CN116129182A - Multi-dimensional medical image classification method based on knowledge distillation and neighbor classification - Google Patents
Multi-dimensional medical image classification method based on knowledge distillation and neighbor classification Download PDFInfo
- Publication number
- CN116129182A CN116129182A CN202310037783.XA CN202310037783A CN116129182A CN 116129182 A CN116129182 A CN 116129182A CN 202310037783 A CN202310037783 A CN 202310037783A CN 116129182 A CN116129182 A CN 116129182A
- Authority
- CN
- China
- Prior art keywords
- model
- medical image
- training
- dimension
- teacher
- 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 38
- 238000013140 knowledge distillation Methods 0.000 title claims abstract description 24
- 238000012549 training Methods 0.000 claims abstract description 77
- 201000010099 disease Diseases 0.000 claims abstract description 33
- 208000037265 diseases, disorders, signs and symptoms Diseases 0.000 claims abstract description 33
- 238000013480 data collection Methods 0.000 claims abstract description 13
- 238000005516 engineering process Methods 0.000 claims abstract description 3
- 238000007781 pre-processing Methods 0.000 claims description 8
- 238000004590 computer program Methods 0.000 claims description 6
- 238000002372 labelling Methods 0.000 claims description 6
- 238000005457 optimization Methods 0.000 claims description 4
- 238000005520 cutting process Methods 0.000 claims description 3
- 238000003860 storage Methods 0.000 claims description 3
- 238000002512 chemotherapy Methods 0.000 claims description 2
- 238000003745 diagnosis Methods 0.000 abstract 1
- 206010018910 Haemolysis Diseases 0.000 description 5
- 206010023126 Jaundice Diseases 0.000 description 5
- 230000008588 hemolysis Effects 0.000 description 5
- 238000013528 artificial neural network Methods 0.000 description 4
- 239000008280 blood Substances 0.000 description 4
- 210000004369 blood Anatomy 0.000 description 4
- 208000014951 hematologic disease Diseases 0.000 description 4
- 238000004422 calculation algorithm Methods 0.000 description 3
- 238000012360 testing method Methods 0.000 description 3
- 238000010276 construction Methods 0.000 description 2
- 238000001514 detection method Methods 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 238000007689 inspection Methods 0.000 description 2
- 208000019838 Blood disease Diseases 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000009534 blood test Methods 0.000 description 1
- 238000007635 classification algorithm Methods 0.000 description 1
- 208000018706 hematopoietic system disease Diseases 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000002360 preparation method Methods 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
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
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02T—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
- Y02T10/00—Road transport of goods or passengers
- Y02T10/10—Internal combustion engine [ICE] based vehicles
- Y02T10/40—Engine management systems
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Artificial Intelligence (AREA)
- General Physics & Mathematics (AREA)
- Computing Systems (AREA)
- Health & Medical Sciences (AREA)
- Evolutionary Computation (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)
- Apparatus For Radiation Diagnosis (AREA)
Abstract
The invention discloses a multi-dimensional medical image classification method based on knowledge distillation and neighbor classification, which comprises a training data collection step, a model training step and a medical image classification step; in the training data collection process, firstly, medical images are collected and preprocessed, and then labels of all disease dimensions are marked on the medical images to obtain training data; in the model training process, firstly, respectively training a corresponding model aiming at each disease dimension, and then converging knowledge into a unified student model through a knowledge distillation technology; and loading a student model in the final medical image classification process, acquiring a case picture to be tested, inputting the preprocessed picture into the model, and feeding back the output result of the case picture in all dimensions. The method can extract the characteristics end to end and make predictions, and a unified model is used for outputting the prediction results in multiple disease dimensions for a specific diagnosis picture at the same time.
Description
Technical Field
The invention relates to a multi-dimensional medical picture classification method based on knowledge distillation and neighbor classification, and belongs to the technical field of image processing and image classification.
Background
Medical image classification related algorithms have been the focus of research in the academia and industry for many years. Because this algorithm is closely related to the daily life of people, for example, assisting doctors in making preliminary diagnoses to save specialist's manpower or to give predictive results faster than some more complex test items. How to build a proper model according to the previous case and make more accurate predictions for the new case data is the final aim of the algorithm.
