CN111444969A - Weakly supervised IVOCT image abnormal region detection method - Google Patents

Weakly supervised IVOCT image abnormal region detection method Download PDF

Info

Publication number
CN111444969A
CN111444969A CN202010238258.0A CN202010238258A CN111444969A CN 111444969 A CN111444969 A CN 111444969A CN 202010238258 A CN202010238258 A CN 202010238258A CN 111444969 A CN111444969 A CN 111444969A
Authority
CN
China
Prior art keywords
image
classification
ivoct
detection
abnormal region
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.)
Granted
Application number
CN202010238258.0A
Other languages
Chinese (zh)
Other versions
CN111444969B (en
Inventor
辛景民
石培文
郑南宁
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.)
Xian Jiaotong University
Original Assignee
Xian Jiaotong 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 Xian Jiaotong University filed Critical Xian Jiaotong University
Priority to CN202010238258.0A priority Critical patent/CN111444969B/en
Publication of CN111444969A publication Critical patent/CN111444969A/en
Application granted granted Critical
Publication of CN111444969B publication Critical patent/CN111444969B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G06F18/2415—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on parametric or probabilistic models, e.g. based on likelihood ratio or false acceptance rate versus a false rejection rate
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
    • 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
    • G06T5/70—
    • 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
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/20—Image preprocessing
    • G06V10/26—Segmentation of patterns in the image field; Cutting or merging of image elements to establish the pattern region, e.g. clustering-based techniques; Detection of occlusion
    • 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
    • G06T2207/10072—Tomographic images
    • G06T2207/10101—Optical tomography; Optical coherence tomography [OCT]
    • 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
    • G06T2207/20024—Filtering details
    • G06T2207/20032—Median filtering
    • 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
    • G06T2207/30101—Blood vessel; Artery; Vein; Vascular
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V2201/00—Indexing scheme relating to image or video recognition or understanding
    • G06V2201/03—Recognition of patterns in medical or anatomical images

Abstract

The invention provides a weakly supervised IVOCT image abnormal region detection method which is reasonable in design, convenient to use, simple to operate and strong in practicability, and only utilizes image level labeling to detect abnormal regions. Firstly, the invention converts a detection task into a task classified according to columns by uniformly dividing according to columns, thereby converting a complex task into a simple task, reducing the number of regions to be processed and greatly accelerating the processing speed. Secondly, an end-to-end abnormal region detection neural network is constructed under the condition of only using the classification label, and optimization is carried out through a classification loss function, so that the dependence on region-level labeling based on a deep learning algorithm is reduced, and the labeling pressure of a professional doctor is relieved. And finally, combining the categories of each column predicted by the detection model and carrying out median filtering to obtain the final abnormal region position, overcoming the defects of complex post-processing, low speed and the like based on a deep learning method, and having convenient use, simple operation and strong practicability.

