CN114821146A - Enhanced weak supervision-based fine-grained Alzheimer's disease classification method - Google Patents
Enhanced weak supervision-based fine-grained Alzheimer's disease classification method Download PDFInfo
- Publication number
- CN114821146A CN114821146A CN202110112318.9A CN202110112318A CN114821146A CN 114821146 A CN114821146 A CN 114821146A CN 202110112318 A CN202110112318 A CN 202110112318A CN 114821146 A CN114821146 A CN 114821146A
- Authority
- CN
- China
- Prior art keywords
- attention
- classification
- pooling
- feature
- map
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 23
- 208000024827 Alzheimer disease Diseases 0.000 title claims abstract description 15
- 238000011176 pooling Methods 0.000 claims abstract description 16
- 238000000605 extraction Methods 0.000 claims abstract description 8
- 239000011159 matrix material Substances 0.000 claims abstract description 7
- 238000010586 diagram Methods 0.000 claims description 22
- 238000012549 training Methods 0.000 claims description 7
- 238000013135 deep learning Methods 0.000 claims description 5
- 238000012360 testing method Methods 0.000 claims description 5
- 238000012935 Averaging Methods 0.000 claims description 2
- 238000005070 sampling Methods 0.000 claims description 2
- 230000006870 function Effects 0.000 claims 1
- 230000004927 fusion Effects 0.000 claims 1
- 238000007781 pre-processing Methods 0.000 claims 1
- 238000012217 deletion Methods 0.000 abstract 1
- 230000037430 deletion Effects 0.000 abstract 1
- 238000004458 analytical method Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000013527 convolutional neural network Methods 0.000 description 1
- 230000003111 delayed effect Effects 0.000 description 1
- 201000010099 disease Diseases 0.000 description 1
- 208000037265 diseases, disorders, signs and symptoms Diseases 0.000 description 1
- 239000003814 drug Substances 0.000 description 1
- 229940079593 drug Drugs 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 239000000463 material Substances 0.000 description 1
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
- 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
-
- 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
- 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
Abstract
The invention discloses a classification method for fine-grained Alzheimer's disease based on enhanced weak supervision. The method comprises the following steps: firstly, performing feature extraction on a trained picture through an improved and enhanced VGG19 network to obtain a feature map; then, an attention map is generated through an attention learning network, and then the picture is enhanced in a mode of randomly selecting one piece for attention clipping or attention deletion through the attention mechanism. And finally, after obtaining the feature map and the attention map, multiplying the feature map and the attention map of each channel according to element points by adopting a bilinear attention pooling mode, and performing pooling dimension reduction and splicing on the multiplied result to obtain a final feature matrix which is used as the input of the linear classification layer. The invention can be applied to the field of general Alzheimer's disease classification.
Description
Technical Field
The invention designs an enhanced weak supervision-based fine-grained Alzheimer's disease classification method, and relates to the field of medical images and deep learning.
Background
In the elderly, Alzheimer's Disease (AD) has a very high prevalence rate, and no cure method is available until late-stage recurrence, and the disease can be delayed only by means of drugs. According to predictions, by 2050, one out of every 85 people was affected by AD. Generally, the method for judging whether the AD patient is a patient by means of comprehensive analysis of experienced clinicians consumes huge manpower, material resources and financial resources, and has the possibility of misdiagnosis; or the AD image is analyzed by applying a machine learning method, which has higher requirement on the accuracy of feature extraction and is beneficial to the classification of data with known features. However, as the features of the AD image have no clear standard, in contrast, deep learning with the advantage of automatic learning and feature extraction is applied more and more widely in the field, such as convolutional neural networks and sparse self-encoders. Only recent deep learning methods cannot easily detect subtle differences in AD images, so an efficient deep learning classification method is needed to identify alzheimer's disease and healthy people.
Disclosure of Invention
The invention provides a method for classifying the Alzheimer's disease based on the enhanced weak supervision fine granularity for solving the problems. The invention obtains higher-level and more refined feature extraction by improving the enhanced VGG19 network, and improves the classification performance of medical pictures.
