Disclosure of Invention
In view of the above-mentioned problems, an object of the present invention is to provide a method, an apparatus and a system for identifying and segmenting a nasopharyngeal carcinoma lymph node area, so as to improve the accuracy of identifying and segmenting the nasopharyngeal carcinoma lymph node area.
The invention provides a method for identifying and segmenting nasopharyngeal carcinoma lymph node regions, which comprises the following steps: acquiring a magnetic resonance image to be identified and segmented; identifying and segmenting the magnetic resonance image through a preset lymph identification and segmentation model so as to obtain a segmented region image; the lymph identification segmentation model is an end-to-end three-dimensional deep supervision convolutional neural network three-dimensional model from coarse to fine.
In one embodiment, the magnetic resonance image is subjected to recognition segmentation through a preset lymph recognition segmentation model, so as to obtain a segmented region image, and the method specifically includes: performing data enhancement processing on the magnetic resonance image to obtain a magnetic resonance enhancement data set; performing a filtering convolution process on the magnetic resonance enhancement data set so as to obtain a characteristic data set of the magnetic resonance enhancement data set; carrying out deconvolution processing on the characteristic data group so as to obtain nasopharyngeal carcinoma metastatic lymph node region images with different segmentation granularities and corresponding existence probabilities; screening out one or more corresponding first nasopharyngeal carcinoma metastatic lymph node regional images of which the existing probability is greater than a preset probability threshold; screening out one or more second nasopharyngeal darcinoma metastatic lymph node regional images corresponding to the first nasopharyngeal darcinoma metastatic lymph node regional image of which the segmentation granularity reaches a preset granularity threshold; and acquiring a third nasopharyngeal carcinoma metastatic lymph node regional image with the highest segmentation granularity in the second nasopharyngeal carcinoma metastatic lymph node regional image, and outputting the third nasopharyngeal carcinoma metastatic lymph node regional image as a segmentation regional image.
In one embodiment, before acquiring the magnetic resonance image to be identified and segmented, the identification segmentation method further comprises: acquiring a preset first training image data set and a preset model to be trained; the model to be trained is an end-to-end, coarse-to-fine and three-dimensional deep supervision convolutional neural network three-dimensional model; processing the training image data set through a preset double-examination data processing method to obtain a second training image data set; the second training image data set comprises a plurality of second training images; and inputting the plurality of second training images into the model to be trained to calculate corresponding predicted values, and updating model parameters of the model to be trained according to each predicted value and the corresponding real value, so as to obtain the lymph identification segmentation model.
In one embodiment, after performing identification segmentation on the magnetic resonance image through a preset lymph identification segmentation model to obtain a segmentation region image, the identification segmentation method further includes: and sending the segmentation area image to a user.
The invention also provides a device for identifying and segmenting the nasopharyngeal darcinoma lymph node area, which comprises a data acquisition unit and an identification and segmentation unit, wherein the data acquisition unit is used for acquiring the magnetic resonance image to be identified and segmented; the identification and segmentation unit is used for identifying and segmenting the nasopharyngeal carcinoma lymph nodes through a preset lymph identification and segmentation model so as to obtain a segmentation area image; the lymph identification segmentation model is an end-to-end three-dimensional deep supervision convolutional neural network three-dimensional model from coarse to fine.
In one embodiment, the recognition segmentation apparatus further includes a model training unit, and the model training unit is configured to: acquiring a preset first training image data set and a preset model to be trained; the model to be trained is an end-to-end, coarse-to-fine and three-dimensional deep supervision convolutional neural network three-dimensional model; processing the training image data set through a preset double-examination data processing method to obtain a second training image data set; the second training image data set comprises a plurality of second training images; and inputting the plurality of second training images into the model to be trained to calculate corresponding predicted values, and updating model parameters of the model to be trained according to each predicted value and the corresponding real value, so as to obtain the lymph identification segmentation model.
In one embodiment, the identification and segmentation apparatus further comprises an image transmission unit configured to: and sending the segmentation area image to a user.
In one embodiment, the identification and segmentation unit is further configured to: performing data enhancement processing on the magnetic resonance image to obtain a magnetic resonance enhancement data set; performing a filtering convolution process on the magnetic resonance enhancement data set so as to obtain a characteristic data set of the magnetic resonance enhancement data set; carrying out deconvolution processing on the characteristic data group so as to obtain nasopharyngeal carcinoma metastatic lymph node region images with different segmentation granularities and corresponding existence probabilities; screening out one or more corresponding first nasopharyngeal carcinoma metastatic lymph node regional images of which the existence probability is greater than a preset probability threshold; screening out one or more second nasopharyngeal darcinoma metastatic lymph node regional images corresponding to the first nasopharyngeal darcinoma metastatic lymph node regional image of which the segmentation granularity reaches a preset granularity threshold; and acquiring a third nasopharyngeal darcinoma metastatic lymph node regional image with the highest segmentation granularity in the second nasopharyngeal darcinoma metastatic lymph node regional image, and outputting the third nasopharyngeal darcinoma metastatic lymph node regional image as a segmentation regional image.
