CN115631386A - Pathological image classification method and system based on machine learning - Google Patents
Pathological image classification method and system based on machine learning Download PDFInfo
- Publication number
- CN115631386A CN115631386A CN202211629416.0A CN202211629416A CN115631386A CN 115631386 A CN115631386 A CN 115631386A CN 202211629416 A CN202211629416 A CN 202211629416A CN 115631386 A CN115631386 A CN 115631386A
- Authority
- CN
- China
- Prior art keywords
- classification
- machine learning
- diagnosis
- images
- pathological
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/764—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using classification, e.g. of video objects
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N20/00—Machine learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/82—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using neural networks
Abstract
The method comprises the steps that for pathological images with diagnosis conclusions, a first classification result is obtained through a first machine learning model and is compared with the diagnosis conclusions, if the pathological images with diagnosis conclusions are the same, the pathological images are placed into a knowledge set, if the pathological images with diagnosis conclusions are different, a second classification result is obtained through a second machine learning model and is compared with the first classification result, if the pathological images with diagnosis conclusions are the same, the pathological images are placed into the knowledge set, if the pathological images with diagnosis conclusions are different, the records are given up, and for pathological images without diagnosis conclusions, classification results obtained through two different machine learning models are compared to judge whether the pathological images enter the knowledge set; meanwhile, when the model is trained, the training set is equally divided to improve the training precision of the model; by the scheme, the classification accuracy is improved, and doctors with less experience can learn to improve the self business literacy.
Description
Technical Field
The invention belongs to the technical field of image processing and artificial intelligence, and particularly relates to a pathological image classification method and system based on machine learning.
Background
At present, the public medical facilities cannot reach the satisfaction degree of people due to uneven medical resource distribution and limited hospitalizing channels, so that the online medical diagnosis platform is more and more widely applied, the online medical diagnosis platform is a medical service combining computer, communication and multimedia technologies with medical technologies, and the purpose of providing a hospitalizing channel selection different from hospital hospitalizing for patients is to meet the requirements of the patients and reduce the running time.
With the wide application of online medical online diagnosis platforms, a large number of medical case images, especially pathological images, containing abundant medical information and medical knowledge are received by the online medical online diagnosis platform every year, and doctors with short working life often have the conditions of poor business literacy and experience due to less contact with actual cases, less diagnosed pathological images and less experience, so that non-professional judgment can be made at a certain probability in the case of the pathological images of patients, and the illness state and medical experience of the patients can be influenced.
Meanwhile, when a machine model is adopted for pathological image recognition, because pathological images relate to privacy, hospitals rarely provide pathological images of patients externally, and the classification model cannot be trained sufficiently due to the fact that sufficient training samples are not available during model training, and the accuracy of classification results is low.
Therefore, there is a need in the art for a technical solution for accurately classifying pathological images for a doctor with a short working life and less experience to learn so as to improve professional literacy.
Disclosure of Invention
The invention aims to solve the technical problem of providing a pathological image classification method and system based on machine learning aiming at the defects of the technical scheme, and the pathological images of the hospitalizing cases received by the medical online diagnosis platform are classified, so that different types of knowledge sets are established for doctors with less experience to learn so as to improve professional literacy.
According to one aspect of the invention, a pathological image classification method based on machine learning comprises the following steps:
step 1: classifying the pathological images of each visit record through a first machine learning model, and obtaining a first classification result;
step 2: judging whether the pathological image in each visit record in the online diagnosis medical platform has a diagnosis conclusion, if so, entering the step 3, and if not, entering the step 5;
and step 3: judging whether the classification result of the step 2 is the same as the diagnosis conclusion in the visit record, and if so, entering a step 4; if not, entering step 7;
and 4, step 4: storing the pathological images and classification results of the treatment records in a knowledge set;
and 5: classifying the pathological images without diagnosis conclusions through a second machine learning model to obtain classification results;
step 6: judging whether the first machine learning model classification result and the second machine learning model classification result of the pathological image without the diagnosis conclusion are the same; if the two are the same, entering step 4; if not, entering step 7;
and 7: the visit record is discarded.
