WO2021115356A1 - 自适应窗宽窗位调节方法、装置、计算机系统及存储介质 - Google Patents

自适应窗宽窗位调节方法、装置、计算机系统及存储介质 Download PDF

Info

Publication number
WO2021115356A1
WO2021115356A1 PCT/CN2020/135076 CN2020135076W WO2021115356A1 WO 2021115356 A1 WO2021115356 A1 WO 2021115356A1 CN 2020135076 W CN2020135076 W CN 2020135076W WO 2021115356 A1 WO2021115356 A1 WO 2021115356A1
Authority
WO
WIPO (PCT)
Prior art keywords
feature vector
truncation
derivable
output feature
value
Prior art date
Application number
PCT/CN2020/135076
Other languages
English (en)
French (fr)
Inventor
徐尚良
张芮溟
Original Assignee
平安科技（深圳）有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 平安科技（深圳）有限公司 filed Critical 平安科技（深圳）有限公司
Publication of WO2021115356A1 publication Critical patent/WO2021115356A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T11/00—2D [Two Dimensional] image generation
    • G06T11/003—Reconstruction from projections, e.g. tomography
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/048—Activation functions
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G06N3/084—Backpropagation, e.g. using gradient descent
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02T—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
    • Y02T10/00—Road transport of goods or passengers
    • Y02T10/10—Internal combustion engine [ICE] based vehicles
    • Y02T10/40—Engine management systems