Multidimensional classification is indeed very common in the medical field, for example in the prediction of blood diseases. The blood sample picture of a patient can correspond to the prediction results of various blood diseases, such as hemolysis, lipidemia and jaundice blood, and the blood diseases can be respectively divided into a plurality of categories according to the degree of severity. The instrument consumes a lot of time to detect the blood sample, which can affect the diagnostic efficiency of the patient. Therefore, whether a relatively accurate prediction result can be rapidly given only according to the picture of the blood sample is a very interesting direction in the auxiliary medical field.
In many cases, the past medical image classification method can only model a certain dimension, but cannot model a plurality of dimensions at the same time, so that the correlation between the plurality of dimensions cannot be considered. In addition, the existing multi-dimensional classification algorithm mostly depends on the extracted sample characteristics, and the classification is performed by using a traditional machine learning model, which is not applicable to medical image classification tasks in the current big data age.
Disclosure of Invention
The invention aims to: aiming at the problems and the shortcomings in the prior art, the invention provides a multidimensional medical image classification method based on knowledge distillation and neighbor classification, which can extract medical image features end to end and classify without extracting the medical image sample features in advance. According to the method provided by the invention, corresponding models are respectively learned in each dimension, then the models are used as teachers, knowledge is gathered on one student model through a knowledge distillation technology, and the student model can respectively extract characteristics in different dimensions for each medical picture and is classified through a nearest neighbor classifier.
The technical scheme is as follows: a multi-dimensional medical image classification method based on knowledge distillation and neighbor classification comprises a training data collection step of medical image classification, a teacher model training step, a student model training step and a medical image classification step; in the step of collecting training data of medical image classification, firstly collecting medical images and preprocessing (the preprocessing on the side mainly comprises the steps of cutting, denoising, centralizing and the like), and then labeling all disease dimensions in the medical images to obtain training data; training a model for each disease dimension respectively to obtain teacher models of each dimension, wherein the models are used as teacher models to teach other student models, training data are input for the teacher models of each dimension in the teacher model training process, output results of the teacher models are compared with actual labels of the corresponding dimension to obtain cross entropy loss, and return loss is used for optimizing the teacher models; in the training process of the student model, training data are respectively input for the student model and the teacher model of each dimension, the output result of the student model is compared with the actual label of the corresponding dimension to obtain cross entropy loss, then the cross entropy loss is compared with the output of the teacher model of the corresponding dimension to obtain KL divergence loss, all losses are returned to optimize the student model, and all losses refer to the sum of all cross entropy losses and all KL divergence losses. For each dimension there is a corresponding cross entropy loss and KL divergence loss; in the medical image classification process, a student model is loaded, medical images in previous training data are firstly input to obtain characteristic representations of class centers of different types in each dimension (the class center characteristic representations can be stored, the step is not repeated in the follow-up test), then the medical images to be classified are obtained, the images are preprocessed and then input to the student model to obtain characteristic representations of the medical images in different dimensions, and then the medical images are classified by a neighbor classifier to obtain prediction results in all disease category dimensions.
In medical problems, for a certain sample, a common classification problem only classifies the sample in one dimension, for example, it is determined which type of blood disease a certain blood test sample belongs to, for example, there are four options of hemolysis, lipidemia, jaundice and no disease, which is a standard multi-class classification problem (multi-class classification). The above-described multi-category classification can select only one of four options, and thus cannot cope with a case where a sample has a plurality of diseases at the same time. Therefore, when we need to judge whether there are three diseases of hemolysis, lipidemia and jaundice for the sample, we have constituted a multi-label classification problem (multi-label classification). However, sometimes, multi-mark learning cannot meet the needs of some specific medical tasks, for example, whether a sample has three diseases, such as hemolysis, lipidemia and jaundice, or not, and whether the three diseases are mild or severe are required to be judged respectively.
The multi-dimensional classification problem (multi-dimensional classification) studied in the present invention is further required to determine whether a disease belongs to a certain dimension (such as hemolysis, lipidemia and jaundice, which is actually a disease category, but is defined as a disease dimension in order to distinguish from a subsequent disease category of a light or heavy degree), and determine an internal category of a disease of each dimension (such as a plurality of categories of normal, light, serious and the like in consideration of the light or heavy degree). Thus, if defined in the dimension of a disease, it is the disease category that has multiple classifications within itself (note that the multiple classifications herein are broad concepts, e.g., the degree of severity is one, the type i, type ii, type iii, etc. of the disease). When we need to consider multiple such disease dimensions, which themselves possess multiple categories, we have constructed a multi-dimensional classification problem, namely the problem of the present invention.