The invention also provides a system for identifying and segmenting the nasopharyngeal darcinoma lymph node area, which comprises an identifying and segmenting module and a data storage module, wherein the identifying and segmenting module is in communication connection with the data storage module and is used for executing the method for identifying and segmenting the nasopharyngeal darcinoma lymph node area according to the data stored in the data storage module.
Compared with the prior art, the embodiment of the invention has the following beneficial effects:
the invention provides a method, a device and a system for identifying and segmenting a nasopharyngeal darcinoma lymph node area.
Furthermore, the method, the device and the system for identifying and segmenting the nasopharyngeal darcinoma lymph node area, provided by the invention, process the first training image data set through a preset double-examination data processing method to obtain a second training image data set, and then input the second training image data set into a model to be trained for training, so that a reasonable model is designed fully according to the morphological characteristics of the lymph node, and the accuracy of identifying and segmenting the nasopharyngeal darcinoma lymph node area is improved.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
Detailed description of the preferred embodiment
The embodiment of the invention firstly describes a method for identifying and segmenting a nasopharyngeal carcinoma lymph node area. Fig. 1 shows a flow chart of an embodiment of a method for identifying and segmenting a nasopharyngeal carcinoma lymph node area according to the present invention.
As shown in fig. 1, the method comprises the steps of:
s1, acquiring a segmented magnetic resonance image to be identified.
And S2, identifying and segmenting the magnetic resonance image through a preset lymph identification and segmentation model so as to obtain a segmentation region image.
The lymph identification segmentation model is an end-to-end, coarse-to-fine and three-dimensional deep supervision convolutional neural network three-dimensional model. In order to improve the accuracy of identification and segmentation of a nasopharyngeal carcinoma lymph node area, the embodiment of the invention designs an end-to-end, coarse-to-fine, three-dimensional deep supervised convolutional neural network three-dimensional model (3D CF-CNN) to identify and segment a magnetic resonance image.
The model starts from a coarse scale, gradually increases the identification segmentation size according to the identification result to further refine positioning, draws the attention of a fine scale in the area detected by the coarse scale, and thus can rapidly and accurately segment the metastatic lymph nodes.
In one embodiment, the magnetic resonance image is identified and segmented by a preset lymph identification and segmentation model, so as to obtain a segmented region image, which specifically includes: performing data enhancement processing on the magnetic resonance image to obtain a magnetic resonance enhancement data set; performing a filtering convolution process on the magnetic resonance enhancement data set so as to obtain a characteristic data set of the magnetic resonance enhancement data set; carrying out deconvolution processing on the characteristic data group so as to obtain nasopharyngeal carcinoma metastatic lymph node region images with different segmentation granularities and corresponding existence probabilities; screening out one or more corresponding first nasopharyngeal carcinoma metastatic lymph node regional images of which the existence probability is greater than a preset probability threshold; screening out one or more second nasopharyngeal darcinoma metastatic lymph node regional images corresponding to the first nasopharyngeal darcinoma metastatic lymph node regional image of which the segmentation granularity reaches a preset granularity threshold; and acquiring a third nasopharyngeal darcinoma metastatic lymph node regional image with the highest segmentation granularity in the second nasopharyngeal darcinoma metastatic lymph node regional image, and outputting the third nasopharyngeal darcinoma metastatic lymph node regional image as a segmentation regional image. In one embodiment, the predetermined probability threshold is 0.5. The nasopharyngeal carcinoma metastasis lymph node area images with different segmentation granularities are characteristic images, and the data representation form of the nasopharyngeal carcinoma metastasis lymph node area images and the corresponding existence probability comprises a probability map.
In practical applications, since images can be represented at different scales, the model can also supervise training at different scales. Where S ∈ {1, …, S } represents a set of different granularity of model output, where S ═ 1 represents the finest granularity, and the resolution at this level is exactly the same as the resolution of the model input image.