Definitions

  • This application relates to the field of artificial intelligence technology, and in particular to an adaptive window width and window level adjustment method, device, computer system and storage medium.
  • window width and window level are commonly used concepts in medical image processing; among them, window width refers to the range of CT values displayed in CT images, which is an important index for truncating the image.
  • the width of the window directly affects the clarity and contrast of the image. If a narrow window width is used, the displayed CT value range is small, the CT value amplitude represented by each gray scale is small, and the contrast is strong, which is suitable for observing tissue structures with close density (such as brain tissue). On the contrary, if a wide window width is used, the displayed CT value range is large, and the CT value amplitude represented by each gray scale is large, the image contrast is poor, but the density is uniform, which is suitable for observing structures with large differences in density (such as bone and soft tissue).
  • Window level refers to the average or center value within the window width, which is an important index for image equalization processing, such as a CT image
  • the window width is 100Hu
  • the window level is selected at 0Hu; then the window level is the center ( 0Hu), upward includes +50Hu, downward includes -50Hu, all tissues within this 100Hu range can be displayed and recognized by the human eye. All the tissues larger than +50Hu are white; all the tissues smaller than -50Hu are black, and the density difference cannot be displayed.
  • the usual method is to analyze the grayscale histogram of the image and perform grayscale cropping from 0.01 to 0.99 or grayscale based on integral transformation. Operations such as equalization.
  • the inventor realizes that the gray scale transformation in this way is the same as the subsequent neural network's task of processing the picture (such as: classification, segmentation, detection, etc.) is segmentation: and this kind of image analysis is based on segmentation. Histogram, the linear transformation operation of gray-level equalization directly using integral transformation will change the structural information of the image, and the generated window width and window level image will not meet the data processing requirements of the neural network due to the linear transformation operation, which will lead to The neural network has low data processing accuracy for window width and level.
  • the purpose of this application is to provide an adaptive window width and window level adjustment method, device, computer system, and storage medium, which are used to solve the problem that the linear transformation operation of gray level equalization directly using integral transformation in the prior art can change the image quality.
  • Structural information makes it difficult for the generated window width and window level images to meet the data processing requirements of the neural network, which in turn leads to the problem of low data processing accuracy of the window width and window level by the neural network; this application can be applied to smart medical scenarios to promote The construction of smart cities.
  • the present application provides an adaptive window width and window level adjustment method based on gradient backpropagation, including:
  • Receive the image to be adjusted sequentially extract the gray value of each pixel in the image to be adjusted and summarize the input feature vector
  • the output feature vector is sent to a preset neural network, and the neural network updates the weight of the derivable truncation model according to the output feature vector to generate an output feature vector that conforms to the loss function of the neural network, and according to The output feature vector generates a window width and window level image.
  • the present application also provides an adaptive window width and window level adjustment device based on gradient backpropagation, including:
  • the gray scale extraction module is used to receive the image to be adjusted, sequentially extract the gray value of each pixel in the image to be adjusted, and summarize to obtain the input feature vector;
  • the derivable truncation module is used to calculate the truncation adjustment coefficients of the gray values in the input feature vector through the derivable truncation model and summarize them to form a truncation adjustment vector, and adjust the input feature vector according to the truncation adjustment vector to generate an output feature vector;
  • the image generation module is configured to send the output feature vector to a preset neural network, and the neural network updates the weight of the derivable truncation model according to the output feature vector, so that it generates a loss function that conforms to the neural network Output a feature vector, and generate a window width and window level image according to the output feature vector.
  • the present application also provides a computer system, which includes a plurality of computer devices, each computer device includes a memory, a processor, and a computer program stored in the memory and running on the processor.
  • processor of the device executes the computer program, the following methods are jointly implemented:
  • Receive the image to be adjusted sequentially extract the gray value of each pixel in the image to be adjusted and summarize the input feature vector
  • the output feature vector is sent to a preset neural network, and the neural network updates the weight of the derivable truncation model according to the output feature vector to generate an output feature vector that conforms to the loss function of the neural network, and according to The output feature vector generates a window width and window level image.
  • the present application also provides a computer-readable storage medium, which includes multiple storage media, each of which stores a computer program, and when the computer program stored in the multiple storage media is executed by a processor Implement the following methods together:
  • Receive the image to be adjusted sequentially extract the gray value of each pixel in the image to be adjusted and summarize the input feature vector
  • the output feature vector is sent to a preset neural network, and the neural network updates the weight of the derivable truncation model according to the output feature vector to generate an output feature vector that conforms to the loss function of the neural network, and according to The output feature vector generates a window width and window level image.
  • the window width and window level image obtained by this application not only meets the user's need for adjusting the window width and window level, but also satisfies the neural network to process the window width and window level due to the weight update of the derivable truncation function.
  • classification requirements enable users to obtain the window width and window level images that meet the requirements, while also ensuring the processing accuracy of the window width and window level images by the neural network.
  • FIG. 1 is a flowchart of Embodiment 1 of an adaptive window width and window level adjustment method according to this application;
  • FIG. 2 is an image of a window width of 125 HU and a bed size of 45 HU in Embodiment 1 of the adaptive window width and window level adjustment method of this application;
  • FIG. 3 is an image with a window width of 80 HU and a window level of 40 HU in the first embodiment of the adaptive window width and window level adjustment method of this application;
  • FIG. 4 is a flowchart of obtaining an input feature vector in Embodiment 1 S1 of an adaptive window width and window level adjustment method of this application;
  • FIG. 5 is a flowchart of forming a truncated adjustment vector in S2 of the first embodiment of the adaptive window width and window level adjustment method of this application;
  • FIG. 6 is a flowchart of generating output feature vectors in S2 of the first embodiment of the adaptive window width and window level adjustment method of this application;
  • FIG. 7 is a flowchart of generating an output feature vector conforming to the neural network loss function in S3 of the first embodiment of the adaptive window width and window level adjustment method of this application;
  • FIG. 8 is a schematic diagram of program modules of Embodiment 2 of an adaptive window width and window level adjustment device according to the present application.
  • FIG. 9 is a schematic diagram of the hardware structure of the computer equipment in the third embodiment of the computer system of this application.
  • the technical solution of the present application can be applied to the fields of artificial intelligence, blockchain and/or big data technology, for example, it can specifically involve neural network technology.
  • the data involved in the application such as coefficients, feature vectors, weights, and/or window width and window level images, can be stored in a database, or can be stored in a blockchain, such as distributed storage through a blockchain, This application is not limited.
  • the adaptive window width and window level adjustment method, device, computer system, and storage medium provided in this application are applicable to the field of artificial intelligence image detection technology, and provide a method based on gray-scale extraction module, derivable truncation module, and image generation module. Adaptive window width and window level adjustment method.
  • This application extracts the gray values of the pixels in the image to be adjusted and summarizes them to obtain the input feature vector; calculates the truncation adjustment coefficients of the gray values in the input feature vector through the derivable truncation model and summarizes them to form the truncation adjustment vector, and according to The truncation adjustment vector adjusts the input feature vector to generate an output feature vector; sends the output feature vector to a preset neural network, and the neural network updates the weight of the derivable truncation model according to the output feature vector, so that It generates an output feature vector conforming to the neural network loss function, and generates a window width and window level image according to the output feature vector.
  • an adaptive window width and window level adjustment method based on gradient backpropagation in this embodiment includes:
  • S1 Receive the image to be adjusted, sequentially extract the gray value of each pixel in the image to be adjusted, and summarize to obtain the input feature vector.
  • S2 Calculate the truncation adjustment coefficients of the gray values in the input feature vector through a derivable truncation model and summarize them to form a truncation adjustment vector, and adjust the input feature vector according to the truncation adjustment vector to generate an output feature vector.
  • S3 Send the output feature vector to a preset neural network, and the neural network updates the weight of the derivable truncation model according to the output feature vector to generate an output feature vector that conforms to the neural network loss function, And according to the output feature vector to generate the window width and window level image.
  • the image to be adjusted sent from the user terminal or extracted from the database is received, the gray value of each pixel in the image to be adjusted is sequentially extracted, and the gray value is arranged according to the position of each pixel, and the result is summarized Enter the feature vector.
  • the truncation adjustment coefficients of the gray values in the input feature vector are sequentially calculated to obtain the truncation adjustment vector, and each of the input feature vectors is adjusted according to the truncation adjustment vector.
  • the gray value of the pixel is used to generate an output feature vector; the output feature vector is output to the neural network through a forward propagation algorithm.
  • the neural network updates the weights of the derivable truncation function in the derivable truncation model according to the output eigenvectors and through a back propagation algorithm; and recalculates the input eigenvectors through the updated derivable truncation model Truncate the adjustment vector, and adjust the input feature vector to generate an output feature vector according to the truncation adjustment vector, until the output feature vector generated by the derivable truncation model is no longer used by the neural network through the backpropagation algorithm to analyze the derivable
  • the derivable truncation function in the truncation model performs weight update, the window width and window level image is generated according to the output feature vector.
  • the window width and window level image because the derivable truncation model not only meets the user's need for adjusting the window width and window level, but also because the weight update of the derivable truncation function satisfies the neural network to process or process the window width and window level.
  • the classification requirements enable users to obtain the window width and window level images that meet the requirements while also ensuring the processing accuracy of the window width and window level images by the neural network.
  • This application can be applied in smart medical scenarios to promote the construction of smart cities.
  • the neural network is a network model used for further processing or processing of window width and window level images, and its type and number can be set according to the needs of users; as we all know, any neural network needs to pass forward propagation
  • the algorithm forwards the received data, and then updates the weight and/or paranoia value of each layer of the neural network through the back-propagation algorithm, which belongs to the prior art; the problem solved by this application is how to make the window wider
  • the bit image can meet the data processing requirements of the neural network; the method adopted in this application is to set the guideable truncation model to make the whole process from the image to be adjusted to the window width and window level image into a guideable calculation process.
  • the weight and/or paranoia value of the derivative truncated model can be updated, so that the derivative truncated model becomes a part of the neural network and can be well matched with every neural network.
  • One-layer matching therefore, from the root, that is, from the input neural network's output vector characteristics, the matching with each layer of the neural network is ensured, and the above-mentioned guarantee of the processing accuracy of the neural network on the window width and window level images is realized. Since the working principle of the neural network belongs to the prior art, it will not be repeated here.
  • the window width in this application refers to the range of CT values displayed in CT images.
  • the CT value is a measurement unit for measuring the density of a certain local tissue or organ of the human body, usually called the Hounsfield unit (HU), for example: air is -1000 (because the color is almost no), dense bone is +1000 (because the color is particularly heavy); window level refers to the average or center value within the window width range, which is an important indicator for the image equalization processing, such as A CT image with a window width of 100Hu and a window level of 0Hu; the window level is the center (0Hu), including +50Hu upward and -50Hu downward. All tissues within this 100Hu range can be displayed and displayed. Recognized by the human eye. All the tissues larger than +50Hu are white; all the tissues smaller than -50Hu are black, and the density difference cannot be displayed.
  • HU Hounsfield unit
  • Figure 2 is an image with a window width of 125 HU and a bed position of 45 HU.
  • Figure 3 is an image with a window width of 80HU and a window level of 40HU.
  • the steps of extracting the gray value of each pixel in the image to be adjusted in sequence in S1 and collectively obtaining the input feature vector include:
  • the gray value of each pixel in the image to be adjusted can be extracted through OpenCV;
  • the OpenCV is a cross-platform computer vision library that can run on Linux, Windows, Android, and Mac OS operating systems. It is lightweight and efficient--consisting of a series of C functions and a small number of C++ classes, it also provides interfaces to languages such as Python, Ruby, and MATLAB, and implements many common algorithms in image processing and computer vision.
  • a pixel matrix can be established according to the arrangement of pixels in the image to be adjusted, and the gray value extracted by OpenCV can be adjusted according to the position of the pixel in the image to be adjusted (such as the number of rows and columns).
  • the degree value is placed in the pixel matrix until the OpenCV extraction is completed; at this time, the pixel matrix will become the gray value of each pixel arranged and summarized to form an input feature vector.
  • the step of calculating the truncation adjustment coefficient of each gray value in the input feature vector through the derivable truncation model in S2 and summarizing the truncation adjustment vector includes:
  • S201 Create a machine learning model and a derivable truncation model with a derivable truncation function, and predict the upper and lower gray levels of the input feature vector through the machine learning model; wherein the derivable truncation function has a derivable Features and used for equalizing and truncating the image to be adjusted.
  • the truncation function can be derived, as shown below:
  • s(x) 1/(1+e- kx ), in the derivable cutoff function, k is a hyperparameter used to adjust the effective activation interval of the function, and x is the gray value in the input feature vector , S(x) is the cutoff adjustment coefficient, windowmin is the lower limit of the gray level of the window width and window level, windowmax is the upper limit of the gray level of the window width and window level, x is the gray value in the input feature vector, M( ⁇ ) is the cutoff adjustment coefficient,
  • the derivable characteristic of the derivable truncated function is realized; the gradient backpropagation of the neural network can be used to derive the derivable truncated function, and then the derivable function can be derived.
