroid nodule detection model by using a Yolov3 network based on a training data set;
fig. 3 is a schematic diagram of the Yolov3 network according to the embodiment of the present invention, as shown in fig. 3, Yolov3 is one of the most popular detection frameworks at present, and as the name suggests, there are 2 versions before Yolov3, where the fundamental improvement of Yolov3 is: the network structure is adjusted, multi-scale features are used for object detection, and object classification Logistic replaces softmax. The new network structure is Darknet-53, the idea of Resnet is borrowed, residual modules are added into the network, and therefore the gradient problem of the deep-level network is solved, each residual module is composed of two convolution layers and one short connection, no pooling layer and a full connection layer are arranged in the whole v3 structure, the down-sampling of the network is achieved by setting the stride of convolution to be 2, and the size of an image is reduced to half after passing through the convolution layer. The multi-scale detection is used for predicting a plurality of scales, and the specific form is realized by performing upsampling and splicing operations on certain final layers of network prediction, so that the characteristics on different scales can be learned. When predicting the object type, softmax is not used, and the output of logistic is used for prediction instead. This enables multi-tagged objects to be supported. When the Yolov3 is used for training a thyroid detection model, only thyroid nodules are detected, and good and malignant identification is not carried out on the nodules, so that the model only focuses on the detection of the nodules, and the nodule detection rate is improved.
Step 102 specifically includes:
mining a training data set by adopting a Focal loss mode, reducing the weight of a simple negative sample in training, and training a thyroid nodule detection model by using a Yolov3 network based on the training data after the Focal loss mining. Specifically, the method comprises the following steps:
samples are encountered in a training detection model framework and are difficult to train, positive samples which are easy to predict by a network in negative samples are often called difficult negative samples, and the training of the difficult negative samples is greatly helpful for improving the classification performance of the network. The method can effectively select the difficult samples by using an initial sample set (to train a network and then to predict the remaining negative samples in the negative sample set by using the trained network, wherein the negative samples with the highest score, namely the negative samples which are most easily judged as positive samples, are selected as the difficult samples, but the efficiency is low, and the training is laborious and time-consuming, so that the method adopts Focal loss to mine, wherein the Focal loss is mainly used for solving the problem of serious imbalance of the proportion of the positive samples and the negative samples in target detection, as shown in formula 1, the loss function reduces the weight of a large number of simple negative samples in the training.
focalloss＝-y′alog(y′)-(1-y′a) log (y') equation 1;
wherein a is a hyper-parameter greater than 0, in order to reduce the loss of easily classifiable samples, so that more attention is paid to difficult, misclassified samples; and y' is the network output result.
Therefore, the network training can pay more attention to the difficult samples, and meanwhile, the samples do not need to be manually selected, so that the network effect is improved.
103, training a thyroid nodule benign and malignant identification model by using a Resnet network based on a training data set; specifically, the method comprises the following steps:
the thyroid recognition model uses a classical Resnet model, the problems of consumption of computing resources, easiness in overfitting of the model, disappearance of gradients and the like are generally accompanied by the increase of the number of network layers in the deep science, and the problems are solved through the technical means such as GPU clustering, Batch Normalization, Dropout and the like. From the perspective of information theory, in the forward transmission process, as the layer number increases, the image information contained in the feature map decreases layer by layer, and Resnet uses a residual structure to do identity mapping to solve the problem perfectly, so that excellent effects are obtained in each visual task, and the Resnet is used as an alphago basic network. The present invention therefore uses Resnet to train thyroid identification models to automatically identify the malignancy and benign of the nodules.
And step 104, fusing the thyroid nodule detection model and the thyroid nodule benign and malignant identification model to generate a thyroid nodule automatic detection model.