Description

Weakly supervised IVOCT image abnormal region detection method
Technical Field
The invention relates to the field of medical image processing, in particular to a weakly supervised IVOCT image abnormal region detection method.
Background
Acute coronary syndrome is one of the most serious atherosclerotic diseases, and the death due to acute coronary syndrome accounts for the majority of cardiovascular disease deaths each year. Therefore, the screening of the acute coronary syndrome can prevent the occurrence of the acute coronary syndrome. Clinically, intravascular optical coherence tomography (IVOCT) is often used to image the wall of a blood vessel to determine whether a lesion occurs. However, each patient can generate thousands of IVOCT images each time the IVOCT is used for detection, so a quick and accurate abnormal area automatic detection algorithm is required, a normal area is filtered out, an abnormal area is identified and marked, and a doctor diagnoses and screens the images after prompt processing.
The existing abnormal region detection algorithm based on deep learning needs an expert to label a large data set. For each IVOCT image in the data set, not only the judgment on whether an abnormal region is contained or not needs to be carried out, but also a specific position needs to be marked. This approach has 2 drawbacks: first, the collection of medical image data is very difficult. Secondly, the labeling of the medical image data set is very difficult, the workload is very large, the identification of the abnormal region takes much time for a professional doctor, and the labeling of the abnormal region is inaccurate due to the subjective activity of the doctor. For the detection task, not only the image hierarchy is required to mark whether the abnormal region exists, but also a bounding box is required to mark the position of the abnormal region at a specific position. In summary, the existing abnormal region detection based on the deep learning IVOCT image excessively depends on a bounding box label, thereby bringing huge label pressure, meanwhile, the detection accuracy is not high due to inaccuracy in the label process, and finally, thousands of regions to be selected are generated in the application process each time, so that the processing speed is very slow.
Disclosure of Invention
Aiming at the problems in the prior art, the invention provides a weakly supervised IVOCT image abnormal region detection method which is reasonable in design, convenient to use, simple to operate and strong in practicability, and only utilizes image level labeling to detect abnormal regions.
The invention is realized by the following technical scheme:
a weakly supervised IVOCT image abnormal region detection method comprises the following steps:
s1, selecting a training atlas in a Cartesian coordinate system, firstly performing noise reduction on an IVOCT image in the Cartesian coordinate system, secondly performing polar coordinate conversion to obtain an IVOCT image in a polar coordinate system, and finally performing image leveling operation to obtain a polar coordinate system image subjected to noise reduction and leveling as a final image;
s2, uniformly dividing the final image in the column direction to obtain K divided images with the same column number, wherein K is an integer larger than 1;
s3, transmitting the segmentation image obtained after uniform segmentation into a pre-trained feature extraction neural network to extract features;
s4, transmitting the extracted features into a detection module;
the detection module consists of two parts, one is a classification branch, and the other is a selection branch; the extracted features are respectively transmitted into a classification branch and a selection branch to obtain a classification probability graph and a selection probability graph;
the classification branch consists of 2 full-connection layers and 1 sigmoid activation layer, and the selection branch consists of 1 full-connection layer and 1 sigmoid activation layer;
s5, carrying out weighted fusion and normalization on the classification probability chart and the selection probability chart of each segmented image to obtain a classification probability chart of the whole image; then, optimizing by using a classification loss function to obtain a trained detection model;
s6, uniformly dividing the input detection image into columns and then transmitting the columns to the trained detection model; and combining the outputs of the classification branches in the detection module to obtain the category of each column, and then obtaining the position of the abnormal region by using a mean filtering algorithm.
Preferably, the specific construction method of the pre-trained feature extraction neural network in S3 is as follows:
r1: deleting the down-sampling layer in the selected basic neural network, and only reserving a convolution layer with the step length of 2 and a maximum pooling layer at the top of the basic neural network;
r2: changing all the reserved convolution layers into void convolution; the void rate of the void convolution is designed by using a sawtooth wave principle, and different convolution rates are adopted in the row direction and the column direction.
Preferably, the classification probability map p of the column-wise segmented image in S4g(c) And selecting a probability chart ps(c) Shown as follows:
Figure BDA0002431734580000031
Figure BDA0002431734580000032
wherein x iss(c) And xg(c) The extracted features are output after passing through the full connection layer.
Further, the probability map x (i) and the loss function L of the whole image in S5 are respectively:
Figure BDA0002431734580000033
Figure BDA0002431734580000034
wherein the content of the first and second substances,
