WO2021027240A1

WO2021027240A1 - Brain atrophy identification method, and device

Info

Publication number: WO2021027240A1
Application number: PCT/CN2019/130863
Authority: WO
Inventors: 倪浩; 郑永升; 石磊; 徐梦迪; 陈思杰
Original assignee: 上海依智医疗技术有限公司
Priority date: 2019-08-09
Filing date: 2019-12-31
Publication date: 2021-02-18
Also published as: CN110517766B; CN110517766A

Abstract

A brain atrophy identification method and a device, pertaining to the technical field of artificial intelligence. The method comprises: determining, by using a keyframe detection module, a keyframe in a brain image sequence (102); detecting, by using a key point detection module, key points in the keyframe, and determining a first-type classification indicator according to the key points in the keyframe (103); segmenting the keyframe by using an image segmentation module, and determining a second-type classification indicator (104); and identifying brain atrophy according to the first-type classification indicator and the second-type classification indicator (105). Different detection manners are used for characteristics of the different classification indicators, thereby improving detection accuracy of the classification indicators. In addition, using the first-type classification indicator and the second-type classification indicator to identify brain atrophy also improves accuracy in brain atrophy identification. Furthermore, a neural network model is used to automatically identify brain atrophy, thereby having low dependency on manual labor and high efficiency compared to manual measurement and calculation.

Description

Method and device for identifying brain atrophy

Cross references to related applications

This application claims the priority of a Chinese patent application filed with the Chinese Patent Office, the application number is 201910736591.1, and the application title is "Method and Device for Recognizing Brain Atrophy" on August 9, 2019, the entire content of which is incorporated into this application by reference .

Technical field

The embodiment of the present invention relates to the field of artificial intelligence technology, in particular to a method and device for identifying brain atrophy.

Background technique

Brain atrophy is an imaging manifestation of qualitative lesions in brain tissue that is smaller than normal. It can be caused by multiple factors such as genetics, neurological diseases, poisoning, and malnutrition. Among them, the most common is cerebral cortex atrophy, which shows flattened gyri, widened sulcus, enlarged ventricles and cisterns, and reduced brain weight. The clinical manifestations are memory loss, decreased thinking ability, emotional instability, and inability to concentrate. In severe cases, it will develop into dementia, language impairment, loss of intelligence, etc. Approximately 15 million people worldwide die from various brain atrophy-related diseases each year, and the mortality rate is increasing year by year. Brain atrophy-related diseases usually have a long course and slow onset, so it is not easy to be detected. Once significant symptoms appear, they cannot be reversed, which seriously affects the work and life of patients. Therefore, early diagnosis and treatment of brain atrophy plays an important role in improving the survival rate and quality of life of patients with brain atrophy.

At present, the main methods for diagnosing brain atrophy based on imaging are brain tissue volume measurement and linear measurement. The linear measurement method reflects the changes in the volume of intracranial cerebrospinal fluid through the measurement of one-dimensional linear indicators of the ventricles, sulci, and split brain, and then indirectly reflects the changes in brain parenchymal volume. The measurement site is clear and fixed, and the method is easy to implement, and is widely used in clinical practice. However, this method relies on doctors for manual measurement and calculation, which is highly subjective and inefficient.

Summary of the invention

Since the current methods for diagnosing brain atrophy based on imaging rely on manual measurement and calculation by doctors, which are subjective and low in efficiency, embodiments of the present invention provide a method and device for identifying brain atrophy.

In one aspect, an embodiment of the present invention provides a method for identifying brain atrophy, including:

Use the key frame detection module to determine the key frames in the brain image sequence;

Use a key point detection module to detect key points in the key frame, and determine the first type of grading index according to the key points in the key frame;

Use an image segmentation module to segment the key frame to determine the second type of grading index;

Recognizing brain atrophy according to the first type grading index and the second type grading index.

Optionally, the first type of grading index includes the largest diameter between the anterior horns, the smallest diameter between the anterior horns, the diameter of the choroid plexus of the lateral ventricle, and the outer diameter of the parietal ventricle;

The adopting the key point detection module to detect the key points in the key frame and determine the first type of classification index according to the key points in the key frame includes:

Use a key point detection module to detect the key points of the anterior angle and the key points of the lateral ventricle in the key frame;

Determine the maximum diameter between the rake angles and the minimum diameter between the rake angles according to the key points of the rake angle;

The inter-choroid plexus diameter of the lateral ventricle and the outer diameter of the parietal ventricle are determined according to the key points of the lateral ventricle.

Optionally, the second type of grading index includes the widest diameter of the third ventricle;

The segmentation of the key frame by the image segmentation module to determine the second type of classification index includes:

Determining the first area according to the key points in the key frame;

Binarize the first area to determine the second area;

Segmenting the second region by using an image segmentation algorithm to determine the third ventricle region;

The widest diameter of the third ventricle is determined according to the area of the third ventricle.