Specifically, the thyroid nodule detection model provided by the embodiment of the invention is only responsible for detecting nodules, so that the nodule detection rate can be improved; the thyroid nodule benign and malignant identification model is only responsible for identifying the nodule benign and malignant, so that the benign and malignant detection accuracy can be improved, but the thyroid nodule benign and malignant identification model and the thyroid nodule benign and malignant identification model are separated, so that the thyroid nodule benign and malignant identification model and the thyroid nodule malignant identification model are inconvenient to use, and therefore the thyroid nodule benign and malignant identification model and the thyroid nodule malignant identification model need to. Firstly, a thyroid nodule detection model is used for detecting whether nodules exist in a test image, and if not, the thyroid nodule detection model is not sent into a thyroid nodule benign and malignant identification model for benign and malignant identification; if the confidence coefficient is higher than 0.7, the thyroid nodule benign and malignant identification model is sent to carry out benign and malignant identification. This allows nodule detection nodule identification to be accomplished in an end-to-end fashion.
In the embodiment of the present invention, the detection of the model is further required, and the specific operations include:
inputting thyroid ultrasonic image test data into a thyroid nodule detection model, if the output result of the thyroid nodule detection model is that a thyroid nodule exists, inputting a thyroid ultrasonic image with the confidence coefficient higher than a preset value into a thyroid nodule benign and malignant identification model, and outputting a final thyroid nodule benign and malignant result; if the output result of the thyroid nodule detection model is that no thyroid nodule exists, directly outputting the result; and detecting the final output result of the thyroid nodule automatic detection model, so as to detect the thyroid nodule automatic detection model, and finally obtaining the trained thyroid nodule automatic detection model if the thyroid nodule automatic detection model passes the detection.
Finally, the thyroid ultrasonic image can be automatically detected through the trained thyroid nodule automatic detection model.
Device embodiment II
The embodiment of the present invention provides a computer-readable storage medium, on which an implementation program for information transmission is stored, and when being executed by the processor 52, the implementation program implements the following method steps:
101, denoising thyroid ultrasound image data to obtain a thyroid ultrasound image training data set; step 101 specifically includes: carrying out graying processing on the thyroid ultrasound image to obtain a binarized image; and (3) performing image opening operation on the basis of the binary image, namely corroding the image and then expanding the image to finish the noise reduction of the thyroid ultrasound image data.
Specifically, the quality of an image preprocessing algorithm directly relates to the effect of subsequent image processing, such as image segmentation, target recognition, edge extraction, and the like, and in order to obtain a high-quality digital image, noise reduction processing is often required to be performed on the image, so that the integrity of original information is maintained as much as possible, and useless information in a signal can be removed. Although various image noise reduction algorithms are added as if the image noise reduction algorithms are new in the spring of the rain, many methods have a general disadvantage that the details or the edge information of the image is often lost while the noise is reduced. The thyroid ultrasound images have a significant portion of noise information such as instrumentation, time, thumbnail images, and others that are noisy, and therefore it is necessary to remove these noises before training the test. Firstly, carrying out gray processing on an image, wherein the threshold value is 0, and obtaining a binarized image; and performing image opening operation on the basis of the binary image. The image opening operation and closing operation are related to the expansion and corrosion operation, and are composed of the combination of the expansion operation and the corrosion operation and the set operation. The opening operation is to corrode the image and then expand the image, so that noise can be removed, and original image information is kept unchanged.
102, training a thyroid nodule detection model by using a Yolov3 network based on a training data set;
fig. 3 is a schematic diagram of the Yolov3 network according to the embodiment of the present invention, as shown in fig. 3, Yolov3 is one of the most popular detection frameworks at present, and as the name suggests, there are 2 versions before Yolov3, where the fundamental improvement of Yolov3 is: the network structure is adjusted, multi-scale features are used for object detection, and object classification Logistic replaces softmax. The new network structure is Darknet-53, the idea of Resnet is borrowed, residual modules are added into the network, and therefore the gradient problem of the deep-level network is solved, each residual module is composed of two convolution layers and one short connection, no pooling layer and a full connection layer are arranged in the whole v3 structure, the down-sampling of the network is achieved by setting the stride of convolution to be 2, and the size of an image is reduced to half after passing through the convolution layer. The multi-scale detection is used for predicting a plurality of scales, and the specific form is realized by performing upsampling and splicing operations on certain final layers of network prediction, so that the characteristics on different scales can be learned. When predicting the object type, softmax is not used, and the output of logistic is used for prediction instead. This enables multi-tagged objects to be supported. When the Yolov3 is used for training a thyroid detection model, only thyroid nodules are detected, and good and malignant identification is not carried out on the nodules, so that the model only focuses on the detection of the nodules, and the nodule detection rate is improved.