  • the weight of the truncation function is updated so that the derivable truncation model can generate window width and window level images that can meet the processing or classification requirements of the neural network.
  • the step of predicting the upper limit and the lower limit of the gray level of the input feature vector through the machine learning model in S201 includes:
  • the training samples used to train the neural network can be used to train the machine learning model so that the machine learning model updates its own weights through the gradient descent method and the backpropagation algorithm to conform to the neural network pairing Data processing requirements.
  • the machine learning model is a computer model obtained through training and convergence, and the machine learning model can use the same training samples as the neural network, so that the machine learning model has an upper limit on the gray scale.
  • the recognition of the lower limit of the gray level and the gray level can meet the data processing requirements of the neural network for data classification, data segmentation, image recognition, etc., so that the machine learning model can analyze the input feature vector, identify the unique image to be adjusted, and is suitable for subsequent follow-up
  • the optimal window width and window level of the image task that is, the data processing task of the neural network.
  • the optimal window width and window level refer to the upper gray level and the lower gray level.
  • the training convergence is to adjust and update the weight of each layer of the neural network through the gradient descent method and the back propagation algorithm to make it converge and meet the training expectations (for example, the loss value of the neural network is less than a preset threshold) Therefore, the gray upper limit and the lower gray limit obtained by the operation of the neural network have the characteristics of gradient back propagation (that is, it meets the training convergence requirements of the machine learning model and the neural network at the same time), so the The upper limit and lower limit of gray level meet the data processing requirements of neural networks.
  • the training of the machine learning model and the obtaining of the upper limit and the lower limit of the gray level by analyzing the image through the machine learning model belong to the prior art.
  • the technical problem solved by this application is: how to obtain a gray with gradient backpropagation characteristics that conform to the neural network
  • the derivable truncation model calculates each gray value in the input feature vector to obtain a lower limit difference and an upper limit difference of each gray value according to the lower gray limit and the upper gray limit.
  • the lower limit of the gray level and the upper limit of the gray level are used to define the gray level of the image to be adjusted to obtain the lower limit difference and the upper limit difference to achieve the truncation effect of the image to be adjusted; where the lower limit difference refers to a certain pixel of the image to be adjusted and the gray level.
  • the difference between the lower limit of the degree, and the upper limit difference refers to the difference between the pixel and the upper limit of the gray scale.
  • S203 Perform derivable equalization processing on the lower limit difference and the upper limit difference to obtain the lower difference of the derivable equilibrium and the upper difference of the derivable equilibrium, and compare the lower difference of the derivable equilibrium and the upper difference of the derivable equilibrium Subtract to obtain the cutoff adjustment coefficient.
  • the lower limit difference is substituted into the Sigmoid-k function to perform logistic regression, so that each discretized lower limit difference is mapped to the range of (0,1), and the value corresponding to the lower limit difference is obtained.
  • the difference under the derivable equilibrium of the Sigmoid-k function; the derivable difference under the derivable equilibrium obtained after regression has derivable characteristics; at the same time, because the Sigmoid-k function is a continuous sum in its domain Smooth function, therefore, continuous smooth equalization processing is performed on the lower limit difference, and the derivative equalization lower difference corresponding to the lower limit difference is obtained.
  • the truncation adjustment coefficient obtained by subtracting the difference between the differential value of the derivable equalization and the differential value of the derivable equalization is used to calculate the gray value of each pixel in the image to be adjusted, so as to realize the truncation and equalization of the image to be adjusted deal with.
  • Each truncation adjustment coefficient is arranged according to the corresponding pixel position to form a truncation adjustment vector.
  • the step of adjusting the input feature vector according to the truncation adjustment vector in S2 to generate an output feature vector includes:
  • the corresponding cut-off adjustment coefficient and the gray value can be multiplied to obtain the output characteristic value.
  • the output feature value corresponding to each gray value in the input feature vector can be calculated by the following formula
  • X is the input feature vector
  • Y is the preliminary feature vector
  • M( ⁇ ) is the truncation adjustment vector
  • S213 Determine in sequence whether the output feature value in the preliminary output feature vector is greater than a preset cutoff threshold; if it is greater than the cutoff threshold, set the output feature value as a retained feature value; if it is not greater than the cutoff threshold, Then the output feature value is subjected to extreme value processing and set as an extreme value feature value.
  • the output feature value greater than the cutoff threshold is regarded as a pixel that should be displayed in the "observation window", and the output feature value not greater than the cutoff threshold is regarded as a pixel that should not be displayed in the "observation window".
  • the cutoff threshold can be set to zero.
  • extreme value processing in S213 includes the following steps:
  • the lower gray scale limit and the upper gray scale limit can be set as required; for example: the gray scale interval of the image to be adjusted is [-1024, 1024]; if the gray scale lower limit is 0, the gray scale upper limit is 80; If the output characteristic value is -20, it is less than the lower gray limit of 0, so the output characteristic value is adjusted to 0; if the output characteristic value is 100, it is greater than the upper gray limit of 80, so the output characteristic value is adjusted to 80.
  • the output feature vector is sent to a preset neural network, and the neural network updates the weight of the derivable truncation model according to the output feature vector to make it
  • the step of generating an output feature vector conforming to the neural network loss function includes:
  • S301 Output the output feature vector to a neural network through a forward propagation algorithm, and each layer in the neural network sequentially performs operations on the output feature vector to obtain an operation result, and calculates the result by the loss function of the neural network.
  • the calculation result obtains the loss value.
  • the loss function is used to estimate the degree of inconsistency between the predicted value f(x) of the model and the true value Y. It is a non-negative real-valued function, usually L(Y, f(x )) means that the smaller the loss function, the better the robustness of the model; the neural network can use the loss function to calculate the difference between the operation result and the real result (such as the expected result of classification, the expected result of segmentation) to generate loss value.
  • S302 Receive the iterative data output by the neural network through the gradient descent method and the backpropagation algorithm according to the loss value, and derive the derivable truncation function in the derivable truncation model according to the iterative data, And update the weight of the derivable truncation model according to the derivation result.
  • the gradient descent method is an algorithm used to iteratively solve the minimized loss function and the parameter values of the neural network in the machine learning algorithm when the goal is to minimize the loss function.
  • the backpropagation algorithm is an algorithm that uses the chain rule to calculate the derivatives of two or more composite functions, backpropagates the gradient of the output unit back to the input unit, and adjusts the learnable parameters of the network according to the calculated gradient; therefore, ,
  • This step is a method in which the neural network calculates the parameter values of each layer in the neural network through the back-propagation algorithm and the gradient descent method; when propagating to the input layer of the neural network, due to the differentiable characteristics of the derivable truncation function, the neural network The network can still propagate iterative data to the derivable truncated model through the gradient descent method and the backpropagation algorithm to derive the derivable truncation function, and update the weight of the derivable truncated model according to the derivation result, that is, adjust
  • the technical problem solved by this application is how to accept the update of the weight value of the derivable truncation function by the neural network so that the output feature vector generated by the derivable truncation module conforms to Neural network data processing requirements; therefore, the working principles of the backpropagation algorithm and gradient descent method will not be repeated here.
  • the loss value of the neural network reaches below the preset loss threshold, it indicates that the output feature vector meets the requirements of the neural network for data processing, so the weight update of the derivable truncation module will no longer be performed, and
  • the output feature vector is the most final available output feature vector.
  • the neural network can not update the weights of its own layers, but directly update the weights of the derivable truncation model through the gradient descent method and the back propagation algorithm, so that the output feature vector generated by the neural network conforms to the loss function of the neural network Requirements.
  • the loss function, the loss value obtained through the calculation result of the loss function, the derivative of the function and the weight update according to the loss value belong to the prior art, and the technical problem solved by this application is how: The output feature vector and the window width and window level image obtained according to the feature vector meet the data processing requirements of the neural network, so the loss function, the loss value is obtained through the calculation result of the loss function, and the function is derived and weighted according to the loss value The value update will not be repeated here.
  • the step of generating a window width and window level image according to the output feature vector in S3 includes:
  • the gray value of each pixel in the image to be adjusted is adjusted, so that the image to be adjusted is converted into the gray value of each pixel, which is respectively compared with the retained feature value or
  • the corresponding summary information is obtained based on the window width and window level image.
  • the summary information is obtained by hashing the window width and window level image, for example, obtained by the sha256s algorithm.
  • Uploading summary information to the blockchain can ensure its security and fairness and transparency to users.
  • the user equipment can download the summary information from the blockchain to verify whether the collation table and/or data table has been tampered with.
  • the blockchain referred to in this example is a new application mode of computer technology such as distributed data storage, point-to-point transmission, consensus mechanism, and encryption algorithm.
  • Blockchain essentially a decentralized database, is a series of data blocks associated with cryptographic methods. Each data block contains a batch of network transaction information for verification. The validity of the information (anti-counterfeiting) and the generation of the next block.
  • the blockchain can include the underlying platform of the blockchain, the platform product service layer, and the application service layer.
  • an adaptive window width and window level adjustment device 1 based on gradient backpropagation in this embodiment includes:
  • the gray scale extraction module 11 is used to receive the image to be adjusted, sequentially extract the gray value of each pixel in the image to be adjusted, and summarize to obtain the input feature vector;
  • the derivable truncation module 12 is used to calculate the truncation adjustment coefficients of the gray values in the input feature vector through the derivable truncation model and summarize them to form a truncation adjustment vector, and adjust the input feature vector according to the truncation adjustment vector to generate an output Feature vector;
  • the image generation module 13 is configured to send the output feature vector to a preset neural network, and the neural network updates the weight of the derivable truncation model according to the output feature vector to generate a loss function that conforms to the neural network The output feature vector of, and generate the window width and window level image according to the output feature vector.
  • This technical solution is based on image detection in the field of artificial intelligence, by sequentially extracting the gray value of each pixel in the image to be adjusted and summarizing the input feature vector; calculating the cutoff adjustment of each gray value in the input feature vector through a derivable cutoff model The coefficients are summed to form a truncation adjustment vector, and the input feature vector is adjusted according to the truncation adjustment vector to generate an output feature vector; the output feature vector is sent to a preset neural network, and the neural network performs processing on the output feature vector according to the output feature vector.
  • the derivable truncation model performs weight update to generate an output feature vector that conforms to the neural network loss function, and generates a window width and window level image according to the output feature vector, so as to realize the grayscale image processing of the window width and window level image .
  • the present application also provides a computer system, which includes a plurality of computer devices 2.
  • the components of the adaptive window width and window level adjustment device 1 of the second embodiment can be dispersed in different computer devices.
  • the device can be a smart phone, a tablet, a laptop, a desktop computer, a rack server, a blade server, a tower server, or a rack server (including an independent server, or a server cluster composed of multiple servers) that executes the program Wait.
  • the computer device in this embodiment at least includes but is not limited to: a memory 21 and a processor 22 that can be communicatively connected to each other through a system bus, as shown in FIG. 9. It should be pointed out that FIG. 9 only shows a computer device with components, but it should be understood that it is not required to implement all the components shown, and more or fewer components may be implemented instead.
  • the memory 21 (ie, readable storage medium) includes flash memory, hard disk, multimedia card, card-type memory (for example, SD or DX memory, etc.), random access memory (RAM), static random access memory (SRAM), Read only memory (ROM), electrically erasable programmable read only memory (EEPROM), programmable read only memory (PROM), magnetic memory, magnetic disk, optical disk, etc.
  • the memory 21 may be an internal storage unit of a computer device, such as a hard disk or memory of the computer device.
  • the memory 21 may also be an external storage device of the computer device, for example, a plug-in hard disk equipped on the computer device, a smart memory card (Smart Media Card, SMC), and a Secure Digital (SD).
  • SD Secure Digital
  • the memory 21 may also include both an internal storage unit of the computer device and an external storage device thereof.
  • the memory 21 is generally used to store an operating system and various application software installed in a computer device, such as the program code of the adaptive window width and window level adjustment device of the first embodiment.
  • the memory 21 can also be used to temporarily store various types of data that have been output or will be output.
  • the processor 22 may be a central processing unit (Central Processing Unit, CPU), a controller, a microcontroller, a microprocessor, or other data processing chips.
  • the processor 22 is generally used to control the overall operation of the computer equipment.
  • the processor 22 is used to run the program code or process data stored in the memory 21, for example, to run an adaptive window width and window level adjustment device, so as to implement the adaptive window width and window level adjustment method of the first embodiment.
  • this application also provides a computer-readable storage system (computer-readable storage medium), which includes multiple storage media, such as flash memory, hard disk, multimedia card, card-type memory (for example, SD or DX memory, etc.) ), random access memory (RAM), static random access memory (SRAM), read only memory (ROM), electrically erasable programmable read only memory (EEPROM), programmable read only memory (PROM), magnetic memory, magnetic disk , CD-ROM, server, App application mall, etc., on which computer programs are stored, and when the programs are executed by the processor 22, the corresponding functions are realized.
  • the computer-readable storage medium of this embodiment is used to store an adaptive window width and window level adjustment device, and when executed by the processor 22, the adaptive window width and window level adjustment method of the first embodiment is implemented.
  • the storage medium involved in this application may be non-volatile or volatile.

