CN112001877A - Thyroid malignant nodule detection method based on deep learning - Google Patents
Thyroid malignant nodule detection method based on deep learning Download PDFInfo
- Publication number
- CN112001877A CN112001877A CN202010417483.0A CN202010417483A CN112001877A CN 112001877 A CN112001877 A CN 112001877A CN 202010417483 A CN202010417483 A CN 202010417483A CN 112001877 A CN112001877 A CN 112001877A
- Authority
- CN
- China
- Prior art keywords
- model
- detection
- target
- training
- deep learning
- 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
- G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
- G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
- G16H30/00—ICT specially adapted for the handling or processing of medical images
- G16H30/20—ICT specially adapted for the handling or processing of medical images for handling medical images, e.g. DICOM, HL7 or PACS
-
- 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
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30004—Biomedical image processing
- G06T2207/30096—Tumor; Lesion
Landscapes
- Engineering & Computer Science (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Radiology & Medical Imaging (AREA)
- Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
- Medical Informatics (AREA)
- General Physics & Mathematics (AREA)
- Computational Linguistics (AREA)
- Biomedical Technology (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Biophysics (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Quality & Reliability (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Epidemiology (AREA)
- Primary Health Care (AREA)
- Public Health (AREA)
- Image Processing (AREA)
- Image Analysis (AREA)
- Ultra Sonic Daignosis Equipment (AREA)
Abstract
The invention relates to a thyroid malignant nodule detection method based on deep learning, which is used for auxiliary diagnosis of thyroid malignant nodules and can realize automatic marking of an interested area after training of a large amount of data under the characteristics of low resolution, low precision and low target and background identification of medical images. Errors caused by subjective factors can be effectively reduced, and the radiologist is helped to diagnose quickly and accurately. Reaching considerable height in detection precision and speed and having potential of clinical application.
Description
Technical Field
The invention belongs to the field of image processing, relates to a deep learning technology and a target detection technology, and particularly relates to a thyroid malignant nodule detection method based on deep learning.
Background
The ultrasonic diagnosis technology is widely applied to clinic due to the advantages of convenient examination, low cost and the like. Research shows that ultrasound is advantageous in identifying benign and malignant thyroid nodules because thyroid nodules in an ultrasound image are obviously different in size, shape, number, cystic changes, calcification, blood supply and the like, so that clinicians can judge the properties of the nodules according to the characteristics. However, the ultrasonic diagnosis requires doctors to manually mark the lesion region, which is a heavy work, and doctors with different experience and level often have subjectivity in diagnosis of the result. Therefore, a new technology is needed to extract the nodule region quickly and accurately, and then extract a series of effective features of the extracted nodule region for discriminating the benign and malignant properties of the nodule.
Deep learning is one of the technical and research fields of machine learning, and artificial intelligence is realized in a computing system by establishing an artificial neural network with a hierarchical structure. Because the hierarchical ANN can extract and screen the input information layer by layer, the deep learning has the characteristic learning capability and can realize end-to-end supervised learning and unsupervised learning. Hierarchical ANNs used for deep learning have a variety of forms, the complexity of their hierarchy being commonly referred to as "depth". The deep learning forms comprise multilayer perceptrons, convolutional neural networks, cyclic neural networks, deep confidence networks and other mixed structures according to the structure types. Deep learning uses data to update parameters in its construction to achieve a training goal, a process commonly referred to as "learning". A common method of learning is the gradient descent algorithm and its variants, some statistical learning theory being used for the optimization of the learning process.
The target detection, also called target extraction, is an image segmentation based on target geometry and statistical characteristics, which combines the segmentation and identification of targets into one, and the accuracy and real-time performance of the method are important capabilities of the whole system. Especially, in a complex scene, when a plurality of targets need to be processed in real time, automatic target extraction and identification are particularly important. Currently, the mainstream target detection algorithm mainly has two directions: one-stage and two-stage.