Step 102 specifically includes:
mining a training data set by adopting a Focal loss mode, reducing the weight of a simple negative sample in training, and training a thyroid nodule detection model by using a Yolov3 network based on the training data after the Focal loss mining. Specifically, the method comprises the following steps:
samples are encountered in a training detection model framework and are difficult to train, positive samples which are easy to predict by a network in negative samples are often called difficult negative samples, and the training of the difficult negative samples is greatly helpful for improving the classification performance of the network. The method can effectively select the difficult samples by using an initial sample set (to train a network and then to predict the remaining negative samples in the negative sample set by using the trained network, wherein the negative samples with the highest score, namely the negative samples which are most easily judged as positive samples, are selected as the difficult samples, but the efficiency is low, and the training is laborious and time-consuming, so that the method adopts Focal loss to mine, wherein the Focal loss is mainly used for solving the problem of serious imbalance of the proportion of the positive samples and the negative samples in target detection, as shown in formula 1, the loss function reduces the weight of a large number of simple negative samples in the training.
focalloss＝-y′alog(y′)-(1-y′a) log (y') equation 1;
wherein a is a hyper-parameter greater than 0, in order to reduce the loss of easily classifiable samples, so that more attention is paid to difficult, misclassified samples; and y' is the network output result.
Therefore, the network training can pay more attention to the difficult samples, and meanwhile, the samples do not need to be manually selected, so that the network effect is improved.
103, training a thyroid nodule benign and malignant identification model by using a Resnet network based on a training data set; specifically, the method comprises the following steps:
the thyroid recognition model uses a classical Resnet model, the problems of consumption of computing resources, easiness in overfitting of the model, disappearance of gradients and the like are generally accompanied by the increase of the number of network layers in the deep science, and the problems are solved through the technical means such as GPU clustering, Batch Normalization, Dropout and the like. From the perspective of information theory, in the forward transmission process, as the layer number increases, the image information contained in the feature map decreases layer by layer, and Resnet uses a residual structure to do identity mapping to solve the problem perfectly, so that excellent effects are obtained in each visual task, and the Resnet is used as an alphago basic network. The present invention therefore uses Resnet to train thyroid identification models to automatically identify the malignancy and benign of the nodules.
And step 104, fusing the thyroid nodule detection model and the thyroid nodule benign and malignant identification model to generate a thyroid nodule automatic detection model.
Specifically, the thyroid nodule detection model provided by the embodiment of the invention is only responsible for detecting nodules, so that the nodule detection rate can be improved; the thyroid nodule benign and malignant identification model is only responsible for identifying the nodule benign and malignant, so that the benign and malignant detection accuracy can be improved, but the thyroid nodule benign and malignant identification model and the thyroid nodule benign and malignant identification model are separated, so that the thyroid nodule benign and malignant identification model and the thyroid nodule malignant identification model are inconvenient to use, and therefore the thyroid nodule benign and malignant identification model and the thyroid nodule malignant identification model need to. Firstly, a thyroid nodule detection model is used for detecting whether nodules exist in a test image, and if not, the thyroid nodule detection model is not sent into a thyroid nodule benign and malignant identification model for benign and malignant identification; if the confidence coefficient is higher than 0.7, the thyroid nodule benign and malignant identification model is sent to carry out benign and malignant identification. This allows nodule detection nodule identification to be accomplished in an end-to-end fashion.
