WO2020224118A1 - 基于图片转换的病灶判断方法、装置、计算机设备 - Google Patents

基于图片转换的病灶判断方法、装置、计算机设备 Download PDF

Info

Publication number
WO2020224118A1
WO2020224118A1 PCT/CN2019/103337 CN2019103337W WO2020224118A1 WO 2020224118 A1 WO2020224118 A1 WO 2020224118A1 CN 2019103337 W CN2019103337 W CN 2019103337W WO 2020224118 A1 WO2020224118 A1 WO 2020224118A1
Authority
WO
WIPO (PCT)
Prior art keywords
picture
model
conversion
value calculation
confidence value
Prior art date
Application number
PCT/CN2019/103337
Other languages
English (en)
French (fr)
Inventor
杨苏辉
高鹏
谢国彤
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
Priority to JP2021500419A priority Critical patent/JP7064050B2/ja
Publication of WO2020224118A1 publication Critical patent/WO2020224118A1/zh

Links

Images

Classifications

    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00Pattern recognition
    • G06F18/20Analysing
    • G06F18/22Matching criteria, e.g. proximity measures
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06TIMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T3/00Geometric image transformations in the plane of the image
    • G06T3/04Context-preserving transformations, e.g. by using an importance map
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06VIMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V2201/00Indexing scheme relating to image or video recognition or understanding
    • G06V2201/03Recognition of patterns in medical or anatomical images