The training data collection step specifically comprises the following steps:
step 101, initializing;
step 102, collecting medical images, such as case pictures;
step 103, preprocessing the collected medical image, including cutting, denoising and centralizing the medical image;
104, labeling the preprocessed medical image with labels on all disease dimensions;
step 105, if the medical image data volume meets the training requirement, continuing to collect the medical image data volume to obtain training data; if the demand is not met, steps 102 to 104 are repeated.
The teacher model training steps are as follows:
step 201, a model is built and initialized for each disease dimension;
step 202, training a model in each dimension by using all training data;
step 203, respectively outputting prediction results in each dimension from all models;
step 204, comparing the prediction result in each dimension with the actual label to obtain loss;
step 205, returning each teacher model of loss optimization respectively;
step 206, judging whether all models are converged, if yes, turning to step 207, otherwise, turning back to step 202;
step 207, save all teacher models.
The student model training steps are as follows:
step 301, initializing a chemo model;
step 302, using the teacher model in each dimension trained in the previous step as a teacher, and training a student model by using knowledge distillation;
step 303, outputting prediction results of all dimensions from the student models, and then outputting prediction results of corresponding dimensions from each teacher model respectively;
step 304, calculating the loss between the output result of the student model and the actual label and the loss between the output result of each teacher model;
step 305, returning all loss optimization student models; all loss refers to the sum of all cross entropy losses and all KL divergence losses;
step 306, judging whether the student model converges, if yes, turning to step 307, otherwise, returning to step 302;
step 307, save student model.
The medical image classification step comprises the following steps:
step 401, loading a student model;
step 402, inputting all training pictures into a student model, and obtaining characteristic representations of the medical pictures in different dimensions;
step 403, calculating the feature representation of class centers of different classes on each dimension respectively;
step 404, obtaining a medical picture to be detected and preprocessing;
step 405, inputting the medical pictures into a student model, and obtaining characteristic representations of the medical pictures in different dimensions;
step 406, respectively calculating the distances from the features of the medical picture to be tested to the class center features of different classes in each dimension, and classifying by using a nearest neighbor classifier;
step 407, feeding back the classification result in all dimensions.
The training workflow of the teacher model is (taking the training process of the teacher model in a specific dimension as an example, and other teacher models are the same as each other):
(1) Randomly initializing teacher model parameters in a certain dimension; (2) Training a teacher model through training data collected in a data collection process, wherein medical images in the training data are provided with labels of each disease dimension; (3) Optimizing a teacher model through cross entropy loss between a prediction result of the teacher model on the medical image and a real label of the medical image; (4) repeating (2) to (3) until the training error converges.
The training workflow of the student model is as follows:
(1) Randomly initializing student model parameters; (2) Training a student model through training data collected in a data collection process and category labels of all dimensions, wherein the attention picture is subjected to pretreatment; (3) Optimizing a student model by means of a loss of cross entropy between a prediction result of the student model on a medical image and a real label of the medical image and a loss of KL divergence between the prediction result and an output of a teacher model in a corresponding dimension; (4) repeating (2) to (3) until the training error converges.
A computer device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, the processor implementing a multi-dimensional medical image classification method based on knowledge distillation and neighbor classification as described above when executing the computer program.
A computer-readable storage medium storing a computer program for performing a multi-dimensional medical image classification method based on knowledge distillation and neighbor classification as described above.
The beneficial effects are that: compared with the prior art, the multidimensional medical image classification method based on knowledge distillation and neighbor classification can extract medical picture features end to end and classify the medical picture features, and can simultaneously predict classification results of all disease category dimensions.
Drawings
FIG. 1 is a flowchart of training data collection work for an embodiment of the present invention;
FIG. 2 is a flow chart of a teacher model training workflow in accordance with an embodiment of the present invention;
FIG. 3 is a student model training workflow diagram of an embodiment of the invention;
fig. 4 is a medical image classification workflow diagram of an embodiment of the invention.
