CN115311491A - Fracture false positive screening method based on course learning and spatial attention - Google Patents

Fracture false positive screening method based on course learning and spatial attention Download PDF

Info

Publication number
CN115311491A
CN115311491A CN202210912896.5A CN202210912896A CN115311491A CN 115311491 A CN115311491 A CN 115311491A CN 202210912896 A CN202210912896 A CN 202210912896A CN 115311491 A CN115311491 A CN 115311491A
Authority
CN
China
Prior art keywords
false positive
fracture
sample
network
true
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN202210912896.5A
Other languages
Chinese (zh)
Inventor
章琛曦
何学才
李铭
宋志坚
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Fudan University
Original Assignee
Fudan University
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Fudan University filed Critical Fudan University
Priority to CN202210912896.5A priority Critical patent/CN115311491A/en
Publication of CN115311491A publication Critical patent/CN115311491A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/764—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using classification, e.g. of video objects
    • 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
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/82—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using neural networks

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Evolutionary Computation (AREA)
  • Physics & Mathematics (AREA)
  • Artificial Intelligence (AREA)
  • General Physics & Mathematics (AREA)
  • Computing Systems (AREA)
  • Health & Medical Sciences (AREA)
  • General Health & Medical Sciences (AREA)
  • Software Systems (AREA)
  • Databases & Information Systems (AREA)
  • Medical Informatics (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Multimedia (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • Computational Linguistics (AREA)
  • Data Mining & Analysis (AREA)
  • Molecular Biology (AREA)
  • General Engineering & Computer Science (AREA)
  • Mathematical Physics (AREA)
  • Image Analysis (AREA)

Abstract

The invention belongs to the technical field of medical image processing, and particularly relates to a false positive screening method based on course learning and spatial attention. Aiming at the problem that a detection model is easy to generate a large number of false positive prediction results due to high bone similarity during fracture detection, the invention provides a false positive screening network Rib-FPR Net, improves the characteristic learning capability of the network on a fracture area by using a space attention mechanism, and effectively improves the false positive screening capability of the network by combining a course learning type training mode. The invention has the characteristics of simple implementation and high efficiency, and effectively solves the problem of excessive false positive results by combining course learning and a space attention mechanism, thereby well improving the auxiliary diagnosis effect of the fracture.