Optionally, the second type of grading index includes the maximum outer diameter of the skull and the maximum inner diameter of the skull;

Segmenting the key frame according to the CT value corresponding to the skull to determine the first boundary;

Segmenting the first boundary by using an image segmentation algorithm to determine the skull boundary;

The maximum outer diameter of the skull and the maximum inner diameter of the skull are determined according to the boundary of the skull.

Optionally, the identifying brain atrophy based on the first type grading index and the second type grading index includes:

Determining a brain atrophy assessment index according to the first type grading index and the second type grading index;

The brain atrophy evaluation index is input into a brain atrophy model to identify brain atrophy.

Optionally, the key point detection module and the key frame detection module are convolutional neural networks.

In one aspect, an embodiment of the present invention provides a device for identifying brain atrophy, including:

Key frame detection module, used to detect key frames in brain image sequences;

The key point detection module is used to detect the key points in the key frame, and determine the first type grading index according to the key points in the key frame;

The image segmentation module is used to segment the key frame and determine the second type of grading index;

The recognition module is used to recognize brain atrophy according to the first type grading index and the second type grading index.

The key point detection module includes:

The first detection module is used to detect the key points of the anterior angle and the key points of the lateral ventricle in the key frame;

A first determining module, configured to determine the maximum diameter between the rake angles and the minimum diameter between the rake angles according to the key points of the rake angle;

The second determining module is used to determine the inter-choroid plexus diameter of the lateral ventricle and the outer diameter of the parietal ventricle according to the key points of the lateral ventricle.

The image segmentation module includes:

The second detection module is configured to determine the first area according to the key points in the key frame;

The third detection module is configured to perform binarization processing on the first area to determine the second area;

The first segmentation module is configured to segment the second region using an image segmentation algorithm to determine the third ventricle region;

The third determining module is used to determine the widest diameter of the third ventricle according to the third ventricle area.

The image segmentation module includes:

The second segmentation module is configured to segment the key frame according to the CT value corresponding to the skull to determine the first boundary;

The third segmentation module is configured to use an image segmentation algorithm to segment the first boundary and determine the skull boundary;

The fourth determining module is used to determine the maximum outer diameter of the skull and the maximum inner diameter of the skull according to the boundary of the skull.

Optionally, the identification module includes:

A fifth determining module, configured to determine a brain atrophy assessment index according to the first type grading index and the second type grading index;

The sixth determining module is used to input the brain atrophy assessment index into a brain atrophy model to identify brain atrophy.

On the one hand, an embodiment of the present invention provides a computer device, including a memory, a processor, and a computer program stored in the memory and running on the processor, and a method for identifying brain atrophy when the processor executes the program A step of.

On the one hand, an embodiment of the present invention provides a computer-readable storage medium that stores a computer program executable by a computer device, and when the program runs on a computer device, the computer device is caused to execute a method for identifying brain atrophy A step of.

In the embodiment of the present invention, the key frames in the brain image sequence are detected first, and then the key points in the key frames are detected, the first type of classification index is determined based on the key points, and the second classification index is determined by segmenting the key frames. The characteristics of the grading index adopt different detection methods to improve the detection accuracy of the grading index. Using the first type of grading index and the second type of grading index to identify brain atrophy also improves the accuracy of identifying brain atrophy. Secondly, the neural network model is used to automatically identify brain atrophy. Compared with manual measurement and calculation, manual dependence is small and the efficiency is high.

Description of the drawings

In order to more clearly describe the technical solutions in the embodiments of the present invention, the following will briefly introduce the drawings needed in the description of the embodiments. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings may be obtained from these drawings without creative labor.

1 is a schematic flowchart of a method for identifying brain atrophy according to an embodiment of the present invention;

Figure 2 is a schematic diagram of a brain image provided by an embodiment of the present invention;

Figure 3a is a schematic structural diagram of a key frame detection module provided by an embodiment of the present invention;

FIG. 3b is a schematic structural diagram of a fast shrinking part in a key frame detection module provided by an embodiment of the present invention;

Figure 3c is a schematic structural diagram of a feature extraction part in a key frame detection module provided by an embodiment of the present invention;

3d is a schematic structural diagram of a feature extraction sub-module in a feature extraction part provided by an embodiment of the present invention;

3e is a schematic diagram of the structure of the classification neural network part of a key frame detection module provided by an embodiment of the present invention;

Figure 4a is a schematic diagram of a key frame provided by an embodiment of the present invention;

Figure 4b is a schematic diagram of a key frame provided by an embodiment of the present invention;

FIG. 5 is a schematic flowchart of a method for detecting a second type of classification index according to an embodiment of the present invention;

Fig. 6 is a schematic diagram of a key frame provided by an embodiment of the present invention;

FIG. 7 is a schematic flowchart of a method for detecting a second type of classification index according to an embodiment of the present invention;

FIG. 8 is a schematic diagram of a key frame provided by an embodiment of the present invention;

9 is a schematic flowchart of a method for determining the level of brain atrophy provided by an embodiment of the present invention;

10 is a schematic structural diagram of a device for identifying brain atrophy provided by an embodiment of the present invention;

FIG. 11 is a schematic structural diagram of a computer device provided by an embodiment of the present invention.

detailed description

In order to make the objectives, technical solutions and beneficial effects of the present invention clearer, the following further describes the present invention in detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described here are only used to explain the present invention, but not to limit the present invention.