Figure BDA0002431734580000035
a prediction probability map, y, representing the entire imageiAnd K is the number of the to-be-selected areas of each image divided according to columns.
Preferably, in S2, the image with the final size of M × N is uniformly divided in the column direction to obtain K images with the same number of columns of M × T, where K is N/T, M and N respectively represent the width and height of the image, and T represents the step size of the column division.
Compared with the prior art, the invention has the following beneficial technical effects:
the invention provides a weakly supervised IVOCT image abnormal region detection method, which comprises the steps of firstly, uniformly dividing according to columns to convert a detection task into a classification task according to the columns, thereby converting a complex task into a simple task, reducing the number of regions to be processed and greatly accelerating the processing speed. Secondly, an end-to-end abnormal region detection neural network is constructed under the condition of only using the classification label, and optimization is carried out through a classification loss function, so that the dependence on region-level labeling based on a deep learning algorithm is reduced, and the labeling pressure of a professional doctor is relieved. And finally, combining the categories of each column predicted by the detection model and carrying out median filtering to obtain the final abnormal region position, overcoming the defects of complex post-processing, low speed and the like based on a deep learning method, and having convenient use, simple operation and strong practicability.
Drawings
FIG. 1a is a schematic diagram of an abnormal region on an IVOCT image in Cartesian coordinates according to an embodiment of the invention.
FIG. 1b is a schematic diagram of an abnormal region on an IVOCT image in Cartesian coordinates as described in the examples of the invention.
Fig. 2 is a schematic flowchart of a weakly supervised IVOCT image abnormal region detection method according to an embodiment of the present invention.
Fig. 3 is a schematic structural diagram of a pre-trained neural network according to an embodiment of the present invention.
FIG. 4a is a schematic diagram illustrating the detection effect of the supervised learning method and the method of the present invention on a picture.
FIG. 4b is a schematic diagram showing the comparison of the detection effect of the supervised learning method and the method of the present invention on another picture.
Detailed Description
The present invention will now be described in further detail with reference to specific examples, which are intended to be illustrative, but not limiting, of the invention.
The invention discloses a weakly supervised IVOCT image abnormal region detection method, which comprises the following steps:
s1: selecting a training atlas under a Cartesian coordinate system, firstly carrying out noise reduction treatment on an IVOCT image under the Cartesian coordinate system, secondly carrying out polar coordinate conversion to obtain the IVOCT image under a polar coordinate system, and finally carrying out image leveling operation to obtain a final image.
And S2, uniformly dividing the polar coordinate system image subjected to noise reduction and flattening into K divided images with the same column number in the column direction, for example, dividing an image with the size of M × N into K images of M × T in columns, wherein K is N/T.
S3: and transmitting the segmentation image obtained after uniform segmentation into a pre-trained feature extraction neural network to extract features.
And S4, transmitting the extracted features to a detection module. The detection module consists of two parts, one is a classification branch, and the other is a selection branch. And the extracted features are respectively transmitted into the classification branch and the selection branch to obtain a classification probability graph and a selection probability graph. The classification branch consists of 2 full connection layers and 1 sigmoid activation layer, and the selection branch consists of 1 full connection layer and 1 sigmoid activation layer. The selection probability map represents the contribution of the image to the classification result.
And S5, performing weighted fusion and normalization on the classification probability graph and the selection probability graph of each image segmented according to columns to obtain the classification probability graph of the whole image. And then, optimizing by using a classification loss function to obtain a finally trained detection model.
S6: in the detection stage, the input detection image is divided uniformly according to columns and then transmitted to a trained detection model. The output of the classification branches in the detection module is combined to obtain the category of each column, and then the position of the abnormal region is obtained by using a mean filtering algorithm.
Wherein, the specific construction method of the pre-trained feature extraction network in S3 is as follows:
r1: and deleting the downsampling layer in the selected basic neural network, and only reserving a convolution layer with the step size of 2 and a maximum pooling layer at the top of the neural network.
R2: all the convolution layers remained later are changed into void convolution. The void rate of the void convolution is designed by using a sawtooth wave principle, and different convolution rates are adopted in the row direction and the column direction.
The classification probability map and the selection probability map of the column-wise segmented image in S4 are represented as:
Figure BDA0002431734580000051