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • Molecular Biology (AREA)
  • Biophysics (AREA)
  • Computational Linguistics (AREA)
  • Artificial Intelligence (AREA)
  • Evolutionary Computation (AREA)
  • General Health & Medical Sciences (AREA)
  • Biomedical Technology (AREA)
  • Computing Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Health & Medical Sciences (AREA)
  • Image Analysis (AREA)

Abstract

一种基于梯度反传的自适应窗宽窗位调节方法、装置、计算机系统及存储介质，包括：提取待调节图像中各像素的灰度值并汇总获得输入特征向量（S1）；通过可导截断模型计算输入特征向量中各灰度值的截断调节系数并汇总形成截断调节向量，及根据截断调节向量调节输入特征向量生成输出特征向量（S2）；将输出特征向量发送预设的神经网络，神经网络根据输出特征向量对可导截断模型进行权值更新，使其生成符合神经网络损失函数的输出特征向量，并根据该输出特征向量生成窗宽窗位图像（S3）。其得到的窗宽窗位图像既符合了用户对窗宽窗位进行调节的需求，又满足了神经网络对窗宽窗位进行处理或分类的要求。

Description

自适应窗宽窗位调节方法、装置、计算机系统及存储介质
本申请要求于2020年5月15日提交中国专利局、申请号为202010412059.7，发明名称为“自适应窗宽窗位调节方法、装置、计算机系统及存储介质”的中国专利申请的优先权，其全部内容通过引用结合在本申请中。
技术领域
本申请涉及人工智能技术领域，尤其涉及一种自适应窗宽窗位调节方法、装置、计算机系统及存储介质。
背景技术
窗宽窗位是医学影像处理中的常用概念；其中，窗宽指CT图像所显示的CT值范围，其为对图像进行截断的重要指标，窗宽的宽窄直接影响图像的清晰度与对比度，如果使用窄的窗宽，则显示的CT值范围小，每一灰阶代表的CT值幅度小，对比度强，适于观察密度接近的组织结构(如脑组织)。反之，如果使用宽的窗宽，则显示的CT值范围大，每一灰阶代表的CT值幅度大，则图像对比度差，但密度均匀，适于观察密度差别大的结构(如骨与软组织)；窗位指窗宽范围内均值或中心值，其是对图像进行均衡化处理的重要指标,比如一幅CT图像，窗宽为100Hu，窗位选在0Hu；则以窗位为中心(0Hu)，向上包括+50Hu，向下包括-50Hu，凡是在这个100Hu范围内的组织均可显示出来并为人眼所识别。凡是大于+50Hu的组织均为白色；凡是小于-50Hu的组织均为黑色，其密度差异无法显示。人眼只能识别土50Hu范围内的CT值，每一个灰阶的CT值范围是100/16＝6.25Hu。
发明人发现，目前的窗宽窗位的调节一般是使用者根据需要或经验自行调节，通常的做法是分析图像的灰度直方图，进行0.01～0.99的灰度裁剪或者基于积分变换的灰度均衡化等操作。
但是，发明人意识到，采用这种方式进行灰度变换是与后续的神经网络对图片进行数据处理的任务(如：分类、分割、检测等任务)是分割的：而这种基于分析图像的直方图，直接用积分变换的灰度均衡化的线性变换操作会改变图像的结构信息，会因为其线性变换操作而导致所生成的窗宽窗位图像难以满足神经网络的数据处理要求，进而导致神经网络对窗宽窗位的数据处理精度低下。
发明内容
本申请的目的是提供一种自适应窗宽窗位调节方法、装置、计算机系统及存储介质，用于解决现有技术存在的直接用积分变换的灰度均衡化的线性变换操作会改变图像的结构信息，导致所生成的窗宽窗位图像难以满足神经网络的数据处理要求，进而导致神经网络对窗宽窗位的数据处理精度低下的问题；本申请可应用于智慧医疗场景中，从而推动智慧城市的建设。
为实现上述目的，本申请提供一种基于梯度反传的自适应窗宽窗位调节方法，包括：
接收待调节图像，依次提取待调节图像中各像素的灰度值并汇总获得输入特征向量；
通过可导截断模型计算所述输入特征向量中各灰度值的截断调节系数并汇总形成截断调节向量，及根据所述截断调节向量调节所述输入特征向量生成输出特征向量；
将所述输出特征向量发送预设的神经网络，所述神经网络根据输出特征向量对所述可导截断模型进行权值更新，使其生成符合所述神经网络损失函数的输出特征向量，并根据该输出特征向量生成窗宽窗位图像。
为实现上述目的，本申请还提供一种基于梯度反传的自适应窗宽窗位调节装置，包括：
灰度提取模块，用于接收待调节图像，依次提取待调节图像中各像素的灰度值并汇总获得输入特征向量；
可导截断模块，用于通过可导截断模型计算所述输入特征向量中各灰度值的截断调节系数并汇总形成截断调节向量，及根据所述截断调节向量调节所述输入特征向量生成输出 特征向量；
图像生成模块，用于将所述输出特征向量发送预设的神经网络，所述神经网络根据输出特征向量对所述可导截断模型进行权值更新，使其生成符合所述神经网络损失函数的输出特征向量，并根据该输出特征向量生成窗宽窗位图像。
为实现上述目的，本申请还提供一种计算机系统，其包括多个计算机设备，各计算机设备包括存储器、处理器以及存储在存储器上并可在处理器上运行的计算机程序，所述多个计算机设备的处理器执行所述计算机程序时共同实现以下方法：
接收待调节图像，依次提取待调节图像中各像素的灰度值并汇总获得输入特征向量；
通过可导截断模型计算所述输入特征向量中各灰度值的截断调节系数并汇总形成截断调节向量，及根据所述截断调节向量调节所述输入特征向量生成输出特征向量；
将所述输出特征向量发送预设的神经网络，所述神经网络根据输出特征向量对所述可导截断模型进行权值更新，使其生成符合所述神经网络损失函数的输出特征向量，并根据该输出特征向量生成窗宽窗位图像。
为实现上述目的，本申请还提供一种计算机可读存储介质，其包括多个存储介质，各存储介质上存储有计算机程序，所述多个存储介质存储的所述计算机程序被处理器执行时共同实现以下方法：
接收待调节图像，依次提取待调节图像中各像素的灰度值并汇总获得输入特征向量；
通过可导截断模型计算所述输入特征向量中各灰度值的截断调节系数并汇总形成截断调节向量，及根据所述截断调节向量调节所述输入特征向量生成输出特征向量；
将所述输出特征向量发送预设的神经网络，所述神经网络根据输出特征向量对所述可导截断模型进行权值更新，使其生成符合所述神经网络损失函数的输出特征向量，并根据该输出特征向量生成窗宽窗位图像。
本申请获取的窗宽窗位图像，因可导截断模型既符合了用户对窗宽窗位进行调节的需求，又因对可导截断函数的权重更新满足了神经网络对窗宽窗位进行处理或分类的要求，使用户在得到符合要求的窗宽窗位图像的同时，还保证了神经网络对窗宽窗位图像的处理精度。
附图说明
图1为本申请自适应窗宽窗位调节方法实施例一的流程图；
图2为本申请自适应窗宽窗位调节方法实施例一中窗宽125HU，床位45HU的图像；
图3为本申请自适应窗宽窗位调节方法实施例一中窗宽80HU,窗位40HU的图像；
图4为本申请自适应窗宽窗位调节方法实施例一S1中获得输入特征向量的流程图；
图5为本申请自适应窗宽窗位调节方法实施例一S2中形成截断调节向量的流程图；
图6为本申请自适应窗宽窗位调节方法实施例一S2中生成输出特征向量的流程图；
图7为本申请自适应窗宽窗位调节方法实施例一S3中生成符合所述神经网络损失函数的输出特征向量的流程图；
图8为本申请自适应窗宽窗位调节装置实施例二的程序模块示意图；
图9为本申请计算机系统实施例三中计算机设备的硬件结构示意图。
附图标记：
1、自适应窗宽窗位调节装置 2、计算机设备 11、灰度提取模块
12、可导截断模块 13、图像生成模块 21、存储器 22、处理器
具体实施方式
为了使本申请的目的、技术方案及优点更加清楚明白，以下结合附图及实施例，对本申请进行进一步详细说明。应当理解，此处所描述的具体实施例仅用以解释本申请，并不用于限定本申请。基于本申请中的实施例，本领域普通技术人员在没有做出创造性劳动前 提下所获得的所有其他实施例，都属于本申请保护的范围。
本申请的技术方案可应用于人工智能、区块链和/或大数据技术领域，如可具体涉及神经网络技术。可选的，本申请涉及的数据如系数、特征向量、权值和/或窗宽窗位图像等可存储于数据库中，或者可以存储于区块链中，比如通过区块链分布式存储，本申请不做限定。
本申请提供的自适应窗宽窗位调节方法、装置、计算机系统及存储介质，适用于人工智能的图像检测技术领域，为提供一种基于灰度提取模块、可导截断模块、图像生成模块的自适应窗宽窗位调节方法。本申请通过提取待调节图像中各像素的灰度值并汇总获得输入特征向量；通过可导截断模型计算所述输入特征向量中各灰度值的截断调节系数并汇总形成截断调节向量，及根据所述截断调节向量调节所述输入特征向量生成输出特征向量；将所述输出特征向量发送预设的神经网络，所述神经网络根据输出特征向量对所述可导截断模型进行权值更新，使其生成符合所述神经网络损失函数的输出特征向量，并根据该输出特征向量生成窗宽窗位图像。
实施例一：
请参阅图1，本实施例的一种基于梯度反传的自适应窗宽窗位调节方法，包括：
S1：接收待调节图像，依次提取待调节图像中各像素的灰度值并汇总获得输入特征向量。
S2：通过可导截断模型计算所述输入特征向量中各灰度值的截断调节系数并汇总形成截断调节向量，及根据所述截断调节向量调节所述输入特征向量生成输出特征向量。
S3：将所述输出特征向量发送预设的神经网络，所述神经网络根据输出特征向量对所述可导截断模型进行权值更新，使其生成符合所述神经网络损失函数的输出特征向量，并根据该输出特征向量生成窗宽窗位图像。