The method for identifying brain atrophy in the embodiment of the present invention can be applied to assist in diagnosing brain atrophy. For example, a CT image of the brain of a patient is obtained, and then the method for identifying brain atrophy in the embodiment of the present invention is used to perform the CT image of the brain of the patient. Perform analysis, output the patient's brain atrophy recognition result, and the doctor will diagnose the patient based on the output recognition result.

Based on the above application scenario, the embodiment of the present invention provides a process of a method for identifying brain atrophy. The process of the method can be executed by an apparatus for identifying brain atrophy. As shown in FIG. 1, it includes the following steps:

Step S101, acquiring a brain image sequence.

Specifically, the brain image sequence includes multiple brain images, and the brain images may be computer tomography (CT) images, nuclear magnetic resonance images, and the like of the brain. Take the CT image sequence as an example. For a given CT modality medical digital imaging and communication (Digital Imaging and Communications in Medicine, DICOM for short) image sequence, read the image information of each frame, and interpolate and scale to a fixed size (for example, 512 *512 pixels), and adjust to a fixed window width and window level (brain window: W=80, L=40) to obtain brain image sequences. Exemplarily, the brain image may be specifically as shown in FIG. 2.

In step S102, the key frame detection module is used to determine the key frame in the brain image sequence.

Specifically, the key frame detection module may be a 2D convolutional neural network (Convolutional Neural Networks, CNN for short) or a 3D CNN. The key frame is a predefined brain image used to detect brain atrophy. The key frame can be the largest and clearest level of the basal ganglia lenticular nucleus and the display level of the ventricular body on both sides. The key frame is one or more brain image sequences. frame.

Take the key frame detection module as 2D CNN as an example, first introduce the network structure of the key frame detection module, as shown in Figure 3a, the network structure of the key frame detection module includes: fast reduction part, feature extraction part, classification neural network part .

As shown in Fig. 3b, the rapid reduction part is composed of a convolutional layer, a batch normalization (BN) layer, an activation function (Rectified Linear Unit, ReLU) layer, and a pooling layer. The size of the convolution kernel of the convolution layer is 5*5, and the interval is 2 pixels. The pooling layer is the maximum pooling of 2*2, and the area of the brain image can be quickly reduced by quickly reducing the part, and the side length becomes 1/4 of the original.

The feature extraction part is shown in Figure 3c, which is composed of N feature extraction sub-modules, where N is an integer greater than zero. Each feature extraction sub-module is shown in Figure 3d, including three bottleneck layers and a down-sampling layer. Both the bottleneck layer and the downsampling layer include three convolutional layers.

The bottleneck layer reduces the number of feature maps of the feature maps that are quickly reduced through the first convolutional layer and the second convolution layer, and then through the third convolution layer, the number of feature maps of the feature maps that are partially output is quickly reduced. Increase the number of feature maps back to the original number, and then directly add the feature maps output by the third convolutional layer and the feature maps output by the quickly reduced part of the output.

Part of the output feature map is quickly reduced through three bottleneck layers for feature extraction and then input to the downsampling layer. The downsampling layer reduces the number of feature maps output by the bottleneck layer through the first convolutional layer and the second convolution layer, and then increases the number of feature maps output by the bottleneck layer through the third convolutional layer Back to the original number of feature maps. At the same time, while increasing the number of feature maps, the third convolutional layer reduces the size of the feature maps to half by setting the convolution step size to 2. The feature map output by the bottleneck layer is reduced to half of the original size through 2*2 average pooling, and finally the feature map output by the third convolution layer and the feature map output by the bottleneck layer after the average pooling are added to the output. .

The classification neural network part is shown in Figure 3e, including a global average pooling layer, a random dropout layer, a fully connected layer, and a softmax layer. The input of the classification neural network part is the feature map output by the feature extraction part, and the output is the predicted category of the brain image. First, through the global average pooling layer, the feature map is extracted into a feature vector, and then the feature vector is input into the dropout layer, fully connected layer and softmax layer to obtain a classification confidence vector, each bit represents the confidence of a category, and The sum of all confidences is 1, and the bit with the highest confidence is output as the prediction category of the brain image.

