CN113902672A - Rare skin lesion classification system based on space transformation optimization element learning - Google Patents

Rare skin lesion classification system based on space transformation optimization element learning Download PDF

Info

Publication number
CN113902672A
CN113902672A CN202111027125.XA CN202111027125A CN113902672A CN 113902672 A CN113902672 A CN 113902672A CN 202111027125 A CN202111027125 A CN 202111027125A CN 113902672 A CN113902672 A CN 113902672A
Authority
CN
China
Prior art keywords
meta
learning
skin lesion
skin
optimization
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
CN202111027125.XA
Other languages
Chinese (zh)
Inventor
李登旺
高祝敏
黄浦
董雪媛
洪亭轩
田伟伟
王建波
朱慧
李婕
吴冰
柴象飞
章桦
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.)
Shandong Normal University
Original Assignee
Shandong Normal 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 Shandong Normal University filed Critical Shandong Normal University
Priority to CN202111027125.XA priority Critical patent/CN113902672A/en
Publication of CN113902672A publication Critical patent/CN113902672A/en
Pending legal-status Critical Current

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
    • 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
    • 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
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • 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
    • G06T2207/30088—Skin; Dermal
    • 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)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • General Physics & Mathematics (AREA)
  • Artificial Intelligence (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Evolutionary Computation (AREA)
  • General Engineering & Computer Science (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Evolutionary Biology (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • General Health & Medical Sciences (AREA)
  • Health & Medical Sciences (AREA)
  • Biomedical Technology (AREA)
  • Computational Linguistics (AREA)
  • Biophysics (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Medical Informatics (AREA)
  • Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
  • Radiology & Medical Imaging (AREA)
  • Quality & Reliability (AREA)
  • Image Analysis (AREA)
  • Measuring And Recording Apparatus For Diagnosis (AREA)

Abstract

The invention discloses a rare skin lesion classification system based on spatial transformation optimization meta-learning, comprising an acquisition module, a classification module and a classification module, wherein the acquisition module is configured to: acquiring a skin image to be classified; a classification module configured to: and processing the skin image to be classified according to the trained meta-learning model based on the space transformation optimization to obtain a skin lesion classification result. The problem of identifying diseases from skin injury images in a low data state is formulated into a simple learning problem, model-independent meta-learning is applied, the capability of a system for rapidly adapting to new tasks and environments is absorbed, only little training is needed, and the influence caused by insufficient sample data is relieved.