Description

Fracture false positive screening method based on course learning and spatial attention
Technical Field
The invention belongs to the technical field of medical image processing, and particularly relates to a fracture false positive screening method based on course learning and spatial attention.
Background
Because the fracture in the CT image has the inherent characteristics of fine structure, high gray level distribution similarity among bones and the like, the existing fracture detection network has insufficient capability of distinguishing true positive fracture candidate regions from false positive fracture candidate regions. On the other hand, from the analysis of the working principle of the fracture detection network, because the detection network inputs image blocks cut by a sliding window, a large number of image blocks only contain partial cutting units of a certain section of skeleton, the image blocks can generate broken lines due to cutting, and the characteristic information of the lines can influence the network to learn the fracture characteristic information. These factors result in more false positive predictions being output, although existing fracture detection networks can achieve higher fracture detection sensitivity. In clinical application, too many false positive prediction results of the detection network may interfere with the diagnosis of doctors, and increase the diagnosis burden of doctors.
Therefore, the invention provides a false positive screening method based on course learning and spatial attention, so that a large number of false positive candidate regions are effectively screened out, and a good clinical auxiliary diagnosis effect is realized.
The prior art related to the present invention is the following references:
[1]Ogul B B,Kosucu P,Ozcam A,et al.Lung Nodule Detection in X-Ray Images:A New Feature Set[C].European Conference of the International Federation for Medical and Biological Engineering,2015.
[2]Hamidian S,Sahiner B,Petrick N,et al.3D Convolutional Neural Network for Automatic Detection of Lung Nodules in Chest CT[C].Medical Imaging 2017:Computer-Aided Diagnosis,2017.
[3]Gong J,Liu J,Wang L,et al.Automatic detection of pulmonary nodules in CT images by incorporating 3D tensor filtering with local image feature analysis[J].Physical Medical European Journal of Medical Physics,2018,46:124–133.
[4] he Xuecai, gold , li Ming, et al rib fracture detection methods based on fully integrated network candidate boxes [ J ]. Anatomical newspaper, 2022,53 (3): 396-401.
Disclosure of Invention
The invention aims to provide a fracture false positive screening method based on course learning and spatial attention so as to screen a large number of fracture false positive candidate region results.
The method for screening false positive of fracture based on course learning and space attention provided by the invention utilizes the space attention mechanism to improve the characteristic learning capability of the 3D ResNet50 to the fracture area; the classification capability of the network is improved by matching with a course learning type training mode; and the false positive screening out is realized efficiently by effectively classifying the true and false positive candidate areas.
In the present invention, the design principle of the spatial attention mechanism (see fig. 2) is as follows: the method comprises the steps of processing an input feature image X by utilizing global average pooling GloAvg () to obtain a feature map M, then processing by utilizing an Adaptive function Adaptive sigmoid () to obtain a spatial attention feature map S, and thus carrying out weighting processing on the feature map X to obtain a spatial attention feature map X', so that a network focuses more on features related to focuses, and a better focus classification effect is realized.
The invention provides a false positive screening method based on course learning and space attention, which comprises the following specific steps.
Step 1, firstly, preprocessing an image; intensity normalization, histogram equalization, image block clipping and the like;
specifically, a rib fracture CT image is divided into a plurality of small pixels, and nonlinear stretching is carried out to enable local gray level histograms to be distributed uniformly; resampling all voxels to a uniform size by a third-order spline interpolation method; adopting a nearest neighbor interpolation method for corresponding segmentation labeling; random transformations including random rotation, clipping, scaling, flipping operations are employed to enhance the data.
Step 2, building a network model; the network model comprises a 3DRESNet50 network and a spatial attention mechanism module SAM (shown in figure 1) and is marked as Rib-FPRNet; the 3 dressnet 50 network consists of convolutional layers "7 × 7conv64", conv, batch normalized and nonlinear layers BN/Relu, pooled layers Pool, global averaged pooled GloAvgPool and fully connected layers FC 2; the network model obtains a feature map of an input image by utilizing operations such as convolution, batch standardization, pooling and the like, obtains an attention feature map of the feature map by combining a spatial attention mechanism, and then obtains a classification confidence degree through global average pooling and full-connection operation.
Wherein, the attention feature map spectrum of the space attention mechanism obtained feature map is represented as:
Figure BDA0003774463370000021
