Method for Detecting Narrow Region of Cerebrovascular CTA Image

Abstract

The present invention application discloses a method for detecting narrow regions in cerebrovascular CTA images. The method includes: inputting multiple cerebrovascular CTA images to be processed; obtaining a first feature map corresponding to the cerebrovascular CTA images to be processed through a first preset convolutional neural network; obtaining a second feature map of the cerebrovascular CTA images to be processed after passing the first feature map through a second preset convolutional neural network; obtaining a third feature map of the cerebrovascular CTA images to be processed after passing the second feature map through an RPN network; and obtaining a final result prediction map after passing the third feature map through a cascade detector. This application solves the technical problem of poor detection and recognition. Through the method of this application, small narrow regions can be located.

Description

Technical Field

[0001] The present invention relates to a medical image processing method, and in particular to a method for detecting narrow areas in cerebral vascular CTA images. Background Art

[0002] Due to the characteristics of rapid examination, less invasiveness and low cost of CTA examination technology, it is currently the recognized preferred examination method for cerebrovascular diseases. However, the lesion sites detected by CTA examination technology are usually not obvious, and in the clinical diagnosis process, it is highly dependent on the experience of physicians. Long-term image diagnosis increases the probability of misdiagnosis or missed diagnosis.

[0003] In addition, since CTA images are horizontal sectional views, cerebral blood vessels can exhibit different shapes, such as circular, oval, fusiform and irregular shapes, on different tomographic levels according to their orientations in different parts. Coupled with the heterogeneity of different patients, there are differences in the location and degree of blood vessel stenosis, and the small narrow area will lead to problems such as sample imbalance during positive and negative sample sampling.

[0004] The existence of the above problems will directly affect the lesion detection effect. At present, no effective solution has been proposed for the problem of poor detection and processing effect in related technologies. Summary of the Invention

[0005] The purpose of the present invention is to overcome the deficiencies of the existing technologies and provide a method for detecting narrow areas in cerebral vascular CTA images with accurate detection effect.

[0006] To achieve the above purpose, the technical solution adopted by the present invention is:

[0007] The method for detecting narrow areas in cerebral vascular CTA images of the present invention includes the following steps:

[0008] Step 101, input multiple cerebral vascular CTA images to be processed through opencv in Python programming software;

[0009] The cerebral vascular CTA images include three-dimensional imaging of transverse, coronal and sagittal positions, and the detection of the narrow area of the cerebral blood vessel is carried out on the transverse position;

[0010] Step 102, input each cerebral vascular CTA image to be processed into a first preset convolutional neural network for processing to obtain a first feature map corresponding to each cerebral vascular CTA image to be processed. Each first feature map includes low-dimensional image information of the contour curves of cerebral blood vessels and their surrounding tissues;

[0011] The first preset convolutional neural network uses the Libra RCNN convolutional neural network. Among them, in the convolutional layer of the backbone network of the Libra RCNN convolutional neural network, a deformable convolutional layer is used. By adding an offset, the convolutional kernel is displaced in different directions to obtain a first feature map containing low-dimensional image information of the contour curves of the cerebral blood vessels and their surrounding tissues;

[0012] Step 103: Obtain multiple global information maps containing the position and morphology of the cerebral blood vessels through image processing. The specific steps are as follows: Process multiple first feature maps through a second preset convolutional neural network to obtain multiple second feature maps of the to-be-processed cerebral CTA image. The multiple first feature maps and the multiple second feature maps maintain the corresponding relationship of the same cross-sectional plane; Each of the second feature maps includes high-dimensional image information of the position of the cerebral blood vessels and their surrounding tissues;

[0013] The second preset convolutional neural network uses the Libra RCNN convolutional neural network. Among them, in the convolutional layer of the balanced pyramid of the Libra RCNN convolutional neural network, a non-local neural network layer is used. By performing repeated convolutional operation operations, global information integration of the image information on multiple second feature maps is realized;

[0014] Step 104: Input the second feature map into the RPN network, and perform convolution and downsampling operations through multiple convolutional layers and pooling layers in the RPN in sequence, and output a third feature map. The third feature map is multi-scale and contains proposed candidate regions of multiple narrow parts;

[0015] Step 105: Determine the position and size of the narrow area through a cascade detector to obtain a prediction result map containing the position and size of the prompted narrow area.