Description

Rare skin lesion classification system based on space transformation optimization element learning
Technical Field
The invention relates to the technical field of medical image processing, in particular to a rare skin lesion classification system based on spatial transformation optimization meta-learning.
Background
The statements in this section merely provide background information related to the present disclosure and may not constitute prior art.
Skin cancer is the most common cancer worldwide, and melanoma is the most fatal cancer. Due to the lack of dermatologists, many patients can only be treated by general practitioners, which can lead to problems with false referrals, delayed care, and errors in diagnosis and treatment.
As a leading artificial intelligence method, the task of deep learning and skin lesion diagnosis is very effective, but in reality, the important problem of lack of medical training sample skin lesion image data exists.
Disclosure of Invention
In order to overcome the defects of the prior art, the invention provides a rare skin lesion classification system based on space transformation optimization meta-learning; the model which is fast adapted is trained by using meta-training data (common diseases), and the network model can be fast adapted by using only a few gradient descent steps of a small amount of meta-test data (rare diseases) to process a new task, so that the accuracy of the rare skin disease classification task is improved.
Rare skin lesion classification system based on spatial transformation optimization meta-learning, comprising:
an acquisition module configured to: acquiring a skin image to be classified;
a classification module configured to: and processing the skin image to be classified according to the trained meta-learning model based on the space transformation optimization to obtain a skin lesion classification result.
Compared with the prior art, the invention has the beneficial effects that:
1. the problem of identifying diseases from skin injury images in a low data state is formulated into a simple learning problem, model-independent meta-learning is applied, the capability of a system for rapidly adapting to new tasks and environments is absorbed, only little training is needed, and the influence caused by insufficient sample data is relieved.
2. The spatial transformation network is combined into the meta-learning framework, so that the convolutional neural network can learn invariance of image transformation, and the problem that the convolutional neural network is limited by the influence of invariance of an input image space is solved.
3. The skin disease diagnosis performance is improved by utilizing a space transformation network and meta-learning, and the rapid adaptability and the classification accuracy of the skin disease classification model under the small sample data are improved.
Advantages of additional aspects of the invention will be set forth in part in the description which follows, or may be learned by practice of the invention.
Drawings
The accompanying drawings, which are incorporated in and constitute a part of this specification, are included to provide a further understanding of the invention, and are incorporated in and constitute a part of this specification, illustrate exemplary embodiments of the invention and together with the description serve to explain the invention and not to limit the invention.
FIG. 1 is a schematic diagram of a rare skin lesion classification system based on spatial transformation optimization meta-learning according to an embodiment of the present invention;
FIG. 2 is a block diagram of a rare skin lesion classification system based on spatial transformation optimization meta-learning according to an embodiment of the present invention;
FIG. 3 is a diagram illustrating integration of spatial transformation networks into different convolutional layers by a rare skin lesion classification system based on spatial transformation optimization meta-learning according to an embodiment of the present invention;
FIG. 4 is a comparison experiment of adding a spatial transformation network to the 0 th layer, the 2 nd layer and the 4 th layer of a convolution frame in the rare skin lesion classification system based on spatial transformation optimization meta-learning according to the embodiment of the present invention;
FIG. 5 is a schematic diagram of a rare skin lesion classification system based on spatial transformation optimization meta-learning according to an embodiment of the present invention;
fig. 6 is a schematic structural diagram of a spatial transformation network according to an embodiment of the present invention.
Detailed Description
It is to be understood that the following detailed description is exemplary and is intended to provide further explanation of the invention as claimed. Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.
As shown in fig. 5, the rare skin lesion classification system based on spatial transformation optimization meta-learning includes:
an acquisition module configured to: acquiring a skin image to be classified;
a classification module configured to: and processing the skin image to be classified according to the trained meta-learning model based on the space transformation optimization to obtain a skin lesion classification result.
Further, the meta-learning model based on spatial transformation optimization; the method specifically comprises the following steps:
the device comprises a first space transformation network, a first convolution module, a second space transformation network, a third convolution module, a fourth convolution module, a third space transformation network, a full connection layer and an output layer which are connected in sequence.
Wherein the internal structures of the first convolution module, the second convolution module, the third convolution module and the fourth convolution module are consistent.
Wherein the first convolution module includes: the device comprises a convolution layer, an activation function layer and a maximum pooling layer which are connected in sequence.