Set the key frames in the brain image sequence to 4 frames, of which key frame 1, key frame 2, and key frame 3 are the largest and clearest display levels of the basal ganglia lenticular nucleus, and key frame 4 is the ventricular body on both sides Display level. When training the above-mentioned key frame detection module, a large number of brain CT image sequences are collected, and then the doctor marks 4 key frames from each CT image sequence. Then the CT image sequence marked with key frames is expanded by data enhancement to obtain training samples. Data enhancement methods include: random up, down, left, and right translation 0-20 pixels, random rotation -20 to 20 degrees, random zoom 0.8 to 1.2 times, etc. Input the training samples into the convolutional neural network for training. During training, the predicted category output by the convolutional neural network is compared with the category marked by the training sample, the cross entropy function is used as the target loss function, and the backpropagation algorithm is used to use the optimization method of sgd to iterate repeatedly until the target The function converges, and the key frame detection module is obtained.

When the key frame detection module is used to detect the key frames in the brain image sequence, first, for each frame of the brain image from the second frame to the second to the last frame in the brain image sequence, each frame of the brain image is taken as the central frame, and One frame of brain images before and after are spliced to determine a brain image sequence including 3 frames of brain images. Then the brain image sequence including 3 frames of brain images is input to the key frame detection module. The key frame detection module performs 5-class prediction on the brain image sequence including 3 frames of brain images, and obtains the confidence that the center frame is 5 categories . The 5 categories are classified into 0 to 4, where 0 means that the center frame is not a key frame, 1 means that the center frame is key frame No. 1, 2 means that the center frame is key frame No. 2, 3 means that the center frame is key frame No. 3, 4 Indicates that the center frame is the No. 4 key frame, and the category with the highest confidence is output as the category to which the center frame belongs.

It should be noted that when the key frame detection module is 3D CNN, it can target each frame of brain image from the third frame to the third to the last in the brain image sequence, with each frame of brain image as the central frame, and the front and back frames. Two frames of brain images are stitched together to determine a brain image sequence including 5 frames of brain images. Then the brain image sequence including 5 frames of brain images is input to the key frame detection module to perform 5 classification prediction, obtain the confidence that the central frame is 5 categories, and output the category with the highest confidence as the category to which the central frame belongs.

In addition, in addition to CNN, the key frame detection module can also be a traditional machine learning model, that is, the mosaic image sequence is input to the key frame detection module, and the key frame detection module calculates the grayscale and texture features of the brain image of each channel and stitches them into features vector. Then take the feature vector as input, use a classifier (such as support vector machine, random forest) to classify, and get the category of the center frame.

In step S103, the key point detection module is used to detect the key points in the key frame, and the first type of classification index is determined according to the key points in the key frame.

Specifically, the key point detection module can be a 2D CNN, which includes: a quick reduction part, a feature extraction part, and a classification neural network part.

The fast reduction part is composed of a convolutional layer, a batch normalization (BN) layer, an activation function (Rectified Linear Unit, ReLU) layer, and a pooling layer. The size of the convolution kernel of the convolution layer is 5*5, and the interval is 2 pixels. The pooling layer is the maximum pooling of 2*2, and the area of the key frame can be quickly reduced by quickly reducing the part, and the side length becomes 1/4 of the original.

The feature extraction part is composed of M feature extraction sub-modules, where M is an integer greater than zero. Each feature extraction sub-module contains three bottleneck layers and a down-sampling layer. Both the bottleneck layer and the downsampling layer include three convolutional layers.

The bottleneck layer reduces the number of feature maps of the feature maps that are quickly reduced through the first convolutional layer and the second convolution layer, and then through the third convolution layer, the number of feature maps of the feature maps that are partially output is quickly reduced. Increase the number of feature maps back to the original. The feature map output by the third convolutional layer and the feature map output by the rapid reduction part are directly added and output.

Quickly reduce part of the output feature maps through three bottleneck layers for feature extraction and then input to the downsampling layer. The downsampling layer passes through the first convolutional layer and the second convolutional layer to output the number of feature maps of the bottleneck layer. Reduce, and then increase the number of feature maps output by the bottleneck layer back to the original number of feature maps through the third convolutional layer. At the same time, while increasing the number of feature maps, the third convolutional layer reduces the size of the feature maps to half by setting the convolution step size to 2. The feature map output by the bottleneck layer is reduced to half of the original size through 2*2 average pooling, and finally the feature map output by the third convolution layer and the feature map output by the bottleneck layer after the average pooling are added to the output. .

The classification neural network part includes a global average pooling layer, a random dropout layer, a fully connected layer, and a linear conversion layer. The input of the classification neural network part is the feature map output by the feature extraction part, and the output is the key point coordinates. First, through the global average pooling layer, the feature map is extracted into a feature vector, and then the feature vector is input into the dropout layer, fully connected layer and linear conversion layer to obtain a two-dimensional coordinate vector. The two-dimensional coordinate vector indicates that the key point is on the X axis And Y axis position.