The invention realizes the purpose through the following technical scheme:
(1) firstly, in the training process, the feature extraction is carried out on the preprocessed pictures through the improved and enhanced VGG19 network to obtain a feature map;
(2) then, generating 32 attention maps from each feature map through an attention learning network, wherein the 32 attention maps correspond to 32 different parts in an original image target;
(3) randomly selecting an attention map to guide an image to be enhanced by using an attention mechanism, wherein the guidance mode comprises attention cutting or attention discarding;
(4) after obtaining the characteristic diagram and the attention diagram, multiplying the characteristic diagram and the attention diagram of each channel according to element points by adopting a bilinear attention pooling mode, and performing pooling dimension reduction and splicing operation on the multiplied result to obtain a final characteristic matrix which is used as the input of a linear classification layer;
(5) inputting the test pictures into a trained model to obtain the rough classification probability and 32 attention diagrams belonging to each category;
(6) averaging the 32 attention diagrams obtained in the fifth step, drawing an interception frame according to the average value, sampling the interception frame, and then putting the interception frame into a trained model to obtain the fine classification probability of each category based on attention;
(7) the final classification result is the average of the coarse classification probability value and the fine classification probability value of the two steps.
Drawings
Fig. 1 is a general schematic diagram of a network model.
Fig. 2 is a schematic block diagram of the improved enhanced VGG19 network structure of the present invention.
Detailed Description
The invention will be further described with reference to the accompanying drawings in which:
fig. 1 is a schematic overall network model, which includes the following steps:
the whole network model is divided into two parts, namely a training process and a testing process.
In the training process, the original image is subjected to a feature extraction network to obtain feature maps, and 32 attention maps are extracted from each feature map in the original text; then, an attention mechanism is used for guiding data enhancement to help the model to obtain more accurate detailed features, and the guiding mode comprises attention cutting and attention discarding; then, the original picture and the picture with enhanced attention are used as input data to be put into a training network; finally, the feature map and the attention map are subjected to a bilinear attention pooling algorithm to obtain a feature matrix, which is used as an input of a linear classification layer, corresponding to (a) in fig. 1.
In the testing process, the total classification probability consists of a coarse classification probability and a fine classification probability; the rough classification probability is that the original picture is subjected to a trained model to obtain a characteristic diagram and an attention diagram, then the characteristic diagram and the attention diagram are subjected to point multiplication by adopting a BAP algorithm to obtain a characteristic matrix, and then pooling and classification are directly carried out, wherein the pooling and classification correspond to a route 1 in the graph 1 (b); after obtaining the attention diagram, the fine classification probability is to perform addition operation on the attention diagram to obtain the attention sum, then use the attention sum to guide data enhancement to realize the attention clipping step, then send the clipped result and the original picture together as the input object into the test network, finally, the feature map and the attention diagram are subjected to point multiplication, pooling and classification to obtain the fine classification probability, which corresponds to the route 2 in the graph (b) of fig. 1.
Fig. 2 is a schematic block diagram of the improved and enhanced VGG19 network structure of the present invention. This is the enhanced feature extraction part, where Block1 contains 2 groups of 64 channels, a 3 × 3 convolution kernel; block2 contains 3 sets of 128 channels, a 3 × 3 convolution kernel; block3 contains 4 sets of 256-channel, 3 × 3 convolution kernels; block4 includes 4 sets of 512 channels, a 3 × 3 convolution kernel; block5 also contains 4 sets of 512-channel, 3 x 3 convolution kernels. It is desirable to extract relatively fine convolution features from the low-level convolutional layer, select Block2 module for enhancement, and add a set of 128-channel, 3 × 3 convolution kernels before pooling in Block2 module to obtain higher-level, more precise features.
Table 1 shows the results of the improved enhanced VGG19 training network compared to the base network, which shows that the improved network is significantly better than the base network.