Wherein the internal structures of the first spatial transformation network, the second spatial transformation network and the third spatial transformation network are identical.
As shown in fig. 6, the working principle of the first spatial transform network includes:
firstly, a positioning network localization net acquires an input skin lesion feature map; outputting transformation parameter theta applied to feature maps：θs＝floc；flocRepresenting a loss function; the skin lesion feature map comprises: height, width, and channel;
then, a Grid generator randomly selects a pixel point of the skin lesion feature map according to the transformation parameters of the feature map so as to generate transformed output;
finally, the skin lesion feature map and the transformed output are used as the input of Sampler to generate an output map.
The Grid generator divides a region of interest into a number of sub-regions (elements), and ideally the shape and distribution of each element in the Grid is determined by an automatic Grid generation algorithm. The spatial transformation network applies a learnable affine transformation to the input skin lesion image based on the input skin lesion feature map.
To perform the transformation of the input feature map, each output pixel is calculated by applying a sampling kernel centered at a set position in the input feature map, a point-by-point transformation formula:
Figure BDA0003243697510000041
wherein the content of the first and second substances,
Figure BDA0003243697510000042
is to input the source coordinates of the defined sample points in the skin condition map,
Figure BDA0003243697510000043
is to output the target coordinates of the grid in the skin condition element map,
Figure BDA0003243697510000044
is an affine transformation matrix.
Finally, the skin lesion feature map and the sampling grid are used as input to the sampler, resulting in an output map sampled from the input at grid points.
The first, second, and third spatial transform networks are added to the convolutional network, which actively transforms the feature maps to help minimize the overall cost function of the network during training.
Each skin lesion image was processed to 80 x 80 size, meta-learning rate of 0.001, batch size of 5, using Adam optimizer, number of iterations 10000 times.
Further, the trained meta-learning model based on space transformation optimization; the specific training process comprises the following steps:
constructing a training set and a test set; the training set and the testing set are skin images of known skin disease lesion labels;
the training set is preprocessed and the image is changed to a set size, for example: 80 by 80;
inputting the preprocessed training set into a meta-learning model based on space transformation optimization, and training the model; and updating parameters by adopting a gradient descending mode to obtain a trained meta-learning model based on space transformation optimization.
Illustratively, an ISIC-2019 skin lesion analysis melanoma detection data set is adopted, 25331 skin lesion images are obtained in total, and a meta-training set and a meta-testing set are constructed for meta-training and meta-testing respectively;
exemplary, data set processing: the 5 types with larger sample size are used as the common diseases as the meta-training data set, and the other 3 types are used as the rare diseases as the meta-testing training set to simulate the problem.
And establishing a meta-test task and a meta-training task by adopting an N-way K-shot mode in meta-learning. N-way means that there are N categories in the training data, and K-shot means that there are K labeled data under each category.
ISIC-2019 skin lesion analysis melanoma detection dataset, comprising eight categories: melanoma, melanocytic nevi, basal cell carcinoma, actinic keratosis, benign keratosis, dermatofibroma, vasculopathy, squamous cell carcinoma. The number of images of the eight categories is 4522, 12875, 3323, 867, 2624, 239, 253, and 628, respectively. Five lesions of melanoma, melanocyte nevus, basal cell carcinoma, actinic keratosis and benign keratosis are used as a meta-training set, and three lesions of dermatofibroma, angiopathy and squamous cell carcinoma are used as a meta-testing set.
Illustratively, fine adjustment is carried out on a meta-test new task support set, testing is carried out on a meta-test new task query set, and the skin lesion classification is realized by using a spatial transformation optimization meta-learning skin lesion classification model.
In the meta-training stage, each task consists of 5 random classes and K samples; in the meta-test phase, each test task consists of 3 random classes, each class having K samples. K is set to 1, 3, 5, respectively.
5 tasks sampled from the metadata set, each task consisting of 5 random samples; fine tuning on the support set of the meta-test task, querying 35 images from each of the 5 classes inferred the accuracy of the classification by randomly sampling samples from the 3 classes in the meta-test dataset. The rapid adaptation and generalization of deep neural networks trained on the basis of common skin disease data are promoted by utilizing meta learning and spatial transformation learning so as to identify rare skin diseases with less annotation data and improve classification performance.
Further, updating parameters in a gradient descending manner; the method specifically comprises the following steps:
the initialization parameters comprise a skin lesion feature map thetas；
In the face of a new task of classifying skin lesions, the parameter theta of the model is updated to theta'i。
When a gradient update is used, it is,
Figure BDA0003243697510000061
wherein, the initialization parameter of theta is theta'iThe updated parameter vector, the alpha-ary learning rate,
Figure BDA0003243697510000062
is a function of the loss as a function of,
Figure BDA0003243697510000063
is a gradient.