The optimization process of the model is as follows:
1. the target pyramid is constructed by expansion and maximum pooling.
target [ s ═ 1] is the label of the lymph node true, then the coarse scale target iteration is derived as follows:
target’[s]=dilate(target[s]);
target[s+1]=maxpool(target’[s]);
the S scale penalty function is weighted by the scale lymph node true label target [ S ] and the coarse scale prediction probability map in pixels, as follows:
prediction’[s+1]=upsample(prediction[s+1]);
weightmap[s]=max(target’[s],prediction’[s+1]);
3. the decoding module of each scale s generates a correction map correction [ s ], and the prediction of one scale is the accumulation of the prediction of its direct coarse scale's [ s +1] and the correction map correction [ s ] of its own scale, as follows:
logit’[s+1]=upsample(logit[s+1]);
logits=merge(logit’[s+1],correction[s]);
and at the 3D CF-CNN model end, a decoding module of each scale outputs a lymph node segmentation probability map. At the target end, a multi-scale pyramid is constructed from manually delineated labels, i.e., lymph node regions, to match the size of the output probability map. The output of each scale is then compared to the corresponding output in the target pyramid. All the losses are added together as a total loss.
In deep attention supervision of 3D CF-CNN, the coarse layer is supervised to segment a superset of the foreground. Its predictive probability map, after upsampling to fine resolution, is used as a weight map to segment the more compact superset of the foreground. Different horizontal weight maps supervise foreground segmentation at different segmentation granularities until the finest resolution is reached. Such a process mimics how humans gradually zoom in to locate an organ and eventually recognize it at the pixel or voxel level. When an organ needs to be searched, rough positioning is firstly carried out; after positioning, the pixel or voxel level can be read carefully; in this process, once the organ is located on a high scale, other parts of the image become irrelevant to the search on a low level. Also in coarse-to-fine depth attention supervision, the coarse-scale predictive probability map is used as a weight map to exclude other irrelevant parts and focus the fine-scale only within the foreground superset that the coarse-scale has identified. In this way, the model can identify lymph nodes with different forms, and the metastatic lymph nodes can be quickly and accurately segmented.
The embodiment of the invention describes a method for identifying and segmenting a nasopharyngeal darcinoma lymph node area, which identifies and segments a magnetic resonance image to be identified and segmented through an end-to-end three-dimensional deep supervised convolutional neural network three-dimensional model from coarse to fine so as to obtain a segmented area comprising the nasopharyngeal darcinoma lymph node.
Detailed description of the invention
Furthermore, the embodiment of the invention also describes a method for identifying and segmenting the nasopharyngeal carcinoma lymph node area. Fig. 2 shows a flow chart of another embodiment of the segmentation method for identifying nasopharyngeal carcinoma lymph node area according to the present invention.
As shown in fig. 2, the method comprises the steps of:
a1, acquiring a preset first training image data set and a preset model to be trained.
The model to be trained is an end-to-end, coarse-to-fine and three-dimensional deep supervision convolutional neural network three-dimensional model. Wherein the first training image dataset comprises nasopharyngeal carcinoma magnetic resonance imaging data collected from a hospital or medical center.
In one embodiment, the model to be trained comprises an input module, an encoding module, a decoding module and an output module; the input module performs data enhancement processing on input data (including data standardization, x-axis and y-axis deformation, rotation along a Z-axis and horizontal overturning); the coding module is used for extracting image characteristics related to metastatic lymph nodes, the coding module comprises a plurality of residual volume blocks with different levels, and each residual volume block adopts a mode of 'activation layer (used for preliminarily filtering data characteristics) in front of and convolution layer behind' (along with continuous deepening of the coding module, the image characteristics extracted by the convolution blocks are more abstract, and the more abstract characteristics can represent input data); the decoding module includes several residual convolution blocks that are used for deconvolution to output different segmentation granularities for nasopharyngeal carcinoma metastatic lymph node regions (each segmentation granularity represents a segmentation accuracy and a specific resolution of the output image) and corresponding probabilities (regions with a probability greater than 0.5 are considered to contain lymph nodes), while the top layer is a probability map of metastatic lymph node regions with the same resolution as the input image.
And A2, processing the first training image data set through a preset double-trial data processing method to obtain a second training image data set.
The second training image data set includes a plurality of second training images. After the first training image data set is obtained, in order to ensure the accuracy and validity of the data used for model training, the first training image data set needs to be checked and screened through a "double-check mechanism" (i.e., an experienced doctor is responsible for sketching and an expert is responsible for checking), and the data obtained through checking and screening needs to be preprocessed. In one embodiment, the pre-processing includes image data cleaning, data normalization, panning and enhanced image registration, among other operations.
A3, inputting the second training images into the model to be trained to calculate corresponding predicted values, and updating model parameters of the model to be trained according to each predicted value and the corresponding real value, so as to obtain the lymph identification segmentation model.