wherein, X represents an input characteristic image, and X' is an attention characteristic diagram obtained by processing X through global average pooling GloAvg () and an Adaptive activation function Adaptive signature ();
the loss function calculation formula is as follows:
Figure BDA0003774463370000022
L CE as a loss function, y i Is e {0,1}, represents the true category of sample i, when the sample is a true positive rib fracture area, y i Is 1, otherwise, y i Is 0;
Figure BDA0003774463370000023
representing the probability that sample i is predicted to be a true positive rib fracture region,
Figure BDA0003774463370000024
n represents the total number of samples.
And 3, dividing the difficulty degrees of all detection results of the training set according to the confidence degrees, and firstly performing 5-fold classification cross training on the training set to obtain a network model Rib-FPRNet, namely, taking different 20% data sets as a verification set every time, and predicting the different 20% data sets for 5 times to obtain the classification confidence degrees of all data samples of the training set. And distinguishing the difficulty degree of the sample by adopting three confidence degree threshold values, wherein the three confidence degree threshold values are 0.8, 0.5 and 0.2 respectively. For the true positive fracture candidate region, when the prediction confidence coefficient is greater than 0.8, the sample is a very simple sample; when the confidence coefficient is between 0.5 and 0.8, the sample is a simple sample; when the confidence coefficient is between 0.2 and 0.5, the sample is a difficult sample; when the confidence is less than 0.2, the sample is very difficult. For false positive fracture candidate regions, the criteria for the ease of division are the opposite of the above criteria.
Wherein, the loss function for training is defined as follows:
Figure BDA0003774463370000031
wherein, y i Is e {0,1}, represents the true category of sample i, when the sample is a true positive rib fracture area, y i Is 1, otherwise, y i Is 0.
Figure BDA0003774463370000032
Representing the probability that sample i is predicted to be a true positive rib fracture region,
Figure BDA0003774463370000033
n represents the total number of samples.
Step 4, training the network Rib-FPRNet by adopting a training method based on a course learning formula; the network firstly takes all very simple sample image blocks X as a training set, and after training fitting, simple samples, difficult samples and very difficult samples are sequentially supplemented for training, so that model parameters are obtained; and (4) carrying out true and false positive classification on the fracture area by using a trained network Rib-FPRNet.
And 5, performing False Positive screening by using the classification result of the True and False Positive candidate regions to obtain a final fracture auxiliary diagnosis result, wherein the final fracture auxiliary diagnosis result comprises statistical indexes such as a recall Rate call, a False Positive screening Rate (FPRR), a True and False Positive candidate region Rate (TFR), the number of False Positive candidate regions (FPR) of each average case and the like. The specific calculation formula is as follows:
Figure BDA0003774463370000034
Figure BDA0003774463370000035
Figure BDA0003774463370000036
Figure BDA0003774463370000037
wherein TP (True Positive) represents the Number of Positive candidates predicted to be Positive (with fracture), FN (False Positive) represents the Number of negative candidates predicted to be negative (without fracture), NFP (Number of False Positive) represents the Number of False Positive candidates before screening out, number of cases, RFP (Remaining False Positive) represents the Number of False Positive candidates Remaining after screening out using the classification result, and RTP (Remaining True Positive) represents the Number of True Positive candidates Remaining after screening out using the classification result.
Corresponding to the above false positive screening method, the invention is a false positive screening system based on course learning and spatial attention, which specifically comprises five modules, respectively: the image preprocessing module is used for building a network module, a confidence difficulty degree dividing module, a Rib-FPR Net network training module and a false positive screening module; the five modules respectively execute five steps of the false positive screening method.
According to the invention, the attention mechanism improves the characteristic learning capacity of the network on the fracture candidate region, and meanwhile, the course learning type training mode improves the distinguishing capacity of the network on the true and false positive candidate regions, so that a good false positive screening effect is realized.
Compared with the prior art, the false positive screening network based on course learning and spatial attention has the following advantages:
(1) The spatial attention mechanism enables the network to pay more attention to the characteristics of the fracture area, and the classification effect is improved;
(2) The course learning type training mode improves the network classification capability, and further improves the false positive screening capability.
Drawings
FIG. 1 is a schematic representation of the Rib-FPRNet network structure of the present invention.
FIG. 2 is a schematic view of a spatial attention mechanism.
FIG. 3 is a schematic diagram of the false positive screening results. Wherein, white arrows represent false positive candidate regions effectively screened out, and white circles represent false positive candidate regions not effectively screened out.