Specifically, the pathological image classified by the first machine learning model is one of an x-ray image, a CT image and a nuclear magnetic resonance image;
further, in step 1, the classification process of the first machine learning model specifically includes the following steps:
step 1.1, pre-training a convolutional neural network model by adopting an open source data set;
specifically, the network model to be trained is selected to be simply trained in an open source big data set with certain correlation, so that the network model learns corresponding prior knowledge in the open source big data set, and then the obtained model is transferred to a task to be processed;
specifically, the open source data set is one of a cifar data set and an imageNET data set;
step 1.2: acquiring images of a training set;
500 CT images with abnormal parts and 500 CT images without abnormal parts are selected from an online medical diagnosis platform to serve as a training set of a first machine learning model, meanwhile, in order to facilitate the training of the model, the formats of the images are unified into a PNG format, and the resolution is uniformly set to be 256 × 256;
step 1.3: training the pre-trained convolutional neural network model of the step 1.1 by adopting the training set;
specifically, the training set in the step 1.2 is divided into five parts, four parts of data are selected as training data in turn, and the remaining part of data is verification data, compared with the prior art in which the whole data set is used as the training set, the method has the advantages that the training set is divided into five parts, so that on one hand, the data operand is reduced, the model training speed is improved, and meanwhile, the four parts of data are used as the training set, which is equivalent to four times of model training, so that the prediction accuracy of the model can be greatly improved;
step 1.4: performing feature extraction on the images in the medical record through a trained convolutional neural network model;
specifically, after the image of the medical record is input into the trained convolutional neural network model, the features of the image in the medical record are extracted through a convolutional layer, a normalized normalization layer and an activation function;
step 1.5: inputting the characteristics of the image obtained in the step 1.4 into a classifier for classification;
furthermore, the classification result is that no abnormal part exists, an abnormal part exists, and one of three classes is judged;
specifically, the diagnosis conclusion is a diagnosis opinion of the pathological image of the online diagnosis medical platform, and specifically includes: abnormal parts exist, and judgment is unclear;
specifically, the knowledge set comprises three types of knowledge subsets, namely an abnormal part knowledge subset, an abnormal part knowledge subset and an unknown part knowledge subset, and the knowledge subsets cannot be judged;
since the second machine learning model is to be compared with the conclusions of the first machine learning model or the conclusions in the visit record, the accuracy of the model classification is prioritized in designing the second machine learning model in the embodiment;
specifically, the second machine learning model is a deep learning convolutional neural network classification model;
further, the deep learning convolutional neural network classification model specifically includes:
step 5.1: training set image acquisition:
in the embodiment, 500 CT images with abnormal parts and 500 CT images without abnormal parts are selected as a training set of a first machine learning model, meanwhile, for facilitating the training of the model, the formats of the images are unified into a PNG format, and the resolution is uniformly set to be 256 × 256;
meanwhile, in order to increase the sample size of the model training set, the real-time embodiment can process the images of the training set by a sample enhancement method, wherein the specific sample enhancement method comprises the steps of image pixel and feature transformation, specifically random noise access, contrast change, protection degree change, brightness change and the like, so that a greater number of sample sizes are generated, and the sample size of the training set is expanded;
step 5.2, constructing a deep learning convolutional neural network classification model;
specifically, the deep learning convolutional neural network model includes three main body layers: the method comprises the steps of (1) convolutional layers, pooling layers and full-connection layers, wherein a deep learning convolutional neural network model with 9 layers is adopted in the embodiment, the front 8 layers perform feature extraction, and the rear 1 layer classifies the extracted features, wherein the convolutional layers comprise 4 convolutional layers, 3 pooling layers, 1 dropout layer and a full-connection layer;
furthermore, the convolution kernels of the convolutional layers are all 3*3 in size;
step 5.3: classifying the pathological images through the second machine learning model obtained by training in the step 5.2;
the classification result is the abnormal part, the abnormal part exists, and one of three types is judged;
it is worth emphasizing that the classification result obtained in this embodiment is only whether a pathological image may have a pathological position, and often in an actual disease diagnosis process, a doctor may not be used for disease diagnosis for one index, but different indications are synthesized for disease diagnosis, so the classification result of this application is an intermediate parameter and does not diagnose a corresponding disease, meanwhile, in this embodiment, the classification process is obtained by a machine in an artificial intelligence manner, and no doctor participates in the whole process, and the classification result of this application is only used for promoting learning by inexperienced doctors.