Definitions

  • This application relates to the field of computer technology, and in particular to a method, device, and computer equipment for judging a lesion based on image conversion.
  • the embodiments of the present application provide a method, device, computer equipment, and storage medium for judging a lesion based on image conversion, aiming to solve the problem in the prior art method that cannot accurately determine whether there is a lesion in a collected picture. problem.
  • an embodiment of the present application provides a method for judging a lesion based on image conversion, which includes:
  • the first picture conversion model and the second picture conversion model are respectively constructed according to the preset conversion template, wherein the first picture conversion model is used to convert the picture collected by the first collection device to the picture collected by the second collection device Pictures with matching styles, the second picture conversion model is used to convert pictures collected by the second collection device into pictures matching the style of the pictures collected by the first collection device;
  • the first confidence value calculation model and the second confidence value calculation model are generated according to the preset calculation template, wherein the first confidence value calculation model is used to compare the picture input to the first confidence value calculation model and the second confidence value calculation model.
  • the similarity of the styles between the pictures collected by the collection device is quantified, and the second confidence value calculation model is used to compare the styles between the pictures input to the second confidence value calculation model and the pictures collected by the first collection device. Similarity is quantified;
  • the acquisition device of the picture to be judged is not the first acquisition device, according to preset image judgment rules and the trained second picture conversion model, judge whether the picture to be judged contains lesions to obtain The result of lesion judgment.
  • an embodiment of the present application provides a device for judging a lesion based on image conversion, which includes:
  • the conversion model construction unit is used to construct a first picture conversion model and a second picture conversion model respectively according to a preset conversion template, wherein the first picture conversion model is used to convert the pictures collected by the first collection device into A picture that matches the style of the picture collected by the second collection device, and the second picture conversion model is used to convert the picture collected by the second collection device into a picture that matches the style of the picture collected by the first collection device;
  • the calculation model generation unit is configured to generate a first confidence value calculation model and a second confidence value calculation model according to a preset calculation template, wherein the first confidence value calculation model is used to input the first confidence value calculation model The style similarity between the picture and the picture collected by the second collection device is quantified, and the second confidence value calculation model is used to compare the picture input to the second confidence value calculation model with the first collection device Quantify the style similarity between the collected pictures;
  • the conversion model training unit is configured to combine the first confidence value calculation model, the second confidence value calculation model, and the first picture conversion model to compare the first confidence value calculation model according to preset model training rules and a preset picture library
  • the second picture conversion model is trained to obtain the second picture conversion model after training;
  • a collecting device determining unit configured to determine whether the collecting device of the picture to be judged is the first collecting device according to the collection source information of the picture to be judged if the picture to be judged input by the user is received;
  • the lesion judgment result acquisition unit is configured to: if the acquisition device of the picture to be judged is not the first acquisition device, compare the picture to be judged according to a preset image judgment rule and the trained second picture conversion model Determine whether the lesion is included in the file to obtain the result of the lesion judgment.
  • an embodiment of the present application provides a computer device, which includes a memory, a processor, and a computer program stored on the memory and running on the processor, and the processor executes the computer
  • the program implements the lesion judgment method based on image conversion described in the first aspect.
  • the embodiments of the present application also provide a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the processor executes the above-mentioned On the one hand, the lesion judgment method based on image conversion.
  • the embodiments of the present application provide a method, device, computer equipment, and storage medium for judging a lesion based on image conversion, which can make the target image obtained by conversion completely match the style of another acquisition device, and improve the efficiency and quality of image conversion. , Thereby greatly increasing the accuracy of judging lesions, and achieved good technical effects in the actual application process.
  • FIG. 1 is a schematic flowchart of a method for judging a lesion based on image conversion according to an embodiment of the application
  • FIG. 2 is a schematic diagram of the effect of the method for judging a lesion based on image conversion provided by an embodiment of the application;
  • FIG. 3 is a schematic diagram of a sub-flow of the method for judging a lesion based on image conversion provided by an embodiment of the application;
  • FIG. 4 is a schematic diagram of another sub-flow of the method for judging a lesion based on image conversion provided by an embodiment of the application;
  • FIG. 5 is a schematic diagram of another sub-flow of the method for judging a lesion based on image conversion according to an embodiment of this application;
  • FIG. 6 is a schematic diagram of another sub-flow of the method for judging a lesion based on image conversion provided by an embodiment of the application;
  • FIG. 7 is a schematic block diagram of an apparatus for judging a lesion based on image conversion according to an embodiment of the application.
  • FIG. 8 is a schematic block diagram of subunits of the apparatus for determining a focus based on image conversion provided by an embodiment of the application;
  • FIG. 9 is a schematic block diagram of another subunit of the apparatus for judging a lesion based on image conversion according to an embodiment of the application.
  • FIG. 10 is a schematic block diagram of another subunit of the apparatus for judging a lesion based on image conversion according to an embodiment of the application;
  • FIG. 11 is a schematic block diagram of another subunit of the apparatus for judging a lesion based on image conversion according to an embodiment of the application;
  • FIG. 12 is a schematic block diagram of a computer device provided by an embodiment of the application.
  • FIG. 1 is a schematic flowchart of a method for judging a lesion based on image conversion provided by an embodiment of the present application.
  • the method for judging lesions based on image conversion is applied to a user terminal, and the method is executed by application software installed in the user terminal.
  • the user terminal is a terminal device used to execute the method for judging lesions based on image conversion to judge lesions , Such as desktop computers, laptops, tablets, or mobile phones.
  • the method includes steps S110 to S150.
  • S110 Construct a first picture conversion model and a second picture conversion model respectively according to a preset conversion template, wherein the first picture conversion model is used to convert the picture collected by the first collection device to the one collected by the second collection device. A picture matching the picture style is collected, and the second picture conversion model is used to convert the picture collected by the second collection device into a picture matching the style of the picture collected by the first collection device.
  • the first picture conversion model and the second picture conversion model are respectively constructed according to the preset conversion template.
  • the conversion template is a template composed of several convolutional layers and deconvolutional layers with steps.
  • the first image conversion can be constructed by converting the template, the first acquisition device, and the second acquisition device.
  • the first collection device and the second collection device are both devices used to collect pictures.
  • the pictures collected by different collection devices have different styles. For example, the above styles include but are not limited to the color distribution, brightness, and contrast in the picture. , Noise, etc.
  • the pictures matching the style of the first collection device can be converted into pictures matching the style of the second collection device.
  • the second picture conversion model can be compared with the first Second, the picture matching the style of the collection device is converted to a picture matching the style of the first collection device.
  • step S110 includes substeps S111 and S112.
  • S111 Construct and obtain a first picture conversion model according to the conversion template and the first format information of the first collection device, where the first format information is used to characterize the quality of the picture collected by the first collection device. format.
  • the first picture conversion model is constructed according to the conversion template and the first format information of the first collection device. Obtain the first format information of the first collection device. Pictures collected by different collection devices have different formats. The first format information is the format information of the picture collected by the first collection device, that is, the specific size of the picture information.
  • the first picture conversion model includes a scaling processing layer, two convolutional layers with a step size of 2 and two deconvolution layers with a step size of 0.5.
  • the scaling processing layer is to process the picture with the first format information. Scaling process to obtain a picture of the corresponding pixel size.
  • Each convolutional layer contains a convolution kernel. Each element of the convolution kernel corresponds to a weight coefficient and a deviation.
  • the same deconvolution layer also contains A deconvolution kernel, each element of the deconvolution kernel corresponds to a weight coefficient and a deviation.
  • the picture collected by the first device is received, the picture is first converted into a 256 ⁇ 256 pixel picture through the scaling processing layer, and the 256 ⁇ 256 pixel picture is converted into a multidimensional vector through convolution processing.
  • the convolution process converts the multi-dimensional vector into another 256 ⁇ 256 pixel picture.
  • a second picture conversion model is constructed according to the conversion template and the second format information of the second collection device.
  • the second format information is the format information of the picture collected by the second acquisition device, that is, the specific size information of the picture.
  • the second picture conversion model includes a zoom processing layer and two convolutional layers with a step size of 2. And two deconvolution layers with a step size of 0.5, where the scaling processing layer is to perform scaling processing on the picture of the second format information to obtain a picture of the corresponding pixel size, and each convolution layer contains a convolution kernel, Each element composing the convolution kernel corresponds to a weight coefficient and a deviation.
  • the same deconvolution layer also contains a deconvolution kernel. Each element composing the deconvolution kernel corresponds to a weight coefficient and a The amount of deviation.
  • S120 Generate a first confidence value calculation model and a second confidence value calculation model according to a preset calculation template, wherein the first confidence value calculation model is used to compare the picture input to the first confidence value calculation model and the The style similarity between the pictures collected by the second collection device is quantified, and the second confidence value calculation model is used to determine the difference between the picture input to the second confidence value calculation model and the picture collected by the first collection device The similarity of styles is quantified.
  • the first confidence value calculation model and the second confidence value calculation model are generated according to the preset calculation template. Copy the calculation template to obtain the first confidence value calculation model and the second confidence value calculation model.
  • Each calculation model includes two convolutional layers with a step length of 2, a fully connected layer, and an output node.
  • a calculation model can process the picture input to the calculation model and calculate the confidence value of the picture. Specifically, if a picture of a specific pixel size is input to a certain calculation model, the input picture is convolved through the calculation model to obtain a multi-dimensional vector, and each dimension in the multi-dimensional vector is a calculation model. Input node, the vector value of each dimension is also the input node value corresponding to the input node.
  • the fully connected layer contains a number of preset feature units. Each feature unit is associated with all input nodes and output nodes. The feature unit That is, it can be used to reflect the relationship between the multi-dimensional vector and the output node, and the feature unit value is the calculated value of the feature unit in the
  • the calculation model also includes formulas from all input nodes to all characteristic units and formulas from all characteristic units to output nodes.
  • the specific value range of the confidence value is [0,1].
  • the confidence value calculated by the first confidence value calculation model can be used to quantify the similarity between the input picture and the style of the picture collected by the second collection device. If the picture input to the first confidence value calculation model is a picture collected by the first acquisition device and converted by the first picture conversion model, the first confidence value calculation model may be used to calculate the picture converted by the first picture conversion model The similarity with the style of the picture collected by the second collecting device, the similarity can be represented by the confidence value.
  • the confidence value calculated by the first confidence value calculation model is 1, it indicates that the style of the input picture is similar to the picture collected by the second collection device; if the confidence value calculated by the first confidence value calculation model is 0, then It indicates that the style of the input picture is not similar to the picture collected by the second collection device.
  • the confidence value calculated by the second confidence value calculation model can be used to quantify the similarity between the input picture and the style of the picture collected by the first collection device. Specifically, if the second confidence value calculation model calculates the confidence A value of 1, indicates that the input picture is similar in style to the picture collected by the first collection device; if the second confidence value calculation model calculates a confidence value of 0, it indicates that the input picture is the same as the picture collected by the first collection device The styles are not similar.
  • the first confidence value calculation model and the second confidence value calculation model can also be trained through preset training data to calculate the first confidence value
  • the parameter values of the formulas in the model and the second confidence value calculation model are adjusted, so that the trained first confidence value calculation model and the second confidence value calculation model meet actual use requirements.
  • the picture library includes a first picture set and a second picture set.
  • the first picture set is a picture set composed of pictures collected by the first collection device
  • the second picture set is a picture set by the second collection device.
  • the first picture set contains multiple first pictures
  • the second picture set contains multiple second pictures.
  • the model training rule is the rule information used to train the first picture conversion model and the second picture conversion model.
  • the model training rules include loss functions and parameter adjustment rules.
  • step S130 includes sub-steps S131, S132, and S133.
  • the first picture is represented by a
  • the first picture is also a picture that matches the style of the first collection device
  • the second picture is represented by b
  • the second picture is also a picture that matches the style of the second collection device.
  • the first picture is converted by the first picture conversion model
  • the second picture is converted by the second picture conversion model
  • the second converted picture can be obtained.
  • the style matching degree of a collection device is higher and the style matching degree of the first conversion picture and the second collection device is higher.
  • the first picture conversion model and the second picture conversion model need to be trained, and the first picture can be compared with the above loss function.
  • the conversion model and the second picture conversion model are trained at the same time to increase the training speed.
  • L is the calculated training loss value
  • the first picture is represented by a
  • the second picture is represented by b
  • the first confidence value calculation model is represented by D_X
  • the second confidence value calculation model is represented by D_Y
  • the first image conversion model It is represented by G_X
  • the second picture conversion model is represented by G_Y.
  • the first converted picture is represented by G_X(a)
  • the second converted picture is represented by G_Y(b).
  • is the specific gravity value in the loss function.
  • the first picture conversion model it is also possible to use the first picture conversion model to convert the second picture after scaling the picture a and picture b, and the picture obtained by converting the second picture is represented by G_X(b), and the first picture is converted using the second picture conversion model
  • the obtained picture is represented by G_Y(a)
  • the confidence value of the first picture calculated by the first confidence value calculation model is represented by G_X(a)
  • the confidence value of the second picture calculated is represented by D_X(b).
  • the confidence value calculation model calculates the confidence value of the first picture using D_Y(a), and the calculated confidence value of the second picture uses D_Y(b).
  • 1 is the norm between picture G_X(b) and picture b.
  • the specific steps for calculating the norm are to convert picture G_X(b) and picture b into unit8 data type
  • the value, that is, the pixel value of each pixel in the picture is obtained.
  • the value of picture G_X(b) and the value of picture b can be subtracted to obtain the norm. The smaller the norm, the more similar the two pictures are.
  • the parameter value in the second picture conversion model is adjusted according to the parameter adjustment rule in the model training rule in combination with the training loss value to complete one training of the second picture conversion model.