In the embodiment of the present invention, the detection of the model is further required, and the specific operations include:
inputting thyroid ultrasonic image test data into a thyroid nodule detection model, if the output result of the thyroid nodule detection model is that a thyroid nodule exists, inputting a thyroid ultrasonic image with the confidence coefficient higher than a preset value into a thyroid nodule benign and malignant identification model, and outputting a final thyroid nodule benign and malignant result; if the output result of the thyroid nodule detection model is that no thyroid nodule exists, directly outputting the result; and detecting the final output result of the thyroid nodule automatic detection model, so as to detect the thyroid nodule automatic detection model, and finally obtaining the trained thyroid nodule automatic detection model if the thyroid nodule automatic detection model passes the detection.
Finally, the thyroid ultrasonic image can be automatically detected through the trained thyroid nodule automatic detection model.
In summary, according to the embodiments of the present invention, the automatic deep learning image recognition and detection technology is applied, and the latest detection and recognition framework is used to complete the automatic thyroid nodule detection task and identify the thyroid nodule, so that the detection task can be completed at the early stage of thyroid cancer, and therefore expensive tasks such as puncturing are not used to help patients to screen, and doctors are assisted to complete the detection and benign and malignant screening of thyroid nodules.
The computer-readable storage medium of this embodiment includes, but is not limited to: ROM, RAM, magnetic or optical disks, and the like.
It will be apparent to those skilled in the art that the modules or steps of the present invention described above may be implemented by a general purpose computing device, they may be centralized on a single computing device or distributed across a network of multiple computing devices, and alternatively, they may be implemented by program code executable by a computing device, such that they may be stored in a storage device and executed by a computing device, and in some cases, the steps shown or described may be performed in an order different than that described herein, or they may be separately fabricated into individual integrated circuit modules, or multiple ones of them may be fabricated into a single integrated circuit module. Thus, the present invention is not limited to any specific combination of hardware and software.
Finally, it should be noted that: the above embodiments are only used to illustrate the technical solution of the present invention, and not to limit the same; while the invention has been described in detail and with reference to the foregoing embodiments, it will be understood by those skilled in the art that: the technical solutions described in the foregoing embodiments may still be modified, or some or all of the technical features may be equivalently replaced; and the modifications or the substitutions do not make the essence of the corresponding technical solutions depart from the scope of the technical solutions of the embodiments of the present invention.
Claims (10)
1. A method for building an automatic thyroid nodule detection model based on a convolutional neural network is characterized by comprising the following steps:
denoising thyroid ultrasound image data to obtain a thyroid ultrasound image training data set;
training a thyroid nodule detection model using a Yolov3 network based on the training dataset;
training a thyroid nodule benign and malignant recognition model using a Resnet network based on the training data set;
and fusing the thyroid nodule detection model and the thyroid nodule benign and malignant identification model to generate a thyroid nodule automatic detection model.
2. The method of claim 1, wherein training a thyroid nodule detection model using a Yolov3 network based on the training dataset specifically comprises:
mining a training data set by means of Focalloss, reducing the weight of a simple negative sample in training, and training a thyroid nodule detection model by using a Yolov3 network based on training data mined by Focal loss.
3. The method of claim 1, wherein de-noising the thyroid ultrasound image data specifically comprises:
carrying out graying processing on the thyroid ultrasound image to obtain a binarized image;
and (3) performing image opening operation on the basis of the binary image, namely corroding the image and then expanding the image to finish the noise reduction of the thyroid ultrasound image data.
4. The method of claim 1, further comprising:
inputting the thyroid ultrasonic image test data into a thyroid nodule detection model, and if the output result of the thyroid nodule detection model is that a thyroid nodule exists, inputting a thyroid ultrasonic image with the confidence coefficient higher than a preset value into the thyroid nodule benign and malignant identification model, and outputting a final thyroid nodule benign and malignant result; if the output result of the thyroid nodule detection model is that no thyroid nodule exists, directly outputting the result;
and detecting the final output result of the thyroid nodule automatic detection model, so as to detect the thyroid nodule automatic detection model, and finally obtaining the trained thyroid nodule automatic detection model if the thyroid nodule automatic detection model passes the detection.