Figure BDA0002431734580000052
wherein x iss(c) And xg(c) The extracted features are output after passing through the full connection layer.
The probability map and the loss function of the whole image in S5 are respectively:
Figure BDA0002431734580000053
Figure BDA0002431734580000061
wherein
Figure BDA0002431734580000062
A prediction probability map, y, representing the entire imageiThe truth label representing the image classification.
Specifically, referring to fig. 2, the invention provides a weakly supervised IVOCT image abnormal region detection method, which includes the following steps:
s1: firstly, the IVOCT image under the Cartesian coordinate system is subjected to noise reduction processing, then the image is converted into the polar coordinate system by utilizing a polar coordinate conversion algorithm, as shown in fig. 1a and 1b, and finally the image is flattened by utilizing a flattening algorithm. The IVOCT image noise reduction under the Cartesian coordinate system is realized through a segmented neural network based on deep learning.
S2, uniformly dividing the noise-reduced and leveled polar coordinate IVOCT image according to columns, wherein the specific method comprises the following steps of uniformly dividing a 352-352 × 720 image according to the column direction, setting the interval T as 10, and dividing the image into 72 352-352 × 10 to-be-processed images, and Table 1 shows the influence of different interval sizes on the performance of the neural network:
TABLE 1 Effect of selection of interval T on neural network Performance
Figure BDA0002431734580000063
S3: and inputting the image to be processed into a pre-trained feature extraction neural network to obtain the extracted features. Fig. 3 is a schematic diagram of a structure of a feature extraction neural network.
S4: sending the extracted features into a detection neural network, and obtaining a classification probability graph and a selection probability graph through a classification branch and a selection branch of the detection neural network, wherein the calculation modes are respectively as follows:
Figure BDA0002431734580000071
Figure BDA0002431734580000072
wherein xg(c)＝Wgψ(c)，xs(c)＝WsPsi (c), phi (c) representing the proposed feature, Wg，WsAnd respectively representing the parameters of the full connection layer in the classification branch and the selection branch.
S5: and weighting and fusing the probability graph of the classification branch and the probability graph of the selection branch, then normalizing to obtain the probability graph of the whole image, and obtaining a final model by using classification label supervision and optimization, wherein the loss function is a classification loss function. The specific implementation mode is as follows:
Figure BDA0002431734580000073
Figure BDA0002431734580000074
wherein
Figure BDA0002431734580000075
A prediction probability map, y, representing the entire imageiAnd K is the number of the to-be-selected areas of each image divided according to columns.
S6: in the prediction stage, the polar coordinate IVOCT image is segmented according to columns and then transmitted into a trained model, a classification probability map is obtained only by using classification branches of a detection neural network, then the classification of each column is obtained by combination, and finally the final detection result can be obtained by using a median filtering algorithm. The specific detection results are shown in table 2, wherein S represents the total score, D represents the contact ratio, P represents the accuracy, and R represents the recall ratio, and the specific calculation method is as follows:
Figure BDA0002431734580000076
wherein, nTP, nFP and nFN respectively represent the number of true positive examples, false positive examples and false negative examples, and Dice represents the overlap ratio coefficient.
Referring to fig. 4a and 4b, two visual graphs for comparing the weak supervised algorithm and the supervised learning algorithm proposed by the present invention are shown, where a box a represents an abnormal region true value, a box b represents a supervised learning algorithm result, and a box c represents the method of the present invention, although the method of the present invention does not use a bounding box label (like a true value box shown in a box a) as in the conventional detection algorithm based on deep learning and only uses an image classification label for training, the abnormal region detected by the method of the present invention is not much different from the supervised learning method and includes an abnormal region true value. Meanwhile, as can be illustrated in table 2, the present invention is not much different from the supervised learning method in the total score in the case of using only the image-level classification labels.
TABLE 2 comparison of the Performance of the weakly supervised algorithm and the supervised learning algorithm proposed by the method of the present invention
Method of producing a composite material S D P R
Two-stage 0.8767 0.8824 0.8077 0.9450
SRCNN 0.888 0.887 0.879 0.900
The invention 0.8459 0.8304 0.8400 0.8842
According to the method for detecting the abnormal area of the weakly supervised IVOCT image, the IVOCT image in the Cartesian coordinate system shown in figure 1a is converted into the polar coordinate system shown in figure 1b, the abnormal area is detected, the normal area is filtered, so that the doctor can save the image discrimination time, the attention can be focused on the discrimination of the abnormal area, only the processing and labeling of the image are provided, and the obtained result cannot be directly subjected to related diagnosis, so that the method is used for the purpose of non-disease diagnosis.