  • the parameter update gradient value can be determined according to the parameter adjustment rules and the training loss value. The larger the training loss value, the larger the corresponding update gradient value, and the smaller the training loss value, the corresponding update gradient The smaller the value.
  • the parameter value in the second image conversion model is also the weight coefficient included in the convolution kernel and deconvolution kernel in the model.
  • the parameter adjustment rule also includes the adjustment direction. Combining the adjustment direction and updating the gradient value, the second The weight coefficients included in the picture conversion model are adjusted, that is, one training of the second picture conversion model is completed.
  • the second picture conversion model is iteratively trained through the above weight coefficient adjustment method, and the number of iterative training can be set by the user. After iterative training, the trained second picture conversion model can be finally obtained.
  • S140 If a picture to be judged input by the user is received, determine whether the collecting device of the picture to be judged is the first collecting device according to the collection source information of the picture to be judged.
  • the collecting device of the picture to be judged is the first collecting device according to the collection source information of the picture to be judged.
  • the picture to be judged is the picture input by the user for lesion judgment.
  • the picture to be judged includes collection source information, and it can be determined by collecting source information whether the collecting device of the picture to be judged is the first collecting device.
  • the collection device of the picture to be judged may be the first collection device or the second collection device.
  • the picture to be judged input by the user may come from a Heidelberg device or an Optovue device. If the collection source information of the picture to be judged is "H”, the collection device of the picture is the Heidelberg device, which is the collection of the picture to be judged The device is the first collection device; if the collection source information of the picture to be judged is "O”, the collection device of the picture is an Optovue device, that is, the collection device of the picture to be judged is not the first collection device.
  • the image judgment rule is the rule information used to judge the picture to be judged.
  • the image judgment rule includes the image segmentation model and the lesion judgment rule.
  • the image segmentation processing model can only be applied to the first collection device, but not to other collection devices.
  • the picture to be judged is processed to obtain the target picture, and some information not contained in the picture to be judged can be obtained from the target picture. This technique is especially suitable for the judgment of some lesions For example, it is judged whether there is a focus of intraomental effusion.
  • Figure 2 is a schematic diagram of the effect of the method for judging a lesion based on image conversion provided by an embodiment of the application, as shown in Figure 2, where Figure (a) is a picture collected by the second collection device; Figure (b) is based on the image
  • the image segmentation model in the judgment rule divides the picture after the image (a) is segmented; the image (c) is the image obtained after the image (a) is converted according to the second image conversion model; the image (d) is judged based on the image
  • the image segmentation model in the rules is the image after segmentation processing of image (c).
  • step S150 includes sub-steps S151, S1523, and S153.
  • the picture to be determined is converted according to the second picture conversion model to obtain the picture to be divided.
  • the second picture conversion model after training can convert the picture to be judged collected by the second collection device into a picture to be segmented that matches the style of the first collection device.
  • the pictures collected by different collection devices have different styles. The above styles include but are not limited to the color distribution, brightness, contrast, noise, etc. in the picture. Since the collection device of the picture to be judged is not the first collection device, the picture to be judged After the image segmentation process is performed, part of the important information cannot be obtained from the segmented image, as shown in Figure 2 (b).
  • the second picture conversion model can be used to convert the picture collected by the second collection device into a picture matching the style of the first collection device, as shown in Figure 2 (c) As shown, the image segmentation processing model is then used to segment the picture that has undergone style conversion to obtain the target picture.
  • S152 Perform segmentation processing on the picture to be segmented according to the image segmentation model in the image judgment rule to obtain a target picture.
  • the image segmentation model is the model used to segment the picture.
  • the image segmentation model can be used to segment the pixel information of some lesions from the picture according to the contrast, brightness, and color distribution of the pixels in the picture, as shown in the figure Figure 2 in (d) shows.
  • the lesion judgment rule is a rule used to judge and analyze the target picture, and the lesion judgment result of whether the target picture contains a lesion can be obtained through the lesion judgment rule.
  • the lesion judgment rule is that the highlighted area in the picture is larger than 50 pixels, then the picture contains the lesion.
  • the target image is contrasted to obtain the highlighted area in the target picture. If the highlighted area is larger than 50 pixels, the result of the lesion judgment is obtained. It means that the target picture contains a lesion; otherwise, the obtained lesion judgment result is that the target picture does not contain a lesion.
  • step S140 includes sub-steps S141 and S142.
  • the acquisition device of the picture to be judged is the first acquisition device
  • segmentation processing is performed on the picture to be judged according to the image segmentation model in the image judgment rule to obtain a target picture. Since the image segmentation processing model can only be applied to the first acquisition device and cannot be applied to other acquisition devices, if the acquisition device of the picture to be judged is the first acquisition device, the picture to be judged can be directly segmented through the image segmentation model to Get the target picture.
  • S142 Determine whether the target picture contains a lesion according to the lesion judgment rule in the image judgment rule to obtain a lesion judgment result.
  • the first image conversion model and the second image conversion model are respectively constructed according to the conversion template, and the first confidence value calculation model and the second confidence value are respectively generated according to the calculation template
  • the calculation model is used to train the second picture conversion model through the model training rules and the picture library and the first picture conversion model, and judge the picture to be judged input by the user according to the trained second picture conversion model and the image judgment rule Obtaining the result of the lesion judgment whether the lesion is included can make the converted target picture completely match the style of another acquisition device, which improves the efficiency and quality of the picture conversion, thereby greatly increasing the accuracy of judging the lesion. Good technical results have been achieved in the application process.
  • FIG. 7 is a schematic block diagram of the apparatus for judging a lesion based on image conversion according to an embodiment of the present application.
  • the device for judging lesions based on image conversion can be configured in user terminals such as desktop computers, notebook computers, tablet computers, or mobile phones.
  • the apparatus 100 for determining a focus based on image conversion includes a conversion model construction unit 110, a calculation model generation unit 120, a conversion model training unit 130, a collection device determination unit 140, and a focus determination result acquisition unit 150.
  • the conversion model construction unit 110 is configured to construct a first picture conversion model and a second picture conversion model respectively according to a preset conversion template, wherein the first picture conversion model is used to convert the pictures collected by the first collection device into A picture that matches the style of the picture collected by the second collection device, and the second picture conversion model is used to convert the picture collected by the second collection device into a picture that matches the style of the picture collected by the first collection device.
  • the conversion model construction unit 110 includes sub-units: a first picture conversion model construction unit 111 and a second picture conversion model construction unit 112.
  • the first picture conversion model construction unit 111 is configured to construct and obtain a first picture conversion model according to the conversion template and the first format information of the first collection device, wherein the first format information is used to characterize the first format information The format of the picture collected by the collecting device.
  • the second picture conversion model construction unit 112 is configured to construct and obtain a second picture conversion model according to the conversion template and the second format information of the second collection device, wherein the second format information is used to characterize the first 2.
  • the format of the picture collected by the collection device is configured to construct and obtain a second picture conversion model according to the conversion template and the second format information of the second collection device, wherein the second format information is used to characterize the first 2.
  • the format of the picture collected by the collection device is configured to construct and obtain a second picture conversion model according to the conversion template and the second format information of the second collection device, wherein the second format information is used to characterize the first 2.
  • the format of the picture collected by the collection device is configured to construct and obtain a second picture conversion model according to the conversion template and the second format information of the second collection device, wherein the second format information is used to characterize the first 2.
  • the format of the picture collected by the collection device is configured to construct and obtain a second picture conversion model according to the conversion template and the second format information of the second collection device, where
  • the calculation model generation unit 120 is configured to generate a first confidence value calculation model and a second confidence value calculation model according to a preset calculation template, wherein the first confidence value calculation model is used to calculate the input first confidence value
  • the style similarity between the picture of the model and the picture collected by the second collection device is quantified
  • the second confidence value calculation model is used to compare the picture input to the second confidence value calculation model with the first collection
  • the similarity of styles between pictures collected by the device is quantified.
  • the conversion model training unit 130 is configured to combine the first confidence value calculation model, the second confidence value calculation model, and the first picture conversion model according to preset model training rules and a preset picture library to the The second picture conversion model is trained to obtain the second picture conversion model after training.
  • the conversion model training unit 130 includes sub-units: a picture acquisition unit 131, a training loss value calculation unit 132 and a parameter value adjustment unit 133.
  • the picture obtaining unit 131 is configured to obtain a first picture in the first picture set in the picture library and a second picture in the second picture set in the picture library, wherein the first picture set is passed A picture set composed of pictures collected by the first collection device, and the second picture set is a picture set composed of pictures collected by a second collection device.
  • the training loss value calculation unit 132 is configured to convert the first picture, the second picture, the first confidence value calculation model, the second confidence value calculation model, the first picture conversion model, and the The second picture conversion model inputs the loss function in the model training rule for calculation to obtain the training loss value.
  • the parameter value adjustment unit 133 is configured to adjust the parameter values in the second picture conversion model according to the parameter adjustment rules in the model training rule and the training loss value to complete the second picture conversion model One training session.
  • the acquisition device determining unit 140 is configured to determine whether the acquisition device of the picture to be determined is the first acquisition device according to the collection source information of the picture to be determined if the picture to be determined input by the user is received.
  • the lesion judgment result acquisition unit 150 is configured to, if the acquisition device of the picture to be judged is not the first acquisition device, perform the judgment on the picture to be judged according to a preset image judgment rule and the trained second picture conversion model. Determine whether the picture contains a lesion to obtain the lesion judgment result.
  • the lesion judgment result acquisition unit 150 includes sub-units: a picture conversion unit 151, a first segmentation processing unit 152, and a first lesion judgment unit 153.
  • the picture conversion unit 151 is configured to convert the picture to be determined according to the second picture conversion model to obtain the picture to be divided.
  • the first segmentation processing unit 152 is configured to perform segmentation processing on the picture to be segmented according to the image segmentation model in the image judgment rule to obtain a target picture.
  • the first lesion judging unit 153 is configured to judge whether the target picture contains a lesion according to the lesion judgment rule in the image judgment rule to obtain a lesion judgment result.
  • the apparatus 100 for determining a lesion based on image conversion further includes sub-units: a second segmentation processing unit 141 and a second lesion determining unit 142.
  • the second segmentation processing unit 141 is configured to, if the acquisition device of the picture to be determined is the first acquisition device, perform segmentation processing on the picture to be determined according to the image segmentation model in the image determination rule to obtain a target picture .
  • the second lesion judgment unit 142 is configured to judge whether the target picture contains a lesion according to the lesion judgment rule in the image judgment rule to obtain a lesion judgment result.
  • the above-mentioned image conversion-based lesion determination method is applied to construct a first image conversion model and a second image conversion model respectively according to the conversion template, and respectively generate the first confidence according to the calculation template
  • the value calculation model and the second confidence value calculation model train the second picture conversion model through the model training rules, the picture library and the first picture conversion model, and determine the collection equipment of the picture to be judged input by the user, according to the image judgment rule
  • the second picture conversion model judges whether the picture to be judged contains a lesion to obtain the result of the lesion judgment, which can make the converted target picture completely match the style of another acquisition device, and improve the efficiency and quality of the picture conversion , Thereby greatly increasing the accuracy of judging lesions, and achieved good technical effects in the actual application process.
  • the above-mentioned apparatus for determining lesions based on image conversion may be implemented in the form of a computer program, and the computer program may be run on a computer device as shown in FIG. 12.
  • FIG. 12 is a schematic block diagram of a computer device according to an embodiment of the present application.
  • the computer device 500 includes a processor 502, a memory, and a network interface 505 connected through a system bus 501, where the memory may include a non-volatile storage medium 503 and an internal memory 504.
  • the non-volatile storage medium 503 can store an operating system 5031 and a computer program 5032.
  • the processor 502 can execute the method for judging a lesion based on image conversion.
  • the processor 502 is used to provide calculation and control capabilities, and support the operation of the entire computer device 500.
  • the internal memory 504 provides an environment for the operation of the computer program 5032 in the non-volatile storage medium 503.
  • the processor 502 can make the processor 502 execute the lesion determination method based on image conversion.
  • the network interface 505 is used for network communication, such as providing data information transmission.
  • the structure shown in FIG. 12 is only a block diagram of part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device 500 to which the solution of the present application is applied.
  • the specific computer device 500 may include more or fewer components than shown in the figure, or combine certain components, or have a different component arrangement.
  • the processor 502 is configured to run a computer program 5032 stored in a memory to implement the method for judging a lesion based on image conversion in an embodiment of the present application.
  • the embodiment of the computer device shown in FIG. 12 does not constitute a limitation on the specific configuration of the computer device.
  • the computer device may include more or less components than those shown in the figure. Or combine certain components, or different component arrangements.
  • the computer device may only include a memory and a processor. In such an embodiment, the structures and functions of the memory and the processor are consistent with the embodiment shown in FIG. 12, and will not be repeated here.
  • the processor 502 may be a central processing unit (Central Processing Unit, CPU), and the processor 502 may also be other general-purpose processors, digital signal processors (Digital Signal Processors, DSPs), Application Specific Integrated Circuit (ASIC), Field-Programmable Gate Array (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc.
  • the general-purpose processor may be a microprocessor or the processor may also be any conventional processor.
  • a computer-readable storage medium may be a non-volatile computer-readable storage medium.
  • the computer-readable storage medium stores a computer program, where the computer program is executed by a processor to implement the method for judging a lesion based on image conversion in an embodiment of the present application.
  • the storage medium is a physical, non-transitory storage medium, such as a U disk, a mobile hard disk, a read-only memory (Read-Only Memory, ROM), a magnetic disk, or an optical disk that can store program codes. medium.
  • a physical, non-transitory storage medium such as a U disk, a mobile hard disk, a read-only memory (Read-Only Memory, ROM), a magnetic disk, or an optical disk that can store program codes. medium.