5. The method of claim 1, further comprising:
and automatically detecting the thyroid ultrasonic image through the trained thyroid nodule automatic detection model.
6. The utility model provides a thyroid nodule automated inspection model construction system based on convolutional neural network which characterized in that specifically includes:
the noise reduction module is used for reducing noise of the thyroid ultrasound image data to obtain a thyroid ultrasound image training data set;
a first training module to train a thyroid nodule detection model using a Yolov3 network based on the training dataset;
the second training module is used for training a thyroid nodule benign and malignant recognition model by using a Resnet network based on the training data set;
and the fusion module is used for fusing the thyroid nodule detection model and the thyroid nodule benign and malignant identification model to generate a thyroid nodule automatic detection model.
7. The method of claim 1,
the first training module is specifically configured to:
mining a training data set by adopting a Focal loss mode, reducing the weight of a simple negative sample in training, and training a thyroid nodule detection model by using a Yolov3 network based on the training data after the Focal loss mining.
The noise reduction module is specifically configured to:
carrying out graying processing on the thyroid ultrasound image to obtain a binarized image;
and (3) performing image opening operation on the basis of the binary image, namely corroding the image and then expanding the image to finish the noise reduction of the thyroid ultrasound image data.
8. The method of claim 1, wherein the system further comprises:
the test module is used for inputting the thyroid ultrasonic image test data into a thyroid nodule detection model, inputting a thyroid ultrasonic image with confidence coefficient higher than a preset value into the thyroid nodule benign and malignant identification model if the output result of the thyroid nodule detection model indicates that thyroid nodules exist, and outputting a final thyroid nodule benign and malignant result; if the output result of the thyroid nodule detection model is that no thyroid nodule exists, directly outputting the result; detecting a final output result of the automatic thyroid nodule detection model so as to detect the automatic thyroid nodule detection model, and finally obtaining a trained automatic thyroid nodule detection model if the detection is passed;
and the detection module is used for automatically detecting the thyroid ultrasonic image through the trained thyroid nodule automatic detection model.
9. A thyroid nodule automatic detection model construction device based on a convolutional neural network is characterized by comprising the following steps: a memory, a processor and a computer program stored on the memory and executable on the processor, the computer program when executed by the processor implementing the steps of the convolutional neural network-based thyroid nodule automatic detection model construction method according to any one of claims 1 to 5.
10. A computer-readable storage medium, on which an information transfer implementing program is stored, which when executed by a processor implements the steps of the convolutional neural network-based thyroid nodule automatic detection model construction method according to any one of claims 1 to 5.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010417220.XA CN111598875A (en) | 2020-05-18 | 2020-05-18 | Method, system and device for building thyroid nodule automatic detection model |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010417220.XA CN111598875A (en) | 2020-05-18 | 2020-05-18 | Method, system and device for building thyroid nodule automatic detection model |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN111598875A true CN111598875A (en) | 2020-08-28 |
Family
ID=72191508
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010417220.XA Pending CN111598875A (en) | 2020-05-18 | 2020-05-18 | Method, system and device for building thyroid nodule automatic detection model |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111598875A (en) |
Cited By (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112417990A (en) * | 2020-10-30 | 2021-02-26 | 四川天翼网络服务有限公司 | Examination student violation behavior identification method and system |
| CN112614133A (en) * | 2021-03-05 | 2021-04-06 | 北京小白世纪网络科技有限公司 | Three-dimensional pulmonary nodule detection model training method and device without anchor point frame |
| CN112819755A (en) * | 2021-01-14 | 2021-05-18 | 北京小白世纪网络科技有限公司 | Thyroid nodule TI-RADS grading system and method |
| CN112861876A (en) * | 2021-01-25 | 2021-05-28 | 北京小白世纪网络科技有限公司 | Automatic liver cancer ultrasonic image identification method and device based on convolutional neural network |
| CN112927808A (en) * | 2021-03-01 | 2021-06-08 | 北京小白世纪网络科技有限公司 | Thyroid ultrasound image-based nodule grading system and method |
| CN113139931A (en) * | 2021-03-17 | 2021-07-20 | 杭州迪英加科技有限公司 | Thyroid slice image classification model training method and device |
| CN113177554A (en) * | 2021-05-19 | 2021-07-27 | 中山大学 | Thyroid nodule identification and segmentation method, system, storage medium and equipment |
| CN113436154A (en) * | 2021-06-11 | 2021-09-24 | 北京小白世纪网络科技有限公司 | Thyroid nodule edge symptom classification method, device and system |
| CN113837224A (en) * | 2021-08-25 | 2021-12-24 | 哈尔滨工业大学（威海） | SPECT image-based thyroid nodule property classification method |
| CN113842166A (en) * | 2021-10-25 | 2021-12-28 | 上海交通大学医学院 | Ultrasonic image acquisition method based on ultrasonic imaging equipment and related device |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109241967A (en) * | 2018-09-04 | 2019-01-18 | 青岛大学附属医院 | Thyroid ultrasound automatic image recognition system, computer equipment, storage medium based on deep neural network |
| CN110335668A (en) * | 2019-05-22 | 2019-10-15 | 台州市中心医院（台州学院附属医院） | Thyroid cancer cell pathological map auxiliary analysis method and system based on deep learning |
| CN110532946A (en) * | 2019-08-28 | 2019-12-03 | 长安大学 | A method of the green vehicle spindle-type that is open to traffic is identified based on convolutional neural networks |
| CN110852190A (en) * | 2019-10-23 | 2020-02-28 | 华中科技大学 | Driving behavior recognition method and system integrating target detection and gesture recognition |
-
2020
- 2020-05-18 CN CN202010417220.XA patent/CN111598875A/en active Pending
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109241967A (en) * | 2018-09-04 | 2019-01-18 | 青岛大学附属医院 | Thyroid ultrasound automatic image recognition system, computer equipment, storage medium based on deep neural network |
| CN110335668A (en) * | 2019-05-22 | 2019-10-15 | 台州市中心医院（台州学院附属医院） | Thyroid cancer cell pathological map auxiliary analysis method and system based on deep learning |
| CN110532946A (en) * | 2019-08-28 | 2019-12-03 | 长安大学 | A method of the green vehicle spindle-type that is open to traffic is identified based on convolutional neural networks |
| CN110852190A (en) * | 2019-10-23 | 2020-02-28 | 华中科技大学 | Driving behavior recognition method and system integrating target detection and gesture recognition |
Non-Patent Citations (3)
| Title |
|---|
| 王彬: "乳腺钼靶图像中的肿块检测和良恶性诊断", 《中国优秀硕士学位论文全文数据库 _医药卫生科技辑》 * |
| 邹奕轩等: "基于深度学习的甲状腺超声图像辅助诊断研究进展", 《医疗卫生装备》 * |
| 陈凯等: "基于YOLOv3与ResNet50的摄影机器人人脸识别跟踪系统", 《计算机与现代化》 * |
Cited By (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112417990A (en) * | 2020-10-30 | 2021-02-26 | 四川天翼网络服务有限公司 | Examination student violation behavior identification method and system |
| CN112819755A (en) * | 2021-01-14 | 2021-05-18 | 北京小白世纪网络科技有限公司 | Thyroid nodule TI-RADS grading system and method |
| CN112861876A (en) * | 2021-01-25 | 2021-05-28 | 北京小白世纪网络科技有限公司 | Automatic liver cancer ultrasonic image identification method and device based on convolutional neural network |
| CN112927808A (en) * | 2021-03-01 | 2021-06-08 | 北京小白世纪网络科技有限公司 | Thyroid ultrasound image-based nodule grading system and method |