[0016] The beneficial effect of the present invention is that the detection effect of the narrow area of the cerebral blood vessels is accurate. Brief Description of the Drawings

[0017] The drawings constituting a part of this application are used to provide a further understanding of this application. The present invention will be further described below in conjunction with the drawings and embodiments. In the drawings:

[0018] Figure 1 It is a schematic diagram of a method for detecting a narrow area of a cerebral CTA image according to an embodiment of the present application; Detailed Embodiment

[0019] To enable those skilled in the art to better understand the solution of this application, the following will clearly and completely describe the technical solution in the embodiments of this application in conjunction with the accompanying drawings in the embodiments of this application. Obviously, the described embodiments are only a part of the embodiments of this application, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments in this application without creative efforts shall fall within the scope of protection of this application.

[0020] The method for detecting the stenosis area of cerebral vascular CTA images of the present invention includes the following steps:

[0021] Step 101, input multiple cerebral vascular CTA images to be processed through opencv in the Python programming software. Those skilled in the art can also input images through other means such as numpy and pillow;

[0022] The cerebral vascular CTA images include three-dimensional imaging in the transverse, coronal, and sagittal planes, and the detection of the cerebral vascular stenosis area is carried out in the transverse plane.

[0023] Step 102, input each cerebral vascular CTA image to be processed into a first preset convolutional neural network for processing, and then obtain a first feature map corresponding to each cerebral vascular CTA image to be processed. Each of the first feature maps includes low-dimensional image information of the contour curves of the cerebral blood vessels and their surrounding tissues; the surrounding tissues include bony structures and soft tissues, etc.;

[0024] The first preset convolutional neural network is improved on the basis of the existing Libra RCNN convolutional neural network. The improvement lies in: using a deformable convolutional layer in the convolutional layer of the backbone network of the Libra RCNN convolutional neural network, and generating displacements of the convolutional kernel in different directions by adding offsets to obtain multiple first feature maps containing low-dimensional image information of the contour curves of the cerebral blood vessels and their surrounding tissues.

[0025] Step 103, obtain multiple global information maps containing the position and shape of the cerebral blood vessels through image processing. The specific steps are as follows: input multiple first feature maps into a second preset convolutional neural network for processing to obtain multiple second feature maps of the cerebral vascular CTA images to be processed. The multiple first feature maps and the multiple second feature maps maintain the corresponding relationship of the same cross-sectional plane, and there is no corresponding relationship in the image order of magnitude; each of the second feature maps includes high-dimensional image information such as the position of the cerebral blood vessels and their surrounding tissues;

[0026] The second preset convolutional neural network is improved on the basis of the existing Libra RCNN convolutional neural network. The improvement lies in that a non-local neural network layer is used in the convolutional layer in the balanced pyramid of the LibraRCNN convolutional neural network, and repeated convolution operations are performed to achieve global information integration of image information on multiple second feature maps. Those skilled in the art may also encode in other ways.

[0027] Step 104, input the second feature map into the RPN network, perform convolution and downsampling operations in turn through multiple convolution layers and pooling layers in the RPN, and output a third feature map, wherein the third feature map is multi-scale and contains multiple proposed candidate regions of narrow parts.

[0028] Step 105, determining the position and size of the narrow area through the cascade detector, and obtaining a prediction result map including the position and size of the narrow area. The specific implementation steps are as follows:

[0029] In the first step, each second feature map is matched with the third feature map and then input into the cascade detector. After three fine-tunings through the fully connected layer and the sigmoid function, the offset and the region of interest output by each detection head are decoded as the input of the region of interest in the next stage. The fine-tuning includes the fine-tuning of the position and category of the cerebrovascular detection box.