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • General Physics & Mathematics (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Evolutionary Biology (AREA)
  • Evolutionary Computation (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • General Engineering & Computer Science (AREA)
  • Artificial Intelligence (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Image Analysis (AREA)
  • Image Processing (AREA)
  • Eye Examination Apparatus (AREA)

Abstract

基于图片转换的病灶判断方法、装置、计算机设备。该方法包括:根据转换模板分别构建第一图片转换模型及第二图片转换模型,根据计算模板分别生成第一置信值计算模型及第二置信值计算模型,通过模型训练规则、图片库及第一图片转换模型对第二图片转换模型进行训练,并确定用户所输入的待判断图片的采集设备,根据图像判断规则及训练后的第二图片转换模型对待判断图片中是否包含病灶进行判断以得到病灶判断结果。

Description

基于图片转换的病灶判断方法、装置、计算机设备
本申请要求于2019年05月05日提交中国专利局、申请号为201910367767.0、申请名称为“基于图片转换的病灶判断方法、装置、计算机设备”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
本申请涉及计算机技术领域,尤其涉及一种基于图片转换的病灶判断方法、装置、计算机设备。
背景技术
在进行病灶判断过程中,需通过采集设备采集图片,然而不同设备所采集到的图片具有各自的风格,例如色彩分布、亮度、对比度、噪声等。部分重要信息无法直接从所采集到的图片中获取,需通过将图片转换为与某一采集设备风格相匹配的图片后才能获取部分重要信息,以方便对是否存在病灶进行判断。传统的病灶判断方法对图片进行风格转换后所得到的图片无法与所需风格完全匹配,进而导致无法对所采集到的图片中是否存在病灶进行准确判断。
发明内容
本申请实施例提供了一种基于图片转换的病灶判断方法、装置、计算机设备及存储介质,旨在解决现有技术方法中所存在的无法对所采集到的图片中是否存在病灶进行准确判断的问题。
第一方面,本申请实施例提供了一种基于图片转换的病灶判断方法,其包括:
根据预设的转换模板分别构建第一图片转换模型及第二图片转换模型,其中,所述第一图片转换模型用于将第一采集设备所采集的图片转换为与第二采集设备所采集图片风格相匹配的图片,所述第二图片转换模型用于将第二采集设备所采集的图片转换为与第一采集设备所采集图片风格相匹配的图片;
根据预设的计算模板生成第一置信值计算模型及第二置信值计算模型,其中,所述第一置信值计算模型用于对输入所述第一置信值计算模型的图片与所述第二采集设备所采集图片之间风格的相似度进行量化,所述第二置信值计算模型用于对输入所述第二置信值计算模型的图片与所述第一采集设备所采集图 片之间风格的相似度进行量化;
根据预设的模型训练规则及预设的图片库结合所述第一置信值计算模型、所述第二置信值计算模型及所述第一图片转换模型对所述第二图片转换模型进行训练,以得到训练后的第二图片转换模型;
若接收到用户所输入的待判断图片,根据所述待判断图片的采集源信息确定所述待判断图片的采集设备是否为第一采集设备;
若所述待判断图片的采集设备不为所述第一采集设备,根据预设的图像判断规则及训练后的所述第二图片转换模型对所述待判断图片中是否包含病灶进行判断以得到病灶判断结果。
第二方面,本申请实施例提供了一种基于图片转换的病灶判断装置,其包括:
转换模型构建单元,用于根据预设的转换模板分别构建第一图片转换模型及第二图片转换模型,其中,所述第一图片转换模型用于将第一采集设备所采集的图片转换为与第二采集设备所采集图片风格相匹配的图片,所述第二图片转换模型用于将第二采集设备所采集的图片转换为与第一采集设备所采集图片风格相匹配的图片;
计算模型生成单元,用于根据预设的计算模板生成第一置信值计算模型及第二置信值计算模型,其中,所述第一置信值计算模型用于对输入所述第一置信值计算模型的图片与所述第二采集设备所采集图片之间风格的相似度进行量化,所述第二置信值计算模型用于对输入所述第二置信值计算模型的图片与所述第一采集设备所采集图片之间风格的相似度进行量化;
转换模型训练单元,用于根据预设的模型训练规则及预设的图片库结合所述第一置信值计算模型、所述第二置信值计算模型及所述第一图片转换模型对所述第二图片转换模型进行训练,以得到训练后的第二图片转换模型;
采集设备确定单元,用于若接收到用户所输入的待判断图片,根据所述待判断图片的采集源信息确定所述待判断图片的采集设备是否为第一采集设备;
病灶判断结果获取单元,用于若所述待判断图片的采集设备不为所述第一采集设备,根据预设的图像判断规则及训练后的所述第二图片转换模型对所述待判断图片中是否包含病灶进行判断以得到病灶判断结果。
第三方面,本申请实施例又提供了一种计算机设备,其包括存储器、处理器及存储在所述存储器上并可在所述处理器上运行的计算机程序,所述处理器执行所述计算机程序时实现上述第一方面所述的基于图片转换的病灶判断方法。
第四方面,本申请实施例还提供了一种计算机可读存储介质,其中所述计算机可读存储介质存储有计算机程序,所述计算机程序当被处理器执行时使所述处理器执行上述第一方面所述的基于图片转换的病灶判断方法。
本申请实施例提供了一种基于图片转换的病灶判断方法、装置、计算机设备及存储介质,能够使转换得到的目标图片与另一采集设备风格完全匹配,提高了对图片进行转换的效率和质量,从而大幅增加了对病灶进行判断的准确率,在实际应用过程中取得了良好的技术效果。
附图说明
为了更清楚地说明本申请实施例技术方案,下面将对实施例描述中所需要使用的附图作简单地介绍,显而易见地,下面描述中的附图是本申请的一些实施例,对于本领域普通技术人员来讲,在不付出创造性劳动的前提下,还可以根据这些附图获得其他的附图。
图1为本申请实施例提供的基于图片转换的病灶判断方法的流程示意图;
图2为本申请实施例提供的基于图片转换的病灶判断方法的效果示意图;
图3为本申请实施例提供的基于图片转换的病灶判断方法的子流程示意图;
图4为本申请实施例提供的基于图片转换的病灶判断方法的另一子流程示意图;
图5为本申请实施例提供的基于图片转换的病灶判断方法的另一子流程示意图;
图6为本申请实施例提供的基于图片转换的病灶判断方法的另一子流程示意图;
图7为本申请实施例提供的基于图片转换的病灶判断装置的示意性框图;
图8为本申请实施例提供的基于图片转换的病灶判断装置的子单元示意性框图;
图9为本申请实施例提供的基于图片转换的病灶判断装置的另一子单元示意性框图;
图10为本申请实施例提供的基于图片转换的病灶判断装置的另一子单元示意性框图;
图11为本申请实施例提供的基于图片转换的病灶判断装置的另一子单元示意性框图;
图12为本申请实施例提供的计算机设备的示意性框图。
具体实施方式
下面将结合本申请实施例中的附图,对本申请实施例中的技术方案进行清楚、完整地描述,显然,所描述的实施例是本申请一部分实施例,而不是全部的实施例。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。
应当理解,当在本说明书和所附权利要求书中使用时,术语“包括”和“包含”指示所描述特征、整体、步骤、操作、元素和/或组件的存在,但并不排除一个或多个其它特征、整体、步骤、操作、元素、组件和/或其集合的存在或添加。
还应当理解,在此本申请说明书中所使用的术语仅仅是出于描述特定实施例的目的而并不意在限制本申请。如在本申请说明书和所附权利要求书中所使用的那样,除非上下文清楚地指明其它情况,否则单数形式的“一”、“一个”及“该”意在包括复数形式。
还应当进一步理解,在本申请说明书和所附权利要求书中使用的术语“和/或”是指相关联列出的项中的一个或多个的任何组合以及所有可能组合,并且包括这些组合。
请参阅图1,图1是本申请实施例提供的基于图片转换的病灶判断方法的流程示意图。该基于图片转换的病灶判断方法应用于用户终端中,该方法通过安装于用户终端中的应用软件进行执行,用户终端即是用于执行基于图片转换的病灶判断方法以对病灶进行判断的终端设备,例如台式电脑、笔记本电脑、平板电脑或手机等。
如图1所示,该方法包括步骤S110~S150。
S110、根据预设的转换模板分别构建第一图片转换模型及第二图片转换模型,其中,所述第一图片转换模型用于将第一采集设备所采集的图片转换为与第二采集设备所采集图片风格相匹配的图片,所述第二图片转换模型用于将第二采集设备所采集的图片转换为与第一采集设备所采集图片风格相匹配的图片。
根据预设的转换模板分别构建第一图片转换模型及第二图片转换模型。具体的,所述转换模板即是由若干带步长的卷积层和反卷积层所组成的模板,通过转换模板、第一采集设备及第二采集设备,即可构建得到第一图片转换模型及第二图片转换模型。第一采集设备及第二采集设备均是用于对图片进行采集的设备,不同采集设备所采集到的图片具有不同的风格,例如,上述风格包括但不限于图片中的色彩分布、亮度、对比度、噪声等。通过第一图片转换模型 的卷积及反卷积作用即可将与第一采集设备风格相匹配的图片转换为与第二采集设备风格相匹配的图片,第二图片转换模型则可将与第二采集设备风格相匹配的图片转换为与第一采集设备风格相匹配的图片。
在一实施例中,如图3所示,步骤S110包括子步骤S111和S112。
S111、根据所述转换模板及所述第一采集设备的第一格式信息构建得到第一图片转换模型,其中,所述第一格式信息用于表征所述第一采集设备所采集到的图片的格式。
根据所述转换模板及所述第一采集设备的第一格式信息构建得到第一图片转换模型。获取第一采集设备的第一格式信息,不同采集设备所采集到的图片具有不同的格式,第一格式信息即是第一采集设备所采集到的图片的格式信息,也即是图片的具体尺寸信息。第一图片转换模型中包括一个缩放处理层、两个步长为2的卷积层及两个步长为0.5的反卷积层,其中,缩放处理层即是将第一格式信息的图片进行缩放处理以得到对应像素大小的图片,每一卷积层中包含一个卷积核,组成卷积核的每个元素都对应一个权重系数和一个偏差量,同样的一个反卷积层中也包含一个反卷积核,组成反卷积核的每个元素都对应一个权重系数和一个偏差量。
例如,若接收到第一设备所采集到的图片,先通过缩放处理层将该图片转换为256×256像素的图片,通过卷积处理将256×256像素的图片转换为一个多维向量,通过反卷积处理将多维向量转换为另一个256×256像素的图片。
S112、根据所述转换模板及所述第二采集设备的第二格式信息构建得到第二图片转换模型,其中,所述第二格式信息用于表征所述第二采集设备所采集到的图片的格式。