TABLE 1 improved enhanced VGG19 training network versus base network
Claims (4)
1. A classification method for weak supervision fine-grained Alzheimer's disease based on enhancement is characterized by comprising the following steps:
the method comprises the following steps: firstly, in the training process, the feature extraction is carried out on the preprocessed pictures through an improved and enhanced VGG19 network to obtain a feature map;
step two: then, generating 32 attention maps from each feature map through an attention learning network, wherein the 32 attention maps correspond to 32 different parts in an original image target;
step three: randomly selecting an attention map to guide an image to be enhanced by using an attention mechanism, wherein the guidance mode comprises attention cutting or attention discarding;
step four: after obtaining the characteristic diagram and the attention diagram, multiplying the characteristic diagram and the attention diagram of each channel according to element points by adopting a bilinear attention pooling mode, and performing pooling dimension reduction and splicing operation on the multiplied result to obtain a final characteristic matrix which is used as the input of a linear classification layer;
step five: inputting the test pictures into a trained model to obtain the rough classification probability and 32 attention diagrams belonging to each category;
step six: averaging the 32 attention diagrams obtained in the fifth step, drawing an interception frame according to the average value, sampling the interception frame, and then putting the interception frame into a trained model to obtain the fine classification probability of each category based on attention;
step seven: the final classification result is the average of the coarse classification probability value and the fine classification probability value of the two steps.
2. The method of claim 1, the method of pre-processing in step one is as follows:
the original image is passed through an M × N weighted mean filter (M and N are both odd numbers) according to the following formula:
in the above formula, x ═ 0, M-1, and y ═ 0, N-1, ensure that all pixels are filtered, and a filter size of 3 × 3 is used, so that each new pixel generated contains the weighted contributions of the 8 original surrounding pixels.
3. The method of claim 1 wherein step one modifies the enhanced VGG19 network as follows:
the network comprises 5 convolution pooling modules, wherein Block1 comprises 2 groups of 64 channels, a convolution kernel of 3 × 3; block2 contains 3 sets of 128 channels, a 33 convolution kernel; block3 contains 4 sets of 256-channel, 3 × 3 convolution kernels; block4 includes 4 sets of 512 channels, a 3 × 3 convolution kernel; block5 also contains 4 sets of 512-channel, 3 x 3 convolution kernels. The invention expects to extract relatively fine convolution characteristics from the low-level convolution layer, selects a Block2 module for enhancement, and adds a group of 128-channel, 3 x 3 convolution kernels before pooling in a Block2 module to obtain higher-level and more precise characteristics.
4. The method of claim 1, wherein the bilinear Attention pooling algorithm BAP (bilinear Attention Pooling) algorithm in step four is as follows:
drawing the feature F and attention A k Dot product generation partial feature map F k ,
F k =A k ⊙F (k=1,2,...,M)
The algorithm can strengthen deep learning of local important characteristic regions, reduce interference of irrelevant information and improve the classification performance of a network model;
f k =g(F k )
in order to solve the problem of overhigh dimension after feature fusion, the pooling dimension reduction operation is carried out according to the formula, and partial features f are extracted k Finally, part of the feature map f k Carrying out summation operation to obtain a feature matrix P consisting of M f k Composition A k Representing the kth attention map, F is the extracted feature map; after the whole BAP process, the feature matrix P is accessed to the central loss function and then is sent to the classification network, and the prediction probability of each class is obtained.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202110112318.9A CN114821146A (en) | 2021-01-27 | 2021-01-27 | Enhanced weak supervision-based fine-grained Alzheimer's disease classification method |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202110112318.9A CN114821146A (en) | 2021-01-27 | 2021-01-27 | Enhanced weak supervision-based fine-grained Alzheimer's disease classification method |
Publications (1)
Publication Number | Publication Date |
---|---|
CN114821146A true CN114821146A (en) | 2022-07-29 |
Family
ID=82525037