According to another aspect of the present invention, there is provided a system for a pathological image classification method based on machine learning, the system using the pathological image classification method based on machine learning includes the following modules:
the first classification module is used for classifying the pathological images of each clinic record through a first machine learning model and obtaining a first classification result;
the diagnosis conclusion judging module is connected with the first classification module and used for judging whether a pathological image in each diagnosis record in the online diagnosis medical platform has a diagnosis conclusion or not, if yes, the pathological image, the diagnosis conclusion and the first classification result are transmitted to the first judging module, and if not, the pathological image and the first classification result are transmitted to the second classification module;
the first judgment module is connected with the diagnosis conclusion judgment module and used for judging whether the first classification result is the same as the diagnosis conclusion in the visit record or not, if so, the pathological image and the diagnosis conclusion are transmitted to the knowledge set building module, and if not, the diagnosis record is abandoned;
the second classification module is connected with the diagnosis conclusion judgment module and used for classifying the pathological images without diagnosis conclusions through a second machine learning model to obtain classification results;
the second judgment module is connected with the second classification module and used for judging whether the first classification result is the same as the second classification result or not, if so, the pathological image and the diagnosis conclusion are transmitted to the knowledge set establishment module, and if not, the diagnosis record is abandoned;
and the knowledge set establishing module is connected with the first classification module and the second classification module and is used for receiving the pathological images with the same conclusion in the first judgment module and the second judgment module.
Based on the technical scheme, the pathological image classification method and system based on machine learning have the following technical effects:
1. in the embodiment, the treatment record with pathological images is judged, the pathological images are classified according to whether a diagnosis conclusion exists, for the pathological images with the diagnosis conclusion, a classification result is obtained through a first machine learning model and is compared with the diagnosis conclusion, if the classification result is the same as the diagnosis conclusion, the pathological images enter a knowledge set, if the classification result is different from the classification result, the pathological images are obtained through a second machine learning model and are compared with the classification result of the first machine learning model, if the classification result is the same as the diagnosis result, the knowledge set is entered, and if the classification result is different from the classification result, the record is abandoned; comparing classification results obtained by two different machine learning models aiming at pathological images without diagnosis conclusions, and judging whether the pathological images enter a knowledge set; according to the technical scheme, pathological images entering the knowledge set are judged and classified twice, so that the accuracy of the knowledge set is improved, and doctors with less experience can learn to improve own business literacy.
2. Because the first machine learning model is a model for classifying the first visit records in the invention, and the classification accuracy of the first machine learning model influences the classification accuracy of the whole technical scheme of the invention, the invention improves the precision of model training by dividing the training set into five equal parts when designing the first machine learning model and training the first machine learning model, thereby improving the accuracy of model classification.
Drawings
In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings needed to be used in the description of the embodiments are briefly introduced below, and it is obvious that the drawings in the following description are only some embodiments of the present application, and it is obvious for those skilled in the art to obtain other drawings based on these drawings without creative efforts.
Fig. 1 is a flowchart of a method for classifying pathology images based on machine learning provided in an embodiment of the present application;
FIG. 2 is a flow diagram of a first machine learning model provided in an embodiment of the present application;
fig. 3 is a flow chart of a second machine learning model provided in an embodiment of the present application.
Detailed Description
The technical solutions in the embodiments of the present application will be described clearly and completely with reference to the drawings in the embodiments of the present application, and it is obvious that the described embodiments are only some embodiments of the present application, and not all embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present application.
The concept to which the present application relates will be first explained below with reference to the drawings. It should be noted that the following descriptions of the concepts are only for the purpose of facilitating understanding of the contents of the present application, and do not represent limitations on the scope of the present application.
The pathological images may be various images that can be used for determining the health condition of a living body (human) in the medical field, for example, X-ray images, CT images, magnetic resonance images, etc., and may be pathological images of various parts such as the brain, the lung, and the hand. The medical image classification is a process of performing classification processing according to the medical image for a specific purpose, and for convenience of description, the present embodiment uniformly adopts a brain pathology CT image as an image to be classified as follows.
The embodiment of the invention aims to solve the technical problem of the prior art, and provides a pathological image classification method and system based on machine learning, which are used for classifying pathological images of medical cases received by a medical online diagnosis platform, so that different types of knowledge sets are established for doctors with less experience to learn and improve professional literacy.