根据所述转换模板及所述第二采集设备的第二格式信息构建得到第二图片转换模型。第二格式信息即是第二采集设备所采集到的图片的格式信息,也即是图片的具体尺寸信息,第二图片转换模型中包括一个缩放处理层、两个步长为2的卷积层及两个步长为0.5的反卷积层,其中,缩放处理层即是将第二格式信息的图片进行缩放处理以得到对应像素大小的图片,每一卷积层中包含一个卷积核,组成卷积核的每个元素都对应一个权重系数和一个偏差量,同样的一个反卷积层中也包含一个反卷积核,组成反卷积核的每个元素都对应一个权重系数和一个偏差量。
S120、根据预设的计算模板生成第一置信值计算模型及第二置信值计算模型,其中,所述第一置信值计算模型用于对输入所述第一置信值计算模型的图 片与所述第二采集设备所采集图片之间风格的相似度进行量化,所述第二置信值计算模型用于对输入所述第二置信值计算模型的图片与所述第一采集设备所采集图片之间风格的相似度进行量化。
根据预设的计算模板生成第一置信值计算模型及第二置信值计算模型。将计算模板进行复制即可得到第一置信值计算模型及第二置信值计算模型,每一个计算模型中均包括两个步长为2的卷积层、一个全连接层、一个输出节点,每一个计算模型均可对输入计算模型的图片进行处理并计算得到该图片的置信值。具体的,若输入一张特定像素大小的图片至某一计算模型,通过该计算模型对输入的图片进行卷积处理以得到一个多维向量,多维向量中的每一个维度即是计算模型中的一个输入节点,每一维度的向量值也即是输入节点对应的输入节点值,全连接层中包含预设的若干个特征单元,每一个特征单元均与所有输入节点和输出节点进行关联,特征单元即可用于反映多维向量与输出节点之间的关系,特征单元值即是全连接隐层中的特征单元的计算值。
计算模型中还包括所有输入节点至所有特征单元的公式以及所有特征单元至输出节点的公式,输入节点至特征单元的公式即是以输入节点值x 1作为输入值、特征单元值y 1作为输出值的计算公式,例如,y 1=i×x 1+j,i及j均是该公式中的参数值;特征单元至输出节点的公式即是以特征单元值作为输入值、输出节点值作为输出值的计算公式,公式具体形式如上述示例所示,通过计算模型对所输入的图片进行计算处理,即可得到该计算模型的输出节点值,也即是该图片对应的置信值。置信值的具体数值范围为[0,1],通过第一置信值计算模型计算得到置信值即可用于对所输入的图片与第二采集设备所采集图片的风格之间的相似度进行量化,若输入第一置信值计算模型的图片为第一采集设备所采集并通过第一图片转换模型转换后的图片,则可通过第一置信值计算模型计算通过第一图片转换模型转换后的该图片与第二采集设备所采集图片的风格之间的相似度,相似度即可通过置信值进行表示。具体的,若第一置信值计算模型计算得到置信值为1,则表明所输入的图片与第二采集设备所采集图片的风格相似;若第一置信值计算模型计算得到置信值为0,则表明所输入的图片与第二采集设备所采集图片的风格不相似。第二置信值计算模型计算得到的置信值即可用于对所输入的图片与第一采集设备所采集图片的风格之间的相似度进行量化,具体的,若第二置信值计算模型计算得到置信值为1,则表明所输入的图片与第一采集设备所采集图片的风格相似;若第二置信值计算模型计算得到置信值为0,则表明所输入的图片与第一采集设备所采集图片的风格不相似。
在构建得到第一置信值计算模型及第二置信值计算模型后,还可通过预设的训练数据对第一置信值计算模型及第二置信值计算模型进行训练,以对第一置信值计算模型及第二置信值计算模型中公式的参数值进行调整,以使训练后的第一置信值计算模型及第二置信值计算模型满足实际使用需求。
S130、根据预设的模型训练规则及预设的图片库结合所述第一置信值计算模型、所述第二置信值计算模型及所述第一图片转换模型对所述第二图片转换模型进行训练,以得到训练后的第二图片转换模型。
根据预设的模型训练规则及预设的图片库结合所述第一置信值计算模型、所述第二置信值计算模型及所述第一图片转换模型对所述第二图片转换模型进行训练,以得到训练后的第二图片转换模型。具体的,图片库中包括第一图片集及第二图片集,第一图片集即为通过第一采集设备所采集到的图片所组成的图片集合,第二图片集即为通过第二采集设备所采集到的图片所组成的图片集合,第一图片集中包含多张第一图片,第二图片集中包含多张第二图片。模型训练规则即是用于对第一图片转换模型及第二图片转换模型进行训练的规则信息,模型训练规则中包括损失函数及参数调整规则。
在一实施例中,如图4所示,步骤S130包括子步骤S131、S132和S133。
S131、获取所述图片库中第一图片集的一张第一图片以及所述图片库中第二图片集的一张第二图片,其中,所述第一图片集为通过所述第一采集设备所采集到的图片所组成的图片集合,所述第二图片集为通过第二采集设备所采集到的图片所组成的图片集合。
随机获取所述图片库中第一图片集的一张第一图片以及所述图片库中第二图片集的一张第二图片。第一图片用a表示,第一图片也即是与第一采集设备风格相匹配的图片,第二图片用b表示,第二图片也即是与第二采集设备风格相匹配的图片。
S132、将所述第一图片、所述第二图片、所述第一置信值计算模型、所述第二置信值计算模型、所述第一图片转换模型及所述第二图片转换模型输入所述模型训练规则中的损失函数进行计算以得到训练损失值。
具体的,损失函数为L=λ×(||G_X(b)-b|| 1+||G_Y(a)-a|| 1+||G_Y(G_X(a))-a|| 1+||G_X(G_Y(b))-b|| 1)+logD_X(b)+log(1-D_X(G_X(a)))+logD_Y(a)+log(1-D_Y(G_Y(b)))。通过第一图片转换模型对第一图片进行转换后即可得到第一转换图片,通过第二图片转换模型对第二图片进行转换后即可得到第二转换图片,为使第二转换图片与第一采集设备风格匹配度更高且使第一转换图片与第二采集设备风 格匹配度更高,需对第一图片转换模型及第二图片转换模型进行训练,可通过上述损失函数对第一图片转换模型及第二图片转换模型同时进行训练,以提高训练的速度。L即是所计算得到的训练损失值,第一图片用a表示,第二图片用b表示,第一置信值计算模型用D_X表示,第二置信值计算模型用D_Y表示,第一图片转换模型用G_X表示,第二图片转换模型用G_Y表示。第一转换图片用G_X(a)表示,第二转换图片用G_Y(b),λ为损失函数中的比重值,例如可设置比重值λ=0.1。此外,还可通过对图片a及图片b进行缩放处理后,使用第一图片转换模型对第二图片进行转换得到的图片用G_X(b)表示,使用第二图片转换模型对第一图片进行转换得到的图片用G_Y(a)表示,通过第一置信值计算模型计算得到第一图片的置信值采用G_X(a)表示、计算得到第二图片的置信值采用D_X(b)表示,通过第二置信值计算模型计算得到第一图片的置信值采用D_Y(a)表示、计算得到第二图片的置信值采用D_Y(b)表示。||G_X(b)-b|| 1即是图片G_X(b)与图片b之间的范数,计算范数的具体步骤为,将图片G_X(b)及图片b转换为unit8数据类型的数值,也即是得到图片中每一像素的像素值,将图片G_X(b)的数值与图片b的数值进行相减即可得到范数,范数越小则表明两张图片越相似。
S133、根据所述模型训练规则中的参数调整规则结合所述训练损失值对所述第二图片转换模型中的参数值进行调整以完成对所述第二图片转换模型进行一次训练。
根据所述模型训练规则中的参数调整规则结合所述训练损失值对所述第二图片转换模型中的参数值进行调整以完成对所述第二图片转换模型进行一次训练。计算得到的训练损失值后,即可根据参数调整规则及训练损失值确定参数更新梯度值,训练损失值越大,则对应的更新梯度值越大,训练损失值越小,则对应的更新梯度值越小。第二图片转换模型中的参数值也即是该模型中卷积核及反卷积核所包含的权重系数,参数调整规则中还包括调整方向,结合调整方向及更新梯度值即可对第二图片转换模型中所包含的权重系数进行调整,也即是完成对第二图片转换模型进行一次训练。
通过上述权重系数调整方法对第二图片转换模型进行迭代训练,迭代训练的次数可由用户设定,经过迭代训练后即可最终得到训练后的第二图片转换模型。
S140、若接收到用户所输入的待判断图片,根据所述待判断图片的采集源信息确定所述待判断图片的采集设备是否为第一采集设备。
若接收到用户所输入的待判断图片,根据所述待判断图片的采集源信息确定所述待判断图片的采集设备是否为第一采集设备。待判断图片即是用户所输入的需进行病灶判断的图片,具体的,待判断图片中包括采集源信息,通过采集源信息即可确定该待判断图片的采集设备是否为第一采集设备,其中,待判断图片的采集设备可以是第一采集设备或第二采集设备。
例如,用户所输入的待判断图片可能来源于海德堡设备或Optovue设备,若待判断图片的采集源信息为“H”,则该图片的采集设备为海德堡设备,也即是该待判断图片的采集设备为第一采集设备;若待判断图片的采集源信息为“O”,则该图片的采集设备为Optovue设备,也即是该待判断图片的采集设备不为第一采集设备。
S150、若所述待判断图片的采集设备不为所述第一采集设备,根据预设的图像判断规则及训练后的所述第二图片转换模型对所述待判断图片中是否包含病灶进行判断以得到病灶判断结果。
若所述待判断图片的采集设备不为所述第一采集设备,根据预设的图像判断规则及训练后的所述第二图片转换模型对所述待判断图片中是否包含病灶进行判断以得到病灶判断结果。图像判断规则即是用于对待判断图片进行判断的规则信息,图像判断规则中包括图像分割模型及病灶判断规则,图像分割处理模型只能适用于第一采集设备,而无法适用于其他采集设备。根据第二图片转换模型及图像分割模型对待判断图片进行处理以得到目标图片,即可从目标图片中获取得到待判断图片中所不包含的一些信息,这一技术尤其适用于对部分病灶的判断中,例如对是否存在网膜内积液这一病灶进行判断。
图2为本申请实施例提供的基于图片转换的病灶判断方法的效果示意图,如图2所示,其中的图(a)是上述第二采集设备所采集的图片;图(b)是根据图像判断规则中的图像分割模型对图(a)进行分割处理后的图片;图(c)是根据第二图片转换模型对图(a)进行转换后得到的图片;图(d)是根据图像判断规则中的图像分割模型对图(c)进行分割处理后的图片。
在一实施例中,如图5所示,步骤S150包括子步骤S151、S1523和S153。
S151、根据所述第二图片转换模型对所述待判断图片进行转换以得到待分割图片。
根据所述第二图片转换模型对所述待判断图片进行转换以得到待分割图片。通过训练后的第二图片转换模型即可将第二采集设备所采集到的待判断图片转换为与第一采集设备风格相匹配的待分割图片。不同采集设备所采集到的图片 具有不同的风格,上述风格包括但不限于图片中的色彩分布、亮度、对比度、噪声等,由于待判断图片的采集设备不是第一采集设备,直接将待判定图片进行图像分割处理后,无法从分割处理后的图片中获取部分重要信息,如图2中的图(b)所示。为获取待判断图片中的部分重要信息,可通过第二图片转换模型将第二采集设备所采集到的图片转换为与第一采集设备风格相匹配的图片,如图2中的图(c)所示,之后再通过图像分割处理模型对已进行风格转换的图片进行分割处理得到目标图片。