Detailed Description
Example 1 false positive screening for Rib fractures Using the method of the invention
The false positive screening network based on course learning and spatial attention provided by the invention is specifically realized by the following steps:
step 1, training set data, total 420 cases, from the RibFracDataset data set. Firstly, intensity normalization and histogram equalization are carried out on the image. The whole image is divided into a plurality of small blocks of pixels for nonlinear stretching, so that the local gray level histogram is uniformly distributed. In order to enable the network to correctly learn the spatial semantics, a third-order spline interpolation method is adopted to resample all CT voxels to a uniform size. And adopting a nearest neighbor interpolation method for corresponding segmentation labeling. Random transformations of random rotation, cut, scale and flip operations are mainly included in each training iteration to enhance the data.
Step 2, building a network model; the network model comprises a 3DRESNet50 network and a spatial attention mechanism module SAM (shown in figure 1) and is marked as Rib-FPRNet; the spatial attention mechanism module SAM (see fig. 2). The 3 dressnet 50 network consists of convolutional layers "7 × 7conv64", conv, batch-normalized and nonlinear layers BN/Relu, pooled layers Pool, global-averaged pooled GloAvgPool and fully-connected layers FC 2. The Rib-FPRNet network obtains a characteristic map of a fracture input CT image by utilizing operations such as convolution, batch standardization and pooling, obtains a Rib fracture attention characteristic map of the characteristic map by combining a spatial attention mechanism, and then obtains Rib fracture classification confidence by means of global average pooling and full-connection operation.
And 3, dividing all data samples into the difficulty degree. First, a 64 × 64 × 64 image block is clipped around the center point of the prediction result generated by the detection network (the network is Rib-Net proposed by reference [4 ]), and a 3 dressnet 50 (spatial attention) network is subjected to 5-fold cross training on the data set, that is, each time a different 20% data set is used as a verification set, the classification confidence of all data samples of the data set can be obtained by predicting the different 20% data set 5 times. We further use 3 confidence thresholds to distinguish sample difficulty levels, i.e. 0.8, 0.5, 0.2. For the true positive fracture candidate region, when the prediction confidence coefficient is greater than 0.8, the sample is a very simple sample; when the confidence coefficient is between 0.5 and 0.8, the sample is a simple sample; when the confidence coefficient is between 0.2 and 0.5, the sample is a difficult sample; when the confidence is less than 0.2, the sample is very difficult. For false positive fracture candidate regions, the criteria for the ease of division are the opposite of the above criteria.
And 4, training a classification network Rib-FPRNet for distinguishing the true and false positive candidate regions, wherein the training adopts a course learning-based training method. Rib-FPR Net requires 4 stages of training learning to be completed. Firstly, all very simple samples of the Rib-FPR Net network are used as a training set, and when a network model reaches an optimal value, the simple samples are added into the training set, so that the learning capacity of the network on characteristic information of slightly difficult fracture is further improved. And then, in the same way, supplementing the difficult samples and the very difficult samples in sequence until the network learns all the data and fits the data to an optimal value. In the training process, the optimization algorithm adopts a classic Adam algorithm, the initial learning rate is set to be 0.001, and the learning rate is reduced along with the increase of the training times.
And 5, statistically analyzing the classification result of the true and false positive candidate regions according to the set classification threshold value, and finally screening out the false positive candidate regions according to the classification result.
The test set consisted of 60 patients, the classification results on the test set are shown in table 1, the false positive screening results are shown in table 2, and the schematic diagram of the false positive screening results is shown in fig. 3.
TABLE 1
AUC ACC
3D ResNet50 0.757 0.712
Rib-FPR Net 0.776 0.731
Rib-FPR Net + course learning strategy (our) 0.791 0.742
TABLE 2
Figure BDA0003774463370000051
The experimental result of the embodiment shows that the invention realizes good true and false positive classification effect; the spatial attention mechanism effectively improves the characteristic learning ability of the network on the fracture area, and the course learning type training mode improves the classification ability of the network. Based on the classification result, a good false positive screening effect is achieved, and the Rib-FPR Net network can screen out most of false positive candidate areas while keeping a high recall rate. When the recall rate is the same as that of the joint diagnosis of two doctors (0.831), the result is slightly better than that of the joint diagnosis of two doctors. The method is simple to implement and can be applied to other types of fracture false positive screening tasks in the same way.