在示例性的实施例中，接收从用户端发送或从数据库中提取的待调节图像，依次提取待调节图像中各像素的灰度值，按照各像素的位置排列所述灰度值并汇总形成输入特征向量。
通过设有可导截断函数的可导截断模型，依次计算所述输入特征向量中各灰度值的截断调节系数以获得截断调节向量，及根据所述截断调节向量调节所述输入特征向量中各像素的灰度值以生成输出特征向量；通过前向传播算法将所述输出特征向量输出至神经网络。
所述神经网络根据所述输出特征向量并通过反向传播算法，对所述可导截断模型中可导截断函数的权值更新；通过更新后的可导截断模型再次计算所述输入特征向量获得截断调节向量，及根据所述截断调节向量调节所述输入特征向量生成输出特征向量，直至所述可导截断模型生成的输出特征向量，不再被神经网络通过反向传播算法对所述可导截断模型中可导截断函数进行权值更新时，根据该输出特征向量生成窗宽窗位图像。此时的窗宽窗位图像，因可导截断模型既符合了用户对窗宽窗位进行调节的需求，又因对可导截断函数的权重更新满足了神经网络对窗宽窗位进行处理或分类的要求，使用户在得到符合要求的窗宽窗位图像的同时，还保证了神经网络对窗宽窗位图像的处理精度。
本申请可应用于智慧医疗场景中，从而推动智慧城市的建设。
需要说明的是，所述神经网络是用于对窗宽窗位图像进行进一步处理或加工的网络模型，其类型和数量可根据用户的需要设置；众所周知，任何的神经网络都需要通过前向传播算法对接收到的数据进行前向传播，再通过反向传播算法对神经网络的每个层进行权重和/或偏执值更新，属于现有技术；本申请所解决的问题是如何使窗宽窗位图像能够满足神经网络在数据处理上的要求；本申请所采取的手段是通过设置可导截断模型使待调节图像到窗宽窗位图像的全过程变成可导的计算过程，当神经网络基于损失函数使用梯度下降法进行反向传播时，可对可导截断模型进行权重和/或偏执值更新，使可导截断模型成为了神 经网络中的一部分并能够很好的与神经网络的每一层匹配，因此从根源上，也就是从输入神经网络的输出向量特征上保证了与神经网络各层的匹配，进而实现了上述的保证了神经网络对窗宽窗位图像的处理精度。由于神经网络的工作原理属于现有技术，因此在此不做赘述。
本申请中的窗宽指CT图像所显示的CT值范围，CT值是测定人体某一局部组织或器官密度大小的一种计量单位，通常称亨氏单位(hounsfield unit，HU)，例如：空气为-1000(因为颜色几乎是无)，致密骨为+1000(因为颜色特别重)；窗位是指：指窗宽范围内均值或中心值，其是对图像进行均衡化处理的重要指标,比如一幅CT图像，窗宽为100Hu，窗位选在0Hu；则以窗位为中心(0Hu)，向上包括+50Hu，向下包括-50Hu，凡是在这个100Hu范围内的组织均可显示出来并为人眼所识别。凡是大于+50Hu的组织均为白色；凡是小于-50Hu的组织均为黑色，其密度差异无法显示。
因此，窗宽窗位的调整，是改变整张图像各像素的色差，以突出病灶部位，减轻无关部位，例如：对于同一张脑部CT图，图2是窗宽125HU，床位45HU的图像，图3是窗宽80HU,窗位40HU的图像。
在一个优选的实施例中，请参阅图4，S1中依次所述提取待调节图像中各像素的灰度值并汇总获得输入特征向量的步骤，包括：
S101：提取所述待调节图像中各像素的灰度值。
本步骤中，可通过OpenCV提取待调节图像中各像素的灰度值；所述OpenCV是一种跨平台计算机视觉库，可以运行在Linux、Windows、Android和Mac OS操作系统上。它轻量级而且高效--由一系列C函数和少量C++类构成，同时提供了Python、Ruby、MATLAB等语言的接口，实现了图像处理和计算机视觉方面的很多通用算法。
S102：按照各像素在待调节图像中的位置，排列所述各像素的灰度值并汇总形成输入特征向量。
本步骤中，可根据待调节图像中像素的排布建立像素矩阵，并将OpenCV提取的灰度值按照其像素在待调节图像中的位置(如：行数和列数)，将所述灰度值放置在所述像素矩阵中，直至OpenCV提取完毕；此时，所述像素矩阵将成为排列有各像素的灰度值并汇总形成输入特征向量。
在一个优选的实施例中，请参阅图5，S2中通过可导截断模型计算所述输入特征向量中各灰度值的截断调节系数并汇总形成截断调节向量的步骤，包括：
S201：创建机器学习模型以及具有可导截断函数的可导截断模型，通过所述机器学习模型预测所述输入特征向量的灰度上限和灰度下限；其中，所述可导截断函数具有可导特性且用于对待调节图像进行均衡化处理和截断处理。
本步骤中，可导截断函数，如下所示：
M(·)＝s(X-window min)-s(X-window max)
其中，s(x)＝1/(1+e -kx)，于所述可导截断函数中，k为超参数，用来调节函数的有效激活区间，x为输入特征向量中的灰度值，s(x)为截断调节系数，windowmin为窗宽窗位的灰度下限，windowmax为窗宽窗位的灰度上限，x为输入特征向量中的灰度值，M(·)为截断调节系数，
需要说明的是，通过Sigmoid-k这种连续可导的函数，实现了可导截断函数的可导特性；使得神经网络的梯度反传可对可导截断函数进行求导，进而实现对可导截断函数的权值更新，以使可导截断模型能够生成能够满足神经网络对的处理或分类要求的窗宽窗位图像。
具体地，S201中通过所述机器学习模型预测所述输入特征向量的灰度上限和灰度下限的步骤，包括：
将所述输入特征向量输出至所述机器学习模型；
接收所述机器学习模型对输入特征向量进行运算所生成的，符合神经网络数据处理要求的灰度上限和灰度下限。
其中，可采用训练所述神经网络所使用的训练样本，训练所述机器学习模型，使所述机器学习模型在通过梯度下降法和反向传播算法对其自身权值的更新，符合神经网络对数据处理的要求。
需要说明的是，所述机器学习模型是一种经过训练并收敛所获得的计算机模型，所述机器学习模型可与所述神经网络使用同样的训练样本，使所述机器学习模型对灰度上限和灰度下限的识别能够符合神经网络对数据分类、数据分割、图像识别等数据处理需求，进而使所述机器学习模型可对输入特征向量进行分析，识别待调节图像特有的，且，适合后续图像任务(即：所述神经网络的数据处理任务)的最佳窗宽窗位，于本实施例中，所述最佳窗宽窗位是指所述灰度上限和灰度下限。其中，所述训练收敛是通过梯度下降法和反向传播算法对神经网络的每个层进行权值调节更新，使其收敛并符合训练预期(如：神经网络的损失值小于预设的阈值)的过程；因此，经过所述神经网络的运算所获得的灰度上限和灰度下限是具有梯度反传特性(即：同时符合机器学习模型和神经网络的训练收敛要求)的，故，所述灰度上限和灰度下限符合神经网络的数据处理需求。
对机器学习模型的训练，以及通过机器学习模型分析图像获得灰度上限和灰度下限属于现有技术，而本申请所解决的技术问题是：如何获得具有符合神经网络的梯度反传特性的灰度上限和灰度下限，以及如何将所述灰度上限与灰度下限与待调节图像均衡化操作关联，使经过均衡化操作及截断操作所获得的窗宽窗位图像，满足神经网络在数据处理上的要求；因此，关于机器模型的训练及其分析图像获得灰度上限和灰度下限的过程在此不做赘述。
S202：所述可导截断模型根据所述灰度下限和灰度上限，计算所述输入特征向量中各灰度值获得各灰度值的下限差值和上限差值。
本步骤中，预设的灰度下限和灰度上限windowmin和windowmax，以获得下限差值和上限差值X-windowmin和X-windowmax。
通过灰度下限和灰度上限对待调节图像进行灰度限定以得到下限差值和上限差值，实现了对待调节图像的截断效果；其中，下限差值是指待调节图像的某一像素与灰度下限之间的差值，而上限差值是指该像素与灰度上限之间的差值。
S203：对所述下限差值和上限差值进行可导均衡化处理，获得可导均衡下差值和可导均衡上差值，将所述可导均衡下差值与可导均衡上差值相减获得截断调节系数。
示例性地，将下限差值代入Sigmoid-k函数对其进行逻辑回归，使每个离散化的下限差值都映射到范围为(0,1)范围内，并获得与所述下限差值对应的可导均衡下差值；因Sigmoid-k函数的可导特性，使得回归后所获得的可导均衡下差值具有可导特性；同时，又因Sigmoid-k函数在其定义域内为连续和光滑的函数，因此，对下限差值进行了连续光滑的均衡化处理，并获得了与下限差值对应的可导均衡下差值。
将上限差值代入Sigmoid-k函数对其进行逻辑回归，使每个离散化的上限差值都映射到范围为(0,1)范围内，并获得与所述上限差值对应的可导均衡上差值；因Sigmoid-k函数的可导特性，使得回归后所获得的可导均衡上差值具有可导特性；同时，又因Sigmoid-k函数在其定义域内为连续和光滑的函数，因此，对上限差值进行了连续光滑的均衡化处理，并获得了与上限差值对应的可导均衡上差值。
将所述可导均衡下差值与可导均衡上差值相减所获得的截断调节系数，用于对待调节图像中各像素的灰度值进行计算，以实现对待调节图像进行截断和均衡化处理。
S204：各截断调节系数按照与其对应的像素位置进行排列，以汇总形成截断调节向量。
在一个优选的实施例中，请参阅图6，S2中根据所述截断调节向量调节所述输入特征向量生成输出特征向量的步骤，包括：
S211：将截断调节向量中各截断调节系数与所述输入特征向量中各像素的灰度值一一对应。
S212：根据所述截断调节系数调整与其对应的灰度值获得初步输出特征向量；
本步骤中，可将相互对应的截断调节系数和灰度值相乘获得输出特征值。
示例性地，可通过以下公式计算输入特征向量中各灰度值所对应的输出特征值；
Y＝X⊙M(·)
其中，X为输入特征向量，Y为初步特征向量，M(·)为截断调节向量。
S213：依次判断所述初步输出特征向量中输出特征值是否大于预设的截断阈值；若大于所述截断阈值，则将所述输出特征值设为保留特征值；若不大于所述截断阈值，则将所述输出特征值进行极值化处理并将其设为极值特征值。