As shown in fig. 1, a pathological image classification method based on machine learning includes:
step 1: classifying the pathological images of each visit record through a first machine learning model, and obtaining a classification result;
specifically, different medical online diagnosis platforms have different medical resources and different advantageous diagnosis directions of different doctors, so that misdiagnosis is inevitable, and a machine learning model may have a classification error when classifying pathological images, but the invention of the present application is primarily aimed at providing learning images for less experienced doctors and needs to ensure classification accuracy, so on one hand, when designing a first machine learning model, when training the first machine learning model, the present embodiment performs five equal divisions on a training set, thereby improving model training precision and further improving model classification accuracy, and on the other hand, the present embodiment designs a step of comparing a machine learning model classification result with a diagnosis conclusion of the online diagnosis platform, so as to improve classification accuracy;
meanwhile, the pathological image classified by the first machine learning model is one of an x-ray image, a CT image and a nuclear magnetic resonance image; for convenience of description, the present embodiment takes a CT image as an example to perform the following description;
the machine learning model takes a neural network model as a framework, so that data characteristics are learned, internal rules contained in a sample are searched, the neural network applies mathematical statistics to simulate biological perception to make a decision, and compared with logical reasoning calculation, the machine learning model has more advantages on problem processing; the working mechanism is that the data transmitted by the upper layer neuron is used as input, then the input and the weight are weighted and summed, and then the weighted sum is mapped by a nonlinear activation function, and a conversion value is used as output to become the input of the lower layer neuron; thereby obtaining a classification result;
the machine learning model of this embodiment is a convolutional neural network model, and it is worth emphasizing that a hardware system operated by the convolutional neural network model of this embodiment is an intel i7 series 8-core CPU, and a 32GB memory and a server equipped with an independent graphics card exist in the memory, such hardware configuration ensures the speed and effectiveness of model training, and meanwhile, a software system adopts a wuban diagram operating system.
As shown in fig. 2, the classification process of the first machine learning model specifically includes the following steps:
step 1.1, pre-training a convolutional neural network model by adopting an open source data set;
in some complex scenes or specific detection target classification, an initial convolutional neural network model is usually not selected to be trained from zero, so that the operation not only needs computer equipment with excellent performance such as computing capacity and storage capacity, and the like, and the research and development cost is too high, but also the feature extraction capacity is reduced and the model prediction result is influenced due to too little data volume which is trained in the convolutional neural network with a complex structure for a long time; in the embodiment, a network model needing to be trained is simply trained in an open source big data set with certain correlation, so that the network model learns corresponding priori knowledge in the open source big data set, and then the obtained model is transferred to a task needing to be processed;
for the classification task of the CT image in this embodiment, on one hand, the image data set under the classification task is not as large as the scale of the open source data set, which results in a less than ideal network model robust index, and on the other hand, for the training of the data set of a certain scale, the cost and the cost are very high, but the hardware device of the general business company at present cannot meet the relevant requirements, so as to serve as an aspect of the improvement of the prior art in this embodiment, the convolutional neural network model is pre-trained in the open source data set, and then is moved to the image data set of this embodiment for training;
specifically, the open source data set is one of a cifar data set and an imageNET data set;
step 1.2: acquiring images of a training set;
in the embodiment, 500 CT images with abnormal parts and 500 CT images without abnormal parts are selected from an online medical diagnosis platform as a training set of a first machine learning model, and meanwhile, for facilitating model training, the formats of the images are unified into a PNG format, and the resolution is uniformly set to 256 × 256;
step 1.3: training the pre-trained convolutional neural network model of the step 1.1 by adopting the training set;
in order to improve the efficiency and precision of model training, in this embodiment, the training set in step 1.2 is divided into five parts, and at the same time, it is ensured that each part of data contains the same number of CT images, wherein four parts of data are selected as training data in turn, and the remaining part of data is verification data, compared with the prior art in which the whole data set is used as the training set, in this embodiment, the training set is divided into five parts, so that on one hand, the data computation is reduced, the speed of model training is improved, and meanwhile, since four parts of data are used as the training set, which is equivalent to four times of model training, the prediction accuracy of the model can be greatly improved;
step 1.4: performing feature extraction on the images in the medical record through a trained convolutional neural network model; specifically, the extraction process is the prior art, and is not discussed in detail herein, it is worth emphasizing that after the image of the medical record is input into the trained convolutional neural network model, the features of the image in the medical record are extracted through the convolutional layer, the normalized normalization layer and the activation function.