Claims (5)

1. A false positive screening method based on course learning and space attention is characterized in that the space attention is utilized to improve the characteristic learning ability of a 3D ResNet50 on a fracture area; the classification capability of the network is improved by matching with a course learning type training mode; the high-efficiency false positive screening effect is realized by effectively classifying the true and false positive candidate areas;
the space attention mechanism simulates a mechanism of observing an object by human visual cells, and focuses attention on an interested area, so that interference of interference information on network performance is avoided; the mechanism carries out weighting processing on the characteristic diagram by utilizing the spatial attention map, so that the network focuses more on the characteristics related to the focus, and a better focus classification effect is realized;
the method comprises the following specific steps:
step 1, firstly, preprocessing a fracture CT image; intensity normalization, histogram equalization and image block clipping are included;
step 2, building a network model; the network model comprises a 3DResNet50 network and a space attention mechanism module SAM, and is marked as Rib-FPRNet; the 3 dressnet 50 network consists of convolutional layers "7 × 7Conv64", conv, batch standardized and nonlinear layers BN/Relu, pooled layers Pool, global averaged pooled GloAvgPool and fully connected layers FC 2; the network model obtains a feature map of an input image by convolution, batch standardization and pooling operation, obtains an attention feature map of the feature map by combining a space attention mechanism, and then obtains a classification confidence degree by global average pooling and full-connection operation;
wherein, the attention feature map spectrum of the space attention mechanism obtained feature map is represented as:
Figure FDA0003774463360000011
wherein, X represents an input characteristic image, and X' is an attention characteristic diagram obtained by processing X through global average pooling GloAvg () and an Adaptive activation function Adaptive signature ();
step 3, dividing the difficulty degrees of all detection results of the training set according to the confidence degrees, and firstly performing 5-fold classification cross training on the training set to obtain a network model Rib-FPRNet, namely, taking different 20% data sets as a verification set each time, and predicting the different 20% data sets for 5 times to obtain the classification confidence degrees of all data samples of the training set; distinguishing the difficulty degree of the sample by adopting three confidence coefficient thresholds, wherein the three confidence coefficient thresholds are respectively 0.8, 0.5 and 0.2; for the true positive fracture candidate region, when the prediction confidence coefficient is greater than 0.8, the sample is a very simple sample; when the confidence coefficient is between 0.5 and 0.8, the sample is a simple sample; when the confidence coefficient is between 0.2 and 0.5, the sample is a difficult sample; when the confidence coefficient is less than 0.2, the sample is a very difficult sample; for the false positive fracture candidate area, the dividing standard of the difficulty degree is opposite to the standard;
step 4, training the network Rib-FPRNet by adopting a training method based on a course learning formula; firstly, taking all very simple sample image blocks X as a training set, and after training fitting, sequentially supplementing simple samples, difficult samples and very difficult samples for training to obtain model parameters; carrying out true and false positive classification on the fracture area by using the trained network Rib-FPRNet;
step 5, carrying out false positive screening by using the classification result of the true and false positive candidate regions to obtain a final fracture auxiliary diagnosis result, wherein the final fracture auxiliary diagnosis result comprises a recall rate recall, a false positive screening proportion FPRR, a true and false positive candidate region proportion TFR and an average number FPR of false positive candidate regions of each case; the specific calculation formula is as follows:
Figure FDA0003774463360000021
Figure FDA0003774463360000022
Figure FDA0003774463360000023
Figure FDA0003774463360000024
wherein TP represents the number of true positive candidate regions predicted to be positive, FN represents the number of true positive candidate regions predicted to be negative, NFP represents the number of false positive candidate regions before screening out, numP represents the number of cases, RFP represents the number of false positive candidate regions remaining after screening out using the classification result, and RTP represents the number of true positive candidate regions remaining after screening out using the classification result.
2. The method for screening false positive of fracture based on curriculum learning and spatial attention as claimed in claim 1, wherein the step 1 of preprocessing the image specifically comprises: dividing the fracture CT image into a plurality of small pixels, and performing nonlinear stretching to uniformly distribute local gray level histograms; resampling all voxels to a uniform size by a third-order spline interpolation method; adopting a nearest neighbor interpolation method for corresponding segmentation labeling; random transformations including random rotation, clipping, scaling, flipping operations are employed to enhance the data.
3. The method for screening false positives based on curriculum learning and spatial attention of claim 1, wherein in step 2, the loss function is calculated as follows:
Figure FDA0003774463360000025
y i ∈{0,1}，represents the true category of the sample i, when the sample is a true positive fracture region, y i Is 1, otherwise, y i Is 0;
Figure FDA0003774463360000026
representing the probability that sample i is predicted to be a true positive rib fracture region,
Figure FDA0003774463360000027
n represents the total number of samples.
4. The method as claimed in claim 1, wherein in step 3, the training loss function is calculated as follows:
Figure FDA0003774463360000028
wherein, y i Is e {0,1}, represents the true category of sample i, when the sample is a true positive rib fracture area, y i Is 1, otherwise, y i Is a non-volatile organic compound (I) with a value of 0,
Figure FDA0003774463360000029
representing the probability that sample i is predicted to be a true positive rib fracture region,
Figure FDA00037744633600000210
n represents the total number of samples.
5. The method for screening false positives based on course learning and spatial attention according to claims 1-4, comprising five modules, respectively: the image preprocessing module is used for constructing a network module, a confidence difficulty degree dividing module, a Rib-FPR Net network training module and a false positive screening module; the five modules respectively execute five steps of the false positive screening method.
CN202210912896.5A 2022-07-31 2022-07-31 Fracture false positive screening method based on course learning and spatial attention Pending CN115311491A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202210912896.5A CN115311491A (en) 2022-07-31 2022-07-31 Fracture false positive screening method based on course learning and spatial attention