本步骤中，大于截断阈值的输出特征值被视为应当在“观察窗”中显示的像素，而不大于截断阈值的输出特征值则被视为不应当在“观察窗”中显示的像素。于本实施例中，所述截断阈值可设为0。
进一步地，S213中极值化处理包括以下步骤：
判断所述初步输出特征向量的输出特征值是否小于灰度下限，若小于所述灰度下限，则将所述输出特征值调整为所述灰度下限；若不小于所述灰度下限，则判断所述输出特征值是否大于灰度上限，若大于所述灰度上限，则将所述输出特征值调整为所述灰度上限。
本步骤中，所述下灰度极限和上灰度极限可根据需要设置；例如：待调节图像的灰度区间为【-1024,1024】；若灰度下限为0，灰度上限为80；若输出特征值为-20，其小于灰度下限0，因此，将该输出特征值调整为0；若输出特征值为100，其大于灰度上限80，因此将该输出特征值调整为80。
S214：将完成对所述初步输出特征向量中所有输出特征值的截断阈值判断，且仅具有保留特征值和/或极值特征值的初步输出特征向量设为输出特征向量。
在一个优选的实施例中，请参阅图7，S3中将所述输出特征向量发送预设的神经网络，所述神经网络根据输出特征向量对所述可导截断模型进行权值更新，使其生成符合所述神经网络损失函数的输出特征向量的步骤包括：
S301：通过前向传播算法将所述输出特征向量输出至神经网络，所述神经网络中每一层依次对所述输出特征向量进行运算获得运算结果，及通过所述神经网络的损失函数计算所述运算结果获得损失值。
本步骤中，所述损失函数(loss function)是用来估量模型的预测值f(x)与真实值Y的不一致程度，它是一个非负实值函数,通常使用L(Y,f(x))来表示，损失函数越小，模型的鲁棒性就越好；神经网络可通过损失函数计算运算结果和真实结果(如：分类期望结果、分割期望结果)之间的差值，生成损失值。
S302：接收所述神经网络根据所述损失值，并通过梯度下降法及反向传播算法输出的迭代数据，根据所述迭代数据对所述可导截断模型中的可导截断函数进行求导，并根据求导结果对所述可导截断模型进行权值更新。
本步骤中，梯度下降法是在机器学习算法中，以最小化损失函数为目标时，用于一步步迭代求解得到最小化的损失函数及其神经网络的参数值的算法。反向传播算法是借助链式法则，计算两个或两个以上复合函数的导数，将输出单元的梯度反向传播回输入单元，根据计算出的梯度，调整网络的可学习参数的算法；因此，本步骤是神经网络通过反向传播算法并利用梯度下降法，计算神经网络中各层的参数值的方法；当传播至神经网络的输入层时，由于可导截断函数的可导特性，神经网络仍然可通过梯度下降法及反向传播算法 向可导截断模型传播迭代数据，以对可导截断函数进行求导，并根据求导结果对所述可导截断模型进行权值更新，即调节可导截断函数中Sigmoid-k函数的k值，使可导截断模型成为神经网络在反向传播过程中的一部分，使其随着神经网络的需求而进行权值更新。
由于反向传播算法和梯度下降法属于现有技术，而本申请所解决的技术问题是，如何能够接受神经网络对可导截断函数的权值更新，使可导截断模块生成的输出特征向量符合神经网络的数据处理需求的问题；因此，反向传播算法和梯度下降法的工作原理在此不做赘述。
S303：通过更新后的可导截断模型，再次计算所述输入特征向量生成输出特征向量并将其发送所述神经网络，直至所述神经网络的损失值达到预设的损失阈值以下，判定所述损失值达到预设的损失阈值以下的输出特征向量，为符合所述神经网络损失函数的输出特征向量。
本步骤中，当神经网络的损失值达到预设的损失阈值以下时，则说明了输出特征向量符合了神经网络对于数据处理的要求，因此将不再对可导截断模块进行权值更新，并将该输出特征向量最为最终可用的输出特征向量。
同时，神经网络还可不对其自身的各层进行权值更新，而直接通过梯度下降法及反向传播算法对可导截断模型进行权值更新，使其生成的输出特征向量符合神经网络损失函数的要求。
需要说明的是，损失函数、通过损失函数计算运算结果获得损失值、以及根据损失值对函数进行求导及权值更新属于现有技术，而本申请所解决的技术问题是如何：使生成的输出特征向量及根据该特征向量所获得的窗宽窗位图像，满足神经网络的数据处理需求，因此损失函数、通过损失函数计算运算结果获得损失值、以及根据损失值对函数进行求导及权值更新在此不做赘述。
在示例性的实施例中，S3中根据该输出特征向量生成窗宽窗位图像的步骤包括：
根据所述输出特征向量中的保留特征值或极值特征值，调整待调节图像中各像素的灰度值，使所述待调节图像转为各像素的灰度值，分别与保留特征值或极值特征值一一对应的窗宽窗位图像；
将所述窗宽窗位图像上传至区块链。
需要说明的是，基于窗宽窗位图像得到对应的摘要信息，具体来说，摘要信息由窗宽窗位图像进行散列处理得到，比如利用sha256s算法处理得到。将摘要信息上传至区块链可保证其安全性和对用户的公正透明性。用户设备可以从区块链中下载得该摘要信息，以便查证整理表和/或数据表是否被篡改。
本示例所指区块链是分布式数据存储、点对点传输、共识机制、加密算法等计算机技术的新型应用模式。区块链(Blockchain)，本质上是一个去中心化的数据库，是一串使用密码学方法相关联产生的数据块，每一个数据块中包含了一批次网络交易的信息，用于验证其信息的有效性(防伪)和生成下一个区块。区块链可以包括区块链底层平台、平台产品服务层以及应用服务层等。
实施例二：
请参阅图8，本实施例的一种基于梯度反传的自适应窗宽窗位调节装置1，包括：
灰度提取模块11，用于接收待调节图像，依次提取待调节图像中各像素的灰度值并汇总获得输入特征向量；
可导截断模块12，用于通过可导截断模型计算所述输入特征向量中各灰度值的截断调节系数并汇总形成截断调节向量，及根据所述截断调节向量调节所述输入特征向量生成输出特征向量；
图像生成模块13，用于将所述输出特征向量发送预设的神经网络，所述神经网络根据 输出特征向量对所述可导截断模型进行权值更新，使其生成符合所述神经网络损失函数的输出特征向量，并根据该输出特征向量生成窗宽窗位图像。
本技术方案基于人工智能领域的图像检测，通过依次提取待调节图像中各像素的灰度值并汇总获得输入特征向量；通过可导截断模型计算所述输入特征向量中各灰度值的截断调节系数并汇总形成截断调节向量，及根据所述截断调节向量调节所述输入特征向量生成输出特征向量；将所述输出特征向量发送预设的神经网络，所述神经网络根据输出特征向量对所述可导截断模型进行权值更新，使其生成符合所述神经网络损失函数的输出特征向量，并根据该输出特征向量生成窗宽窗位图像，实现对窗宽窗位图像的灰度化图像处理。
实施例三：
为实现上述目的，本申请还提供一种计算机系统，该计算机系统包括多个计算机设备2，实施例二的自适应窗宽窗位调节装置1的组成部分可分散于不同的计算机设备中，计算机设备可以是执行程序的智能手机、平板电脑、笔记本电脑、台式计算机、机架式服务器、刀片式服务器、塔式服务器或机柜式服务器(包括独立的服务器，或者多个服务器所组成的服务器集群)等。本实施例的计算机设备至少包括但不限于：可通过系统总线相互通信连接的存储器21、处理器22，如图9所示。需要指出的是，图9仅示出了具有组件-的计算机设备，但是应理解的是，并不要求实施所有示出的组件，可以替代的实施更多或者更少的组件。
本实施例中，存储器21(即可读存储介质)包括闪存、硬盘、多媒体卡、卡型存储器(例如，SD或DX存储器等)、随机访问存储器(RAM)、静态随机访问存储器(SRAM)、只读存储器(ROM)、电可擦除可编程只读存储器(EEPROM)、可编程只读存储器(PROM)、磁性存储器、磁盘、光盘等。在一些实施例中，存储器21可以是计算机设备的内部存储单元，例如该计算机设备的硬盘或内存。在另一些实施例中，存储器21也可以是计算机设备的外部存储设备，例如该计算机设备上配备的插接式硬盘，智能存储卡(Smart Media Card,SMC)，安全数字(Secure Digital,SD)卡，闪存卡(Flash Card)等。当然，存储器21还可以既包括计算机设备的内部存储单元也包括其外部存储设备。本实施例中，存储器21通常用于存储安装于计算机设备的操作系统和各类应用软件，例如实施例一的自适应窗宽窗位调节装置的程序代码等。此外，存储器21还可以用于暂时地存储已经输出或者将要输出的各类数据。
处理器22在一些实施例中可以是中央处理器(Central Processing Unit，CPU)、控制器、微控制器、微处理器、或其他数据处理芯片。该处理器22通常用于控制计算机设备的总体操作。本实施例中，处理器22用于运行存储器21中存储的程序代码或者处理数据，例如运行自适应窗宽窗位调节装置，以实现实施例一的自适应窗宽窗位调节方法。
实施例四：
为实现上述目的，本申请还提供一种计算机可读存储系统(计算机可读存储介质)，其包括多个存储介质，如闪存、硬盘、多媒体卡、卡型存储器(例如，SD或DX存储器等)、随机访问存储器(RAM)、静态随机访问存储器(SRAM)、只读存储器(ROM)、电可擦除可编程只读存储器(EEPROM)、可编程只读存储器(PROM)、磁性存储器、磁盘、光盘、服务器、App应用商城等等，其上存储有计算机程序，程序被处理器22执行时实现相应功能。本实施例的计算机可读存储介质用于存储自适应窗宽窗位调节装置，被处理器22执行时实现实施例一的自适应窗宽窗位调节方法。
可选的，本申请涉及的存储介质可以是非易失性的，也可以是易失性的。
上述本申请实施例序号仅仅为了描述，不代表实施例的优劣。
通过以上的实施方式的描述，本领域的技术人员可以清楚地了解到上述实施例方法可借助软件加必需的通用硬件平台的方式来实现，当然也可以通过硬件，但很多情况下前者 是更佳的实施方式。
以上仅为本申请的优选实施例，并非因此限制本申请的专利范围，凡是利用本申请说明书及附图内容所作的等效结构或等效流程变换，或直接或间接运用在其他相关的技术领域，均同理包括在本申请的专利保护范围内。