S152、根据所述图像判断规则中的图像分割模型对所述待分割图片进行分割处理以得到目标图片。
根据所述图像判断规则中的图像分割模型对所述待分割图片进行分割处理以得到目标图片。通过图像分割处理模型对已进行风格转换的图片进行分割处理得到目标图片后,即可从目标图片中获取其所包含的部分重要信息。具体的,图像分割模型即是用于对图片进行分割处理的模型,可通过图像分割模型根据图片中像素的对比度、亮度、色彩分布等信息,从图片中分割出部分病灶的像素信息,如图2中的图(d)所示。
S153、根据所述图像判断规则中的病灶判断规则对所述目标图片中是否包含病灶进行判断以得到病灶判断结果。
根据所述图像判断规则中的病灶判断规则对所述目标图片中是否包含病灶进行判断以得到病灶判断结果。其中,病灶判断规则即是用于对目标图片进行判断分析的规则,通过病灶判断规则即可获取目标图片中是否包含病灶的病灶判断结果。
例如,病灶判断规则为图片中高亮区域大于50个像素则图片包含病灶,通过对目标图片进行对比度分析以获取目标图片中的高亮区域,若高亮区域大于50个像素则得到的病灶判断结果为该目标图片中包含病灶;否则得到的病灶判断结果为该目标图片中不包含病灶。
在一实施例中,如图6所示,步骤S140之后包括子步骤S141和S142。
S141、若所述待判断图片的采集设备为所述第一采集设备,根据所述图像判断规则中的图像分割模型对所述待判断图片进行分割处理以得到目标图片。
若所述待判断图片的采集设备为所述第一采集设备,根据所述图像判断规则中的图像分割模型对所述待判断图片进行分割处理以得到目标图片。由于图像分割处理模型只能适用于第一采集设备,而无法适用于其他采集设备,若待判断图片的采集设备为第一采集设备,则可直接通过图像分割模型对待判断图 片进行分割处理,以获取目标图片。
S142、根据所述图像判断规则中的病灶判断规则对所述目标图片中是否包含病灶进行判断以得到病灶判断结果。
根据所述图像判断规则中的病灶判断规则对所述目标图片中是否包含病灶进行判断以得到病灶判断结果。通过病灶判断规则即可获取目标图片中是否包含病灶的病灶判断结果。
在本申请实施例所提供的基于图片转换的病灶判断方法中,根据转换模板分别构建第一图片转换模型及第二图片转换模型,根据计算模板分别生成第一置信值计算模型及第二置信值计算模型,通过模型训练规则及图片库及第一图片转换模型对第二图片转换模型进行训练,并根据训练后的第二图片转换模型及图像判断规则对用户所输入的待判断图片进行判断以得到是否包含病灶的病灶判断结果,能够使转换得到的目标图片与另一采集设备风格完全匹配,提高了对图片进行转换的效率和质量,从而大幅增加了对病灶进行判断的准确率,在实际应用过程中取得了良好的技术效果。
本申请实施例还提供一种基于图片转换的病灶判断装置,该基于图片转换的病灶判断装置用于执行前述基于图片转换的病灶判断方法的任一实施例。具体地,请参阅图7,图7是本申请实施例提供的基于图片转换的病灶判断装置的示意性框图。该基于图片转换的病灶判断装置可以配置于台式电脑、笔记本电脑、平板电脑或手机等用户终端中。
如图7所示,基于图片转换的病灶判断装置100包括转换模型构建单元110、计算模型生成单元120、转换模型训练单元130、采集设备确定单元140和病灶判断结果获取单元150。
转换模型构建单元110,用于根据预设的转换模板分别构建第一图片转换模型及第二图片转换模型,其中,所述第一图片转换模型用于将第一采集设备所采集的图片转换为与第二采集设备所采集图片风格相匹配的图片,所述第二图片转换模型用于将第二采集设备所采集的图片转换为与第一采集设备所采集图片风格相匹配的图片。
其他申请实施例中,如图8所示,所述转换模型构建单元110包括子单元:第一图片转换模型构建单元111和第二图片转换模型构建单元112。
第一图片转换模型构建单元111,用于根据所述转换模板及所述第一采集设备的第一格式信息构建得到第一图片转换模型,其中,所述第一格式信息用于表征所述第一采集设备所采集到的图片的格式。
第二图片转换模型构建单元112,用于根据所述转换模板及所述第二采集设备的第二格式信息构建得到第二图片转换模型,其中,所述第二格式信息用于表征所述第二采集设备所采集到的图片的格式。
计算模型生成单元120,用于根据预设的计算模板生成第一置信值计算模型及第二置信值计算模型,其中,所述第一置信值计算模型用于对输入所述第一置信值计算模型的图片与所述第二采集设备所采集图片之间风格的相似度进行量化,所述第二置信值计算模型用于对输入所述第二置信值计算模型的图片与所述第一采集设备所采集图片之间风格的相似度进行量化。
转换模型训练单元130,用于根据预设的模型训练规则及预设的图片库结合所述第一置信值计算模型、所述第二置信值计算模型及所述第一图片转换模型对所述第二图片转换模型进行训练,以得到训练后的第二图片转换模型。
其他申请实施例中,如图9所示,所述转换模型训练单元130包括子单元:图片获取单元131、训练损失值计算单元132和参数值调整单元133。
图片获取单元131,用于获取所述图片库中第一图片集的一张第一图片以及所述图片库中第二图片集的一张第二图片,其中,所述第一图片集为通过所述第一采集设备所采集到的图片所组成的图片集合,所述第二图片集为通过第二采集设备所采集到的图片所组成的图片集合。训练损失值计算单元132,用于将所述第一图片、所述第二图片、所述第一置信值计算模型、所述第二置信值计算模型、所述第一图片转换模型及所述第二图片转换模型输入所述模型训练规则中的损失函数进行计算以得到训练损失值。参数值调整单元133,用于根据所述模型训练规则中的参数调整规则结合所述训练损失值对所述第二图片转换模型中的参数值进行调整以完成对所述第二图片转换模型进行一次训练。
采集设备确定单元140,用于若接收到用户所输入的待判断图片,根据所述待判断图片的采集源信息确定所述待判断图片的采集设备是否为第一采集设备。
病灶判断结果获取单元150,用于若所述待判断图片的采集设备不为所述第一采集设备,根据预设的图像判断规则及训练后的所述第二图片转换模型对所述待判断图片中是否包含病灶进行判断以得到病灶判断结果。
其他申请实施例中,如图10所示,所述病灶判断结果获取单元150包括子单元:图片转换单元151、第一分割处理单元152和第一病灶判断单元153。
图片转换单元151,用于根据所述第二图片转换模型对所述待判断图片进行转换以得到待分割图片。第一分割处理单元152,用于根据所述图像判断规则中的图像分割模型对所述待分割图片进行分割处理以得到目标图片。第一病灶判 断单元153,用于根据所述图像判断规则中的病灶判断规则对所述目标图片中是否包含病灶进行判断以得到病灶判断结果。
其他申请实施例中,如图11所示,所述基于图片转换的病灶判断装置100还包括子单元:第二分割处理单元141和第二病灶判断单元142。
第二分割处理单元141,用于若所述待判断图片的采集设备为所述第一采集设备,根据所述图像判断规则中的图像分割模型对所述待判断图片进行分割处理以得到目标图片。第二病灶判断单元142,用于根据所述图像判断规则中的病灶判断规则对所述目标图片中是否包含病灶进行判断以得到病灶判断结果。
在本申请实施例所提供的基于图片转换的病灶判断装置应用上述基于图片转换的病灶判断方法,根据转换模板分别构建第一图片转换模型及第二图片转换模型,根据计算模板分别生成第一置信值计算模型及第二置信值计算模型,通过模型训练规则、图片库及第一图片转换模型对第二图片转换模型进行训练,并确定用户所输入的待判断图片的采集设备,根据图像判断规则及训练后的第二图片转换模型对待判断图片中是否包含病灶进行判断以得到病灶判断结果,能够使转换得到的目标图片与另一采集设备风格完全匹配,提高了对图片进行转换的效率和质量,从而大幅增加了对病灶进行判断的准确率,在实际应用过程中取得了良好的技术效果。
上述基于图片转换的病灶判断装置可以实现为计算机程序的形式,该计算机程序可以在如图12所示的计算机设备上运行。
请参阅图12,图12是本申请实施例提供的计算机设备的示意性框图。
参阅图12,该计算机设备500包括通过系统总线501连接的处理器502、存储器和网络接口505,其中,存储器可以包括非易失性存储介质503和内存储器504。
该非易失性存储介质503可存储操作系统5031和计算机程序5032。该计算机程序5032被执行时,可使得处理器502执行基于图片转换的病灶判断方法。
该处理器502用于提供计算和控制能力,支撑整个计算机设备500的运行。
该内存储器504为非易失性存储介质503中的计算机程序5032的运行提供环境,该计算机程序5032被处理器502执行时,可使得处理器502执行基于图片转换的病灶判断方法。
该网络接口505用于进行网络通信,如提供数据信息的传输等。本领域技术人员可以理解,图12中示出的结构,仅仅是与本申请方案相关的部分结构的框图,并不构成对本申请方案所应用于其上的计算机设备500的限定,具体的 计算机设备500可以包括比图中所示更多或更少的部件,或者组合某些部件,或者具有不同的部件布置。
其中,所述处理器502用于运行存储在存储器中的计算机程序5032,以实现本申请实施例的基于图片转换的病灶判断方法。
本领域技术人员可以理解,图12中示出的计算机设备的实施例并不构成对计算机设备具体构成的限定,在其他实施例中,计算机设备可以包括比图示更多或更少的部件,或者组合某些部件,或者不同的部件布置。例如,在一些实施例中,计算机设备可以仅包括存储器及处理器,在这样的实施例中,存储器及处理器的结构及功能与图12所示实施例一致,在此不再赘述。
应当理解,在本申请实施例中,处理器502可以是中央处理单元(Central Processing Unit,CPU),该处理器502还可以是其他通用处理器、数字信号处理器(Digital Signal Processor,DSP)、专用集成电路(Application Specific Integrated Circuit,ASIC)、现成可编程门阵列(Field-Programmable Gate Array,FPGA)或者其他可编程逻辑器件、分立门或者晶体管逻辑器件、分立硬件组件等。其中,通用处理器可以是微处理器或者该处理器也可以是任何常规的处理器等。
在本申请的另一实施例中提供计算机可读存储介质。该计算机可读存储介质可以为非易失性的计算机可读存储介质。该计算机可读存储介质存储有计算机程序,其中计算机程序被处理器执行时实现本申请实施例的基于图片转换的病灶判断方法。
所述存储介质为实体的、非瞬时性的存储介质,例如可以是U盘、移动硬盘、只读存储器(Read-Only Memory,ROM)、磁碟或者光盘等各种可以存储程序代码的实体存储介质。
所属领域的技术人员可以清楚地了解到,为了描述的方便和简洁,上述描述的设备、装置和单元的具体工作过程,可以参考前述方法实施例中的对应过程,在此不再赘述。
以上所述,仅为本申请的具体实施方式,但本申请的保护范围并不局限于此,任何熟悉本技术领域的技术人员在本申请揭露的技术范围内,可轻易想到各种等效的修改或替换,这些修改或替换都应涵盖在本申请的保护范围之内。因此,本申请的保护范围应以权利要求的保护范围为准。