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202110112318.9A Pending CN114821146A (en) | 2021-01-27 | 2021-01-27 | Enhanced weak supervision-based fine-grained Alzheimer's disease classification method |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN114821146A (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN115830402A (en) * | 2023-02-21 | 2023-03-21 | 华东交通大学 | Fine-grained image recognition classification model training method, device and equipment |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111191737A (en) * | 2020-01-05 | 2020-05-22 | 天津大学 | Fine-grained image classification method based on multi-scale repeated attention mechanism |
CN111583109A (en) * | 2020-04-23 | 2020-08-25 | 华南理工大学 | Image super-resolution method based on generation countermeasure network |
CN111738363A (en) * | 2020-07-24 | 2020-10-02 | 温州大学 | Alzheimer disease classification method based on improved 3D CNN network |
CN112257601A (en) * | 2020-10-22 | 2021-01-22 | 福州大学 | Fine-grained vehicle identification method based on data enhancement network of weak supervised learning |
-
2021
- 2021-01-27 CN CN202110112318.9A patent/CN114821146A/en active Pending
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111191737A (en) * | 2020-01-05 | 2020-05-22 | 天津大学 | Fine-grained image classification method based on multi-scale repeated attention mechanism |
CN111583109A (en) * | 2020-04-23 | 2020-08-25 | 华南理工大学 | Image super-resolution method based on generation countermeasure network |
CN111738363A (en) * | 2020-07-24 | 2020-10-02 | 温州大学 | Alzheimer disease classification method based on improved 3D CNN network |
CN112257601A (en) * | 2020-10-22 | 2021-01-22 | 福州大学 | Fine-grained vehicle identification method based on data enhancement network of weak supervised learning |
Non-Patent Citations (4)
Title |
---|
TSUNG-YU LIN 等: "Bilinear CNNs for Fine-grained Visual Recognition", 《 COMPUTER VISION AND PATTERN RECOGNITION》 * |
崔秀明: "基于深度神经网络的阿尔兹海默病分类算法研究", 《中国优秀硕士学位论文全文数据库医药卫生科技辑》 * |
邓爽 等: "基于改进VGG网络的弱监督细粒度阿尔兹海默症分类方法", 《计算机应用》 * |
陆鑫伟 等: "基于注意力自身线性融合的弱监督细粒度图像分类算法", 《计算机应用》 * |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN115830402A (en) * | 2023-02-21 | 2023-03-21 | 华东交通大学 | Fine-grained image recognition classification model training method, device and equipment |
CN115830402B (en) * | 2023-02-21 | 2023-09-12 | 华东交通大学 | Fine-granularity image recognition classification model training method, device and equipment |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Jifara et al. | Medical image denoising using convolutional neural network: a residual learning approach | |
CN107220990B (en) | Hair segmentation method based on deep learning | |
CN112116605A (en) | Pancreas CT image segmentation method based on integrated depth convolution neural network | |
CN108052523A (en) | Gambling site recognition methods and system based on convolutional neural networks | |
CN113782190B (en) | Image processing method based on multistage space-time characteristics and mixed attention network | |
CN110807139A (en) | Picture identification method and device, computer readable storage medium and computer equipment | |
CN111695430A (en) | Multi-scale face detection method based on feature fusion and visual receptive field network | |
CN111709914B (en) | Non-reference image quality evaluation method based on HVS characteristics | |
CN106295501A (en) | The degree of depth based on lip movement study personal identification method | |
CN112163488B (en) | Video virtual dummy face detection method and electronic device | |
CN114187311A (en) | Image semantic segmentation method, device, equipment and storage medium | |
CN112836756B (en) | Image recognition model training method, system and computer equipment | |
CN112149526B (en) | Lane line detection method and system based on long-distance information fusion | |
CN114612714A (en) | Curriculum learning-based non-reference image quality evaluation method | |
CN110163855B (en) | Color image quality evaluation method based on multi-path deep convolutional neural network | |
CN114387270B (en) | Image processing method, image processing device, computer equipment and storage medium | |
CN111882516A (en) | Image quality evaluation method based on visual saliency and deep neural network | |
CN114821146A (en) | Enhanced weak supervision-based fine-grained Alzheimer's disease classification method | |
CN104820988B (en) | One kind is without with reference to objective evaluation method for quality of stereo images | |
CN108304916B (en) | Convolutional neural network optimization method combining attention mechanism and depth separable convolution | |
CN111652238A (en) | Multi-model integration method and system | |
CN113963427B (en) | Method and system for rapid in-vivo detection | |
CN115761888A (en) | Tower crane operator abnormal behavior detection method based on NL-C3D model | |
CN114596287A (en) | Image quality determination method, device, equipment and storage medium | |
CN114785978A (en) | Video image quality determination method for video conference |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20220729 |
|
RJ01 | Rejection of invention patent application after publication |