Claims (20)

  1. 一种基于梯度反传的自适应窗宽窗位调节方法，其中，包括：
    接收待调节图像，依次提取待调节图像中各像素的灰度值并汇总获得输入特征向量；
    通过可导截断模型计算所述输入特征向量中各灰度值的截断调节系数并汇总形成截断调节向量，及根据所述截断调节向量调节所述输入特征向量生成输出特征向量；
    将所述输出特征向量发送预设的神经网络，所述神经网络根据输出特征向量对所述可导截断模型进行权值更新，使其生成符合所述神经网络损失函数的输出特征向量，并根据该输出特征向量生成窗宽窗位图像。
  2. 根据权利要求1所述的自适应窗宽窗位调节方法，其中，依次提取待调节图像中各像素的灰度值并汇总获得输入特征向量的步骤，包括：
    提取所述待调节图像中各像素的灰度值；
    按照各像素在待调节图像中的位置，排列所述各像素的灰度值并汇总形成输入特征向量。
  3. 根据权利要求1所述的自适应窗宽窗位调节方法，其中，通过可导截断模型计算所述输入特征向量中各灰度值的截断调节系数并汇总形成截断调节向量的步骤，包括：
    创建机器学习模型以及具有可导截断函数的可导截断模型，通过所述机器学习模型预测所述输入特征向量的灰度上限和灰度下限；其中，所述可导截断函数具有可导特性且用于对待调节图像进行均衡化处理和截断处理；
    所述可导截断模型根据所述灰度下限和灰度上限，计算所述输入特征向量中各灰度值获得各灰度值的下限差值和上限差值；
    对所述下限差值和上限差值进行可导均衡化处理，获得可导均衡下差值和可导均衡上差值，将所述可导均衡下差值与可导均衡上差值相减获得截断调节系数；
    各截断调节系数按照与其对应的像素位置进行排列，以汇总形成截断调节向量。
  4. 根据权利要求1所述的自适应窗宽窗位调节方法，其中，根据所述截断调节向量调节所述输入特征向量生成输出特征向量的步骤，包括：
    将截断调节向量中各截断调节系数与所述输入特征向量中各像素的灰度值一一对应；
    根据所述截断调节系数调整与其对应的灰度值获得初步输出特征向量；
    依次判断所述初步输出特征向量中输出特征值是否大于预设的截断阈值；若大于所述截断阈值，则将所述输出特征值设为保留特征值；若不大于所述截断阈值，则将所述输出特征值进行极值化处理并将其设为极值特征值；
    将完成对所述初步输出特征向量中所有输出特征值的截断阈值判断，且仅具有保留特征值和/或极值特征值的初步输出特征向量设为输出特征向量。
  5. 根据权利要求4所述的自适应窗宽窗位调节方法，其中，所述极值化处理包括以下步骤：
    判断所述初步输出特征向量的输出特征值是否小于灰度下限，若小于所述灰度下限，则将所述输出特征值调整为所述灰度下限；若不小于所述灰度下限，则判断所述输出特征值是否大于灰度上限，若大于所述灰度上限，则将所述输出特征值调整为所述灰度上限。
  6. 根据权利要求1所述的自适应窗宽窗位调节方法，其中，将所述输出特征向量发送预设的神经网络，所述神经网络根据输出特征向量对所述可导截断模型进行权值更新，使其生成符合所述神经网络损失函数的输出特征向量的步骤，包括：
    通过前向传播算法将所述输出特征向量输出至神经网络，所述神经网络中每一层依次对所述输出特征向量进行运算获得运算结果，及通过所述神经网络的损失函数计算所述运算结果获得损失值；
    接收所述神经网络根据所述损失值，并通过梯度下降法及反向传播算法输出的迭代数 据，根据所述迭代数据对所述可导截断模型中的可导截断函数进行求导，并根据求导结果对所述可导截断模型进行权值更新；
    通过更新后的可导截断模型，再次计算所述输入特征向量生成输出特征向量并将其发送所述神经网络，直至所述神经网络的损失值达到预设的损失阈值以下，判定所述损失值达到预设的损失阈值以下的输出特征向量，为符合所述神经网络损失函数的输出特征向量。
  7. 根据权利要求1所述的自适应窗宽窗位调节方法，其中，根据该输出特征向量生成窗宽窗位图像的步骤，包括：
    根据所述输出特征向量中的保留特征值或极值特征值，调整待调节图像中各像素的灰度值，使所述待调节图像转为各像素的灰度值，分别与保留特征值或极值特征值一一对应的窗宽窗位图像；
    将所述窗宽窗位图像上传至区块链。
  8. 一种基于梯度反传的自适应窗宽窗位调节装置，其中，包括：
    灰度提取模块，用于接收待调节图像，依次提取待调节图像中各像素的灰度值并汇总获得输入特征向量；
    可导截断模块，用于通过可导截断模型计算所述输入特征向量中各灰度值的截断调节系数并汇总形成截断调节向量，及根据所述截断调节向量调节所述输入特征向量生成输出特征向量；
    图像生成模块，用于将所述输出特征向量发送预设的神经网络，所述神经网络根据输出特征向量对所述可导截断模型进行权值更新，使其生成符合所述神经网络损失函数的输出特征向量，并根据该输出特征向量生成窗宽窗位图像。
  9. 一种计算机系统，其包括多个计算机设备，各计算机设备包括存储器、处理器以及存储在存储器上并可在处理器上运行的计算机程序，其中，所述多个计算机设备的处理器执行所述计算机程序时共同实现以下方法：
    接收待调节图像，依次提取待调节图像中各像素的灰度值并汇总获得输入特征向量；
    通过可导截断模型计算所述输入特征向量中各灰度值的截断调节系数并汇总形成截断调节向量，及根据所述截断调节向量调节所述输入特征向量生成输出特征向量；
    将所述输出特征向量发送预设的神经网络，所述神经网络根据输出特征向量对所述可导截断模型进行权值更新，使其生成符合所述神经网络损失函数的输出特征向量，并根据该输出特征向量生成窗宽窗位图像。
  10. 根据权利要求9所述的计算机系统，其中，依次提取待调节图像中各像素的灰度值并汇总获得输入特征向量时，具体实现：
    提取所述待调节图像中各像素的灰度值；
    按照各像素在待调节图像中的位置，排列所述各像素的灰度值并汇总形成输入特征向量。
  11. 根据权利要求9所述的计算机系统，其中，通过可导截断模型计算所述输入特征向量中各灰度值的截断调节系数并汇总形成截断调节向量时，具体实现：
    创建机器学习模型以及具有可导截断函数的可导截断模型，通过所述机器学习模型预测所述输入特征向量的灰度上限和灰度下限；其中，所述可导截断函数具有可导特性且用于对待调节图像进行均衡化处理和截断处理；
    所述可导截断模型根据所述灰度下限和灰度上限，计算所述输入特征向量中各灰度值获得各灰度值的下限差值和上限差值；
    对所述下限差值和上限差值进行可导均衡化处理，获得可导均衡下差值和可导均衡上差值，将所述可导均衡下差值与可导均衡上差值相减获得截断调节系数；
    各截断调节系数按照与其对应的像素位置进行排列，以汇总形成截断调节向量。
  12. 根据权利要求9所述的计算机系统，其中，根据所述截断调节向量调节所述输入特征向量生成输出特征向量时，具体实现：
    将截断调节向量中各截断调节系数与所述输入特征向量中各像素的灰度值一一对应；
    根据所述截断调节系数调整与其对应的灰度值获得初步输出特征向量；
    依次判断所述初步输出特征向量中输出特征值是否大于预设的截断阈值；若大于所述截断阈值，则将所述输出特征值设为保留特征值；若不大于所述截断阈值，则将所述输出特征值进行极值化处理并将其设为极值特征值；
    将完成对所述初步输出特征向量中所有输出特征值的截断阈值判断，且仅具有保留特征值和/或极值特征值的初步输出特征向量设为输出特征向量。
  13. 根据权利要求9所述的计算机系统，其中，将所述输出特征向量发送预设的神经网络，所述神经网络根据输出特征向量对所述可导截断模型进行权值更新，使其生成符合所述神经网络损失函数的输出特征向量时，具体实现：
    通过前向传播算法将所述输出特征向量输出至神经网络，所述神经网络中每一层依次对所述输出特征向量进行运算获得运算结果，及通过所述神经网络的损失函数计算所述运算结果获得损失值；
    接收所述神经网络根据所述损失值，并通过梯度下降法及反向传播算法输出的迭代数据，根据所述迭代数据对所述可导截断模型中的可导截断函数进行求导，并根据求导结果对所述可导截断模型进行权值更新；
    通过更新后的可导截断模型，再次计算所述输入特征向量生成输出特征向量并将其发送所述神经网络，直至所述神经网络的损失值达到预设的损失阈值以下，判定所述损失值达到预设的损失阈值以下的输出特征向量，为符合所述神经网络损失函数的输出特征向量。
  14. 根据权利要求9所述的计算机系统，其中，根据该输出特征向量生成窗宽窗位图像时，具体实现：
    根据所述输出特征向量中的保留特征值或极值特征值，调整待调节图像中各像素的灰度值，使所述待调节图像转为各像素的灰度值，分别与保留特征值或极值特征值一一对应的窗宽窗位图像；
    将所述窗宽窗位图像上传至区块链。
  15. 一种计算机可读存储介质，其包括多个存储介质，各存储介质上存储有计算机程序，其中，所述多个存储介质存储的所述计算机程序被处理器执行时共同实现以下方法：
    接收待调节图像，依次提取待调节图像中各像素的灰度值并汇总获得输入特征向量；
    通过可导截断模型计算所述输入特征向量中各灰度值的截断调节系数并汇总形成截断调节向量，及根据所述截断调节向量调节所述输入特征向量生成输出特征向量；
    将所述输出特征向量发送预设的神经网络，所述神经网络根据输出特征向量对所述可导截断模型进行权值更新，使其生成符合所述神经网络损失函数的输出特征向量，并根据该输出特征向量生成窗宽窗位图像。
  16. 根据权利要求15所述的计算机可读存储介质，其中，依次提取待调节图像中各像素的灰度值并汇总获得输入特征向量时，具体实现：
    提取所述待调节图像中各像素的灰度值；
    按照各像素在待调节图像中的位置，排列所述各像素的灰度值并汇总形成输入特征向量。
  17. 根据权利要求15所述的计算机可读存储介质，其中，通过可导截断模型计算所述输入特征向量中各灰度值的截断调节系数并汇总形成截断调节向量时，具体实现：
    创建机器学习模型以及具有可导截断函数的可导截断模型，通过所述机器学习模型预测所述输入特征向量的灰度上限和灰度下限；其中，所述可导截断函数具有可导特性且用 于对待调节图像进行均衡化处理和截断处理；
    所述可导截断模型根据所述灰度下限和灰度上限，计算所述输入特征向量中各灰度值获得各灰度值的下限差值和上限差值；
    对所述下限差值和上限差值进行可导均衡化处理，获得可导均衡下差值和可导均衡上差值，将所述可导均衡下差值与可导均衡上差值相减获得截断调节系数；
    各截断调节系数按照与其对应的像素位置进行排列，以汇总形成截断调节向量。
  18. 根据权利要求15所述的计算机可读存储介质，其中，根据所述截断调节向量调节所述输入特征向量生成输出特征向量时，具体实现：
    将截断调节向量中各截断调节系数与所述输入特征向量中各像素的灰度值一一对应；
    根据所述截断调节系数调整与其对应的灰度值获得初步输出特征向量；
    依次判断所述初步输出特征向量中输出特征值是否大于预设的截断阈值；若大于所述截断阈值，则将所述输出特征值设为保留特征值；若不大于所述截断阈值，则将所述输出特征值进行极值化处理并将其设为极值特征值；
    将完成对所述初步输出特征向量中所有输出特征值的截断阈值判断，且仅具有保留特征值和/或极值特征值的初步输出特征向量设为输出特征向量。
  19. 根据权利要求15所述的计算机可读存储介质，其中，将所述输出特征向量发送预设的神经网络，所述神经网络根据输出特征向量对所述可导截断模型进行权值更新，使其生成符合所述神经网络损失函数的输出特征向量时，具体实现：
    通过前向传播算法将所述输出特征向量输出至神经网络，所述神经网络中每一层依次对所述输出特征向量进行运算获得运算结果，及通过所述神经网络的损失函数计算所述运算结果获得损失值；
    接收所述神经网络根据所述损失值，并通过梯度下降法及反向传播算法输出的迭代数据，根据所述迭代数据对所述可导截断模型中的可导截断函数进行求导，并根据求导结果对所述可导截断模型进行权值更新；
    通过更新后的可导截断模型，再次计算所述输入特征向量生成输出特征向量并将其发送所述神经网络，直至所述神经网络的损失值达到预设的损失阈值以下，判定所述损失值达到预设的损失阈值以下的输出特征向量，为符合所述神经网络损失函数的输出特征向量。
  20. 根据权利要求15所述的计算机可读存储介质，其中，根据该输出特征向量生成窗宽窗位图像时，具体实现：
    根据所述输出特征向量中的保留特征值或极值特征值，调整待调节图像中各像素的灰度值，使所述待调节图像转为各像素的灰度值，分别与保留特征值或极值特征值一一对应的窗宽窗位图像；
    将所述窗宽窗位图像上传至区块链。
PCT/CN2020/135076 2020-05-15 2020-12-10 自适应窗宽窗位调节方法、装置、计算机系统及存储介质 WO2021115356A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202010412059.7A CN111696164B (zh) 2020-05-15 2020-05-15 自适应窗宽窗位调节方法、装置、计算机系统及存储介质
CN202010412059.7 2020-05-15