Step 1.5: inputting the characteristics of the image obtained in the step 1.4 into a classifier for classification;
the classification result is that there is no abnormal part, there is abnormal part, judge one of three types;
illustratively, the record with the conclusion in the medical platform is verified by using the machine learning model, and by using the model of the embodiment, the detection accuracy is 92.73%, while the accuracy of using the EFFICIENT-NET network commonly used in the medical field is 89.14%, and the accuracy of using the VGG14 network is 78.15%, so that the model of the embodiment has advantages in both detection accuracy and detection speed;
step 2: judging whether the pathological image in each diagnosis record in the online diagnosis medical platform has a diagnosis conclusion, if so, entering step 3, and if not, entering step 5;
specifically, the diagnosis conclusion is a diagnosis opinion of the pathological image of the online diagnosis medical platform, and specifically includes: abnormal parts exist, and judgment is unclear;
and step 3: judging whether the classification result of the step 2 is the same as the diagnosis conclusion in the visit record, and if so, entering a step 4; if not, entering step 7;
and 4, step 4: storing the pathological images and classification results of the treatment records in a knowledge set;
specifically, the knowledge set comprises three types of knowledge subsets, namely an abnormal part knowledge subset, an abnormal part knowledge subset and an unknown part knowledge subset, and the knowledge subsets cannot be judged;
and 5: classifying the pathological images without diagnosis conclusions through a second machine learning model to obtain classification results;
since the second machine learning model is to be compared with the conclusion of the first machine learning model or the conclusion in the visit record, the accuracy of the model classification is preferentially considered in designing the second machine learning model in the embodiment;
specifically, the second machine learning model is a deep learning convolutional neural network classification model;
further, as shown in fig. 3, the deep learning convolutional neural network classification model specifically includes:
step 5.1: training set image acquisition:
in the embodiment, 500 CT images with abnormal parts and 500 CT images without abnormal parts are selected as a training set of a first machine learning model, meanwhile, for facilitating the training of the model, the formats of the images are unified into a PNG format, and the resolution is uniformly set to be 256 × 256;
meanwhile, in order to increase the sample size of the model training set, the real-time embodiment can process the images of the training set by a sample enhancement method, wherein the specific sample enhancement method comprises the steps of image pixel and feature transformation, specifically random noise access, contrast change, protection degree change, brightness change and the like, so that a greater number of sample sizes are generated, and the sample size of the training set is expanded;
step 5.2, constructing a deep learning convolutional neural network classification model;
specifically, the deep learning convolutional neural network model includes three main body layers: the method comprises the following steps of (1) convolutional layers, pooling layers and full-connection layers, wherein a deep learning convolutional neural network model with 9 layers is adopted in the embodiment, the front 8 layers perform feature extraction, and the rear 1 layer classifies the extracted features, wherein the convolutional layers comprise 4 convolutional layers, 3 pooling layers, 1 dropout layer and one full-connection layer;
furthermore, the convolution kernels of the convolutional layers are all 3*3 in size;
step 5.3: classifying the pathological images through the second machine learning model obtained by training in the step 5.2;
the classification result is the abnormal part, the abnormal part exists, and one of three types is judged;
and 6: judging whether the first machine learning model classification result and the second machine learning model classification result of the pathological image without the diagnosis conclusion are the same; if the two are the same, entering step 4; if not, entering step 7;
and 7: the visit record is discarded.
According to another embodiment of the present application, the present invention further provides a system of a pathological image classification method based on machine learning, where the system adopts the pathological image classification method based on machine learning, and the system includes the following modules:
the first classification module is used for classifying the pathological images of each clinic record through a first machine learning model and obtaining a first classification result;
the diagnosis conclusion judging module is connected with the first classification module and used for judging whether a pathological image in each diagnosis record in the online diagnosis medical platform has a diagnosis conclusion or not, if yes, the pathological image, the diagnosis conclusion and the first classification result are transmitted to the first judging module, and if not, the pathological image and the first classification result are transmitted to the second classification module;
the first judgment module is connected with the diagnosis conclusion judgment module and used for judging whether the first classification result is the same as the diagnosis conclusion in the visit record or not, if so, the pathological image and the diagnosis conclusion are transmitted to the knowledge set building module, and if not, the diagnosis record is abandoned;
the second classification module is connected with the diagnosis conclusion judgment module and used for classifying the pathological images without diagnosis conclusions through a second machine learning model to obtain classification results;
the second judgment module is connected with the second classification module and is used for judging whether the first classification result is the same as the second classification result or not, if so, the pathological image and the diagnosis conclusion are transmitted to the knowledge set establishment module, and if not, the diagnosis record is abandoned;
and the knowledge set establishing module is connected with the first classification module and the second classification module and is used for receiving the pathological images with the same conclusion in the first judgment module and the second judgment module.
It is worth emphasizing that the classification result obtained in this embodiment is only whether a pathological image may have a pathological position, and often in an actual disease diagnosis process, a doctor may not be used for disease diagnosis for one index, but different indications are synthesized for disease diagnosis, so the classification result of this application is an intermediate parameter and does not diagnose a corresponding disease, meanwhile, in this embodiment, the classification process is obtained by a machine in an artificial intelligence manner, and no doctor participates in the whole process, and the classification result of this application is only used for promoting learning by inexperienced doctors.