When training the above-mentioned key point detection module, a large number of brain CT image sequences are collected, and the doctor marks the key frames in each CT image sequence, and then marks the key points in the key frames. The CT image sequence marked with key frames and key points is expanded by data enhancement to obtain training samples. Data enhancement methods include: random up, down, left, and right translation 0-20 pixels, random rotation -20 to 20 degrees, random zoom 0.8 to 1.2 times, etc. Input the training samples into the convolutional neural network for training. During training, the key point coordinates predicted by the convolutional neural network are compared with the key point coordinates marked by the training sample. The Mean Square Error (MSE) function is used as the target loss function, and the back propagation algorithm is used to use the sgd Optimization method, iterate repeatedly until the objective function converges, and obtain the key point detection module. When the key point detection module is used to detect the key points in the key frame, the key frame is input to the key point detection module, and the coordinates of the key points in the key frame are output.

Optionally, the first type of grading index includes the largest diameter between the anterior horns, the smallest diameter between the anterior horns, the diameter of the choroid plexus of the lateral ventricle, and the outer diameter of the parietal ventricle. When detecting the first type of grading index, the key point detection module is used to detect the key points of the anterior angle and the key points of the lateral ventricle in the key frame. Then determine the maximum diameter between the anterior angles and the minimum diameter between the anterior angles according to the key points of the anterior angle, and determine the diameter of the choroid plexus of the lateral ventricle and the outer diameter of the lateral ventricle according to the key points of the lateral ventricle.

Exemplarily, after the key frame detection module is set to detect the brain image sequence, 4 key frames are determined, and the key point detection module performs key point detection on the 4 key frames respectively to determine the key points in each key frame. Set two key frames as shown in Figure 4a and Figure 4b. The key points of the front corner are the key point a ₁ , key point a ₂ , key point b ₁ , and key point b _{2 in} Figure 4a. The key point d ₁ and key point d _{2 in} Fig. 4a and the key point e ₁ and key point e _{2 in} Fig. 4b. And a ₁ and between key Key ₂ a comparator 4 for a key frame _2, _the distance between the key and the key point a ₁ key each keyframe detection of a ₁ and a key The maximum distance is determined as the maximum diameter A between the rake angles. The key point for b _2, the key point is detected in each frame of the key frame and key points b ₁ ₁ b and the distance between the keys b _2, and 4 compare the key frame and the key Key ₁ b ₂ b between The maximum distance is determined as the minimum diameter B between the rake angles. The key point for d ₁ and d ₂ key, the key is detected for each keyframe point d ₁ and a distance between key points between ₂ d _2, then compare four key frame and key key d ₁ d The maximum distance is determined as the diameter D of the choroid plexus of the lateral ventricle. _2, for the key from the key points and ₁ point e e e each detected key frame ₁ key frame and the key between the point e _2, 4 and Comparative key frame ₁ key e and the key between the point e ₂ The maximum distance is determined as the outer diameter E between the roof of the lateral ventricle.

Determine the first type of grading index by detecting the key points in each key frame, and then comparing the distance between the key points in the key frames of each frame, compared to determining the first type of grading index based on the key points in a single frame key frame , Improve the detection accuracy.

Step S104: Use the image segmentation module to segment the key frame to determine the second type of grading index.

In a possible implementation, the second type of grading index includes the widest diameter of the third ventricle, and detecting the second type of grading index includes the following steps, as shown in FIG. 5:

Step S501: Determine the first area according to the key points in the key frame.

Specifically, the key point detection module may be used to detect the key points of the third ventricle in the key frame, and then determine the first region based on the key points of the third ventricle and the key points of the anterior angle. Exemplarily, as shown in FIG. 6, the key points of the three ventricles are the key point c ₁ and the key point c ₂ shown in FIG. 6, combining key point b ₁ , key point b ₂ , key point c ₁ and key point c ₂ Determine the first area.

Step S502: Binarize the first area to determine the second area.

Specifically, for each pixel in the first region, when the intensity of the pixel is greater than the preset threshold, the pixel is determined as a part of the third ventricle, otherwise the pixel is determined as a part of the background, and the pixel in the first region Pixels with dot intensity greater than the preset threshold constitute the second area.

Step S503: Use an image segmentation algorithm to segment the second area to determine the third ventricle area.

In specific implementation, image segmentation algorithms include threshold-based segmentation methods, edge-based segmentation methods, region-based segmentation methods, and specific theories-based segmentation methods.

The basic idea of the threshold-based segmentation method is to calculate one or more gray-level thresholds based on the gray-level characteristics of the image, and compare the gray-level value of each pixel in the image with the threshold, and finally divide the pixels according to the comparison result. Into the appropriate category. Therefore, the most critical step of this type of method is to solve the optimal gray threshold according to a certain criterion function.

The edge-based segmentation method refers to the collection of continuous pixels on the boundary line of two different regions in the image, which reflects the discontinuity of the local features of the image, and reflects the sudden change of image characteristics such as grayscale, color, and texture. Generally, the edge-based segmentation method refers to the edge detection based on the gray value, which is a method based on the observation that the edge gray value will show a step-shaped or roof-shaped change.