In fact, the spatial transformation optimization meta-learning skin lesion classification model aims at optimizing the model parameters such that one or a small number of gradient steps on a new task will yield the most efficient behavior on that task.
The meta-optimization across tasks is performed by adaptive moment estimation, such that the model parameters θ are updated as follows:
Figure BDA0003243697510000064
wherein, beta is the element step size,
Figure BDA0003243697510000065
an updated function;
the spatial transformation optimization element learning skin lesion classification is modeled as a combination of two functions:
Figure BDA0003243697510000071
wherein the content of the first and second substances,
Figure BDA0003243697510000072
representing a convolution function;
Figure BDA0003243697510000073
representing a spatial transfer function;
x → Y denotes that given an image of skin lesion to predict the relevant disease label,
Figure BDA0003243697510000074
x → X', estimating and applying the affine transformation to the input image X,
Figure BDA0003243697510000075
x' → Y designate a prediction label for the transformed skin lesion image.
The model utilizes the optimal initialization parameters to finely adjust a support set on a meta-test task of skin fibroma, angiopathy and squamous cell carcinoma membrane lesion, and completes spatial transformation optimization meta-learning skin lesion classification on a query set.
Further, processing the skin image to be classified according to the trained meta-learning model based on space transformation optimization to obtain a skin lesion classification result; the specific process comprises the following steps:
the Model-independent Meta-learning MAML (Model-Agnostic Meta-learning) adapts to rare skin disease classification by using the prior knowledge learned by classification tasks on common skin diseases, and solves the problem of unbalanced data set categories; the space conversion module enables the network to learn and actively convert the characteristic diagram.
The space transformation network inserts each layer in the convolution of the meta-learning MAML network irrelevant to the model, thereby improving the capability of the convolution network for learning space invariant features and improving the classification performance. In meta-learning, the goal of the trained model is to quickly learn the classification of new task rare cases from data classifications of common skin cases.
Referring to fig. 1, a schematic diagram of a rare skin lesion classification system based on spatial transformation optimization meta-learning according to the present invention is shown, five lesions, namely melanoma, melanocyte nevus, basal cell carcinoma, actinic keratosis, and benign keratosis, are used as a meta-training set, three lesions, namely, dermatofibroma, angiopathy, and squamous cell carcinoma, are used as a meta-testing set, and a meta-training task and a meta-testing task are constructed in a 5-way K-shot manner, where K is 1, 3, and 5, respectively.
Referring to fig. 2, a block diagram of a rare skin lesion classification system based on spatial transformation optimization element learning according to the present invention is shown. Referring to fig. 3, a rare skin lesion classification system based on spatial transformation optimization meta-learning according to the present invention integrates a spatial transformation network into different convolutional layers.
Referring to fig. 4, a graph of the results of comparative experiments in which a spatial transformation network is added to the 0 th layer, the 2 nd layer and the 4 th layer in a convolution frame in the rare skin lesion classification system based on spatial transformation optimization meta-learning according to the present invention is shown.
In summary, the rare skin lesion classification system based on spatial transformation optimization meta-learning utilizes the ISIC-2019 skin lesion analysis melanoma detection data set, uses five lesions of melanoma, melanocyte nevus, basal cell carcinoma, actinic keratosis and benign keratosis as a meta-training set, and uses three lesions of dermatofibroma, angiopathy and squamous cell carcinoma as a meta-testing set. The invention combines the space transformation network into the meta-learning framework, so that the convolutional neural network can learn the invariance of the image transformation, and the problem that the convolutional neural network is limited by the influence on the invariance of the input image space is solved. The capability of the system to adapt to new tasks and environments quickly is absorbed, and the influence caused by insufficient sample data is relieved. The skin disease diagnosis performance is improved by utilizing a space transformation network and meta-learning, and the rapid adaptability and the classification accuracy of the skin lesion classification model under low data are improved.
And selecting a model-independent meta-learning small sample classification method. The model independent learning trains the initialization parameters of the model, so that the model can obtain maximum performance by performing gradient descent once or several times on a new task with few samples.
The rapid adaptation and generalization of deep neural networks trained on the basis of common disease data are promoted by utilizing meta learning and spatial transformation learning so as to identify rare diseases with less annotation data and improve the classification performance.
The key idea is to train initial parameters of the optimization model so that after updating the parameters through one or more gradient steps, which are calculated with a small amount of data from the new task, the model has the maximum performance on the new task, so that one or a small number of gradient steps on the new task will produce the most efficient behavior on that task.
The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention, and various modifications and changes may be made by those skilled in the art. Any modification, equivalent replacement, or improvement made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims (10)