In the embodiment, the medical record with pathological images is judged, the pathological images are classified according to whether the pathological images have a diagnosis conclusion or not, the classification result of the pathological images with the diagnosis conclusion is obtained through the first machine learning model and is compared with the diagnosis conclusion, if the classification result is the same as the diagnosis conclusion, the pathological images enter the knowledge set, if the classification result is different from the diagnosis result, the pathological images are obtained through the second machine learning model and are compared with the classification result of the first machine learning model, if the classification result is the same as the diagnosis result, the pathological images enter the knowledge set, and if the classification result is different from the diagnosis conclusion, the record is abandoned; comparing classification results obtained by two different machine learning models aiming at pathological images without diagnosis conclusions, and judging whether the pathological images enter a knowledge set; according to the technical scheme, pathological images entering the knowledge set are judged and classified twice, so that the accuracy of the knowledge set is improved, and doctors with less experience can learn to improve self business literacy.
The above-described embodiments and/or implementations are only illustrative of the preferred embodiments and/or implementations for implementing the present technology, and are not intended to limit the embodiments of the present technology in any way, and those skilled in the art can make modifications or changes without departing from the scope of the technical means disclosed in the present disclosure, but should be regarded as the technical means or implementations that are substantially the same as the present invention.
Claims (10)
1. A pathological image classification method based on machine learning is characterized by comprising the following steps:
step 1: classifying pathological images of each visit record of the online diagnosis medical platform through a first machine learning model, and obtaining a first classification result;
step 2: judging whether the pathological image in each visit record in the online diagnosis medical platform has a diagnosis conclusion, if so, entering the step 3, and if not, entering the step 5;
and 3, step 3: judging whether the first classification result is the same as the diagnosis conclusion in the visit record, and if so, entering the step 4; if not, entering step 7;
and 4, step 4: storing the pathology image of the visit record and the first classification result in a knowledge set;
and 5: classifying the pathological images without diagnosis conclusions through a second machine learning model to obtain a second classification result;
step 6: judging whether the first classification result of the first machine learning model of the pathological image without the diagnosis conclusion is the same as the second classification result of the second machine learning model; if the two are the same, entering the step 4; if not, entering step 7;
and 7: the visit record is discarded.
2. The method for classifying pathological images based on machine learning according to claim 1, wherein in the step 2, the diagnosis conclusion is a diagnosis opinion of pathological images of the online diagnosis medical platform, which specifically includes: there were no abnormal parts, and there were abnormal parts, which were not clearly judged.
3. The method for classifying pathology images based on machine learning according to claim 1, wherein in the step 1, the first machine learning model specifically comprises the following steps:
step 1.1: pre-training a convolutional neural network model by adopting an open source data set;
step 1.2: acquiring images of a training set;
step 1.3: equally dividing the training set in the step 1.2 into five parts, selecting four parts of data as training data in turn, using the remaining part of data as verification data, and training the pre-trained convolutional neural network model in the step 1.1;
step 1.4: carrying out feature extraction on the images in the visit record through a trained convolutional neural network model;
step 1.5: and (5) inputting the characteristics of the image acquired in the step (1.4) into a classifier for classification.
4. The method of classifying pathology images based on machine learning of claim 3, wherein 500 pathology images with abnormal parts and 500 pathology images without abnormal parts are selected as a training set of the first machine learning model, the format of the images in the training set is PNG format, and the resolution is set to 256 × 256.
5. The method of classifying pathology images based on machine learning according to claim 3, wherein the first classification result is no abnormal portion, an abnormal portion exists, and one of three unknown types is determined.
6. The method of classifying pathology images based on machine learning according to claim 1, wherein in step 4, the knowledge set comprises three types of knowledge subsets, namely, an abnormal part knowledge subset and an uncertain knowledge subset.
7. The method for classifying pathology images based on machine learning according to claim 1, wherein the step 5 specifically comprises:
step 5.1: training set image acquisition:
step 5.2: constructing a deep learning convolutional neural network classification model as a second machine learning model;
the deep learning convolutional neural network classification model comprises three main body layers: the deep learning convolutional neural network classification model adopts a 9-layer deep learning convolutional neural network model, the front 8 layers execute feature extraction, and the rear 1 layer classifies the extracted features, wherein the deep learning convolutional neural network classification model comprises 4 convolutional layers, 3 pooling layers, 1 dropout layer and a full connection layer; the convolution kernels of the convolutional layers are all 3*3;
step 5.3: and classifying the pathological images through the second machine learning model obtained through the training in the step 5.2 to obtain a second classification result.
8. The method of classifying pathology images according to claim 7, wherein the second classification result is no abnormal part, existence of abnormal part, and judgment of one of three unknown classes.
9. The method of machine learning based pathology image classification of claim 7, characterized in that training set images are processed by sample enhancement method, the sample enhancement method includes random noise access, contrast variation, protection variation, brightness variation.