Claims (20)

  1. 一种基于图片转换的病灶判断方法,包括:
    根据预设的转换模板分别构建第一图片转换模型及第二图片转换模型,其中,所述第一图片转换模型用于将第一采集设备所采集的图片转换为与第二采集设备所采集图片风格相匹配的图片,所述第二图片转换模型用于将第二采集设备所采集的图片转换为与第一采集设备所采集图片风格相匹配的图片;
    根据预设的计算模板生成第一置信值计算模型及第二置信值计算模型,其中,所述第一置信值计算模型用于对输入所述第一置信值计算模型的图片与所述第二采集设备所采集图片之间风格的相似度进行量化,所述第二置信值计算模型用于对输入所述第二置信值计算模型的图片与所述第一采集设备所采集图片之间风格的相似度进行量化;
    根据预设的模型训练规则及预设的图片库结合所述第一置信值计算模型、所述第二置信值计算模型及所述第一图片转换模型对所述第二图片转换模型进行训练,以得到训练后的第二图片转换模型;
    若接收到用户所输入的待判断图片,根据所述待判断图片的采集源信息确定所述待判断图片的采集设备是否为第一采集设备;
    若所述待判断图片的采集设备不为所述第一采集设备,根据预设的图像判断规则及训练后的所述第二图片转换模型对所述待判断图片中是否包含病灶进行判断以得到病灶判断结果。
  2. 根据权利要求1所述的基于图片转换的病灶判断方法,其中,所述根据预设的转换模板分别构建第一图片转换模型及第二图片转换模型,包括:
    根据所述转换模板及所述第一采集设备的第一格式信息构建得到第一图片转换模型,其中,所述第一格式信息用于表征所述第一采集设备所采集到的图片的格式;
    根据所述转换模板及所述第二采集设备的第二格式信息构建得到第二图片转换模型,其中,所述第二格式信息用于表征所述第二采集设备所采集到的图片的格式。
  3. 根据权利要求1所述的基于图片转换的病灶判断方法,其中,所述根据预设的模型训练规则及预设的图片库结合所述第一置信值计算模型、所述第二置信值计算模型及所述第一图片转换模型对所述第二图片转换模型进行训练,以得到训练后的第二图片转换模型,包括:
    获取所述图片库中第一图片集的一张第一图片以及所述图片库中第二图片集的一张第二图片,其中,所述第一图片集为通过所述第一采集设备所采集到的图片所组成的图片集合,所述第二图片集为通过第二采集设备所采集到的图片所组成的图片集合;
    将所述第一图片、所述第二图片、所述第一置信值计算模型、所述第二置信值计算模型、所述第一图片转换模型及所述第二图片转换模型输入所述模型训练规则中的损失函数进行计算以得到训练损失值;
    根据所述模型训练规则中的参数调整规则结合所述训练损失值对所述第二图片转换模型中的参数值进行调整以完成对所述第二图片转换模型进行一次训练。
  4. 根据权利要求1所述的基于图片转换的病灶判断方法,其中,根据预设的图像判断规则及训练后的所述第二图片转换模型对所述待判断图片中是否包含病灶进行判断以得到病灶判断结果,包括:
    根据所述第二图片转换模型对所述待判断图片进行转换以得到待分割图片;
    根据所述图像判断规则中的图像分割模型对所述待分割图片进行分割处理以得到目标图片;
    根据所述图像判断规则中的病灶判断规则对所述目标图片中是否包含病灶进行判断以得到病灶判断结果。
  5. 根据权利要求4所述的基于图片转换的病灶判断方法,其中,所述根据所述待判断图片的采集源信息确定所述待判断图片的采集设备是否为第一采集设备之后,还包括:
    若所述待判断图片的采集设备为所述第一采集设备,根据所述图像判断规则中的图像分割模型对所述待判断图片进行分割处理以得到目标图片;
    根据所述图像判断规则中的病灶判断规则对所述目标图片中是否包含病灶进行判断以得到病灶判断结果。
  6. 根据权利要求1所述的基于图片转换的病灶判断方法,其中,所述第一图片转换模型包括缩放处理层、卷积层及反卷积层,所述第二图片转换模型包括缩放处理层、卷积层及反卷积层。
  7. 根据权利要求1所述的基于图片转换的病灶判断方法,其中,所述第一置信值计算模型包括卷积层、全连接层及输出节点,所述第二置信值计算模型包括卷积层、全连接层及输出节点。
  8. 一种基于图片转换的病灶判断装置,包括:
    转换模型构建单元,用于根据预设的转换模板分别构建第一图片转换模型及第二图片转换模型,其中,所述第一图片转换模型用于将第一采集设备所采集的图片转换为与第二采集设备所采集图片风格相匹配的图片,所述第二图片转换模型用于将第二采集设备所采集的图片转换为与第一采集设备所采集图片风格相匹配的图片;
    计算模型生成单元,用于根据预设的计算模板生成第一置信值计算模型及第二置信值计算模型,其中,所述第一置信值计算模型用于对输入所述第一置信值计算模型的图片与所述第二采集设备所采集图片之间风格的相似度进行量化,所述第二置信值计算模型用于对输入所述第二置信值计算模型的图片与所述第一采集设备所采集图片之间风格的相似度进行量化;
    转换模型训练单元,用于根据预设的模型训练规则及预设的图片库结合所述第一置信值计算模型、所述第二置信值计算模型及所述第一图片转换模型对所述第二图片转换模型进行训练,以得到训练后的第二图片转换模型;
    采集设备确定单元,用于若接收到用户所输入的待判断图片,根据所述待判断图片的采集源信息确定所述待判断图片的采集设备是否为第一采集设备;
    病灶判断结果获取单元,用于若所述待判断图片的采集设备不为所述第一采集设备,根据预设的图像判断规则及训练后的所述第二图片转换模型对所述待判断图片中是否包含病灶进行判断以得到病灶判断结果。
  9. 根据权利要求8所述的基于图片转换的病灶判断装置,其中,所述转换模型构建单元,包括:
    第一图片转换模型构建单元,用于根据所述转换模板及所述第一采集设备的第一格式信息构建得到第一图片转换模型,其中,所述第一格式信息用于表征所述第一采集设备所采集到的图片的格式;
    第二图片转换模型构建单元,用于根据所述转换模板及所述第二采集设备的第二格式信息构建得到第二图片转换模型,其中,所述第二格式信息用于表征所述第二采集设备所采集到的图片的格式。
  10. 根据权利要求8所述的基于图片转换的病灶判断装置,其中,所述转换模型训练单元,包括:
    图片获取单元,用于获取所述图片库中第一图片集的一张第一图片以及所述图片库中第二图片集的一张第二图片,其中,所述第一图片集为通过所述第一采集设备所采集到的图片所组成的图片集合,所述第二图片集为通过第二采集设备所采集到的图片所组成的图片集合;
    训练损失值计算单元,用于将所述第一图片、所述第二图片、所述第一置信值计算模型、所述第二置信值计算模型、所述第一图片转换模型及所述第二图片转换模型输入所述模型训练规则中的损失函数进行计算以得到训练损失值;
    参数值调整单元,用于根据所述模型训练规则中的参数调整规则结合所述训练损失值对所述第二图片转换模型中的参数值进行调整以完成对所述第二图片转换模型进行一次训练。
  11. 一种计算机设备,包括存储器、处理器及存储在所述存储器上并可在所述处理器上运行的计算机程序,所述处理器执行所述计算机程序时实现以下步骤:
    根据预设的转换模板分别构建第一图片转换模型及第二图片转换模型,其中,所述第一图片转换模型用于将第一采集设备所采集的图片转换为与第二采集设备所采集图片风格相匹配的图片,所述第二图片转换模型用于将第二采集设备所采集的图片转换为与第一采集设备所采集图片风格相匹配的图片;
    根据预设的计算模板生成第一置信值计算模型及第二置信值计算模型,其中,所述第一置信值计算模型用于对输入所述第一置信值计算模型的图片与所述第二采集设备所采集图片之间风格的相似度进行量化,所述第二置信值计算模型用于对输入所述第二置信值计算模型的图片与所述第一采集设备所采集图片之间风格的相似度进行量化;
    根据预设的模型训练规则及预设的图片库结合所述第一置信值计算模型、所述第二置信值计算模型及所述第一图片转换模型对所述第二图片转换模型进行训练,以得到训练后的第二图片转换模型;
    若接收到用户所输入的待判断图片,根据所述待判断图片的采集源信息确定所述待判断图片的采集设备是否为第一采集设备;
    若所述待判断图片的采集设备不为所述第一采集设备,根据预设的图像判断规则及训练后的所述第二图片转换模型对所述待判断图片中是否包含病灶进行判断以得到病灶判断结果。
  12. 根据权利要求11所述的基于图片转换的计算机设备,其中,所述根据预设的转换模板分别构建第一图片转换模型及第二图片转换模型,包括:
    根据所述转换模板及所述第一采集设备的第一格式信息构建得到第一图片转换模型,其中,所述第一格式信息用于表征所述第一采集设备所采集到的图片的格式;
    根据所述转换模板及所述第二采集设备的第二格式信息构建得到第二图片 转换模型,其中,所述第二格式信息用于表征所述第二采集设备所采集到的图片的格式。
  13. 根据权利要求11所述的基于图片转换的计算机设备,其中,所述根据预设的模型训练规则及预设的图片库结合所述第一置信值计算模型、所述第二置信值计算模型及所述第一图片转换模型对所述第二图片转换模型进行训练,以得到训练后的第二图片转换模型,包括:
    获取所述图片库中第一图片集的一张第一图片以及所述图片库中第二图片集的一张第二图片,其中,所述第一图片集为通过所述第一采集设备所采集到的图片所组成的图片集合,所述第二图片集为通过第二采集设备所采集到的图片所组成的图片集合;
    将所述第一图片、所述第二图片、所述第一置信值计算模型、所述第二置信值计算模型、所述第一图片转换模型及所述第二图片转换模型输入所述模型训练规则中的损失函数进行计算以得到训练损失值;
    根据所述模型训练规则中的参数调整规则结合所述训练损失值对所述第二图片转换模型中的参数值进行调整以完成对所述第二图片转换模型进行一次训练。
  14. 根据权利要求11所述的基于图片转换的计算机设备,其中,根据预设的图像判断规则及训练后的所述第二图片转换模型对所述待判断图片中是否包含病灶进行判断以得到病灶判断结果,包括:
    根据所述第二图片转换模型对所述待判断图片进行转换以得到待分割图片;
    根据所述图像判断规则中的图像分割模型对所述待分割图片进行分割处理以得到目标图片;
    根据所述图像判断规则中的病灶判断规则对所述目标图片中是否包含病灶进行判断以得到病灶判断结果。
  15. 根据权利要求14所述的基于图片转换的计算机设备,其中,所述根据所述待判断图片的采集源信息确定所述待判断图片的采集设备是否为第一采集设备之后,还包括:
    若所述待判断图片的采集设备为所述第一采集设备,根据所述图像判断规则中的图像分割模型对所述待判断图片进行分割处理以得到目标图片;
    根据所述图像判断规则中的病灶判断规则对所述目标图片中是否包含病灶进行判断以得到病灶判断结果。
  16. 根据权利要求11所述的基于图片转换的计算机设备,其中,所述第一 图片转换模型包括缩放处理层、卷积层及反卷积层,所述第二图片转换模型包括缩放处理层、卷积层及反卷积层。
  17. 根据权利要求11所述的基于图片转换的计算机设备,其中,所述第一置信值计算模型包括卷积层、全连接层及输出节点,所述第二置信值计算模型包括卷积层、全连接层及输出节点。
  18. 一种计算机可读存储介质,所述计算机可读存储介质存储有计算机程序,所述计算机程序当被处理器执行时使所述处理器执行以下操作:根据预设的转换模板分别构建第一图片转换模型及第二图片转换模型,其中,所述第一图片转换模型用于将第一采集设备所采集的图片转换为与第二采集设备所采集图片风格相匹配的图片,所述第二图片转换模型用于将第二采集设备所采集的图片转换为与第一采集设备所采集图片风格相匹配的图片;
    根据预设的计算模板生成第一置信值计算模型及第二置信值计算模型,其中,所述第一置信值计算模型用于对输入所述第一置信值计算模型的图片与所述第二采集设备所采集图片之间风格的相似度进行量化,所述第二置信值计算模型用于对输入所述第二置信值计算模型的图片与所述第一采集设备所采集图片之间风格的相似度进行量化;
    根据预设的模型训练规则及预设的图片库结合所述第一置信值计算模型、所述第二置信值计算模型及所述第一图片转换模型对所述第二图片转换模型进行训练,以得到训练后的第二图片转换模型;
    若接收到用户所输入的待判断图片,根据所述待判断图片的采集源信息确定所述待判断图片的采集设备是否为第一采集设备;
    若所述待判断图片的采集设备不为所述第一采集设备,根据预设的图像判断规则及训练后的所述第二图片转换模型对所述待判断图片中是否包含病灶进行判断以得到病灶判断结果。
  19. 根据权利要求18所述的基于图片转换的存储介质,其中,所述根据预设的转换模板分别构建第一图片转换模型及第二图片转换模型,包括:
    根据所述转换模板及所述第一采集设备的第一格式信息构建得到第一图片转换模型,其中,所述第一格式信息用于表征所述第一采集设备所采集到的图片的格式;
    根据所述转换模板及所述第二采集设备的第二格式信息构建得到第二图片转换模型,其中,所述第二格式信息用于表征所述第二采集设备所采集到的图片的格式。
  20. 根据权利要求18所述的基于图片转换的存储介质,其中,所述根据预设的模型训练规则及预设的图片库结合所述第一置信值计算模型、所述第二置信值计算模型及所述第一图片转换模型对所述第二图片转换模型进行训练,以得到训练后的第二图片转换模型,包括:
    获取所述图片库中第一图片集的一张第一图片以及所述图片库中第二图片集的一张第二图片,其中,所述第一图片集为通过所述第一采集设备所采集到的图片所组成的图片集合,所述第二图片集为通过第二采集设备所采集到的图片所组成的图片集合;
    将所述第一图片、所述第二图片、所述第一置信值计算模型、所述第二置信值计算模型、所述第一图片转换模型及所述第二图片转换模型输入所述模型训练规则中的损失函数进行计算以得到训练损失值;
    根据所述模型训练规则中的参数调整规则结合所述训练损失值对所述第二图片转换模型中的参数值进行调整以完成对所述第二图片转换模型进行一次训练。
PCT/CN2019/103337 2019-05-05 2019-08-29 基于图片转换的病灶判断方法、装置、计算机设备 WO2020224118A1 (zh)