The region-based segmentation method divides the image into different regions according to the similarity criterion, and mainly includes several types such as seed region growth method, region split and merge method and watershed method. Among them, the watershed method is a mathematical morphological segmentation method based on topological theory. The basic idea is to regard the image as a geodetic topological topography, and the gray value of each pixel in the image represents the altitude of the point. , Each local minimum and its affected area is called a catchment basin, and the boundary of the catchment basin forms a watershed. The realization of this algorithm can be simulated as a flooding process, the lowest point of the image is first submerged, and then the water gradually submerges the entire valley. When the water level reaches a certain height, it will overflow. At this time, build a dam where the water overflows. Repeat this process until all the pixels on the image are submerged. At this time, the built series of dams become the watershed that separates the basins. . The watershed algorithm has a good response to weak edges, but the noise in the image will cause the watershed algorithm to produce over-segmentation.

Step S504: Determine the widest diameter of the third ventricle according to the area of the third ventricle.

Specifically, the maximum width of the third ventricle region is determined as the widest diameter C of the third ventricle. In a possible implementation manner, when the key frames are multiple frames, the above method can be used to detect the three ventricle area in each key frame, and determine the maximum width of the three ventricle area in each key frame. Then sort the maximum width of the third ventricle region in each key frame in the order from largest to smallest, and determine the largest width ranked first as the widest diameter of the third ventricle. By combining key point detection and image segmentation algorithms to detect the second type of grading index, the accuracy of detecting the second type of grading index is effectively improved.

In a possible implementation, the second type of grading index includes the maximum outer diameter of the skull and the maximum inner diameter of the skull, and detecting the second type of grading index includes the following steps, as shown in FIG. 7:

Step S701: Segment the key frame according to the CT value corresponding to the skull to determine the first boundary.

Specifically, tissues of different densities correspond to different CT values. The CT value corresponding to the skull is generally greater than 400HU, and its density is generally greater than that of other brain tissues. Therefore, the CT value corresponding to the skull can be used to segment the key frames and filter out other brains. Tissue, obtain a first boundary, and the first boundary includes at least the inner boundary of the skull and the outer boundary of the skull.

Step S702: Use an image segmentation algorithm to segment the first boundary to determine the skull boundary.

Step S703: Determine the maximum outer diameter of the skull and the maximum inner diameter of the skull according to the boundary of the skull.

Illustratively, the skull boundary in the key frame is set as shown in FIG. 8, the maximum outer diameter of the skull is the distance F, and the maximum inner diameter of the skull is the distance G. In a possible implementation, when the key frames are multiple frames, the above method can be used to detect the skull boundary in each key frame, and then determine the maximum outer diameter of the skull and the maximum inner diameter of the skull in each key frame. Then compare the size of the maximum outer diameter of the skull in each key frame, and use the largest maximum outer diameter of the skull as the second type of grading index. Compare the size of the largest inner diameter of the skull in each key frame, and use the largest inner diameter of the skull as the second type of grading index. Because the density of the skull is greater than that of other brain tissues, when the CT value is used to segment the key frames, a more accurate first boundary of the skull can be obtained, and then the image segmentation algorithm is used for segmentation to obtain the accurate boundary of the skull, and then based on The precise boundary of the skull determines the maximum outer diameter of the skull and the maximum inner diameter of the skull, effectively improving the accuracy of detecting the second type of grading index.

Step S105: Recognizing brain atrophy according to the first type grading index and the second type grading index.

Optionally, when recognizing brain atrophy based on the first type grading index and the second type grading index, the following steps are specifically included, as shown in FIG. 9:

Step S901: Determine a brain atrophy assessment index according to the first type grading index and the second type grading index.

Specifically, the brain atrophy assessment index includes Hastelloy value, ventricle index, lateral ventricle body index, lateral ventricle body width index, anterior horn index, and third ventricle width.

The Hastelloy value is the sum of the largest diameter between the rake angles and the smallest diameter between the rake angles. Generally speaking, the normal Hastelloy range for men is 3 to 6.9, and the normal range for women is 2.6 to 5.2.

The ventricular index is the ratio of the diameter of the choroid plexus of the lateral ventricle to the largest diameter of the anterior horn. Generally speaking, the normal ventricular index for men ranges from 1.1 to 3.3, and the normal range for women is 1.1 to 2.9.

The lateral ventricle body index is the ratio of the maximum outer diameter of the skull to the outer diameter of the lateral ventricle roof. Generally speaking, the normal lateral ventricle body index for men ranges from 4.3 to 7.4, and the normal lateral ventricle body index for women ranges from 3.9 to 7.7.

The lateral ventricle body width index is the ratio of the maximum inner diameter of the skull to the outer diameter of the roof of the lateral ventricle. Generally speaking, the normal lateral ventricle body width index for men ranges from 3.1 to 6.7, and the normal lateral ventricle body width index for women ranges from 3.5～6.8.