10. A pathological image classification system characterized by: the method for machine learning based pathology image classification according to any one of claims 1-9, further comprising the following modules:
the first classification module is used for classifying the pathological images of each clinic record through a first machine learning model and obtaining a first classification result;
the diagnosis conclusion judging module is connected with the first classification module and used for judging whether a pathological image in each diagnosis record in the online diagnosis medical platform has a diagnosis conclusion or not, if yes, the pathological image, the diagnosis conclusion and the first classification result are transmitted to the first judging module, and if not, the pathological image and the first classification result are transmitted to the second classification module;
the first judgment module is connected with the diagnosis conclusion judgment module and used for judging whether the first classification result is the same as the diagnosis conclusion in the visit record or not, if so, the pathological image and the diagnosis conclusion are transmitted to the knowledge set building module, and if not, the diagnosis record is abandoned;
the second classification module is connected with the diagnosis conclusion judgment module and used for classifying the pathological images without diagnosis conclusions through a second machine learning model to obtain classification results;
the second judgment module is connected with the second classification module and is used for judging whether the first classification result is the same as the second classification result or not, if so, the pathological image and the diagnosis conclusion are transmitted to the knowledge set establishment module, and if not, the diagnosis record is abandoned;
and the knowledge set establishing module is connected with the first classification module and the second classification module and is used for receiving the pathological images with the same conclusion in the first judgment module and the second judgment module.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211629416.0A CN115631386B (en) | 2022-12-19 | 2022-12-19 | Pathological image classification method and system based on machine learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211629416.0A CN115631386B (en) | 2022-12-19 | 2022-12-19 | Pathological image classification method and system based on machine learning |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN115631386A true CN115631386A (en) | 2023-01-20 |
| CN115631386B CN115631386B (en) | 2023-03-10 |
Family
ID=84909701
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211629416.0A Active CN115631386B (en) | 2022-12-19 | 2022-12-19 | Pathological image classification method and system based on machine learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115631386B (en) |
Citations (14)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103690240A (en) * | 2013-09-16 | 2014-04-02 | 上海华美络信息技术有限公司 | Medical system |
| CN105975793A (en) * | 2016-05-23 | 2016-09-28 | 麦克奥迪（厦门）医疗诊断系统有限公司 | Auxiliary cancer diagnosis method based on digital pathological images |
| CN106682446A (en) * | 2017-01-24 | 2017-05-17 | 宁波江丰生物信息技术有限公司 | Pathological diagnosis method |
| CN107909095A (en) * | 2017-11-07 | 2018-04-13 | 江苏大学 | A kind of image-recognizing method based on deep learning |
| CN110335256A (en) * | 2019-06-18 | 2019-10-15 | 广州智睿医疗科技有限公司 | A kind of pathology aided diagnosis method |
| CN110459300A (en) * | 2019-07-22 | 2019-11-15 | 中国石油大学(华东) | A kind of lung cancer pathology type diagnostic method based on computer vision and CT images |
| CN110929807A (en) * | 2019-12-06 | 2020-03-27 | 腾讯科技（深圳）有限公司 | Training method of image classification model, and image classification method and device |
| CN111666993A (en) * | 2020-05-28 | 2020-09-15 | 平安科技（深圳）有限公司 | Medical image sample screening method and device, computer equipment and storage medium |
| KR102162683B1 (en) * | 2020-01-31 | 2020-10-07 | 주식회사 에프앤디파트너스 | Reading aid using atypical skin disease image data |
| CN111863237A (en) * | 2020-05-29 | 2020-10-30 | 东莞理工学院 | Intelligent auxiliary diagnosis system for mobile terminal diseases based on deep learning |
| CN112365981A (en) * | 2020-11-26 | 2021-02-12 | 中国联合网络通信集团有限公司 | Intelligent medical information processing method and device |
| CN113793305A (en) * | 2021-08-23 | 2021-12-14 | 上海派影医疗科技有限公司 | Pathological image classification and identification method and system integrating multiple information |
| CN113935969A (en) * | 2021-10-18 | 2022-01-14 | 太原理工大学 | Diagnosis system for new coronary pneumonia specific cases based on domain knowledge guidance |
| CN115239993A (en) * | 2022-07-06 | 2022-10-25 | 杭州丝跃科技有限公司 | Human body alopecia type and stage identification system based on cross-domain semi-supervised learning |
-
2022