Priority Applications (1)

Application Number Priority Date Filing Date Title
JP2021500419A JP7064050B2 (ja) 2019-05-05 2019-08-29 画像変換に基づく病巣判定方法、装置、コンピュータ機器及び記憶媒体

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201910367767.0 2019-05-05
CN201910367767.0A CN110232401B (zh) 2019-05-05 2019-05-05 基于图片转换的病灶判断方法、装置、计算机设备

Publications (1)

Publication Number Publication Date
WO2020224118A1 true WO2020224118A1 (zh) 2020-11-12

Family

ID=67860610

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2019/103337 WO2020224118A1 (zh) 2019-05-05 2019-08-29 基于图片转换的病灶判断方法、装置、计算机设备

Country Status (3)

Country Link
JP (1) JP7064050B2 (zh)
CN (1) CN110232401B (zh)
WO (1) WO2020224118A1 (zh)

Families Citing this family (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112766277A (zh) * 2021-02-07 2021-05-07 普联技术有限公司 卷积神经网络模型的通道调整方法、装置和设备

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20070014457A1 (en) * 2005-07-13 2007-01-18 Marie-Pierre Jolly Method for knowledge based image segmentation using shape models
CN108596180A (zh) * 2018-04-09 2018-09-28 深圳市腾讯网络信息技术有限公司 图像中的参数识别、参数识别模型的训练方法及装置
CN109189973A (zh) * 2018-08-30 2019-01-11 清华大学 基于策略梯度的大规模图像检索方法及装置
CN109389135A (zh) * 2017-08-03 2019-02-26 杭州海康威视数字技术股份有限公司 一种图像筛选方法及装置

Family Cites Families (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP2001258044A (ja) 2000-03-14 2001-09-21 Matsushita Research Institute Tokyo Inc 医療用画像処理装置
JP4169954B2 (ja) 2000-09-18 2008-10-22 富士フイルム株式会社 異常陰影候補の検出方法
CN107665333A (zh) * 2017-08-28 2018-02-06 平安科技(深圳)有限公司 一种基于卷积神经网络的不雅图片识别方法、终端、设备及计算机可读存储介质
CN108564127B (zh) * 2018-04-19 2022-02-18 腾讯科技(深圳)有限公司 图像转换方法、装置、计算机设备及存储介质
CN109308679B (zh) 2018-08-13 2022-08-30 深圳市商汤科技有限公司 一种图像风格转换方法及装置、设备、存储介质
CN109241318B (zh) * 2018-09-21 2023-06-13 平安科技(深圳)有限公司 图片推荐方法、装置、计算机设备及存储介质
CN109166087A (zh) 2018-09-29 2019-01-08 上海联影医疗科技有限公司 医学图像的风格转换方法、装置、医学设备、影像系统及存储介质
CN109685102B (zh) * 2018-11-13 2024-07-09 平安科技(深圳)有限公司 胸部病灶图像分类方法、装置、计算机设备及存储介质

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20070014457A1 (en) * 2005-07-13 2007-01-18 Marie-Pierre Jolly Method for knowledge based image segmentation using shape models
CN109389135A (zh) * 2017-08-03 2019-02-26 杭州海康威视数字技术股份有限公司 一种图像筛选方法及装置
CN108596180A (zh) * 2018-04-09 2018-09-28 深圳市腾讯网络信息技术有限公司 图像中的参数识别、参数识别模型的训练方法及装置
CN109189973A (zh) * 2018-08-30 2019-01-11 清华大学 基于策略梯度的大规模图像检索方法及装置

Also Published As

Publication number Publication date
CN110232401B (zh) 2023-08-04
JP7064050B2 (ja) 2022-05-09
CN110232401A (zh) 2019-09-13
JP2021530780A (ja) 2021-11-11

Similar Documents

Publication Publication Date Title
Zeng et al. Learning image-adaptive 3d lookup tables for high performance photo enhancement in real-time
JP7373554B2 (ja) クロスドメイン画像変換
US11586464B2 (en) Techniques for workflow analysis and design task optimization
US10019823B2 (en) Combined composition and change-based models for image cropping
US9299004B2 (en) Image foreground detection
US9330334B2 (en) Iterative saliency map estimation
CN111833430B (zh) 基于神经网络的光照数据预测方法、系统、终端及介质
WO2019165949A1 (zh) 图像处理方法、设备、存储介质和计算机程序产品
US9311756B2 (en) Image group processing and visualization
US9747526B2 (en) Using machine learning to define user controls for photo adjustments
WO2022147964A1 (zh) 图像扭曲渲染方法及装置
GB2587833A (en) Image modification styles learned from a limited set of modified images
WO2021223677A1 (en) Dense 3d modelling method for ifc bim object production from rgbd videos
TWI711004B (zh) 圖片處理方法和裝置
CN110443874A (zh) 基于卷积神经网络的视点数据生成方法和装置
WO2021109867A1 (zh) 图像处理方法及装置、计算机可读介质和电子设备
CN110807293B (zh) 一种基于度量标准的海流场几何可视化方法
WO2020224118A1 (zh) 基于图片转换的病灶判断方法、装置、计算机设备
CN115457054A (zh) 图像分割方法、装置、设备及可读存储介质
CN108966042B (zh) 一种基于最短路径的视频摘要生成方法及装置
JP2015179426A (ja) 情報処理装置、パラメータの決定方法、及びプログラム
CN109388784A (zh) 最小熵核密度估计器生成方法、装置和计算机可读存储介质
CN106469437B (zh) 图像处理方法和图像处理装置
WO2021000495A1 (zh) 一种图像处理方法以及装置
Singaraju et al. Estimation of alpha mattes for multiple image layers

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: 19928218

Country of ref document: EP

Kind code of ref document: A1

ENP Entry into the national phase

Ref document number: 2021500419

Country of ref document: JP

Kind code of ref document: A

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 19928218

Country of ref document: EP

Kind code of ref document: A1

122 Ep: pct application non-entry in european phase

Ref document number: 19928218

Country of ref document: EP

Kind code of ref document: A1