Detailed Description
The present invention is further illustrated below in conjunction with specific embodiments, it being understood that these embodiments are meant to be illustrative of the invention only and not limiting the scope of the invention, and that modifications of the invention, which are equivalent to those skilled in the art to which the invention pertains, will fall within the scope of the invention as defined in the claims appended hereto.
As shown in fig. 1, a certain number of case images need to be collected in the training stage, which is a training data collection process. Specifically, the mobile device is initialized first, and generally includes steps of hardware device inspection, program running environment inspection, etc. (step 101); then collecting case images from past detection cases in the hospital or other ways (step 102); preprocessing such as denoising, size adjustment, centering and the like is carried out on the case image (step 103); labeling all disease dimensions of the case image, wherein the process can be expert labeling or labeling the detection result of the instrument (step 104); judging whether a preset number of training samples are acquired (step 105), if so, entering step 106, and if not, entering step 102; the labeled training samples are saved as training data in preparation for subsequent training (step 106).
The teacher model training phase workflow is shown in figure 2.
Specifically, firstly, a model is initialized for each disease dimension, mainly the construction of a deep neural network and the random initialization of model parameters, the teacher model on this side can uniformly adopt a residual neural network, for example, resnet18 as a feature extractor, and then a full connection layer is connected as a classifier (step 201). All models are then trained, and the preprocessed training pictures are input into all teacher models, respectively (step 202). All models output predictions in each disease dimension separately (step 203). The prediction results of each teacher model are compared with the actual labels of the corresponding dimensions, respectively, and a penalty, such as a cross entropy penalty, is calculated (step 204). Loss is returned separately, and each teacher model is optimized by gradient descent (step 205). A determination is made as to whether each teacher model is converging, and the non-converging training process of steps 202-205 may continue (step 206). All trained teacher models are saved (step 207).
The student model training phase workflow is shown in figure 3.
Specifically, a student model is initialized firstly, mainly the construction of a deep neural network and the random initialization of model parameters. The student model on this side can also use a residual neural network, such as Resnet18, as a feature extractor, but the extracted features need to be projected into multiple parts, one for each disease dimension, and the projection layer on this side can be implemented using a fully connected network. Finally, a full connection layer is respectively connected to the characteristic of each dimension and used as a classifier of each dimension. (step 301). The student model is then trained using knowledge distillation using the teacher model in each dimension trained in the previous stage as a teacher (step 302). The training pictures are input into the student model and all the teacher models respectively, prediction results of all dimensions are output from the student model, and prediction results in corresponding dimensions are output from the teacher models respectively (step 303). Comparing the prediction results of the student model in all dimensions with the actual labels of the corresponding dimensions respectively, and calculating losses, such as cross entropy losses; at the same time, the output of the teacher model of the corresponding dimension is compared, and a loss, such as a KL divergence loss, is calculated (step 304). All losses are passed back and the student model is optimized by gradient descent. Note that this loss is not passed back to the teacher model, i.e., all teacher models are fixed and not updated at this stage (step 305). A determination is made as to whether the student model is converging and if not, the training process of steps 302-305 may continue (step 206). The trained student model is saved (step 307).
The medical image classification stage workflow is shown in fig. 4. Specifically, a previously trained student model is first loaded (step 401); inputting all training pictures into a student model, and obtaining the output of the case pictures after each projection layer, so as to obtain the characteristic representation of the case pictures in different disease dimensions (step 402); feature representations of class centers of different classes are calculated in each dimension respectively, for example, feature representations of all pictures of the same class are directly averaged to serve as the class center of the class. The resulting feature representations of all class centers may be stored here, steps 402 and 403 may be skipped during subsequent multi-batch testing (step 403); obtaining a case picture to be predicted and performing preprocessing identical to the training phase (step 404); inputting the pictures into a student model, and obtaining characteristic representations of the pictures in different dimensions, namely output after each projection layer (step 405); calculating the distance from the feature of the case picture to be tested to the feature of the center of different classes in each dimension, classifying by using a nearest neighbor classifier, namely, predicting the class to which class center is nearest (step 406); the prediction results in all dimensions are fed back (step 307).