The anterior angle index is the ratio of the largest inner diameter of the skull to the largest diameter between the anterior horns. Generally speaking, the normal anterior angle index for men ranges from 2.8 to 8.2, and the normal anterior angle index for women ranges from 3.0 to 8.5.

Generally speaking, the normal width of the third ventricle in men ranges from 1 to 6.7, and the normal width of the third ventricle in women ranges from 0 to 7.

Step S902: Input the brain atrophy evaluation index into the brain atrophy model, and identify the brain atrophy.

Specifically, the brain atrophy model can only be used to identify whether there is brain atrophy, or it can be used to identify whether there is brain atrophy and the level of brain atrophy.

When the brain atrophy model is used to identify whether there is brain atrophy, the brain atrophy model can be a logistic regression model, a Bayesian model, etc.

In a possible implementation, the brain atrophy model is a logistic regression model, which specifically conforms to the following formula (1):

y ₁ =a ₁ x ₁ +a ₂ x ₂ +a ₃ x ₃ +a ₄ x ₄ +a ₅ x ₅ +a ₆ x ₆ ……………(1)

Among them, y ₁ is the value of brain atrophy, x _i is the evaluation index of brain atrophy, i=1, 2, 3, 4, 5, 6, a _j is a weighting coefficient, j=1, 2, 3, 4, 5, 6, 0≤a _j ≤1.

When the brain atrophy value y _{1 is} greater than the first threshold, it is determined that there is brain atrophy, and when the brain atrophy value y _{1 is} not greater than the first threshold, it is determined that there is no brain atrophy.

In a possible implementation, the brain atrophy model may be a Bayesian model, which specifically conforms to the following formula (2):

Among them, y ₁ is the brain atrophy category, x _i is the brain atrophy assessment index, i = 1, 2, 3, 4, 5, 6, C _k is the category item, C ₀ , C ₁ are specific categories, divided into: For brain atrophy and non-brain atrophy, 1 can be used to indicate brain atrophy, and 0 to indicate no brain atrophy.

When the brain atrophy model is used to identify whether there is brain atrophy and the level of brain atrophy, the brain atrophy model includes a brain atrophy determination module and a brain atrophy classification module. The brain atrophy determination module is first used to determine whether there is brain atrophy. When it is determined that there is brain atrophy, the brain atrophy classification module is used to further determine the level of brain atrophy. The brain atrophy determination module may be a logistic regression model, a Bayesian model, etc., and the brain atrophy classification module may be a logistic regression model, a Bayesian model, etc.

In a possible embodiment, the brain atrophy determination module is a logistic regression model, which specifically conforms to the above formula (1). When the value of the degree of brain atrophy y _{1 is} greater than the first threshold, it is determined that there is brain atrophy. When the value y _{1 is} not greater than the first threshold, it is determined that there is no brain atrophy.

When it is determined that there is brain atrophy, the brain atrophy grading module is used to determine the level of brain atrophy, where the brain atrophy grading module is a logistic regression model, which specifically conforms to the following formula (3):

y ₂ = b ₁ x ₁ + b ₂ x ₂ + b ₃ x ₃ + b ₄ x ₄ + b ₅ x ₅ + b ₆ x ₆ ……………(3)

Among them, y ₂ is the brain atrophy grading value, x _i is the brain atrophy assessment index, i = 1, 2, 3, 4, 5, 6, b _k is the weighting coefficient, k = 1, 2, 3, 4, 5, 6, 0≤b _k ≤1.

A comparison table of the relationship between the brain atrophy grading value and the brain atrophy level is preset. After the brain atrophy grading value is determined by the brain atrophy grading module, the brain atrophy level can be directly obtained by querying the comparison table.

In a possible implementation, the brain atrophy determination module may be a Bayesian model, which specifically conforms to the above formula (2).

When it is determined that there is brain atrophy, the brain atrophy grading module is used to determine the level of brain atrophy, where the brain atrophy grading module is a Bayesian model, which specifically conforms to the following formula (4):

Among them, y ₂ is the level of brain atrophy, x _i is the evaluation index of brain atrophy, i = 1, 2, 3, 4, 5, 6, D _k is the category item, D ₀ , D ₁ , and D ₂ are specific categories, Divided into mild brain atrophy, moderate brain atrophy and severe brain atrophy, you can use 0 to indicate mild brain atrophy, 1 to indicate moderate brain atrophy, and 2 to indicate severe brain atrophy.

In the embodiment of the present invention, the key frames in the brain image sequence are detected first, and then the key points in the key frames are detected, the first type of classification index is determined based on the key points, and the second classification index is determined by segmenting the key frames. The characteristics of the grading index adopt different detection methods to improve the detection accuracy of the grading index. Using the first type of grading index and the second type of grading index to identify brain atrophy and determine the level of brain atrophy also improves the accuracy of identifying brain atrophy and determining the classification of brain atrophy. Secondly, the neural network model is used to automatically identify brain atrophy and determine the level of brain atrophy. Compared with manual measurement and calculation, manual dependence is small and the efficiency is high.