In order to train the model to be trained, dividing the obtained second training image into a training set, a verification set and a test set according to a preset division ratio, wherein the model is trained by the training set, the model is evaluated and selected by the verification set, and the robustness and the generalization ability of the model are tested by the test set.
A4, acquiring a segmented magnetic resonance image to be identified.
And A5, carrying out identification segmentation on the magnetic resonance image through a preset lymph identification segmentation model so as to obtain a segmentation region image.
The lymph identification segmentation model is an end-to-end, coarse-to-fine and three-dimensional deep supervision convolutional neural network three-dimensional model. In order to improve the accuracy of identification and segmentation of a nasopharyngeal carcinoma lymph node area, the embodiment of the invention designs an end-to-end, coarse-to-fine, three-dimensional deep supervised convolutional neural network three-dimensional model (3D CF-CNN) to identify and segment a magnetic resonance image.
The model starts from a coarse scale, gradually increases the identification segmentation size according to the identification result to further refine positioning, draws the attention of a fine scale in the area detected by the coarse scale, and thus can rapidly and accurately segment the metastatic lymph nodes.
In one embodiment, the magnetic resonance image is subjected to recognition segmentation through a preset lymph recognition segmentation model, so as to obtain a segmented region image, and the method specifically includes: performing data enhancement processing on the magnetic resonance image to obtain a magnetic resonance enhancement data set; performing a filtering convolution process on the magnetic resonance enhancement data set so as to obtain a characteristic data set of the magnetic resonance enhancement data set; carrying out deconvolution processing on the characteristic data group so as to obtain nasopharyngeal carcinoma metastatic lymph node region images with different segmentation granularities and corresponding existence probabilities; screening out one or more corresponding first nasopharyngeal carcinoma metastatic lymph node regional images of which the existence probability is greater than a preset probability threshold; screening out one or more second nasopharyngeal darcinoma metastatic lymph node regional images corresponding to the first nasopharyngeal darcinoma metastatic lymph node regional image of which the segmentation granularity reaches a preset granularity threshold; and acquiring a third nasopharyngeal darcinoma metastatic lymph node regional image with the highest segmentation granularity in the second nasopharyngeal darcinoma metastatic lymph node regional image, and outputting the third nasopharyngeal darcinoma metastatic lymph node regional image as a segmentation regional image.
A6: and sending the segmentation area image to a user.
In one embodiment, the form of sending to the user comprises displaying the segmented region image to the user on a display screen, and sending the segmented region image to the user terminal of the user through the communication module.
The embodiment of the invention describes a method for identifying and segmenting a nasopharyngeal darcinoma lymph node area, which identifies and segments a magnetic resonance image to be identified and segmented through an end-to-end three-dimensional deep supervised convolutional neural network three-dimensional model from coarse to fine so as to obtain a segmented area comprising the nasopharyngeal darcinoma lymph node; furthermore, the method for identifying and segmenting the nasopharyngeal darcinoma lymph node area described in the embodiment of the present invention further processes the first training image data set by a preset double-examination data processing method to obtain the second training image data set, and inputs the second training image data set into the model to be trained for training, so as to design a reasonable model according to the morphological characteristics of the lymph node, thereby improving the accuracy of identifying and segmenting the nasopharyngeal darcinoma lymph node area.
Detailed description of the invention
Besides the method, the embodiment of the invention also describes a device for identifying and segmenting the nasopharyngeal carcinoma lymph node area. Fig. 3 is a block diagram showing an embodiment of a nasopharyngeal carcinoma lymph node region identification and segmentation apparatus according to the present invention.
As shown in fig. 3, the identification division apparatus includes a data acquisition unit 11 and an identification division unit 12.
The data acquisition unit 11 is configured to acquire a magnetic resonance image of the segment to be identified.
The identifying and segmenting unit 12 is configured to identify and segment the nasopharyngeal carcinoma lymph nodes through a preset lymph identifying and segmenting model, so as to obtain segmented region images; the lymph identification segmentation model is an end-to-end three-dimensional deep supervision convolutional neural network three-dimensional model from coarse to fine. The model starts from a coarse scale, gradually increases the identification segmentation size according to the identification result to further refine positioning, draws the attention of a fine scale in the area detected by the coarse scale, and thus can rapidly and accurately segment the metastatic lymph nodes.