1. Rare skin lesion classification system based on spatial transformation optimization meta-learning, characterized by including:
an acquisition module configured to: acquiring a skin image to be classified;
a classification module configured to: and processing the skin image to be classified according to the trained meta-learning model based on the space transformation optimization to obtain a skin lesion classification result.
2. The rare skin lesion classification system based on spatial transform optimization meta-learning of claim 1, wherein the meta-learning model based on spatial transform optimization; the method specifically comprises the following steps:
the device comprises a first space transformation network, a first convolution module, a second space transformation network, a third convolution module, a fourth convolution module, a third space transformation network, a full connection layer and an output layer which are connected in sequence.
3. The rare skin lesion classification system based on spatial transform optimization meta-learning of claim 2, wherein the first convolution module comprises: the device comprises a convolution layer, an activation function layer and a maximum pooling layer which are connected in sequence.
4. The rare skin lesion classification system based on spatial transform optimization meta-learning as claimed in claim 2 wherein the first spatial transform network operates on the principle comprising:
firstly, a positioning network localization net acquires an input skin lesion feature map; outputting transformation parameter theta applied to feature maps：θs＝floc；flocRepresenting a loss function; the skin lesion feature map comprises: height, width, and channel;
then, a Grid generator randomly selects a pixel point of the skin lesion feature map according to the transformation parameters of the feature map so as to generate transformed output;
finally, the skin lesion feature map and the transformed output are used as the input of Sampler to generate an output map.
5. The rare skin lesion classification system based on spatial transform optimization meta-learning of claim 1, wherein the trained meta-learning model based on spatial transform optimization; the specific training process comprises the following steps:
constructing a training set and a test set; the training set and the testing set are skin images of known skin disease lesion labels;
preprocessing the training set, and changing the image into a set size;
inputting the preprocessed training set into a meta-learning model based on space transformation optimization, and training the model; and updating parameters by adopting a gradient descending mode to obtain a trained meta-learning model based on space transformation optimization.
6. The rare skin lesion classification system based on spatial transform optimization meta-learning of claim 5, wherein the training set is preprocessed to change the image to a set size; the method specifically comprises the following steps:
each skin lesion image was processed to 80 x 80 size, meta-learning rate 0.001, batch size 5, using Adam optimizer, number of iterations 10000 times.
7. The rare skin lesion classification system based on spatial transform optimization meta-learning as claimed in claim 5, wherein the parameter updating is performed in a gradient descent manner; the method specifically comprises the following steps:
the initialization parameters comprise a skin lesion feature map thetas；
In the face of a new task of classifying skin lesions, the parameter theta of the model is updated to theta'i；
When a gradient update is used, it is,
Figure FDA0003243697500000021
wherein, the initialization parameter of theta is theta'iThe updated parameter vector, the alpha-ary learning rate,
Figure FDA0003243697500000022
is a function of the loss as a function of,
Figure FDA0003243697500000023
is a gradient.
8. The rare skin lesion classification system based on spatial transform optimization meta-learning of claim 5, wherein a meta-test task and a meta-training task are established in an N-way K-shot manner in meta-learning; n-way means that there are N categories in the training data, and K-shot means that there are K labeled data under each category.
9. The rare skin lesion classification system based on spatial transform optimization meta-learning of claim 4, wherein to perform the transformation of the input feature map, each output pixel is computed by applying a sampling kernel centered at a set position in the input feature map, the point-by-point transformation formula:
Figure FDA0003243697500000031
wherein the content of the first and second substances,
Figure FDA0003243697500000032
is to input the source coordinates of the defined sample points in the skin condition map,
Figure FDA0003243697500000033
is to output the target coordinates of the grid in the skin condition element map,
Figure FDA0003243697500000034
is an affine transformation matrix.
10. The rare skin lesion classification system based on spatial transform optimization meta-learning of claim 5, wherein the first spatial transform network, the second spatial transform network and the third spatial transform network are added to the convolutional network, such that the convolutional network actively transforms the feature map to help minimize the overall cost function of the network during training.
CN202111027125.XA 2021-09-02 2021-09-02 Rare skin lesion classification system based on space transformation optimization element learning Pending CN113902672A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202111027125.XA CN113902672A (en) 2021-09-02 2021-09-02 Rare skin lesion classification system based on space transformation optimization element learning