Disclosure of Invention
The invention aims to overcome the defects of the prior art, provides a thyroid malignant nodule detection method based on deep learning, is used for auxiliary diagnosis of thyroid malignant nodules, and can realize automatic marking of an interested area after training of a large amount of data under the characteristics of low resolution, low precision and low target and background identification of medical images. Errors caused by subjective factors can be effectively reduced, and the radiologist is helped to diagnose quickly and accurately. Reaching considerable height in detection precision and speed and having potential of clinical application.
The technical problem to be solved by the invention is realized by the following technical scheme:
a thyroid malignant nodule detection method based on deep learning is characterized in that: the method comprises the following steps:
1) reading ultrasonic image data by a python writing program, and preprocessing;
2) adopting a fast-RCNN algorithm to construct a target detection network, and adjusting and training a model;
3) constructing a target detection network by adopting a RetinaNet algorithm, and adjusting and training a model;
4) and detecting and evaluating the target detection network constructed by the two algorithms, and comparing the effect and the analysis result.
Moreover, the specific operation of the step 1) for preprocessing the ultrasonic image data is as follows:
a. cutting off pictures, namely cutting off additional marks such as equipment name models and patient privacy information contained in the input ultrasonic images;
b. storing the pictures into a JPEGImages file and naming the pictures in a unified format to generate a txt file under a Main folder, wherein the txt file comprises picture numbers of a verification set, a training set and a test set;
c. and (3) labeling the picture by using a labelImg tool, establishing a box frame for the target object, and storing and generating the xml file, wherein the xml file comprises the class name of the target and the coordinates of the predicted frame.
Moreover, the step 2) adjusts and trains the model specifically to:
a. training on the basis of VGG16 weight pre-trained on ImageNet by adopting a BackBone network as VGG16, and making an image classification data set aiming at the thyroid ultrasound nodule by taking the format of a VOC2007 data set as a standard;
b. changing the target categories in the pascal _ voc.py and demo.py into background and jiejie, changing the category number in demo.py into 2, calculating the pixel average value of a data set, and updating a pix _ mean list in the config.py;
c. setting initial parameters of a model, namely, learning, the Batchsize, the maximum iteration number and stepsize, adjusting the size of a data set, the iteration number and the anchor according to a detection result, and formally training the detection model after a prediction frame of the model is closer to a real target area when the intersection ratio of the prediction target area and the real target area is closer to 100 percent and the real target area is closer to the real target area.
Moreover, the step 3) adjusts and trains the model specifically to:
a. the adopted BackBone network is ResNet50, a model of ResNet50 which is pre-trained on ImageNet is downloaded and stored in a corresponding folder;
b. setting initial parameters Scorethreshold and IoUthreshold of the model, adjusting the size of the data set, the iteration times and the anchor according to the detection result, and training the detection model formally after multiple adjustments.
In the step 4), the effect of the algorithm is evaluated and verified by using the mAP as an evaluation index in the detection evaluation, and meanwhile, the detection accuracy and the recall rate of the model and the intersection ratio (IoU) of the predicted target frame and the real target frame need to be calculated, and the calculation formula is as follows:
wherein: TP means that the positive sample is correctly identified as a positive sample;
FP is a negative sample that was misidentified as a positive sample;
FN is that positive samples were misidentified as negative samples;
AP is the area enclosed by the curves formed by Precision in the formula (1) and Recall in the formula (2);
k represents k categories;
mAP is the average of k classes of APs;
m represents mean, and the average of the APs of each class is obtained to obtain the value of mAP.
The invention has the advantages and beneficial effects that:
1. according to the thyroid malignant nodule detection method based on deep learning, disclosed by the invention, the thyroid malignant nodule is subjected to auxiliary detection by training a set data training model and adjusting model parameters through verification set data.