It will be apparent to those skilled in the art that the steps of the multi-dimensional medical image classification method based on knowledge distillation and neighbor classification of the embodiments of the invention described above may be implemented in a general purpose computing device, they may be centralized in a single computing device, or distributed over a network of computing devices, or alternatively, they may be implemented in program code executable by computing devices, such that they may be stored in a memory device, executed by computing devices, and in some cases, the steps shown or described may be executed in a different order than what is shown or described herein, or they may be individually fabricated as individual integrated circuit modules, or a plurality of the modules or steps may be fabricated as a single integrated circuit module. Thus, embodiments of the invention are not limited to any specific combination of hardware and software.
Claims (8)
1. A multi-dimensional medical image classification method based on knowledge distillation and neighbor classification is characterized by comprising a training data collection step of medical image classification, a teacher model training step, a student model training step and a medical image classification step; in the training data collection step of medical image classification, firstly, medical images are collected and preprocessed, and then all disease dimensions in the medical images are marked to obtain training data; in the training process of the teacher model, training a teacher model aiming at each disease dimension, and converging knowledge into a unified student model through a knowledge distillation technology; and loading a student model in the final medical image classification process, acquiring a case picture to be tested, inputting the preprocessed picture into the model, and feeding back output results in all dimensions.
2. The multi-dimensional medical image classification method based on knowledge distillation and neighbor classification according to claim 1, wherein the training data collection step specifically comprises:
step 101, initializing;
step 102, collecting medical images, such as case pictures;
step 103, preprocessing the collected medical image, including cutting, denoising and centralizing the medical image;
104, labeling all disease dimension labels on the preprocessed medical image;
step 105, if the medical image data volume meets the training requirement, continuing to collect the medical image data volume to obtain training data; if the demand is not met, steps 102 to 104 are repeated.
3. The multi-dimensional medical image classification method based on knowledge distillation and neighbor classification according to claim 1, wherein said teacher model training step is:
step 201, a model is built and initialized for each disease dimension;
step 202, training a model in each dimension by using all training data;
step 203, respectively outputting prediction results in each dimension from all models;
step 204, comparing the prediction result in each dimension with the actual label to obtain loss;
step 205, returning each teacher model of loss optimization respectively;
step 206, judging whether all models are converged, if yes, turning to step 207, otherwise, turning back to step 202;
step 207, save all teacher models.
4. The multi-dimensional medical image classification method based on knowledge distillation and neighbor classification according to claim 1, wherein the student model training step is:
step 301, initializing a chemo model;
step 302, using the teacher model in each dimension trained in the previous step as a teacher, and training a student model by using knowledge distillation;
step 303, outputting prediction results of all dimensions from the student models, and then outputting prediction results of corresponding classification dimensions from each teacher model respectively;
step 304, calculating the loss between the output result of the student model and the actual label and the loss between the output result of each teacher model;
step 305, returning all loss optimization student models;
step 306, judging whether the student model converges, if yes, turning to step 307, otherwise, returning to step 302;
step 307, save student model;
the medical image classification step comprises the following steps:
step 401, loading a student model;
step 402, inputting all training pictures into a student model, and obtaining characteristic representations of the medical pictures in different dimensions;
step 403, calculating the feature representation of class centers of different classes on each dimension respectively;
step 404, obtaining a medical picture to be detected and preprocessing;
step 405, inputting the medical pictures into a model, and obtaining characteristic representations of the medical pictures in different dimensions;
step 406, respectively calculating the distances from the features of the medical picture to be tested to the class center features of different classes in each dimension, and classifying by using a nearest neighbor classifier;
step 407, feeding back the classification result in all dimensions.
5. The multi-dimensional medical image classification method based on knowledge distillation and neighbor classification according to claim 1, wherein the training workflow of the teacher model is:
(1) Randomly initializing teacher model parameters in a certain dimension; (2) Training a teacher model through training data collected in a data collection process, wherein a picture is required to be preprocessed; (3) Optimizing a teacher model through cross entropy loss between a prediction result of the teacher model on the medical image and a real label of the medical image; (4) repeating (2) to (3) until the training error converges.