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202111027125.XA CN113902672A (en) 2021-09-02 2021-09-02 Rare skin lesion classification system based on space transformation optimization element learning

Publications (1)

Publication Number Publication Date
CN113902672A true CN113902672A (en) 2022-01-07

Family

ID=79188541

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202111027125.XA Pending CN113902672A (en) 2021-09-02 2021-09-02 Rare skin lesion classification system based on space transformation optimization element learning

Country Status (1)

Country Link
CN (1) CN113902672A (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114663679A (en) * 2022-05-25 2022-06-24 山东师范大学 Blood coagulation index abnormity classification method based on feature fusion meta-learning

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114663679A (en) * 2022-05-25 2022-06-24 山东师范大学 Blood coagulation index abnormity classification method based on feature fusion meta-learning

Similar Documents

Publication Publication Date Title
Saxena et al. Predictive modeling of brain tumor: a deep learning approach
Jifara et al. Medical image denoising using convolutional neural network: a residual learning approach
Zhang et al. Identification of maize leaf diseases using improved deep convolutional neural networks
Tan et al. Intelligent alerting for fruit-melon lesion image based on momentum deep learning
Khan et al. Intelligent fusion-assisted skin lesion localization and classification for smart healthcare
Prabhu et al. Few-shot learning for dermatological disease diagnosis
CN110838108A (en) Medical image-based prediction model construction method, prediction method and device
CN113693563A (en) Brain function network classification method based on hypergraph attention network
Gao et al. GQ-GCN: group quadratic graph convolutional network for classification of histopathological images
Tanwar et al. Deep learning-based hybrid model for severity prediction of leaf smut rice infection
Huang et al. Visual object tracking with discriminative correlation filtering and hybrid color feature
CN113902672A (en) Rare skin lesion classification system based on space transformation optimization element learning
Raja et al. A Novel Fuzzy-Based Modified GAN and Faster RCNN for Classification of Banana Leaf Disease
Prakash et al. Meningioma brain tumor detection and classification using hybrid CNN method and RIDGELET transform
CN114140437A (en) Fundus hard exudate segmentation method based on deep learning
Shenbagam et al. Corn Leaf Disease Detection; A Survey
Deari et al. A hybrid multistage model based on YOLO and modified inception network for rice leaf disease analysis
CN114926745B (en) Domain feature mapping small sample SAR target recognition method
Deshpande et al. Train small, generate big: Synthesis of colorectal cancer histology images
Ali et al. Effective iris recognition for distant images using log-gabor wavelet based contourlet transform features
Fahrudin et al. SkinMate: Mobile-Based Application for Detecting Multi-Class Skin Diseases Classification Using Pre-Trained MobileNetV2 on CNN Architecture
Pushpa An efficient internet of things (iot)-enabled skin lesion detection model using hybrid feature extraction with extreme machine learning model
Babu et al. Leaf Disease Detection Using Machine Learning (ML)
Suguna et al. An Efficient Water Strider Algorithm with Auto Encoder for Cervical Cancer Diagnosis using Pap Smear Images
Irmawati et al. Classification of human blastocyst quality using wavelets and transfer learning

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