- 2022-12-19 CN CN202211629416.0A patent/CN115631386B/en active Active
Patent Citations (14)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103690240A (en) * | 2013-09-16 | 2014-04-02 | 上海华美络信息技术有限公司 | Medical system |
| CN105975793A (en) * | 2016-05-23 | 2016-09-28 | 麦克奥迪（厦门）医疗诊断系统有限公司 | Auxiliary cancer diagnosis method based on digital pathological images |
| CN106682446A (en) * | 2017-01-24 | 2017-05-17 | 宁波江丰生物信息技术有限公司 | Pathological diagnosis method |
| CN107909095A (en) * | 2017-11-07 | 2018-04-13 | 江苏大学 | A kind of image-recognizing method based on deep learning |
| CN110335256A (en) * | 2019-06-18 | 2019-10-15 | 广州智睿医疗科技有限公司 | A kind of pathology aided diagnosis method |
| CN110459300A (en) * | 2019-07-22 | 2019-11-15 | 中国石油大学(华东) | A kind of lung cancer pathology type diagnostic method based on computer vision and CT images |
| CN110929807A (en) * | 2019-12-06 | 2020-03-27 | 腾讯科技（深圳）有限公司 | Training method of image classification model, and image classification method and device |
| KR102162683B1 (en) * | 2020-01-31 | 2020-10-07 | 주식회사 에프앤디파트너스 | Reading aid using atypical skin disease image data |
| CN111666993A (en) * | 2020-05-28 | 2020-09-15 | 平安科技（深圳）有限公司 | Medical image sample screening method and device, computer equipment and storage medium |
| CN111863237A (en) * | 2020-05-29 | 2020-10-30 | 东莞理工学院 | Intelligent auxiliary diagnosis system for mobile terminal diseases based on deep learning |
| CN112365981A (en) * | 2020-11-26 | 2021-02-12 | 中国联合网络通信集团有限公司 | Intelligent medical information processing method and device |
| CN113793305A (en) * | 2021-08-23 | 2021-12-14 | 上海派影医疗科技有限公司 | Pathological image classification and identification method and system integrating multiple information |
| CN113935969A (en) * | 2021-10-18 | 2022-01-14 | 太原理工大学 | Diagnosis system for new coronary pneumonia specific cases based on domain knowledge guidance |
| CN115239993A (en) * | 2022-07-06 | 2022-10-25 | 杭州丝跃科技有限公司 | Human body alopecia type and stage identification system based on cross-domain semi-supervised learning |
Also Published As
| Publication number | Publication date |
|---|---|
| CN115631386B (en) | 2023-03-10 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112017198B (en) | Right ventricle segmentation method and device based on self-attention mechanism multi-scale features | |
| WO2016192612A1 (en) | Method for analysing medical treatment data based on deep learning, and intelligent analyser thereof | |
| CN109559300A (en) | Image processing method, electronic equipment and computer readable storage medium | |
| CN110660478A (en) | Cancer image prediction and discrimination method and system based on transfer learning | |
| Moreira et al. | Evolutionary radial basis function network for gestational diabetes data analytics | |
| CN113808738B (en) | Disease identification system based on self-identification image | |
| US20220121902A1 (en) | Method and apparatus for quality prediction | |
| CN114372564A (en) | Model training method for object classification, object classification method and device | |
| EP3769270A1 (en) | A method, an apparatus and a computer program product for an interpretable neural network representation | |
| CN114549469A (en) | Deep neural network medical image diagnosis method based on confidence degree calibration | |
| CN114743037A (en) | Deep medical image clustering method based on multi-scale structure learning | |
| CN115579141A (en) | Interpretable disease risk prediction model construction method and disease risk prediction device | |
| Huang et al. | TMED 2: a dataset for semi-supervised classification of echocardiograms | |
| Elayaraja et al. | An efficient approach for detection and classification of cancer regions in cervical images using optimization based CNN classification approach | |
| Guangyu et al. | Tongue color classification based on convolutional neural network | |
| CN117457192A (en) | Intelligent remote diagnosis method and system | |
| CN115631386B (en) | Pathological image classification method and system based on machine learning | |
| CN114743619B (en) | Questionnaire quality evaluation method and system for disease risk prediction | |
| CN111882551B (en) | Pathological image cell counting method, system and device | |
| US11580390B2 (en) | Data processing apparatus and method | |
| CN114649092A (en) | Auxiliary diagnosis method and device based on semi-supervised learning and multi-scale feature fusion | |
| Brown et al. | Pitfalls and opportunities in the development and evaluation of artificial intelligence systems | |
| CN112509052A (en) | Method and device for detecting fovea maculata, computer equipment and storage medium | |
| CN112086174A (en) | Three-dimensional knowledge diagnosis model construction method and system | |
| Cao et al. | Alzheimer’s Disease Stage Detection Method Based on Convolutional Neural Network |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |