CN110859642A - Method, device, equipment and storage medium for realizing medical image auxiliary diagnosis based on AlexNet network model - Google Patents
Method, device, equipment and storage medium for realizing medical image auxiliary diagnosis based on AlexNet network model Download PDFInfo
- Publication number
- CN110859642A CN110859642A CN201911176103.2A CN201911176103A CN110859642A CN 110859642 A CN110859642 A CN 110859642A CN 201911176103 A CN201911176103 A CN 201911176103A CN 110859642 A CN110859642 A CN 110859642A
- Authority
- CN
- China
- Prior art keywords
- image
- layer
- medical image
- activation function
- sample
- 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
- 238000003745 diagnosis Methods 0.000 title claims abstract description 83
- 238000000034 method Methods 0.000 title claims abstract description 44
- 238000003860 storage Methods 0.000 title claims abstract description 11
- 230000006870 function Effects 0.000 claims description 103
- 230000004913 activation Effects 0.000 claims description 102
- 238000012549 training Methods 0.000 claims description 73
- 238000007781 pre-processing Methods 0.000 claims description 46
- 238000010606 normalization Methods 0.000 claims description 42
- 210000002569 neuron Anatomy 0.000 claims description 36
- 238000011176 pooling Methods 0.000 claims description 36
- 238000004891 communication Methods 0.000 claims description 24
- 238000012545 processing Methods 0.000 claims description 15
- 238000004364 calculation method Methods 0.000 claims description 12
- 230000000415 inactivating effect Effects 0.000 claims description 12
- 238000004590 computer program Methods 0.000 claims description 11
- 238000010586 diagram Methods 0.000 claims description 11
- 238000011478 gradient descent method Methods 0.000 claims description 8
- 238000005520 cutting process Methods 0.000 claims description 6
- 230000011218 segmentation Effects 0.000 claims description 6
- 230000007480 spreading Effects 0.000 claims description 6
- 238000003892 spreading Methods 0.000 claims description 6
- 238000005070 sampling Methods 0.000 claims description 3
- 230000001902 propagating effect Effects 0.000 claims description 2
- 230000008569 process Effects 0.000 abstract description 7
- 238000013527 convolutional neural network Methods 0.000 abstract description 6
- 238000013135 deep learning Methods 0.000 abstract description 4
- 238000005516 engineering process Methods 0.000 abstract description 4
- 238000013473 artificial intelligence Methods 0.000 abstract description 3
- 230000002349 favourable effect Effects 0.000 abstract description 2
- 238000005096 rolling process Methods 0.000 description 5
- 230000000694 effects Effects 0.000 description 4
- 238000002583 angiography Methods 0.000 description 3
- 238000013528 artificial neural network Methods 0.000 description 2
- 238000002591 computed tomography Methods 0.000 description 2
- 238000002059 diagnostic imaging Methods 0.000 description 2
- 201000010099 disease Diseases 0.000 description 2
- 208000037265 diseases, disorders, signs and symptoms Diseases 0.000 description 2
- 238000010801 machine learning Methods 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000013421 nuclear magnetic resonance imaging Methods 0.000 description 2
- 238000002604 ultrasonography Methods 0.000 description 2
- 238000012795 verification Methods 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000004422 calculation algorithm Methods 0.000 description 1
- 230000000747 cardiac effect Effects 0.000 description 1
- 210000000245 forearm Anatomy 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 238000009607 mammography Methods 0.000 description 1
- 238000002600 positron emission tomography Methods 0.000 description 1
- 238000012805 post-processing Methods 0.000 description 1
- 238000003672 processing method Methods 0.000 description 1
- 230000003252 repetitive effect Effects 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
- 238000013526 transfer learning Methods 0.000 description 1
- 210000001364 upper extremity Anatomy 0.000 description 1
Images
Classifications
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B6/00—Apparatus for radiation diagnosis, e.g. combined with radiation therapy equipment
- A61B6/52—Devices using data or image processing specially adapted for radiation diagnosis
- A61B6/5211—Devices using data or image processing specially adapted for radiation diagnosis involving processing of medical diagnostic data
- A61B6/5217—Devices using data or image processing specially adapted for radiation diagnosis involving processing of medical diagnostic data extracting a diagnostic or physiological parameter from medical diagnostic data
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/05—Detecting, measuring or recording for diagnosis by means of electric currents or magnetic fields; Measuring using microwaves or radio waves
- A61B5/055—Detecting, measuring or recording for diagnosis by means of electric currents or magnetic fields; Measuring using microwaves or radio waves involving electronic [EMR] or nuclear [NMR] magnetic resonance, e.g. magnetic resonance imaging
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/72—Signal processing specially adapted for physiological signals or for diagnostic purposes
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B8/00—Diagnosis using ultrasonic, sonic or infrasonic waves
- A61B8/52—Devices using data or image processing specially adapted for diagnosis using ultrasonic, sonic or infrasonic waves
- A61B8/5215—Devices using data or image processing specially adapted for diagnosis using ultrasonic, sonic or infrasonic waves involving processing of medical diagnostic data
- A61B8/5223—Devices using data or image processing specially adapted for diagnosis using ultrasonic, sonic or infrasonic waves involving processing of medical diagnostic data for extracting a diagnostic or physiological parameter from medical diagnostic data
-
- 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
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Engineering & Computer Science (AREA)
- Medical Informatics (AREA)
- Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- Surgery (AREA)
- Veterinary Medicine (AREA)
- Public Health (AREA)
- Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
- Animal Behavior & Ethology (AREA)
- Pathology (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Heart & Thoracic Surgery (AREA)
- Molecular Biology (AREA)
- Radiology & Medical Imaging (AREA)
- Physiology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- High Energy & Nuclear Physics (AREA)
- Optics & Photonics (AREA)
- Artificial Intelligence (AREA)
- Psychiatry (AREA)
- Signal Processing (AREA)
- Measuring And Recording Apparatus For Diagnosis (AREA)
- Medical Treatment And Welfare Office Work (AREA)
Abstract
The invention relates to the technical field of medical equipment, and discloses a method, a device, equipment and a storage medium for realizing medical image auxiliary diagnosis based on an AlexNet network model. The invention provides a new method for realizing medical image auxiliary diagnosis by using the latest deep learning technology in artificial intelligence, namely, in an AlexNet network model with good generalization capability, a convolutional neural network is used for extracting various complex characteristics, and then a full-connection network is used for comprehensively judging the characteristics, so that the model can cover various complex conditions, the method is favorable for realizing end-to-end automatic diagnosis in the whole process of medical image auxiliary diagnosis, and the efficiency and the accuracy are extremely high. In addition, newly added parts or images can be migrated and learned, so that model upgrading becomes simple and easy to operate, the maximum compatibility of original knowledge can be ensured, and the method is convenient for practical application and popularization.
Description
Technical Field
The invention belongs to the technical field of medical equipment, and particularly relates to a method, a device, equipment and a storage medium for realizing medical image auxiliary diagnosis based on an AlexNet network model.
Background
Medical imaging devices on the market of Medical instruments are available to view Medical images, such as Angiography (Angiography), cardioangiography (Cardiac Angiography), Computed Tomography (CT), Mammography (mammogram), Positron Emission Tomography (PET), Nuclear Magnetic Resonance Imaging (NMRI), and Medical ultrasonography (Medical ultrasonography). However, the current image diagnosis industry has several problems as follows: (1) the culture period of the diagnosis doctor is long, the cost is high, and the gap of the current high-end talents is large; (2) the subjective difference of doctors is very large, and the diagnosis standard is difficult to be completely unified; (3) the doctor only observes by naked eyes, a lot of bottom layer information is invisible, and the information utilization rate is low; (4) too many similar diseases exist, and a great deal of repetitive labor exists in the process of reading the film; (5) the time spent for doctor to read the film is long, at least from tens of minutes to hours.
In view of the above problems, there are two current solutions: firstly, by utilizing a traditional image processing method, aiming at a certain specific disease and a specific type of image film, characteristics are manually extracted, and then judgment is made according to a threshold value, the method needs a large amount of medical and image processing experience and has no universality, and after the part or the image type is changed, the method is difficult to have higher accuracy rate and even cannot work at all; and secondly, the machine learning model or the statistical learning model is utilized, automation is realized to a certain extent, certain universality is achieved, and compared with the first method, the accuracy is greatly improved, but the model still has the problem of insufficient expression capability, cannot cover various complex conditions, and is difficult to perform transfer learning.
Disclosure of Invention
The invention aims to solve the problem that the conventional medical equipment cannot cover various complex conditions when the machine learning model is used for image diagnosis, and provides a method, a device, equipment and a storage medium for realizing medical image auxiliary diagnosis based on an AlexNet network model.
The technical scheme adopted by the invention is as follows:
a method for realizing medical image auxiliary diagnosis based on an AlexNet network model comprises the following steps:
s101, obtaining a plurality of sample medical images and diagnosis labels corresponding to the sample medical images, wherein the diagnosis labels are positive examples or negative examples;
s102, carrying out image preprocessing on each sample medical image to obtain a corresponding standard sample medical image which is square, consistent in size and normalized, and carrying out digital coding on a corresponding diagnosis label in the following form; if positive, it is marked as 1, and if negative, it is marked as 0;
s103, extracting 2 from the training sample data setnAnd then, introducing the standard sample medical images into an AlexNet network model for forward propagation to obtain a first prediction probability corresponding to each standard sample medical image, wherein the training sample data set comprises not less than 2nN is a natural number between 4 and 8, and the first prediction probability comprises a probability of identifying the image as a positive example and a probability of identifying the image as a negative example;
the AlexNet network model sequentially comprises a first convolution layer, a first batch of normalization layers, a first maximum pooling layer, a second convolution layer, a second batch of normalization layers, a second maximum pooling layer, a third convolution layer, a fourth convolution layer, a fifth convolution layer, the flattening layer, a first full-connection layer, a first discarding layer, a second full-connection layer, a second discarding layer and a third full-connection layer along the forward propagation direction, wherein the convolution kernel of the first convolution layer is (11, 11), the step length is (4, 4), the number of output channels is 96, the activation function is a relu activation function, and the padding type is valid; the first normalization layer is used for accelerating network convergence, preventing overfitting and improving generalization capability; the pooling core of the first maximum pooling layer is (3, 3), the step length is (2, 2), and the padding type is valid; the convolution kernel of the second convolution layer is (5, 5), the step length is (1, 1), the number of output channels is 256, the activation function is a relu activation function, and the padding type is same; the second batch of normalization layers are also used for accelerating network convergence, preventing overfitting and improving generalization capability; the pooling core of the second maximum pooling layer is (3, 3), the step length is (2, 2), and the padding type is valid; the convolution kernel of the third convolution layer is (3, 3), the step length is (1, 1), the number of output channels is 384, the activation function is a relu activation function, and the padding type is same; the convolution kernel of the fourth convolution layer is (3, 3), the step length is (1, 1), the number of output channels is 256, the activation function is a relu activation function, and the padding type is same; the convolution kernel of the fifth convolution layer is (3, 3), the step length is (1, 1), the number of output channels is 384, the activation function is a relu activation function, and the padding type is same; the flattening layer is used for spreading the characteristic diagram into a line, so that a first full-connection layer is connected to the rear side of the flattening layer conveniently; the number of the neurons of the first full-junction layer is 1024, and the activation function is a relu activation function; the first discarding layer is used for randomly inactivating half of the neurons, so that overfitting is avoided, and generalization capability is improved; the number of neurons of the second fully-connected layer is half of the number of neurons of the first fully-connected layer, and the activation function is a relu activation function; the second discarding layer is used for randomly inactivating half of the neurons, so that overfitting is avoided, and generalization capability is improved; the number of neurons of the third full junction layer is 2, and the activation function is a softmax activation function, wherein the softmax activation function is used for identifying the probability that the image is a positive example and the probability that the image is a negative example;
s104, calculating the average loss value loss of the training according to the following formula:
wherein i is 1 to 2nNatural number between, yiThe diagnostic label code value of the ith standard sample medical image,
probability of identifying the image as a positive case of the ith standard sample medical image, wherein k is a constant;
s105, reversely propagating the average loss value loss layer by layer through a gradient descent method, and updating model parameters;
s106, returning to execute the steps S103-S105 until the average loss value loss is reduced to a preset threshold value, and finishing model training;
s107, acquiring a medical image to be diagnosed;
s108, image preprocessing is carried out on the medical image to be diagnosed, and a standard medical image to be diagnosed which is square and has the size consistent with that of the standard sample medical image and the normalization mode consistent with that of the standard sample medical image is obtained;
s109, inputting the standard medical image to be diagnosed into an AlexNet network model which is trained to obtain a second prediction probability corresponding to the standard medical image to be diagnosed, wherein the second prediction probability comprises a probability of identifying an image as a positive example and a probability of identifying an image as a negative example;
s110, obtaining a diagnosis result according to the second prediction probability: and if the probability of identifying the positive example of the image in the second prediction probability is not less than the segmentation threshold of the positive example and the negative example, judging that the diagnosis result is positive example and/or positive, otherwise, judging that the diagnosis result is negative example and/or negative.
Optimally, in the step S102, the sample medical image in the square shape is obtained through image preprocessing in the following manner:
when the original shape of the sample medical image is a non-square rectangle, pixel points with pixel values of 0 or pixel mean values are symmetrically filled at two sides in the width direction, so that the final image width is equal to the image length, wherein the pixel mean value is the pixel mean value of all the pixel points in the original medical image.
Further optimally, after the square sample medical image is obtained in step S102, the sample medical images with consistent sizes are obtained through image preprocessing in the following manner:
the sample medical image is changed into the size of 2 by interpolation or pressure samplingk*2kWherein k is a natural number between 7 and 12.
In detail, after the sample medical image with the consistent size is obtained in step S102, the normalized sample medical image is obtained through image preprocessing in the following manner:
aiming at each pixel point on the sample medical image, obtaining a normalized pixel value P according to the following formulaNew：
In the formula, POldThe value is the pixel value before normalization, mu is the pixel average value of all pixel points before normalization, and sigma is the pixel value standard deviation of all pixel points before normalization.
Preferably, before the step S103, the training sample data set is subjected to data enhancement processing in any one or any combination of the following manners (a) to (D):
(A) randomly turning the standard sample medical image up and down and/or left and right, and then adding the obtained image as a new sample into the training sample data set;
(B) carrying out random angular rotation processing on the standard sample medical image, and then adding the obtained image serving as a new sample into the training sample data set;
(C) randomly adding Gaussian noise to the medical image of the standard sample, and then adding the obtained image serving as a new sample to the training sample data set;
(D) and randomly cutting the image of the standard sample medical image, and then adding the obtained image as a new sample into the training sample data set, wherein the image cutting part accounts for no more than 5% of the whole image.
The other technical scheme adopted by the invention is as follows:
a device for realizing medical image auxiliary diagnosis based on an AlexNet network model comprises a sample acquisition module, a sample preprocessing module, a model iterative training module, an image acquisition module, an image preprocessing module, an image recognition module and an image diagnosis module, wherein the model iterative training module comprises a model training submodule, a loss calculation submodule, a parameter updating submodule and an iterative control submodule;
the sample acquisition module is used for acquiring a plurality of sample medical images and diagnosis labels corresponding to the sample medical images, wherein the diagnosis labels are positive examples or negative examples;
the sample preprocessing module is in communication connection with the sample acquisition module and is used for preprocessing images of all sample medical images to obtain corresponding standard sample medical images which are square, consistent in size and normalized, and simultaneously digitally encoding corresponding diagnosis labels in the following forms; if positive, it is marked as 1, and if negative, it is marked as 0;
the model training submodule is in communication connection with the sample preprocessing module and is used for extracting 2 from a training sample data setnAnd then, introducing the standard sample medical images into an AlexNet network model for forward propagation to obtain a first prediction probability corresponding to each standard sample medical image, wherein the training sample data set comprises not less than 2nN is a natural number between 4 and 8, and the first prediction probability comprises a probability of identifying the image as a positive example and a probability of identifying the image as a negative example;
the AlexNet network model sequentially comprises a first convolution layer, a first batch of normalization layers, a first maximum pooling layer, a second convolution layer, a second batch of normalization layers, a second maximum pooling layer, a third convolution layer, a fourth convolution layer, a fifth convolution layer, the flattening layer, a first full-connection layer, a first discarding layer, a second full-connection layer, a second discarding layer and a third full-connection layer along the forward propagation direction, wherein the convolution kernel of the first convolution layer is (11, 11), the step length is (4, 4), the number of output channels is 96, the activation function is a relu activation function, and the padding type is valid; the first normalization layer is used for accelerating network convergence, preventing overfitting and improving generalization capability; the pooling core of the first maximum pooling layer is (3, 3), the step length is (2, 2), and the padding type is valid; the convolution kernel of the second convolution layer is (5, 5), the step length is (1, 1), the number of output channels is 256, the activation function is a relu activation function, and the padding type is same; the second batch of normalization layers are also used for accelerating network convergence, preventing overfitting and improving generalization capability; the pooling core of the second maximum pooling layer is (3, 3), the step length is (2, 2), and the padding type is valid; the convolution kernel of the third convolution layer is (3, 3), the step length is (1, 1), the number of output channels is 384, the activation function is a relu activation function, and the padding type is same; the convolution kernel of the fourth convolution layer is (3, 3), the step length is (1, 1), the number of output channels is 256, the activation function is a relu activation function, and the padding type is same; the convolution kernel of the fifth convolution layer is (3, 3), the step length is (1, 1), the number of output channels is 384, the activation function is a relu activation function, and the padding type is same; the flattening layer is used for spreading the characteristic diagram into a line, so that a first full-connection layer is connected to the rear side of the flattening layer conveniently; the number of the neurons of the first full-junction layer is 1024, and the activation function is a relu activation function; the first discarding layer is used for randomly inactivating half of the neurons, so that overfitting is avoided, and generalization capability is improved; the number of neurons of the second fully-connected layer is half of the number of neurons of the first fully-connected layer, and the activation function is a relu activation function; the second discarding layer is used for randomly inactivating half of the neurons, so that overfitting is avoided, and generalization capability is improved; the number of neurons of the third full junction layer is 2, and the activation function is a softmax activation function, wherein the softmax activation function is used for identifying the probability that the image is a positive example and the probability that the image is a negative example;
the loss calculation submodule is respectively in communication connection with the model training submodule and the sample preprocessing module, and is used for calculating the average loss value loss of the training according to the following formula:
wherein i is 1 to 2nNatural number between, yiIs the ith standard sampleThe diagnostic label code value of the medical image,
probability of identifying the image as a positive case of the ith standard sample medical image, wherein k is a constant;
the parameter updating submodule is in communication connection with the loss calculating submodule and is used for reversely transmitting the average loss value loss layer by layer through a gradient descent method to update the model parameters;
the iteration control submodule is in communication connection with the loss calculation submodule and is used for circularly and sequentially starting the model training submodule, the loss calculation submodule and the parameter updating submodule until the average loss value loss is reduced to a preset threshold value, and completing model training;
the image acquisition module is used for acquiring medical images to be diagnosed;
the image preprocessing module is in communication connection with the image acquisition module and is used for preprocessing the medical image to be diagnosed to obtain a standard medical image to be diagnosed, which is square, has the size consistent with that of the standard sample medical image and has the normalization mode consistent with that of the standard sample medical image;
the image identification module is respectively in communication connection with the model iteration training module and the image preprocessing module, and is used for inputting the standard medical image to be diagnosed into an AlexNet network model which is trained to obtain a second prediction probability corresponding to the standard medical image to be diagnosed, wherein the second prediction probability comprises the probability that the image is recognized as a positive example and the probability that the image is recognized as a negative example;
the image diagnosis module is in communication connection with the image identification module and is used for obtaining a diagnosis result according to the second prediction probability: and if the probability of identifying the positive example of the image in the second prediction probability is not less than the segmentation threshold of the positive example and the negative example, judging that the diagnosis result is positive example and/or positive, otherwise, judging that the diagnosis result is negative example and/or negative.
The other technical scheme adopted by the invention is as follows:
an apparatus for implementing medical image-assisted diagnosis based on AlexNet network model, comprising a memory and a processor, wherein the memory is used for storing a computer program, and the processor is used for executing the computer program to implement the method steps for implementing medical image-assisted diagnosis based on AlexNet network model as described above.
The other technical scheme adopted by the invention is as follows:
a storage medium having stored thereon a computer program which, when being executed by a processor, carries out the method steps of carrying out a medical image-assisted diagnosis based on an AlexNet network model as described above.
The invention has the beneficial effects that:
(1) the invention has created and provided a new method for utilizing the latest deep learning technology in the artificial intelligence to realize the auxiliary diagnosis of medical image, namely in AlexNet network model with very good generalization ability, extract various complicated characteristics with the neural network of convolution first, then utilize the full-connection network to carry on the comprehensive judgement to the characteristic, and then make the model cover various complicated situations, can help in the whole course of auxiliary diagnosis of medical image, realize the automatic diagnosis of end-to-end, efficiency and accuracy are all extremely high;
(2) and newly added parts or images can be migrated and learned, so that the model upgrading becomes simple and easy to operate, the maximum compatibility of the original knowledge can be ensured, and the method is convenient for practical application and popularization.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described below, it is obvious that the drawings in the following description are only some embodiments of the present invention, and for those skilled in the art, other drawings can be obtained according to the drawings without creative efforts.
Fig. 1 is a schematic flow chart of a method for implementing medical image-assisted diagnosis based on an AlexNet network model provided by the present invention.
FIG. 2 is an exemplary diagram of a pre-and post-processing image contrast for image fill and uniform size image pre-processing according to the present invention.
Fig. 3 is a schematic diagram of a multi-layer structure of an AlexNet network model provided by the present invention.
Fig. 4 is a schematic structural diagram of a device for implementing medical image auxiliary diagnosis based on an AlexNet network model provided by the invention.
Fig. 5 is a schematic structural diagram of the device for implementing medical image-assisted diagnosis based on the AlexNet network model provided by the invention.
Detailed Description
The invention is further described with reference to the following figures and specific embodiments. It should be noted that the description of the embodiments is provided to help understanding of the present invention, but the present invention is not limited thereto. Specific structural and functional details disclosed herein are merely illustrative of example embodiments of the invention. This invention may, however, be embodied in many alternate forms and should not be construed as limited to the embodiments set forth herein.
It will be understood that, although the terms first, second, etc. may be used herein to describe various elements, these elements should not be limited by these terms. These terms are only used to distinguish one element from another. For example, a first element could be termed a second element, and, similarly, a second element could be termed a first element, without departing from the scope of example embodiments of the present invention.
It should be understood that, for the term "and/or" as may appear herein, it is merely an associative relationship that describes an associated object, meaning that three relationships may exist, e.g., a and/or B may mean: a exists alone, B exists alone, and A and B exist at the same time; for the term "/and" as may appear herein, which describes another associative object relationship, it means that two relationships may exist, e.g., a/and B, may mean: a exists independently, and A and B exist independently; in addition, for the character "/" that may appear herein, it generally means that the former and latter associated objects are in an "or" relationship.
It will be understood that when an element is referred to herein as being "connected," "connected," or "coupled" to another element, it can be directly connected or coupled to the other element or intervening elements may be present. Conversely, if a unit is referred to herein as being "directly connected" or "directly coupled" to another unit, it is intended that no intervening units are present. In addition, other words used to describe the relationship between elements should be interpreted in a similar manner (e.g., "between … …" versus "directly between … …", "adjacent" versus "directly adjacent", etc.).
It is to be understood that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of example embodiments of the invention. As used herein, the singular forms "a", "an" and "the" are intended to include the plural forms as well, unless the context clearly indicates otherwise. It will be further understood that the terms "comprises," "comprising," "includes" and/or "including," when used herein, specify the presence of stated features, integers, steps, operations, elements, and/or components, but do not preclude the presence or addition of one or more other features, numbers, steps, operations, elements, components, and/or groups thereof.
It should also be noted that, in some alternative implementations, the functions/acts noted may occur out of the order noted in the figures. For example, two figures shown in succession may, in fact, be executed substantially concurrently, or the figures may sometimes be executed in the reverse order, depending upon the functionality/acts involved.
It should be understood that specific details are provided in the following description to facilitate a thorough understanding of example embodiments. However, it will be understood by those of ordinary skill in the art that the example embodiments may be practiced without these specific details. For example, systems may be shown in block diagrams in order not to obscure the examples in unnecessary detail. In other instances, well-known processes, structures and techniques may be shown without unnecessary detail in order to avoid obscuring example embodiments.
Example one
As shown in fig. 1 to 3, the method for implementing medical image aided diagnosis based on AlexNet network model provided in this embodiment may include, but is not limited to, the following steps S101 to S110.
S101, obtaining a plurality of sample medical images and diagnosis labels corresponding to the sample medical images, wherein the diagnosis labels are positive examples or negative examples.
In step S101, the sample medical image and the corresponding diagnosis label (positive case corresponds to positive case, negative case corresponds to negative case, which is common medical knowledge) can be obtained from a database of history files, or from a database of network open sources, such as MURA database (e.g., (MURA database) open source by stanford wurda team: (https:// stanfordmlgroup.github.io/competitions/mura/) It comprises an upper limb musculoskeletal X-ray picture of at least 14982 cases, each case comprising one or more images, each manually marked (i.e., forming a corresponding diagnostic label) by a radiologist, and the MURA database can be used for compensation in order to encourage improvement of the medical image diagnostic model.
S102, carrying out image preprocessing on each sample medical image to obtain a corresponding standard sample medical image which is square, consistent in size and normalized, and carrying out digital coding on a corresponding diagnosis label in the following form; if positive, it is marked as 1, and if negative, it is marked as 0.
In step S102, the square sample medical image may be obtained by, but not limited to, image preprocessing in the following manner: when the original shape of the sample medical image is a non-square rectangle, pixel points with pixel values of 0 or pixel mean values are symmetrically filled at two sides in the width direction, so that the final image width is equal to the image length, wherein the pixel mean value is the pixel mean value of all the pixel points in the original medical image. As shown in fig. 2, the sample medical images are single-channel images, and the length and the width of the single-channel images can be arbitrary, and if the length and the width are not equal, the length and the width need to be equal in the manner described above for subsequent model training. In addition, the symmetrical filling is carried out on two sides in the width direction, so that the original image can be ensured to be positioned in the middle area of the new image.
After the square sample medical image is obtained in step S102, the sample medical image with a uniform size may be obtained by, but not limited to, image preprocessing in the following manner: the sample medical image is changed into the size of 2 by interpolation or pressure samplingk*2kWherein k is a natural number between 7 and 12. For example, an image with a size of 256 × 256 pixels (k is 8 in this case), an image with a size of 512 × 512 pixels (k is 9 in this case, the usual size), or an image with a size of 1024 × 1024 pixels (k is 10 in this case) is obtained.
After obtaining the sample medical image with the consistent size in step S102, the normalized sample medical image may be obtained by, but is not limited to, image preprocessing in the following manner: aiming at each pixel point on the sample medical image, obtaining a normalized pixel value P according to the following formulaNew：
In the formula, POldThe value is the pixel value before normalization, mu is the pixel average value of all pixel points before normalization, and sigma is the pixel value standard deviation of all pixel points before normalization. By the normalization processing of the images, convergence can be accelerated during later training, the model training time and the recognition time are shortened, and the diagnostic result can be obtained quickly.
S103, extracting 2 from the training sample data setnAnd then, introducing the standard sample medical images into an AlexNet network model for forward propagation to obtain a first prediction probability corresponding to each standard sample medical image, wherein the training sample data set comprises not less than 2nN is a natural number between 4 and 8, and the first prediction probability comprises a probability of identifying the image as a positive example and a probability of identifying the image as a negative example.
In said step S103, 2nGenerally is16. 32, 64, 128, etc., depending on the hardware. If the standard sample medical image in the training sample data set is less than 2nIn order to perform step S103 and simulate a real scene more truly, it is necessary to perform data enhancement processing on the training sample data set before step S103 so as to enrich the number of images of the training sample data set. Optimally, for the training sample data set, the data enhancement processing may be performed in any one or any combination of the following manners (a) to (D): (A) randomly turning the standard sample medical image up and down and/or left and right, and then adding the obtained image as a new sample into the training sample data set; (B) carrying out random angular rotation processing on the standard sample medical image, and then adding the obtained image serving as a new sample into the training sample data set; (C) randomly adding Gaussian noise to the medical image of the standard sample, and then adding the obtained image serving as a new sample to the training sample data set; (D) and randomly cutting the image of the standard sample medical image, and then adding the obtained image as a new sample into the training sample data set, wherein the image cutting part accounts for no more than 5% of the whole image. The foregoing flipping processing, rotation processing, gaussian noise adding processing, image cropping processing, and the like are all conventional image processing manners, wherein, for example, the mean value of gaussian noise is 0, and the standard deviation is 0.01.
In step S103, the AlexNet network model is designed by the ImageNet race champion acquirer Hinton and the student AlexKrizhevsky in 2012, includes several new technical points, and also successfully applies the locks such as ReLU, Dropout, and LRN to the CNN for the first time, so as to raise the idea of the LeNet network model, apply the basic principle of CNN (Convolutional Neural network) to a very deep and wide network, and finally detonate the application hot tide of the Neural network, win the champion of the 2012' S image recognition champion, and make the CNN become the core algorithm model in image classification. Thus, an improved AlexNet network model suitable for sample medical image training and recognition can be constructed based on the layer structure principle of the AlexNet network model, that is, as shown in fig. 3, the improved AlexNet network model sequentially includes, in the forward propagation direction, a first rolling layer (Conv2D), a first batch of normalization layers (batch normalization), a first maximum pooling layer (MaxPooling2D), a second rolling layer (Conv2D), a second batch of normalization layers (batch normalization), a second maximum pooling layer (MaxPooling2D), a third rolling layer (Conv2D), a fourth rolling layer (Conv2D), a fifth rolling layer (Conv2D), the flattening layer (flatting), a first fully connected layer (density), a first discarded layer (Dropout), a second fully connected layer (density), a second discarded layer (Dropout), and a third fully connected layer (density); the convolution kernel of the first convolution layer is (11, 11), the step length is (4, 4), the number of output channels is 96, the activation function is a relu activation function, and the padding type is valid; the first normalization layer is used for accelerating network convergence, preventing overfitting and improving generalization capability; the pooling core of the first maximum pooling layer is (3, 3), the step length is (2, 2), and the padding type is valid; the convolution kernel of the second convolution layer is (5, 5), the step length is (1, 1), the number of output channels is 256, the activation function is a relu activation function, and the padding type is same; the second batch of normalization layers are also used for accelerating network convergence, preventing overfitting and improving generalization capability; the pooling core of the second maximum pooling layer is (3, 3), the step length is (2, 2), and the padding type is valid; the convolution kernel of the third convolution layer is (3, 3), the step length is (1, 1), the number of output channels is 384, the activation function is a relu activation function, and the padding type is same; the convolution kernel of the fourth convolution layer is (3, 3), the step length is (1, 1), the number of output channels is 256, the activation function is a relu activation function, and the padding type is same; the convolution kernel of the fifth convolution layer is (3, 3), the step length is (1, 1), the number of output channels is 384, the activation function is a relu activation function, and the padding type is same; the flattening layer is used for spreading the characteristic diagram into a line, so that a first full-connection layer is connected to the rear side of the flattening layer conveniently; the number of the neurons of the first full-junction layer is 1024, and the activation function is a relu activation function; the first discarding layer is used for randomly inactivating half of the neurons, so that overfitting is avoided, and generalization capability is improved; the number of neurons of the second fully-connected layer is half of the number of neurons of the first fully-connected layer, and the activation function is a relu activation function; the second discarding layer is used for randomly inactivating half of the neurons, so that overfitting is avoided, and generalization capability is improved; the number of neurons in the third full junction layer is 2, and the activation function is a softmax activation function, wherein the softmax activation function is used for identifying the probability that the image is a positive example and the probability that the image is a negative example. All the technical terms are common terms in the prior deep learning technology, and are not described in detail herein.
S104, calculating the average loss value loss of the training according to the following formula:
wherein i is 1 to 2nNatural number between, yiThe diagnostic label code value of the ith standard sample medical image,
is the probability of the ith standard sample medical image and identifying the image as a positive case, and κ is a constant.
In step S104, the diagnostic tag code value represents the true probability of identifying the image as a positive example, if the diagnostic tag code value is marked as 1, the true probability of identifying the image as a positive example is 100%, and if the diagnostic tag code value is marked as 0, the true probability of identifying the image as a positive example is 0%, so that the core of the above formula is to make the prediction probability and the true probability cross entropy. Further, κ may be exemplified by a constant of 2, e or 10.
And S105, reversely transmitting the average loss value loss layer by layer through a gradient descent method, and updating model parameters.
In step S105, the mode of updating the model parameters by the gradient descent method is the conventional mode.
S106, returning to execute the steps S103-S105 until the average loss value loss is reduced to a preset threshold value, and finishing model training.
In said step S106 andwhen step S103 is executed again, 2 may be extracted from the training sample data setnAnd (3) drawing a standard sample medical image which is the same as, partially the same as or completely different from the previous time. In addition, when the training precision meets the requirement, the return execution can be stopped, and the model training can be completed.
S107, medical images to be diagnosed are obtained.
In step S107, the medical image to be diagnosed may be derived from an output interface (e.g., a USB interface or an HDMI interface) of an existing medical imaging device.
And S108, carrying out image preprocessing on the medical image to be diagnosed to obtain a standard medical image to be diagnosed, which is square, has the size consistent with that of the standard sample medical image and has the normalization mode consistent with that of the standard sample medical image.
In the step S108, the image preprocessing method may be completely consistent with the image preprocessing method in the step S102.
S109, inputting the standard medical image to be diagnosed into the AlexNet network model which is trained, and obtaining a second prediction probability corresponding to the standard medical image to be diagnosed, wherein the second prediction probability comprises the probability of identifying the image as a positive example and the probability of identifying the image as a negative example.
S110, obtaining a diagnosis result according to the second prediction probability: and if the probability of identifying the positive example of the image in the second prediction probability is not less than the segmentation threshold of the positive example and the negative example, judging that the diagnosis result is positive example and/or positive, otherwise, judging that the diagnosis result is negative example and/or negative.
In the step S110, the dividing threshold of the positive example and the negative example is generally 0.5.
Following the aforementioned steps S101 to S110, the present embodiment also performs a test using a FOREARM (forward) medical image in the MURA database by way of example: firstly, 1825 medical images (single-channel images) are subjected to the sample processing in the step S102, so as to obtain standard sample medical images with the size of 512 × 512 pixels, and form the training sample data set; then, the image processing in step S108 is performed on 301 medical images (single-channel images) to obtain a standard medical image to be diagnosed with a size of 512 × 512 pixels, and a verification set is formed; then configuring parameters of the AlexNet network model as follows: optimizer-sgd, stochastic gradient descent method, loss function-binary _ cross, learning rate-0.01; then, the model training is completed by iterating for 20 times; and finally, the verification set is verified one by applying the trained AlexNet network model, and the diagnosis errors are found to occur only a few times, which greatly exceeds the diagnosis accuracy of human experts. In addition, even after each data enhancement, the training sample data set may vary and the final result may vary slightly, but all exceed the diagnostic accuracy of human experts.
In summary, the method for implementing medical image auxiliary diagnosis based on the AlexNet network model provided by the embodiment has the following technical effects:
(1) the embodiment provides a new method for realizing medical image auxiliary diagnosis by using the latest deep learning technology in artificial intelligence, namely, in an AlexNet network model with good generalization capability, a convolutional neural network is used for extracting various complex characteristics, and then a full-connection network is used for comprehensively judging the characteristics, so that the model can cover various complex conditions, the method is favorable for realizing end-to-end automatic diagnosis in the whole process of medical image auxiliary diagnosis, and the efficiency and the accuracy are extremely high;
(2) and newly added parts or images can be migrated and learned, so that the model upgrading becomes simple and easy to operate, the maximum compatibility of the original knowledge can be ensured, and the method is convenient for practical application and popularization.
Example two
As shown in fig. 4, the present embodiment provides a hardware device for implementing the method for implementing medical image auxiliary diagnosis based on the AlexNet network model according to the first embodiment, including a sample acquisition module, a sample preprocessing module, a model iteration training module, an image acquisition module, an image preprocessing module, an image recognition module, and an image diagnosis module, where the model iteration training module includes a model training submodule, a loss calculation submodule, a parameter updating submodule, and an iteration control submodule;
the sample acquisition module is used for acquiring a plurality of sample medical images and diagnosis labels corresponding to the sample medical images, wherein the diagnosis labels are positive examples or negative examples;
the sample preprocessing module is in communication connection with the sample acquisition module and is used for preprocessing images of all sample medical images to obtain corresponding standard sample medical images which are square, consistent in size and normalized, and simultaneously digitally encoding corresponding diagnosis labels in the following forms; if positive, it is marked as 1, and if negative, it is marked as 0;
the model training submodule is in communication connection with the sample preprocessing module and is used for extracting 2 from a training sample data setnAnd then, introducing the standard sample medical images into an AlexNet network model for forward propagation to obtain a first prediction probability corresponding to each standard sample medical image, wherein the training sample data set comprises not less than 2nN is a natural number between 4 and 8, and the first prediction probability comprises a probability of identifying the image as a positive example and a probability of identifying the image as a negative example;
the AlexNet network model sequentially comprises a first convolution layer, a first batch of normalization layers, a first maximum pooling layer, a second convolution layer, a second batch of normalization layers, a second maximum pooling layer, a third convolution layer, a fourth convolution layer, a fifth convolution layer, the flattening layer, a first full-connection layer, a first discarding layer, a second full-connection layer, a second discarding layer and a third full-connection layer along the forward propagation direction, wherein the convolution kernel of the first convolution layer is (11, 11), the step length is (4, 4), the number of output channels is 96, the activation function is a relu activation function, and the padding type is valid; the first normalization layer is used for accelerating network convergence, preventing overfitting and improving generalization capability; the pooling core of the first maximum pooling layer is (3, 3), the step length is (2, 2), and the padding type is valid; the convolution kernel of the second convolution layer is (5, 5), the step length is (1, 1), the number of output channels is 256, the activation function is a relu activation function, and the padding type is same; the second batch of normalization layers are also used for accelerating network convergence, preventing overfitting and improving generalization capability; the pooling core of the second maximum pooling layer is (3, 3), the step length is (2, 2), and the padding type is valid; the convolution kernel of the third convolution layer is (3, 3), the step length is (1, 1), the number of output channels is 384, the activation function is a relu activation function, and the padding type is same; the convolution kernel of the fourth convolution layer is (3, 3), the step length is (1, 1), the number of output channels is 256, the activation function is a relu activation function, and the padding type is same; the convolution kernel of the fifth convolution layer is (3, 3), the step length is (1, 1), the number of output channels is 384, the activation function is a relu activation function, and the padding type is same; the flattening layer is used for spreading the characteristic diagram into a line, so that a first full-connection layer is connected to the rear side of the flattening layer conveniently; the number of the neurons of the first full-junction layer is 1024, and the activation function is a relu activation function; the first discarding layer is used for randomly inactivating half of the neurons, so that overfitting is avoided, and generalization capability is improved; the number of neurons of the second fully-connected layer is half of the number of neurons of the first fully-connected layer, and the activation function is a relu activation function; the second discarding layer is used for randomly inactivating half of the neurons, so that overfitting is avoided, and generalization capability is improved; the number of neurons of the third full junction layer is 2, and the activation function is a softmax activation function, wherein the softmax activation function is used for identifying the probability that the image is a positive example and the probability that the image is a negative example;
the loss calculation submodule is respectively in communication connection with the model training submodule and the sample preprocessing module, and is used for calculating the average loss value loss of the training according to the following formula:
wherein i is 1 to 2nNatural number between, yiThe diagnostic label code value of the ith standard sample medical image,
is as followsProbability of i standard sample medical images and identifying the images as positive examples, and k is a constant;
the parameter updating submodule is in communication connection with the loss calculating submodule and is used for reversely transmitting the average loss value loss layer by layer through a gradient descent method to update the model parameters;
the iteration control submodule is in communication connection with the loss calculation submodule and is used for circularly and sequentially starting the model training submodule, the loss calculation submodule and the parameter updating submodule until the average loss value loss is reduced to a preset threshold value, and completing model training;
the image acquisition module is used for acquiring medical images to be diagnosed;
the image preprocessing module is in communication connection with the image acquisition module and is used for preprocessing the medical image to be diagnosed to obtain a standard medical image to be diagnosed, which is square, has the size consistent with that of the standard sample medical image and has the normalization mode consistent with that of the standard sample medical image;
the image identification module is respectively in communication connection with the model iteration training module and the image preprocessing module, and is used for inputting the standard medical image to be diagnosed into an AlexNet network model which is trained to obtain a second prediction probability corresponding to the standard medical image to be diagnosed, wherein the second prediction probability comprises the probability that the image is recognized as a positive example and the probability that the image is recognized as a negative example;
the image diagnosis module is in communication connection with the image identification module and is used for obtaining a diagnosis result according to the second prediction probability: and if the probability of identifying the positive example of the image in the second prediction probability is not less than the segmentation threshold of the positive example and the negative example, judging that the diagnosis result is positive example and/or positive, otherwise, judging that the diagnosis result is negative example and/or negative.
The working process, working details and technical effects of the foregoing apparatus provided in this embodiment may be referred to in the first embodiment, and are not described herein again.
EXAMPLE III
As shown in fig. 5, the present embodiment provides a hardware device for implementing the method for implementing medical image-assisted diagnosis based on AlexNet network model according to the first embodiment, which includes a memory and a processor, which are communicatively connected, wherein the memory is used for storing a computer program, and the processor is used for executing the computer program to implement the method steps for implementing medical image-assisted diagnosis based on AlexNet network model according to the first embodiment.
The working process, the working details and the technical effects of the foregoing device provided in this embodiment may be referred to as embodiment one, and are not described herein again.
Example four
The present embodiment provides a storage medium storing a computer program comprising the method for implementing medical image-assisted diagnosis based on AlexNet network model according to the first embodiment, that is, a computer program is stored on the storage medium, and when being executed by a processor, the computer program implements the method steps for implementing medical image-assisted diagnosis based on AlexNet network model according to the first embodiment. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices, or may be a mobile intelligent device (such as a smart phone, a PAD, or an ipad).
For the working process, the working details, and the technical effects of the storage medium provided in this embodiment, reference may be made to embodiment one, which is not described herein again.
The various embodiments described above are merely illustrative, and may or may not be physically separate, as they relate to elements illustrated as separate components; if reference is made to a component displayed as a unit, it may or may not be a physical unit, and may be located in one place or distributed over a plurality of network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of the embodiment. One of ordinary skill in the art can understand and implement it without inventive effort.
The above examples are only intended to illustrate the technical solution of the present invention, but not to limit it; although the present invention has been described in detail with reference to the foregoing embodiments, it will be understood by those of ordinary skill in the art that: modifications of the technical solutions described in the embodiments or equivalent replacements of some technical features may still be made. And such modifications or substitutions do not depart from the spirit and scope of the corresponding technical solutions of the embodiments of the present invention.
Finally, it should be noted that the present invention is not limited to the above alternative embodiments, and that various other forms of products can be obtained by anyone in light of the present invention. The above detailed description should not be taken as limiting the scope of the invention, which is defined in the claims, and which the description is intended to be interpreted accordingly.
Claims (8)
1. A method for realizing medical image auxiliary diagnosis based on an AlexNet network model is characterized by comprising the following steps:
s101, obtaining a plurality of sample medical images and diagnosis labels corresponding to the sample medical images, wherein the diagnosis labels are positive examples or negative examples;
s102, carrying out image preprocessing on each sample medical image to obtain a corresponding standard sample medical image which is square, consistent in size and normalized, and carrying out digital coding on a corresponding diagnosis label in the following form; if positive, it is marked as 1, and if negative, it is marked as 0;
s103, extracting 2 from the training sample data setnAnd then, introducing the standard sample medical images into an AlexNet network model for forward propagation to obtain a first prediction probability corresponding to each standard sample medical image, wherein the training sample data set comprises not less than 2nN is a natural number between 4 and 8, and the first prediction probability comprises a probability of identifying the image as a positive example and a probability of identifying the image as a negative example;
the AlexNet network model sequentially comprises a first convolution layer, a first batch of normalization layers, a first maximum pooling layer, a second convolution layer, a second batch of normalization layers, a second maximum pooling layer, a third convolution layer, a fourth convolution layer, a fifth convolution layer, the flattening layer, a first full-connection layer, a first discarding layer, a second full-connection layer, a second discarding layer and a third full-connection layer along the forward propagation direction, wherein the convolution kernel of the first convolution layer is (11, 11), the step length is (4, 4), the number of output channels is 96, the activation function is a relu activation function, and the padding type is valid; the first normalization layer is used for accelerating network convergence, preventing overfitting and improving generalization capability; the pooling core of the first maximum pooling layer is (3, 3), the step length is (2, 2), and the padding type is valid; the convolution kernel of the second convolution layer is (5, 5), the step length is (1, 1), the number of output channels is 256, the activation function is a relu activation function, and the padding type is same; the second batch of normalization layers are also used for accelerating network convergence, preventing overfitting and improving generalization capability; the pooling core of the second maximum pooling layer is (3, 3), the step length is (2, 2), and the padding type is valid; the convolution kernel of the third convolution layer is (3, 3), the step length is (1, 1), the number of output channels is 384, the activation function is a relu activation function, and the padding type is same; the convolution kernel of the fourth convolution layer is (3, 3), the step length is (1, 1), the number of output channels is 256, the activation function is a relu activation function, and the padding type is same; the convolution kernel of the fifth convolution layer is (3, 3), the step length is (1, 1), the number of output channels is 384, the activation function is a relu activation function, and the padding type is same; the flattening layer is used for spreading the characteristic diagram into a line, so that a first full-connection layer is connected to the rear side of the flattening layer conveniently; the number of the neurons of the first full-junction layer is 1024, and the activation function is a relu activation function; the first discarding layer is used for randomly inactivating half of the neurons, so that overfitting is avoided, and generalization capability is improved; the number of neurons of the second fully-connected layer is half of the number of neurons of the first fully-connected layer, and the activation function is a relu activation function; the second discarding layer is used for randomly inactivating half of the neurons, so that overfitting is avoided, and generalization capability is improved; the number of neurons of the third full junction layer is 2, and the activation function is a softmax activation function, wherein the softmax activation function is used for identifying the probability that the image is a positive example and the probability that the image is a negative example;
s104, calculating the average loss value loss of the training according to the following formula:
wherein i is 1 to 2nNatural number between, yiThe diagnostic label code value of the ith standard sample medical image,
probability of identifying the image as a positive case of the ith standard sample medical image, wherein k is a constant;
s105, reversely propagating the average loss value loss layer by layer through a gradient descent method, and updating model parameters;
s106, returning to execute the steps S103-S105 until the average loss value loss is reduced to a preset threshold value, and finishing model training;
s107, acquiring a medical image to be diagnosed;
s108, image preprocessing is carried out on the medical image to be diagnosed, and a standard medical image to be diagnosed which is square and has the size consistent with that of the standard sample medical image and the normalization mode consistent with that of the standard sample medical image is obtained;
s109, inputting the standard medical image to be diagnosed into an AlexNet network model which is trained to obtain a second prediction probability corresponding to the standard medical image to be diagnosed, wherein the second prediction probability comprises a probability of identifying an image as a positive example and a probability of identifying an image as a negative example;
s110, obtaining a diagnosis result according to the second prediction probability: and if the probability of identifying the positive example of the image in the second prediction probability is not less than the segmentation threshold of the positive example and the negative example, judging that the diagnosis result is positive example and/or positive, otherwise, judging that the diagnosis result is negative example and/or negative.
2. The method according to claim 1, wherein in step S102, the square sample medical image is obtained by image preprocessing in the following manner:
when the original shape of the sample medical image is a non-square rectangle, pixel points with pixel values of 0 or pixel mean values are symmetrically filled at two sides in the width direction, so that the final image width is equal to the image length, wherein the pixel mean value is the pixel mean value of all the pixel points in the original medical image.
3. The method according to claim 2, wherein after the square sample medical image is obtained in step S102, the sample medical image with the same size is obtained by image preprocessing as follows:
the sample medical image is changed into the size of 2 by interpolation or pressure samplingk*2kWherein k is a natural number between 7 and 12.
4. The method according to claim 3, wherein after the sample medical images with consistent sizes are obtained in step S102, the normalized sample medical images are obtained by image preprocessing in the following manner:
aiming at each pixel point on the sample medical image, obtaining a normalized pixel value P according to the following formulaNew：
In the formula, POldThe value is the pixel value before normalization, mu is the pixel average value of all pixel points before normalization, and sigma is the pixel value standard deviation of all pixel points before normalization.
5. The method according to claim 1, wherein before the step S103, the training sample data set is subjected to data enhancement in any one or any combination of the following manners (a) to (D):
(A) randomly turning the standard sample medical image up and down and/or left and right, and then adding the obtained image as a new sample into the training sample data set;
(B) carrying out random angular rotation processing on the standard sample medical image, and then adding the obtained image serving as a new sample into the training sample data set;
(C) randomly adding Gaussian noise to the medical image of the standard sample, and then adding the obtained image serving as a new sample to the training sample data set;
(D) and randomly cutting the image of the standard sample medical image, and then adding the obtained image as a new sample into the training sample data set, wherein the image cutting part accounts for no more than 5% of the whole image.
6. A device for realizing medical image auxiliary diagnosis based on an AlexNet network model is characterized by comprising a sample acquisition module, a sample preprocessing module, a model iteration training module, an image acquisition module, an image preprocessing module, an image recognition module and an image diagnosis module, wherein the model iteration training module comprises a model training submodule, a loss calculation submodule, a parameter updating submodule and an iteration control submodule;
the sample acquisition module is used for acquiring a plurality of sample medical images and diagnosis labels corresponding to the sample medical images, wherein the diagnosis labels are positive examples or negative examples;
the sample preprocessing module is in communication connection with the sample acquisition module and is used for preprocessing images of all sample medical images to obtain corresponding standard sample medical images which are square, consistent in size and normalized, and simultaneously digitally encoding corresponding diagnosis labels in the following forms; if positive, it is marked as 1, and if negative, it is marked as 0;
the model training submodule is in communication connection with the sample preprocessing module and is used for extracting 2 from a training sample data setnAnd then, introducing the standard sample medical images into an AlexNet network model for forward propagation to obtain a first prediction probability corresponding to each standard sample medical image, wherein the training sample data set comprises not less than 2nN is a natural number between 4 and 8, and the first prediction probability comprises a probability of identifying the image as a positive example and a probability of identifying the image as a negative example;
the AlexNet network model sequentially comprises a first convolution layer, a first batch of normalization layers, a first maximum pooling layer, a second convolution layer, a second batch of normalization layers, a second maximum pooling layer, a third convolution layer, a fourth convolution layer, a fifth convolution layer, the flattening layer, a first full-connection layer, a first discarding layer, a second full-connection layer, a second discarding layer and a third full-connection layer along the forward propagation direction, wherein the convolution kernel of the first convolution layer is (11, 11), the step length is (4, 4), the number of output channels is 96, the activation function is a relu activation function, and the padding type is valid; the first normalization layer is used for accelerating network convergence, preventing overfitting and improving generalization capability; the pooling core of the first maximum pooling layer is (3, 3), the step length is (2, 2), and the padding type is valid; the convolution kernel of the second convolution layer is (5, 5), the step length is (1, 1), the number of output channels is 256, the activation function is a relu activation function, and the padding type is same; the second batch of normalization layers are also used for accelerating network convergence, preventing overfitting and improving generalization capability; the pooling core of the second maximum pooling layer is (3, 3), the step length is (2, 2), and the padding type is valid; the convolution kernel of the third convolution layer is (3, 3), the step length is (1, 1), the number of output channels is 384, the activation function is a relu activation function, and the padding type is same; the convolution kernel of the fourth convolution layer is (3, 3), the step length is (1, 1), the number of output channels is 256, the activation function is a relu activation function, and the padding type is same; the convolution kernel of the fifth convolution layer is (3, 3), the step length is (1, 1), the number of output channels is 384, the activation function is a relu activation function, and the padding type is same; the flattening layer is used for spreading the characteristic diagram into a line, so that a first full-connection layer is connected to the rear side of the flattening layer conveniently; the number of the neurons of the first full-junction layer is 1024, and the activation function is a relu activation function; the first discarding layer is used for randomly inactivating half of the neurons, so that overfitting is avoided, and generalization capability is improved; the number of neurons of the second fully-connected layer is half of the number of neurons of the first fully-connected layer, and the activation function is a relu activation function; the second discarding layer is used for randomly inactivating half of the neurons, so that overfitting is avoided, and generalization capability is improved; the number of neurons of the third full junction layer is 2, and the activation function is a softmax activation function, wherein the softmax activation function is used for identifying the probability that the image is a positive example and the probability that the image is a negative example;
the loss calculation submodule is respectively in communication connection with the model training submodule and the sample preprocessing module, and is used for calculating the average loss value loss of the training according to the following formula:
wherein i is 1 to 2nNatural number between, yiThe diagnostic label code value of the ith standard sample medical image,
probability of identifying the image as a positive case of the ith standard sample medical image, wherein k is a constant;
the parameter updating submodule is in communication connection with the loss calculating submodule and is used for reversely transmitting the average loss value loss layer by layer through a gradient descent method to update the model parameters;
the iteration control submodule is in communication connection with the loss calculation submodule and is used for circularly and sequentially starting the model training submodule, the loss calculation submodule and the parameter updating submodule until the average loss value loss is reduced to a preset threshold value, and completing model training;
the image acquisition module is used for acquiring medical images to be diagnosed;
the image preprocessing module is in communication connection with the image acquisition module and is used for preprocessing the medical image to be diagnosed to obtain a standard medical image to be diagnosed, which is square, has the size consistent with that of the standard sample medical image and has the normalization mode consistent with that of the standard sample medical image;
the image identification module is respectively in communication connection with the model iteration training module and the image preprocessing module, and is used for inputting the standard medical image to be diagnosed into an AlexNet network model which is trained to obtain a second prediction probability corresponding to the standard medical image to be diagnosed, wherein the second prediction probability comprises the probability that the image is recognized as a positive example and the probability that the image is recognized as a negative example;
the image diagnosis module is in communication connection with the image identification module and is used for obtaining a diagnosis result according to the second prediction probability: and if the probability of identifying the positive example of the image in the second prediction probability is not less than the segmentation threshold of the positive example and the negative example, judging that the diagnosis result is positive example and/or positive, otherwise, judging that the diagnosis result is negative example and/or negative.
7. An apparatus for implementing medical image aided diagnosis based on an AlexNet network model, comprising a memory and a processor, wherein the memory is used for storing a computer program, and the processor is used for executing the computer program to implement the method steps of implementing medical image aided diagnosis based on the AlexNet network model according to any one of claims 1 to 5.
8. A storage medium, characterized in that the storage medium has stored thereon a computer program which, when being executed by a processor, implements the method steps of implementing medical image-assisted diagnosis based on AlexNet network model according to any one of claims 1 to 5.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911176103.2A CN110859642B (en) | 2019-11-26 | 2019-11-26 | Method, device, equipment and storage medium for realizing medical image auxiliary diagnosis based on AlexNet network model |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911176103.2A CN110859642B (en) | 2019-11-26 | 2019-11-26 | Method, device, equipment and storage medium for realizing medical image auxiliary diagnosis based on AlexNet network model |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110859642A true CN110859642A (en) | 2020-03-06 |
| CN110859642B CN110859642B (en) | 2024-01-23 |
Family
ID=69656658
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201911176103.2A Active CN110859642B (en) | 2019-11-26 | 2019-11-26 | Method, device, equipment and storage medium for realizing medical image auxiliary diagnosis based on AlexNet network model |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110859642B (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111462079A (en) * | 2020-03-31 | 2020-07-28 | 上海全景医学影像诊断中心有限公司 | Automatic migratable artificial intelligence medical image recognition system and recognition method |
| CN111931626A (en) * | 2020-08-03 | 2020-11-13 | 天津理工大学 | Automatic operation method of vascular intervention robot based on deep learning |
| CN111986117A (en) * | 2020-08-31 | 2020-11-24 | 南京大学 | System and method for correcting arithmetic operation |
| CN113057647A (en) * | 2021-03-25 | 2021-07-02 | 山东省人工智能研究院 | Quality evaluation method of electrocardiosignal |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2018028255A1 (en) * | 2016-08-11 | 2018-02-15 | 深圳市未来媒体技术研究院 | Image saliency detection method based on adversarial network |
| CN107909095A (en) * | 2017-11-07 | 2018-04-13 | 江苏大学 | A kind of image-recognizing method based on deep learning |
| CN108257135A (en) * | 2018-02-01 | 2018-07-06 | 浙江德尚韵兴图像科技有限公司 | The assistant diagnosis system of medical image features is understood based on deep learning method |
| CN109978778A (en) * | 2019-03-06 | 2019-07-05 | 浙江工业大学 | Convolutional neural networks medicine CT image denoising method based on residual error study |
| CN110232383A (en) * | 2019-06-18 | 2019-09-13 | 湖南省华芯医疗器械有限公司 | A kind of lesion image recognition methods and lesion image identifying system based on deep learning model |
-
2019
- 2019-11-26 CN CN201911176103.2A patent/CN110859642B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2018028255A1 (en) * | 2016-08-11 | 2018-02-15 | 深圳市未来媒体技术研究院 | Image saliency detection method based on adversarial network |
| CN107909095A (en) * | 2017-11-07 | 2018-04-13 | 江苏大学 | A kind of image-recognizing method based on deep learning |
| CN108257135A (en) * | 2018-02-01 | 2018-07-06 | 浙江德尚韵兴图像科技有限公司 | The assistant diagnosis system of medical image features is understood based on deep learning method |
| CN109978778A (en) * | 2019-03-06 | 2019-07-05 | 浙江工业大学 | Convolutional neural networks medicine CT image denoising method based on residual error study |
| CN110232383A (en) * | 2019-06-18 | 2019-09-13 | 湖南省华芯医疗器械有限公司 | A kind of lesion image recognition methods and lesion image identifying system based on deep learning model |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111462079A (en) * | 2020-03-31 | 2020-07-28 | 上海全景医学影像诊断中心有限公司 | Automatic migratable artificial intelligence medical image recognition system and recognition method |
| CN111462079B (en) * | 2020-03-31 | 2021-01-08 | 上海全景医学影像诊断中心有限公司 | Automatic migratable artificial intelligence medical image recognition system and recognition method |
| CN111931626A (en) * | 2020-08-03 | 2020-11-13 | 天津理工大学 | Automatic operation method of vascular intervention robot based on deep learning |
| CN111931626B (en) * | 2020-08-03 | 2023-12-26 | 深圳市爱博医疗机器人有限公司 | Vascular intervention robot automatic operation method based on deep learning |
| CN111986117A (en) * | 2020-08-31 | 2020-11-24 | 南京大学 | System and method for correcting arithmetic operation |
| CN113057647A (en) * | 2021-03-25 | 2021-07-02 | 山东省人工智能研究院 | Quality evaluation method of electrocardiosignal |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110859642B (en) | 2024-01-23 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110689038B (en) | Training method and device for neural network model and medical image processing system | |
| CN110859642B (en) | Method, device, equipment and storage medium for realizing medical image auxiliary diagnosis based on AlexNet network model | |
| CN110232383B (en) | Focus image recognition method and focus image recognition system based on deep learning model | |
| CN110298383B (en) | Multi-mode deep learning-based pathology classification method and system | |
| CN109166130B (en) | Image processing method and image processing device | |
| CN109754403A (en) | Tumour automatic division method and system in a kind of CT image | |
| CN111291825B (en) | Focus classification model training method, apparatus, computer device and storage medium | |
| CN110544264A (en) | Temporal bone key anatomical structure small target segmentation method based on 3D deep supervision mechanism | |
| CN110310287A (en) | It is neural network based to jeopardize the automatic delineation method of organ, equipment and storage medium | |
| CN111145147B (en) | Multi-mode medical image segmentation method and terminal equipment | |
| CN110706214B (en) | Three-dimensional U-Net brain tumor segmentation method fusing condition randomness and residual error | |
| CN114494296A (en) | Brain glioma segmentation method and system based on fusion of Unet and Transformer | |
| CN111080654B (en) | Image lesion region segmentation method and device and server | |
| CN111028248A (en) | Method and device for separating static and dynamic pulses based on CT (computed tomography) image | |
| CN110570394B (en) | Medical image segmentation method, device, equipment and storage medium | |
| CN116310076A (en) | Three-dimensional reconstruction method, device, equipment and storage medium based on nerve radiation field | |
| CN110458852A (en) | Segmentation of lung parenchyma method, apparatus, equipment and storage medium based on capsule network | |
| CN110826581A (en) | Animal number identification method, device, medium and electronic equipment | |
| CN115018809A (en) | Target area segmentation and identification method and system of CT image | |
| CN111582437B (en) | Construction method of parallax regression depth neural network | |
| CN111091560A (en) | Nasopharyngeal carcinoma primary tumor image identification method and system | |
| CN112634308B (en) | Nasopharyngeal carcinoma target area and organ-at-risk delineating method based on different receptive fields | |
| CN111210398A (en) | White blood cell recognition system based on multi-scale pooling | |
| CN110930391A (en) | Method, device and equipment for realizing medical image auxiliary diagnosis based on VggNet network model and storage medium | |
| CN110930392A (en) | Method, device, equipment and storage medium for realizing medical image auxiliary diagnosis based on GoogLeNet network model |
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 |