[0030] In the second step, the offset and region of interest outputted by the third fine-tuning are decoded and visualized through the matplotlib library in python. Based on the cerebrovascular CTA as the original image, colored rectangular boxes and labels are drawn as the prediction result map. The colored rectangular boxes represent narrow cerebrovascular vessels, and the label information is the confidence interval of the category to which it belongs.

[0031] Experimental analysis

[0032] The experimental data in this application consists of 79 patients, with a total of 1021 cerebrovascular CTA transverse images. The data set is divided into a training set and a test set, with 821 images as the training set and 200 images as the test set.

[0033] In order to evaluate and compare model performance, this application uses AP (Average Precision), FPS (Frames Per Second), AP 50 , A.P. 75 and AP S , where AP represents the average accuracy of the predicted box in the test set, and the frame rate per second (Frame Per Second, FPS) represents the number of predicted pictures per second. 50Represents the average precision at IoU = 0.50, AP 75 Represents the average precision at IoU = 0.75, and APS represents the average precision of small target objects (area < 32×32). For objects of class C, the calculation formula for its AP is: and The experimental results are shown in Table 1.

[0034] Table 1. Experimental results of this application and different object detection models.

[0035]

[0036] The above are only the preferred embodiments of the present invention and are not used to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions recorded in the foregoing embodiments or perform equivalent replacements for some of the technical features. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. A method for detecting the stenosis region of cerebral CTA images, characterized in that It includes the following steps: Step 101: Input multiple cerebral vascular CTA images to be processed through opencv in Python programming software; The cerebral vascular CTA images include three-dimensional imaging of transverse, coronal and sagittal planes, and the detection of cerebral vascular stenosis areas is carried out on the transverse plane; Step 102: Input each cerebral vascular CTA image to be processed into a first preset convolutional neural network for processing, and obtain a first feature map corresponding to each cerebral vascular CTA image to be processed. Each of the first feature maps includes low-dimensional image information of the contour curves of cerebral blood vessels and their surrounding tissues; The first preset convolutional neural network uses the Libra RCNN convolutional neural network. Among them, in the convolutional layer of the backbone network of the Libra RCNN convolutional neural network, a deformable convolutional layer is used to displace the convolutional kernel in different directions by adding offsets, and obtain multiple first feature maps containing low-dimensional image information of the contour curves of cerebral blood vessels and their surrounding tissues; Step 103: Obtain multiple global information maps containing the positions and morphologies of cerebral blood vessels through image processing. The specific steps are as follows: Process multiple first feature maps through a second preset convolutional neural network to obtain multiple second feature maps of the cerebral vascular CTA images to be processed. The multiple first feature maps and the multiple second feature maps maintain the corresponding relationship of the same cross-sectional plane. Each of the second feature maps includes high-dimensional image information of the positions of cerebral blood vessels and their surrounding tissues; The second preset convolutional neural network uses the Libra RCNN convolutional neural network. Among them, a non-local neural network layer is used in the convolutional layer of the balanced pyramid of the Libra RCNN convolutional neural network, and through repeated convolutional operation operations, the global information integration of the image information on multiple second feature maps is realized; Step 104: Input the second feature map into the RPN network, and perform convolution and downsampling operations through multiple convolutional layers and pooling layers in the RPN in sequence, and output a third feature map. The third feature map is multi-scale and contains proposed candidate regions of multiple stenosis sites; Step 105: Determine the positions and sizes of the stenosis regions through a cascade detector, and obtain a prediction result map containing the positions and sizes of the stenosis regions. The specific implementation steps are as follows: First step: Match each second feature map and the third feature map and input them into the cascade detector. Through three fine-tunings of the fully connected layer and the sigmoid function, the offsets and regions of interest output by each detection head are decoded as the input of the region of interest in the next stage. The fine-tuning includes the fine-tuning of the positions and categories of the cerebral vascular detection boxes; Second step: Decode the offsets and regions of interest output by the third fine-tuning, and perform visualization operations through the matplotlib library in Python. On the basis of the cerebral vascular CTA as the original image, draw a colored rectangular box and label as the prediction result map. The narrow cerebral blood vessels are in the colored rectangular box, and the label information is the confidence interval of the category to which it belongs.

Images