2. According to the thyroid malignant nodule detection method based on deep learning, the experimental effect shows that both the fast-RCNN algorithm and the RetinaNet algorithm have the feasibility of assisting the clinical application of thyroid malignant nodule detection, wherein the RetinaNet algorithm is adopted, and after the number of training sets and the number of iterative training are adjusted, a target detection network based on deep learning for thyroid malignant nodule detection has higher detection accuracy; by adopting the fast-RCNN algorithm, the overfitting phenomenon is not easy to occur, and the robustness of the trained model is good.
3. The invention relates to a thyroid malignant nodule detection method based on deep learning, which is used for auxiliary diagnosis of thyroid malignant nodules and can realize automatic marking of an interested area after training of a large amount of data under the characteristics of low resolution, low precision and low identification degree of a target and a background of a medical image; errors caused by subjective factors can be effectively reduced, and a radiologist is helped to diagnose quickly and accurately; reaching considerable height in detection precision and speed and having potential of clinical application.
Drawings
Fig. 1 is a diagram of the network architecture of the present invention.
Detailed Description
The present invention is further illustrated by the following specific examples, which are intended to be illustrative, not limiting and are not intended to limit the scope of the invention.
A thyroid malignant nodule detection method based on deep learning is characterized in that: the method comprises the following steps:
s0101: inputting an ultrasonic medical image for model training and detection, wherein the program removes some additional marks contained in the image data, including information such as the size of a nodule, the name and the model of equipment, the privacy of a patient and the like;
s0102: storing the pictures into a JPEGImages file and naming the pictures in a unified format;
s0103: dividing a data set to generate a txt file under a Main folder, wherein the txt file comprises a verification set, a training set and a test set picture number, a labelImg tool is used for labeling pictures, a box frame is established for a target object, and the box frame is stored to generate an xml file which comprises a class name of the target and a coordinate of a prediction frame;
s0201: the model using the fast-RCNN algorithm is adjusted. The adopted BackBone network is VGG16, training is carried out on the basis of pre-trained VGG16 weight on ImageNet, an image classification data set aiming at thyroid ultrasound nodules is made by taking the format of a VOC2007 data set as a standard, the target classes in pascal _ voc.py and demo.py are changed into background and jiejie, the class number in demo.py is changed into 2, the pixel average value of the data set is calculated, and the pix _ mean table in config.py is updated;
firstly, the learningsite is set to 0.001, the batch size is set to 100, the maximum iteration number is set to 500, and the stepsize is set to 400, so that the target recognition degree of the model is extremely low, almost no feature of the target is learned, the overall characteristic of the target cannot be grasped, the data set segmentation ratio is adjusted, and the data set is enlarged. If the effect is still not good after adjustment, iteration is increased, the anchor size is changed, and the prediction frame of the model is closer to the real target area along with the parameter adjustment step by step;
s0202: gradually increasing the iterative training times of the adjusted model of the fast-RCNN algorithm, 1000 times, 2000 times until 10000 times, learning the training process by obtaining the training information printed and output by the console, saving the console information as a txt format file, searching a matching character string in the txt file, such as totalloss, extracting the subsequent numerical value, storing the numerical value into an array to obtain a total loss value, drawing the loss value array through drawing software to obtain a loss curve, and testing the model by using test set data;
s0301: the model using the RetinaNet algorithm is adjusted initially. The adopted BackBone network is ResNet50, and a model of ResNet50 which is pre-trained on ImageNet is downloaded firstly and stored into a corresponding folder. Setting the initial parameters Scorethreshold of the model to be 0.05 and IoUthreshold to be 0.5, wherein the two parameters are related to the calculation of the mAP, and the prediction target box with the confidence score threshold lower than 0.05 is discarded without participating in the calculation of the mAP;
s0302: and increasing the number of iterative training times for the adjusted model of the RetinaNet algorithm, wherein 10000 times of iterative training, 20000 times of iterative training, 30000 times of iterative training and 40000 times of iterative training are performed. Learning a training process by acquiring training information printed and output by a console, storing the console information as a txt format file, searching a matching character string such as totalloss in the txt file by a programming program, extracting a value after the searching and storing the value in an array to obtain a total loss value, drawing the loss value array by drawing software to obtain a loss curve, and testing a model by using test set data;
s0401: by calculating and comparing the mAP index, the effect of the target detection network constructed by the two algorithms can be evaluated and verified, a proper model is selected,
meanwhile, the detection accuracy and the recall rate of the model and the intersection ratio (IoU) of the predicted target frame and the real target frame need to be calculated, and the calculation formula is as follows:
wherein: TP means that the positive sample is correctly identified as a positive sample;
FP is a negative sample that was misidentified as a positive sample;
FN is that positive samples were misidentified as negative samples;
AP is the area enclosed by the curves formed by Precision in the formula (1) and Recall in the formula (2);
k represents k categories;
mAP is the average of k classes of APs;
m represents mean, and the average of the APs of each class is obtained to obtain the value of mAP.