Claims (5)

1. A weakly supervised IVOCT image abnormal region detection method is characterized by comprising the following steps:
s1, selecting a training atlas in a Cartesian coordinate system, firstly performing noise reduction on an IVOCT image in the Cartesian coordinate system, secondly performing polar coordinate conversion to obtain an IVOCT image in a polar coordinate system, and finally performing image leveling operation to obtain a polar coordinate system image subjected to noise reduction and leveling as a final image;
s2, uniformly dividing the final image in the column direction to obtain K divided images with the same column number, wherein K is an integer larger than 1;
s3, transmitting the segmentation image obtained after uniform segmentation into a pre-trained feature extraction neural network to extract features;
s4, transmitting the extracted features into a detection module;
the detection module consists of two parts, one is a classification branch, and the other is a selection branch; the extracted features are respectively transmitted into a classification branch and a selection branch to obtain a classification probability graph and a selection probability graph;
the classification branch consists of 2 full-connection layers and 1 sigmoid activation layer, and the selection branch consists of 1 full-connection layer and 1 sigmoid activation layer;
s5, carrying out weighted fusion and normalization on the classification probability chart and the selection probability chart of each segmented image to obtain a classification probability chart of the whole image; then, optimizing by using a classification loss function to obtain a trained detection model;
s6, uniformly dividing the input detection image into columns and then transmitting the columns to the trained detection model; and combining the outputs of the classification branches in the detection module to obtain the category of each column, and then obtaining the position of the abnormal region by using a mean filtering algorithm.
2. The method for detecting abnormal areas of weakly supervised IVOCT images as recited in claim 1, wherein the pre-trained feature extraction neural network in S3 is constructed by the following specific method:
r1: deleting the down-sampling layer in the selected basic neural network, and only reserving a convolution layer with the step length of 2 and a maximum pooling layer at the top of the basic neural network;
r2: changing all the reserved convolution layers into void convolution; the void rate of the void convolution is designed by using a sawtooth wave principle, and different convolution rates are adopted in the row direction and the column direction.
3. The method for detecting abnormal regions of weakly supervised IVOCT image of claim 1, wherein the classification probability map p of the column-wise segmented image in S4g(c) And selecting a probability chart ps(c) Shown as follows:
Figure FDA0002431734570000021
Figure FDA0002431734570000022
wherein x iss(c) And xg(c) The extracted features are output after passing through the full connection layer.
4. The method for detecting abnormal areas of weakly supervised IVOCT image of claim 3, wherein the probability map x (I) and the loss function L of the whole image in S5 are respectively:
Figure FDA0002431734570000023
Figure FDA0002431734570000024
wherein the content of the first and second substances,
Figure FDA0002431734570000025
a prediction probability map, y, representing the entire imageiTo representAnd K is the number of the to-be-selected areas of each image divided according to columns.
5. The method for detecting abnormal regions of a weakly supervised IVOCT image as claimed in claim 1, wherein in S2, the image with the final size of M × N is divided uniformly in the column direction to obtain K images with the same number of columns of M × T, where K is N/T, M and N respectively represent the width and height of the image, and T represents the step size of column division.
CN202010238258.0A 2020-03-30 2020-03-30 Weakly supervised IVOCT image abnormal region detection method Active CN111444969B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202010238258.0A CN111444969B (en) 2020-03-30 2020-03-30 Weakly supervised IVOCT image abnormal region detection method

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202010238258.0A CN111444969B (en) 2020-03-30 2020-03-30 Weakly supervised IVOCT image abnormal region detection method

Publications (2)

Publication Number Publication Date
CN111444969A true CN111444969A (en) 2020-07-24
CN111444969B CN111444969B (en) 2022-02-01

Family

ID=71652624

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202010238258.0A Active CN111444969B (en) 2020-03-30 2020-03-30 Weakly supervised IVOCT image abnormal region detection method

Country Status (1)

Country Link
CN (1) CN111444969B (en)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112669274A (en) * 2020-12-23 2021-04-16 山东大学 Multi-task detection method for pixel-level segmentation of surface abnormal region
CN113034516A (en) * 2021-03-24 2021-06-25 联通(上海)产业互联网有限公司 Deep learning algorithm for segmenting abnormal regions in image

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20160217344A1 (en) * 2015-01-22 2016-07-28 Microsoft Technology Licensing, Llc. Optimizing multi-class image classification using patch features
CN108492272A (en) * 2018-03-26 2018-09-04 西安交通大学 Cardiovascular vulnerable plaque recognition methods based on attention model and multitask neural network and system
CN108764292A (en) * 2018-04-27 2018-11-06 北京大学 Deep learning image object mapping based on Weakly supervised information and localization method
US20190102646A1 (en) * 2017-10-02 2019-04-04 Xnor.ai Inc. Image based object detection
CN110188654A (en) * 2019-05-27 2019-08-30 东南大学 A kind of video behavior recognition methods not cutting network based on movement
CN110738229A (en) * 2018-07-20 2020-01-31 杭州海康威视数字技术股份有限公司 fine-grained image classification method and device and electronic equipment