6. The multi-dimensional medical image classification method based on knowledge distillation and neighbor classification according to claim 1, wherein the training workflow of the student model is:
(1) Randomly initializing student model parameters; (2) Training a student model through training data collected in a data collection process, wherein the picture is firstly preprocessed; (3) Optimizing a student model by means of a loss of cross entropy between a prediction result of the student model on a medical image and a real label of the medical image and a loss of KL divergence between the prediction result and an output of a teacher model in a corresponding dimension; (4) repeating (2) to (3) until the training error converges.
7. A computer device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, the processor implementing a multi-dimensional medical image classification method based on knowledge distillation and neighbor classification as claimed in any one of claims 1-6 when executing the computer program.
8. A computer readable storage medium, characterized in that the computer readable storage medium stores a computer program for performing the multi-dimensional medical image classification method based on knowledge distillation and neighbor classification as claimed in any one of claims 1-6.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310037783.XA CN116129182A (en) | 2023-01-10 | 2023-01-10 | Multi-dimensional medical image classification method based on knowledge distillation and neighbor classification |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310037783.XA CN116129182A (en) | 2023-01-10 | 2023-01-10 | Multi-dimensional medical image classification method based on knowledge distillation and neighbor classification |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN116129182A true CN116129182A (en) | 2023-05-16 |
Family
ID=86304249
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310037783.XA Pending CN116129182A (en) | 2023-01-10 | 2023-01-10 | Multi-dimensional medical image classification method based on knowledge distillation and neighbor classification |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116129182A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117393156A (en) * | 2023-12-12 | 2024-01-12 | 珠海灏睿科技有限公司 | Multi-dimensional remote auscultation and diagnosis intelligent system based on cloud computing |
-
2023
- 2023-01-10 CN CN202310037783.XA patent/CN116129182A/en active Pending
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117393156A (en) * | 2023-12-12 | 2024-01-12 | 珠海灏睿科技有限公司 | Multi-dimensional remote auscultation and diagnosis intelligent system based on cloud computing |
| CN117393156B (en) * | 2023-12-12 | 2024-04-05 | 珠海灏睿科技有限公司 | Multi-dimensional remote auscultation and diagnosis intelligent system based on cloud computing |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Jakhar et al. | Big data deep learning framework using keras: A case study of pneumonia prediction | |
| US10198431B2 (en) | Information relation generation | |
| US20230162049A1 (en) | Artificial intelligence (ai) method for cleaning data for training ai models | |
| CN110660478A (en) | Cancer image prediction and discrimination method and system based on transfer learning | |
| WO2015148304A1 (en) | Method and system for large scale data curation | |
| CN109685765B (en) | X-ray film pneumonia result prediction device based on convolutional neural network | |
| JPH07295989A (en) | Device that forms interpreter to analyze data | |
| CN110969015B (en) | Automatic label identification method and equipment based on operation and maintenance script | |
| Godbin et al. | Screening of COVID-19 based on GLCM features from CT images using machine learning classifiers | |
| CN116842194A (en) | Electric power semantic knowledge graph system and method | |
| CN116129182A (en) | Multi-dimensional medical image classification method based on knowledge distillation and neighbor classification | |
| Caicedo et al. | A semantic content-based retrieval method for histopathology images | |
| CN116883768A (en) | Lung nodule intelligent grading method and system based on multi-modal feature fusion | |
| CN113555110B (en) | Method and equipment for training multi-disease referral model | |
| CN111898528B (en) | Data processing method, device, computer readable medium and electronic equipment | |
| Kotiyal et al. | Diabetic retinopathy binary image classification using PySpark | |
| CN115456693A (en) | Automatic evaluation method for automobile exterior design driven by big data | |
| CN114022698A (en) | Multi-tag behavior identification method and device based on binary tree structure | |
| BalaKrishna et al. | Autism spectrum disorder detection using machine learning | |
| CN117521673B (en) | Natural language processing system with analysis training performance | |
| US20240062907A1 (en) | Predicting an animal health result from laboratory test monitoring | |
| KR102388184B1 (en) | Collection, classification and management system of public information for detecting and controlling undeclared nuclear fuel cycle-related activities and method thereof | |
| Ojetunmibi et al. | Pneumonia disease detection and classification system using naive Bayesian technique | |
| CN116503674B (en) | Small sample image classification method, device and medium based on semantic guidance | |
| US20230316143A1 (en) | Methods and systems for creating training libraries and training AI processors |
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 |