The experimental effects of the comparison experiment are shown in tables 1 and 2, the target detection network adopting the Faster-RCNN algorithm has the advantages that the mAP (maximum likelihood ratio) detection effect is stably improved along with the increase of the iteration times, the mAP detection effect reaches 0.765 after 10000 times of iterative training, and the feasibility is realized; the target detection network adopting the Retina Net algorithm has an obvious better detection effect, but an overfitting phenomenon appears after the iteration times exceed 10000 times, so that the model has poor robustness, the adaptability to new data is poor, and the detection performance is influenced.
TABLE 1 evaluation index Table for fast-RCNN experiment
| Number of iterations | 1000 | 2000 | 3000 | 4000 | 5000 | 6000 | 7000 | 8000 | 9000 | 10000 |
| mAP | 0.091 | 0.104 | 0.359 | 0.474 | 0.488 | 0.549 | 0.689 | 0.661 | 0.723 | 0.765 |
TABLE 2 Retina Net experiment evaluation index table
| Number of iterations | 10000 | 20000 | 30000 | 40000 |
| mAP | 0.9750 | 0.9796 | 0.9770 | 0.9740 |
By combining the table 1 and the table 2, the two algorithms have the advantages when applied to the construction of the deep learning-based target detection network for detecting the thyroid malignant nodules, and both have the feasibility of assisting the clinical application of the thyroid malignant nodule detection.
The thyroid malignant nodule detection method based on deep learning detects an ultrasonic image of a thyroid malignant nodule, processes and learns ultrasonic image information, automatically marks an interested area by adopting the deep learning method, identifies the thyroid malignant nodule and assists diagnosis; two algorithms with advantages are adopted to construct a model, feature extraction can be automatically realized, the value of a convolution kernel is adjusted in a self-adaptive manner, the error between the predicted value and the true value of the model is minimized, and a proper model can be selected according to actual requirements; ultrasonic image data of thyroid malignant nodules are largely learned through a deep learning method, characteristics of the thyroid malignant nodules are captured comprehensively and in detail, and diagnosis accuracy and diagnosis speed are improved.
Although the embodiments of the present invention and the accompanying drawings are disclosed for illustrative purposes, those skilled in the art will appreciate that: various substitutions, changes and modifications are possible without departing from the spirit and scope of the invention and the appended claims, and therefore the scope of the invention is not limited to the disclosure of the embodiments and the accompanying drawings.
Claims (5)
1. A thyroid malignant nodule detection method based on deep learning is characterized in that: the method comprises the following steps:
1) reading ultrasonic image data by a python writing program, and preprocessing;
2) adopting a fast-RCNN algorithm to construct a target detection network, and adjusting and training a model;
3) constructing a target detection network by adopting a RetinaNet algorithm, and adjusting and training a model;
4) and detecting and evaluating the target detection network constructed by the two algorithms, and comparing the effect and the analysis result.