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20160217344A1 (en) * 2015-01-22 2016-07-28 Microsoft Technology Licensing, Llc. Optimizing multi-class image classification using patch features
US20190102646A1 (en) * 2017-10-02 2019-04-04 Xnor.ai Inc. Image based object detection
CN108492272A (en) * 2018-03-26 2018-09-04 西安交通大学 Cardiovascular vulnerable plaque recognition methods based on attention model and multitask neural network and system
CN108764292A (en) * 2018-04-27 2018-11-06 北京大学 Deep learning image object mapping based on Weakly supervised information and localization method
CN110738229A (en) * 2018-07-20 2020-01-31 杭州海康威视数字技术股份有限公司 fine-grained image classification method and device and electronic equipment
CN110188654A (en) * 2019-05-27 2019-08-30 东南大学 A kind of video behavior recognition methods not cutting network based on movement

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
LIMIN WANG 等，: "UntrimmedNets for Weakly Supervised Action Recognition and Detection", 《2017 IEEE CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION (CVPR)》 *
缪月红，: "基于IVOCT的冠状动脉粥样硬化斑块识别算法研究", 《中国优秀博硕士学位论文全文数据库(硕士)医药卫生科技辑》 *

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112669274A (en) * 2020-12-23 2021-04-16 山东大学 Multi-task detection method for pixel-level segmentation of surface abnormal region
CN113034516A (en) * 2021-03-24 2021-06-25 联通(上海)产业互联网有限公司 Deep learning algorithm for segmenting abnormal regions in image

Also Published As

Publication number Publication date
CN111444969B (en) 2022-02-01

Similar Documents

Publication Publication Date Title
Wu et al. Predict pneumonia with chest X-ray images based on convolutional deep neural learning networks
CN111667456B (en) Method and device for detecting vascular stenosis in coronary artery X-ray sequence radiography
CN110503630B (en) Cerebral hemorrhage classifying, positioning and predicting method based on three-dimensional deep learning model
US6549646B1 (en) Divide-and-conquer method and system for the detection of lung nodule in radiological images
CN113902761B (en) Knowledge distillation-based unsupervised segmentation method for lung disease focus
CN111444969B (en) Weakly supervised IVOCT image abnormal region detection method
CN112365973B (en) Pulmonary nodule auxiliary diagnosis system based on countermeasure network and fast R-CNN
CN113223005B (en) Thyroid nodule automatic segmentation and grading intelligent system
Waheed et al. Hybrid features and mediods classification based robust segmentation of blood vessels
CN111161287A (en) Retinal vessel segmentation method based on symmetric bidirectional cascade network deep learning
CN114529516A (en) Pulmonary nodule detection and classification method based on multi-attention and multi-task feature fusion
CN112464701A (en) Method for detecting whether people wear masks or not based on light weight characteristic fusion SSD
CN111798437A (en) Novel coronavirus pneumonia AI rapid diagnosis method based on CT image
CN115984555A (en) Coronary artery stenosis identification method based on depth self-encoder composition
CN112862756A (en) Method for identifying pathological change type and gene mutation in thyroid tumor pathological image
Khan et al. Benchmark methodological approach for the application of artificial intelligence to lung ultrasound data from covid-19 patients: From frame to prognostic-level
CN114266794B (en) Pathological section image cancer region segmentation system based on full convolution neural network
Dhindsa et al. Grading prenatal hydronephrosis from ultrasound imaging using deep convolutional neural networks
CN103169506A (en) Ultrasonic diagnosis device and method capable of recognizing liver cancer automatically
Manikandan et al. Segmentation and Detection of Pneumothorax using Deep Learning
CN113902702A (en) Pulmonary nodule benign and malignant auxiliary diagnosis system based on computed tomography
CN117351307A (en) Model training method, device, equipment and storage medium
CN117152433A (en) Medical image segmentation method based on multi-scale cross-layer attention fusion network
CN111127400A (en) Method and device for detecting breast lesions
CN116310569A (en) Mammary gland lesion detection and classification device based on robust texture features

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
GR01 Patent grant
GR01 Patent grant