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202210912896.5A CN115311491A (en) 2022-07-31 2022-07-31 Fracture false positive screening method based on course learning and spatial attention

Publications (1)

Publication Number Publication Date
CN115311491A true CN115311491A (en) 2022-11-08

Family

ID=83859615

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202210912896.5A Pending CN115311491A (en) 2022-07-31 2022-07-31 Fracture false positive screening method based on course learning and spatial attention

Country Status (1)

Country Link
CN (1) CN115311491A (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116128940A (en) * 2023-01-13 2023-05-16 北京医准智能科技有限公司 Method, device and equipment for inhibiting false positive detection of rib fracture

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116128940A (en) * 2023-01-13 2023-05-16 北京医准智能科技有限公司 Method, device and equipment for inhibiting false positive detection of rib fracture
CN116128940B (en) * 2023-01-13 2023-08-15 北京医准智能科技有限公司 Method, device and equipment for inhibiting false positive detection of rib fracture

Similar Documents

Publication Publication Date Title
CN110120040B (en) Slice image processing method, slice image processing device, computer equipment and storage medium
Ahmmed et al. Classification of tumors and it stages in brain MRI using support vector machine and artificial neural network
CN109544518B (en) Method and system applied to bone maturity assessment
CN111882560B (en) Lung parenchyma CT image segmentation method based on weighted full convolution neural network
CN110516759B (en) Soft tissue sarcoma transfer risk prediction system based on machine learning
CN113344864A (en) Ultrasonic thyroid nodule benign and malignant prediction method based on deep learning
CN112529870B (en) Multi-scale CNNs lung nodule false positive eliminating method based on combination of source domain and frequency domain
CN112365973B (en) Pulmonary nodule auxiliary diagnosis system based on countermeasure network and fast R-CNN
CN115601602A (en) Cancer tissue pathology image classification method, system, medium, equipment and terminal
CN113706434B (en) Post-processing method for chest enhancement CT image based on deep learning
CN116884623B (en) Medical rehabilitation prediction system based on laser scanning imaging
CN114842238A (en) Embedded mammary gland ultrasonic image identification method
CN111524140B (en) Medical image semantic segmentation method based on CNN and random forest method
CN114445356A (en) Multi-resolution-based full-field pathological section image tumor rapid positioning method
CN113421228A (en) Thyroid nodule identification model training method and system based on parameter migration
CN117036288A (en) Tumor subtype diagnosis method for full-slice pathological image
CN115311491A (en) Fracture false positive screening method based on course learning and spatial attention
Manikandan et al. Segmentation and Detection of Pneumothorax using Deep Learning
CN114565786A (en) Tomography image classification device and method based on channel attention mechanism
CN110363240B (en) Medical image classification method and system
CN115239695A (en) Pulmonary nodule identification system and method based on time sequence image
Alrahhal et al. Deep learning-based system for detection of lung cancer using fusion of features
Smelyakov et al. Lung X-Ray Images Preprocessing Algorithms for COVID-19 Diagnosing Intelligent Systems.
CN114359308A (en) Aortic dissection method based on edge response and nonlinear loss
Shekerbek et al. APPLICATION OF MATHEMATICAL METHODS AND MACHINE LEARNING ALGORITHMS FOR CLASSIFICATION OF X-RAY IMAGES.

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