2. The deep learning-based thyroid malignant nodule detection method according to claim 1, wherein: the specific operation of the step 1) for preprocessing the ultrasonic image data is as follows:
a. cutting off pictures, namely cutting off additional marks such as equipment name models and patient privacy information contained in the input ultrasonic images;
b. storing the pictures into a JPEGImages file and naming the pictures in a unified format to generate a txt file under a Main folder, wherein the txt file comprises picture numbers of a verification set, a training set and a test set;
c. and (3) labeling the picture by using a labelImg tool, establishing a box frame for the target object, and storing and generating the xml file, wherein the xml file comprises the class name of the target and the coordinates of the predicted frame.
3. The deep learning-based thyroid malignant nodule detection method according to claim 1, wherein: the step 2) adjusts and trains the specific operation of the model as follows:
a. training on the basis of VGG16 weight pre-trained on ImageNet by adopting a BackBone network as VGG16, and making an image classification data set aiming at the thyroid ultrasound nodule by taking the format of a VOC2007 data set as a standard;
b. changing the target categories in the pascal _ voc.py and demo.py into background and jiejie, changing the category number in demo.py into 2, calculating the pixel average value of a data set, and updating a pix _ mean list in the config.py;
c. setting initial parameters of a model, namely, learning, the Batchsize, the maximum iteration number and stepsize, adjusting the size of a data set, the iteration number and the anchor according to a detection result, and formally training the detection model after a prediction frame of the model is closer to a real target area when the intersection ratio of the prediction target area and the real target area is closer to 100 percent and the real target area is closer to the real target area.
4. The deep learning-based thyroid malignant nodule detection method according to claim 1, wherein: the step 3) adjusts and trains the specific operation of the model as follows:
a. the adopted BackBone network is ResNet50, a model of ResNet50 which is pre-trained on ImageNet is downloaded and stored in a corresponding folder;
b. setting initial parameters Scorethreshold and IoUthreshold of the model, adjusting the size of the data set, the iteration times and the anchor according to the detection result, and training the detection model formally after multiple adjustments.
5. The deep learning-based thyroid malignant nodule detection method according to claim 1, wherein: in the step 4), mAP is used as an evaluation index to evaluate and verify the effect of the algorithm in the detection evaluation, and meanwhile, the detection accuracy and the recall rate of the model and the intersection ratio (IoU) of the predicted target frame and the real target frame need to be calculated, and the calculation formula is as follows:
wherein: TP means that the positive sample is correctly identified as a positive sample;
FP is a negative sample that was misidentified as a positive sample;
FN is that positive samples were misidentified as negative samples;
AP is the area enclosed by the curves formed by Precision in the formula (1) and Recall in the formula (2);
k represents k categories;
mAP is the average of k classes of APs;
m represents mean, and the average of the APs of each class is obtained to obtain the value of mAP.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010417483.0A CN112001877A (en) | 2020-05-18 | 2020-05-18 | Thyroid malignant nodule detection method based on deep learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010417483.0A CN112001877A (en) | 2020-05-18 | 2020-05-18 | Thyroid malignant nodule detection method based on deep learning |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN112001877A true CN112001877A (en) | 2020-11-27 |
Family
ID=73461765
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010417483.0A Pending CN112001877A (en) | 2020-05-18 | 2020-05-18 | Thyroid malignant nodule detection method based on deep learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112001877A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| 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 |
| CN114820455A (en) * | 2022-03-31 | 2022-07-29 | 华中科技大学同济医学院附属同济医院 | Ovarian cancer ultrasonic image processing method, system, medium, equipment and terminal |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109447940A (en) * | 2018-08-28 | 2019-03-08 | 天津医科大学肿瘤医院 | Convolutional neural networks training method, ultrasound image recognition positioning method and system |