Publications (1)

Publication Number Publication Date
WO2021115356A1 true WO2021115356A1 (zh) 2021-06-17

Family

ID=72477802

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2020/135076 WO2021115356A1 (zh) 2020-05-15 2020-12-10 自适应窗宽窗位调节方法、装置、计算机系统及存储介质

Country Status (2)

Country Link
CN (1) CN111696164B (zh)
WO (1) WO2021115356A1 (zh)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115482248A (zh) * 2022-09-22 2022-12-16 推想医疗科技股份有限公司 图像分割方法、装置、电子设备及存储介质
CN116109644A (zh) * 2023-04-14 2023-05-12 东莞市佳超五金科技有限公司 用于铜铝转接排的表面缺陷检测方法

Families Citing this family (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111696164B (zh) * 2020-05-15 2023-08-25 平安科技（深圳）有限公司 自适应窗宽窗位调节方法、装置、计算机系统及存储介质
CN112529818B (zh) * 2020-12-25 2022-03-29 万里云医疗信息科技（北京）有限公司 基于神经网络的骨影抑制方法、装置、设备及存储介质
CN112686874A (zh) * 2020-12-31 2021-04-20 苏州朗润医疗系统有限公司 一种基于卷积神经网络的mri自适应窗宽窗位提取方法
CN112700445B (zh) * 2021-03-23 2021-06-29 上海市东方医院（同济大学附属东方医院） 图像处理方法、装置及系统
CN113177948B (zh) * 2021-04-15 2023-02-03 中国海洋大学 一种基于语义分割的医学图像自动窗口调节方法及应用
CN114937019B (zh) * 2022-05-30 2022-12-23 杭州健培科技有限公司 基于自适应局部灰度均衡的关键点检测方法、装置及应用

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US6175643B1 (en) * 1997-12-18 2001-01-16 Siemens Corporate Research, Inc. Neural network based auto-windowing system for MR images
CN102855617A (zh) * 2011-07-01 2013-01-02 中国科学院深圳先进技术研究院 自适应图像处理方法及系统
CN103310227A (zh) * 2012-03-16 2013-09-18 上海联影医疗科技有限公司 基于神经网络的自动窗宽窗位提取方法
US20180165551A1 (en) * 2016-12-08 2018-06-14 Intel Corporation Technologies for improved object detection accuracy with multi-scale representation and training
CN109273074A (zh) * 2018-11-27 2019-01-25 数坤（北京）网络科技有限公司 一种用于医疗图像的网络模型调整方法及设备
CN111696164A (zh) * 2020-05-15 2020-09-22 平安科技（深圳）有限公司 自适应窗宽窗位调节方法、装置、计算机系统及存储介质

Family Cites Families (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104323789B (zh) * 2014-09-25 2017-08-25 沈阳东软医疗系统有限公司 一种ct扫描图像重建方法及ct扫描仪
WO2017013514A1 (en) * 2015-07-23 2017-01-26 Koninklijke Philips N.V. Computed tomography visualization adjustment
CN105184759B (zh) * 2015-09-22 2018-02-27 中国科学院西安光学精密机械研究所 一种基于直方图紧致性变换的图像自适应增强方法
CN108537794B (zh) * 2018-04-19 2021-09-21 上海联影医疗科技股份有限公司 医学图像数据处理方法、装置和计算机可读存储介质

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US6175643B1 (en) * 1997-12-18 2001-01-16 Siemens Corporate Research, Inc. Neural network based auto-windowing system for MR images
CN102855617A (zh) * 2011-07-01 2013-01-02 中国科学院深圳先进技术研究院 自适应图像处理方法及系统
CN103310227A (zh) * 2012-03-16 2013-09-18 上海联影医疗科技有限公司 基于神经网络的自动窗宽窗位提取方法
US20180165551A1 (en) * 2016-12-08 2018-06-14 Intel Corporation Technologies for improved object detection accuracy with multi-scale representation and training
CN109273074A (zh) * 2018-11-27 2019-01-25 数坤（北京）网络科技有限公司 一种用于医疗图像的网络模型调整方法及设备
CN111696164A (zh) * 2020-05-15 2020-09-22 平安科技（深圳）有限公司 自适应窗宽窗位调节方法、装置、计算机系统及存储介质

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
LEI LV, XUN-JIE ZHAO: "Auto-window Algorithm for 16-bit Grayscale Image", ELECTRO-OPTIC TECHNOLOGY APPLICATION, vol. 31, no. 4, 1 August 2016 (2016-08-01), pages 27 - 30+45, XP055821118 *

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115482248A (zh) * 2022-09-22 2022-12-16 推想医疗科技股份有限公司 图像分割方法、装置、电子设备及存储介质
CN115482248B (zh) * 2022-09-22 2023-12-08 推想医疗科技股份有限公司 图像分割方法、装置、电子设备及存储介质
CN116109644A (zh) * 2023-04-14 2023-05-12 东莞市佳超五金科技有限公司 用于铜铝转接排的表面缺陷检测方法
CN116109644B (zh) * 2023-04-14 2023-06-16 东莞市佳超五金科技有限公司 用于铜铝转接排的表面缺陷检测方法

Also Published As

Publication number Publication date
CN111696164B (zh) 2023-08-25
CN111696164A (zh) 2020-09-22

Similar Documents

Publication Publication Date Title
WO2021115356A1 (zh) 自适应窗宽窗位调节方法、装置、计算机系统及存储介质
US11487995B2 (en) Method and apparatus for determining image quality
WO2021155650A1 (zh) 图片识别模型的训练方法、装置、计算机系统及存储介质
US20210174264A1 (en) Training tree-based machine-learning modeling algorithms for predicting outputs and generating explanatory data
CN111754596B (zh) 编辑模型生成、人脸图像编辑方法、装置、设备及介质
US10719693B2 (en) Method and apparatus for outputting information of object relationship
EP3779774A1 (en) Training method for image semantic segmentation model and server
WO2018166114A1 (zh) 图片识别的方法、系统、电子装置及介质
CN107507153B (zh) 图像去噪方法和装置
CN112381782B (zh) 人脸图像质量评估方法、装置、计算机设备及存储介质
CN112270686B (zh) 图像分割模型训练、图像分割方法、装置及电子设备
WO2021135449A1 (zh) 基于深度强化学习的数据分类方法、装置、设备及介质
EP4322056A1 (en) Model training method and apparatus
WO2018120723A1 (zh) 视频压缩感知重构方法、系统、电子装置及存储介质
CN113435583A (zh) 基于联邦学习的对抗生成网络模型训练方法及其相关设备
CN115050064A (zh) 人脸活体检测方法、装置、设备及介质
US11756288B2 (en) Image processing method and apparatus, electronic device and storage medium
CN112668482A (zh) 人脸识别训练方法、装置、计算机设备及存储介质
CN111382791B (zh) 深度学习任务处理方法、图像识别任务处理方法和装置
CN113627361B (zh) 人脸识别模型的训练方法、装置及计算机程序产品
CN113469091B (zh) 人脸识别方法、训练方法、电子设备及存储介质
CN109614780B (zh) 生物信息认证方法及装置、存储介质、电子设备
US9679219B2 (en) Image feature classification
CN109101984B (zh) 一种基于卷积神经网络的图像识别方法及装置
CN112801107A (zh) 一种图像分割方法和电子设备

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 20898949

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 20898949

Country of ref document: EP

Kind code of ref document: A1