In one embodiment, the identification partitioning unit 12 is further configured to: performing data enhancement processing on the magnetic resonance image to obtain a magnetic resonance enhancement data set; performing a filtering convolution process on the magnetic resonance enhancement data set so as to obtain a characteristic data set of the magnetic resonance enhancement data set; carrying out deconvolution processing on the characteristic data group so as to obtain nasopharyngeal carcinoma metastatic lymph node region images with different segmentation granularities and corresponding existence probabilities; screening out one or more corresponding first nasopharyngeal carcinoma metastatic lymph node regional images of which the existing probability is greater than a preset probability threshold; screening out one or more second nasopharyngeal darcinoma metastatic lymph node regional images corresponding to the first nasopharyngeal darcinoma metastatic lymph node regional image of which the segmentation granularity reaches a preset granularity threshold; and acquiring a third nasopharyngeal darcinoma metastatic lymph node regional image with the highest segmentation granularity in the second nasopharyngeal darcinoma metastatic lymph node regional image, and outputting the third nasopharyngeal darcinoma metastatic lymph node regional image as a segmentation regional image.
In one embodiment, the recognition segmentation apparatus further includes a model training unit, and the model training unit is configured to: acquiring a preset first training image data set and a preset model to be trained; the model to be trained is an end-to-end coarse-to-fine three-dimensional deep supervision convolutional neural network three-dimensional model; processing the training image data set through a preset double-examination data processing method to obtain a second training image data set; the second training image data set comprises a plurality of second training images; and inputting the plurality of second training images into the model to be trained to calculate corresponding predicted values, and updating model parameters of the model to be trained according to each predicted value and the corresponding real value, so as to obtain the lymph identification segmentation model.
In one embodiment, the identification and segmentation apparatus further comprises an image transmission unit configured to: and sending the segmentation area image to a user.
The embodiment of the invention describes a device for identifying and segmenting a nasopharyngeal darcinoma lymph node area, which identifies and segments a magnetic resonance image to be identified and segmented through an end-to-end three-dimensional deep supervised convolutional neural network three-dimensional model from coarse to fine so as to obtain a segmented area comprising the nasopharyngeal darcinoma lymph node; furthermore, the apparatus for identifying and segmenting a nasopharyngeal darcinoma lymph node area described in the embodiment of the present invention further processes the first training image data set by a preset double-examination data processing method to obtain the second training image data set, and then inputs the second training image data set into the model to be trained for training, so as to design a reasonable model according to the morphological characteristics of the lymph node, thereby improving the accuracy of identifying and segmenting the nasopharyngeal darcinoma lymph node area.
Detailed description of the invention
In addition to the above method and apparatus, the present invention also describes a nasopharyngeal carcinoma lymph node region identification and segmentation system. Fig. 4 shows a block diagram of an embodiment of a nasopharyngeal carcinoma lymph node area identification segmentation system according to the present invention.
As shown in fig. 4, the identification and segmentation system comprises an identification and segmentation module 1 and a data storage module 2, wherein the identification and segmentation module 1 is in communication connection with the data storage module 2, the data storage module 2 is used for storing all data, and the identification and segmentation module 1 is used for executing the identification and segmentation method of the nasopharyngeal carcinoma lymph node region according to the data stored in the data storage module 2.
In one embodiment, the identification and segmentation system further comprises a user interaction module, wherein the user interaction module is used for interacting with the identification and segmentation module 1 and the data storage module 2 according to instructions input by a user, and sending finally obtained segmented region images and other information to the user.
In one embodiment, the user interaction module includes a touch display/non-touch display, an input keyboard, a virtual keyboard, an indicator light, a microphone, a speaker, and combinations of one or more of the foregoing.
The embodiment of the invention describes a nasopharyngeal darcinoma lymph node area identifying and segmenting system, which identifies and segments a magnetic resonance image to be identified and segmented through an end-to-end three-dimensional depth supervision convolutional neural network three-dimensional model from coarse to fine so as to obtain segmented areas comprising nasopharyngeal darcinoma lymph nodes, and the identifying and segmenting system improves the accuracy of identifying and segmenting the nasopharyngeal darcinoma lymph node areas; furthermore, the system for identifying and segmenting the nasopharyngeal darcinoma lymph node area described in the embodiment of the present invention further processes the first training image data set by a preset double-examination data processing method to obtain the second training image data set, and then inputs the second training image data set into the model to be trained for training, so as to design a reasonable model according to the morphological characteristics of the lymph node, thereby improving the accuracy of identifying and segmenting the nasopharyngeal darcinoma lymph node area.
The above-mentioned embodiments are provided to further explain the objects, technical solutions and advantages of the present invention in detail, and it should be understood that the above-mentioned embodiments are only examples of the present invention and are not intended to limit the scope of the present invention. It should be understood that any modifications, equivalents, improvements and the like, which come within the spirit and principle of the invention, may occur to those skilled in the art and are intended to be included within the scope of the invention.