| CN109726658A (en) * | 2018-12-21 | 2019-05-07 | 上海科技大学 | Crowd counts and localization method, system, electric terminal and storage medium |
| CN110223287A (en) * | 2019-06-13 | 2019-09-10 | 首都医科大学北京友谊医院 | A method of early diagnosing mammary cancer rate can be improved |
| CN110390691A (en) * | 2019-06-12 | 2019-10-29 | 合肥合工安驰智能科技有限公司 | A kind of ore scale measurement method and application system based on deep learning |
| CN110490892A (en) * | 2019-07-03 | 2019-11-22 | 中山大学 | A kind of Thyroid ultrasound image tubercle automatic positioning recognition methods based on USFaster R-CNN |
-
2020
- 2020-05-18 CN CN202010417483.0A patent/CN112001877A/en active Pending
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109447940A (en) * | 2018-08-28 | 2019-03-08 | 天津医科大学肿瘤医院 | Convolutional neural networks training method, ultrasound image recognition positioning method and system |
| CN109726658A (en) * | 2018-12-21 | 2019-05-07 | 上海科技大学 | Crowd counts and localization method, system, electric terminal and storage medium |
| CN110390691A (en) * | 2019-06-12 | 2019-10-29 | 合肥合工安驰智能科技有限公司 | A kind of ore scale measurement method and application system based on deep learning |
| CN110223287A (en) * | 2019-06-13 | 2019-09-10 | 首都医科大学北京友谊医院 | A method of early diagnosing mammary cancer rate can be improved |
| CN110490892A (en) * | 2019-07-03 | 2019-11-22 | 中山大学 | A kind of Thyroid ultrasound image tubercle automatic positioning recognition methods based on USFaster R-CNN |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| 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 |
| CN114820455A (en) * | 2022-03-31 | 2022-07-29 | 华中科技大学同济医学院附属同济医院 | Ovarian cancer ultrasonic image processing method, system, medium, equipment and terminal |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111476292B (en) | Small sample element learning training method for medical image classification processing artificial intelligence | |
| EP3876190B1 (en) | Endoscopic image processing method and system and computer device | |
| CN107610087B (en) | Tongue coating automatic segmentation method based on deep learning | |
| CN107464250B (en) | Automatic breast tumor segmentation method based on three-dimensional MRI (magnetic resonance imaging) image | |
| CN111985536B (en) | Based on weak supervised learning gastroscopic pathology image Classification method | |
| CN107862694A (en) | A kind of hand-foot-and-mouth disease detecting system based on deep learning | |
| CN112101451B (en) | Breast cancer tissue pathological type classification method based on generation of antagonism network screening image block | |
| EP3610456A1 (en) | Recist assessment of tumour progression | |
| WO2017207138A1 (en) | Method of training a deep neural network | |
| CN108765392B (en) | Digestive tract endoscope lesion detection and identification method based on sliding window | |
| CN112001877A (en) | Thyroid malignant nodule detection method based on deep learning | |
| CN111524144A (en) | Intelligent pulmonary nodule diagnosis method based on GAN and Unet network | |
| CN114092450A (en) | Real-time image segmentation method, system and device based on gastroscopy video | |
| CN110634132A (en) | Method for automatically generating tuberculosis quantitative diagnosis report based on deep learning 3D CT image | |
| Chitnis et al. | Domain-specific pre-training improves confidence in whole slide image classification | |
| CN110837844A (en) | Pancreatic cystic tumor benign and malignant classification method based on CT image dissimilarity characteristics | |
| CN110969156A (en) | Convolutional neural network model for detecting milk cow nipple and construction method thereof | |
| CN113205153B (en) | Training method of pediatric pneumonia auxiliary diagnosis model and model obtained by training | |
| CN113177602B (en) | Image classification method, device, electronic equipment and storage medium | |
| CN114565762A (en) | Weakly supervised liver tumor segmentation based on ROI and split fusion strategy | |
| CN109948706B (en) | Micro-calcification cluster detection method combining deep learning and feature multi-scale fusion | |
| CN106228180A (en) | The identification model parameter training method of a kind of lung image focus and recognition methods | |
| CN112259199A (en) | Medical image classification model training method, system, storage medium and medical image processing device | |
| CN111598144A (en) | Training method and device of image recognition model | |
| CN110705570A (en) | Image feature identification method |
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 |