| CN112614133A (en) * | 2021-03-05 | 2021-04-06 | 北京小白世纪网络科技有限公司 | Three-dimensional pulmonary nodule detection model training method and device without anchor point frame |
| CN113139931B (en) * | 2021-03-17 | 2022-06-03 | 杭州迪英加科技有限公司 | Thyroid section image classification model training method and device |
| CN113139931A (en) * | 2021-03-17 | 2021-07-20 | 杭州迪英加科技有限公司 | Thyroid slice image classification model training method and device |
| CN113177554A (en) * | 2021-05-19 | 2021-07-27 | 中山大学 | Thyroid nodule identification and segmentation method, system, storage medium and equipment |
| CN113177554B (en) * | 2021-05-19 | 2023-01-20 | 中山大学 | Thyroid nodule identification and segmentation method, system, storage medium and equipment |
| CN113436154A (en) * | 2021-06-11 | 2021-09-24 | 北京小白世纪网络科技有限公司 | Thyroid nodule edge symptom classification method, device and system |
| CN113837224A (en) * | 2021-08-25 | 2021-12-24 | 哈尔滨工业大学（威海） | SPECT image-based thyroid nodule property classification method |
| CN113837224B (en) * | 2021-08-25 | 2022-06-24 | 哈尔滨工业大学（威海） | SPECT image-based thyroid nodule property classification method |
| CN113842166A (en) * | 2021-10-25 | 2021-12-28 | 上海交通大学医学院 | Ultrasonic image acquisition method based on ultrasonic imaging equipment and related device |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111598875A (en) | Method, system and device for building thyroid nodule automatic detection model | |
| CN108464840B (en) | Automatic detection method and system for breast lumps | |
| CN111488921B (en) | Intelligent analysis system and method for panoramic digital pathological image | |
| US20220092789A1 (en) | Automatic pancreas ct segmentation method based on a saliency-aware densely connected dilated convolutional neural network | |
| CN111553892B (en) | Lung nodule segmentation calculation method, device and system based on deep learning | |
| CN112862830A (en) | Multi-modal image segmentation method, system, terminal and readable storage medium | |
| CN112365973A (en) | Pulmonary nodule auxiliary diagnosis system based on countermeasure network and fast R-CNN | |
| CN113269799A (en) | Cervical cell segmentation method based on deep learning | |
| CN111028940B (en) | Multi-scale lung nodule detection method, device, equipment and medium | |
| CN115187566A (en) | Intracranial aneurysm detection method and device based on MRA image | |
| CN112508884A (en) | Comprehensive detection device and method for cancerous region | |
| CN116309571A (en) | Three-dimensional cerebrovascular segmentation method and device based on semi-supervised learning | |
| Ahmad et al. | Optimized lung nodule prediction model for lung cancer using contour features extraction | |
| CN113177554B (en) | Thyroid nodule identification and segmentation method, system, storage medium and equipment | |
| CN111401102A (en) | Deep learning model training method and device, electronic equipment and storage medium | |
| CN112819765A (en) | Liver image processing method | |
| CN116862836A (en) | System and computer equipment for detecting extensive organ lymph node metastasis cancer | |
| CN116129184A (en) | Multi-phase focus classification method, device, equipment and readable storage medium | |
| CN112862785B (en) | CTA image data identification method, device and storage medium | |
| CN112862786B (en) | CTA image data processing method, device and storage medium | |
| CN115661029A (en) | Pulmonary nodule detection and identification system based on YOLOv5 | |
| CN115409812A (en) | CT image automatic classification method based on fusion time attention mechanism | |
| Shijitha et al. | Efficient Morphological Segmentation of Brain Hemorrhage Stroke Lesion Through MultiResUNet. | |
| CN112086174A (en) | Three-dimensional knowledge diagnosis model construction method and system | |
| CN112614092A (en) | Spine detection method and device |
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 | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20200828 |
|
| RJ01 | Rejection of invention patent application after publication |