Based on the same technical concept, an embodiment of the present invention provides a device for identifying brain atrophy. As shown in FIG. 10, the device can execute the process of the method for identifying brain atrophy. The device 1000 includes:

The key frame detection module 1001 is used to detect key frames in the brain image sequence;

The key point detection module 1002 is configured to detect the key points in the key frame, and determine the first type of classification index according to the key points in the key frame;

The image segmentation module 1003 is configured to segment the key frame and determine the second type of classification index;

The identification module 1004 is configured to identify brain atrophy according to the first type grading index and the second type grading index.

The key frame detection module 1001 includes:

The image segmentation module 1003 includes:

Optionally, the identification module 1004 includes:

Optionally, the key point detection module 1002 and the key frame detection module 1001 are convolutional neural networks.

Based on the same technical concept, an embodiment of the present invention provides a computer device. As shown in FIG. 11, it includes at least one processor 1101 and a memory 1102 connected to the at least one processor. The embodiment of the present invention does not limit the processor. The specific connection medium between 1101 and the memory 1102, the connection between the processor 1101 and the memory 1102 in FIG. 11 is taken as an example. The bus can be divided into address bus, data bus, control bus, etc.

In the embodiment of the present invention, the memory 1102 stores instructions that can be executed by at least one processor 1101. By executing the instructions stored in the memory 1102, the at least one processor 1101 can execute the steps included in the aforementioned method for identifying brain atrophy.

Among them, the processor 1101 is the control center of the computer equipment. It can use various interfaces and lines to connect various parts of the computer equipment, and recognize the brain by running or executing instructions stored in the memory 1102 and calling data stored in the memory 1102. Shrinking. Optionally, the processor 1101 may include one or more processing units, and the processor 1101 may integrate an application processor and a modem processor. The application processor mainly processes an operating system, a user interface, and an application program. The adjustment processor mainly deals with wireless communication. It is understandable that the foregoing modem processor may not be integrated into the processor 1101. In some embodiments, the processor 1101 and the memory 1102 may be implemented on the same chip, and in some embodiments, they may also be implemented on separate chips.

The processor 1101 may be a general-purpose processor, such as a central processing unit (CPU), a digital signal processor, an application specific integrated circuit (ASIC), a field programmable gate array or other programmable logic devices, discrete gates or transistors Logic devices and discrete hardware components can implement or execute the methods, steps, and logic block diagrams disclosed in the embodiments of the present invention. The general-purpose processor may be a microprocessor or any conventional processor. The steps of the method disclosed in the embodiments of the present invention may be directly embodied as being executed and completed by a hardware processor, or executed and completed by a combination of hardware and software modules in the processor.

As a non-volatile computer-readable storage medium, the memory 1102 can be used to store non-volatile software programs, non-volatile computer-executable programs, and modules. The memory 1102 may include at least one type of storage medium, for example, it may include flash memory, hard disk, multimedia card, card-type memory, random access memory (Random Access Memory, RAM), static random access memory (Static Random Access Memory, SRAM), Programmable Read Only Memory (PROM), Read Only Memory (ROM), Electrically Erasable Programmable Read-Only Memory (EEPROM), magnetic memory, disk , CD, etc. The memory 1102 is any other medium that can be used to carry or store desired program codes in the form of instructions or data structures and that can be accessed by a computer, but is not limited thereto. The memory 1102 in the embodiment of the present invention may also be a circuit or any other device capable of realizing a storage function, for storing program instructions and/or data.

Based on the same technical concept, the embodiments of the present invention provide a computer-readable storage medium, which stores a computer program executable by a computer device. When the program runs on the computer device, the computer device executes the recognition process. Steps of shrinking method.

Those skilled in the art should understand that the embodiments of the present invention may be provided as methods or computer program products. Therefore, the present invention may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention may adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program codes.

The present invention is described with reference to flowcharts and/or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and/or block in the flowchart and/or block diagram, and the combination of processes and/or blocks in the flowchart and/or block diagram can be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing equipment to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing equipment are generated It is a device that realizes the functions specified in one process or multiple processes in the flowchart and/or one block or multiple blocks in the block diagram.

These computer program instructions can also be stored in a computer-readable memory that can guide a computer or other programmable data processing equipment to work in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture including the instruction device. The device implements the functions specified in one process or multiple processes in the flowchart and/or one block or multiple blocks in the block diagram.

These computer program instructions can also be loaded on a computer or other programmable data processing equipment, so that a series of operation steps are executed on the computer or other programmable equipment to produce computer-implemented processing, so as to execute on the computer or other programmable equipment. The instructions provide steps for implementing functions specified in a flow or multiple flows in the flowchart and/or a block or multiple blocks in the block diagram.

Although the preferred embodiments of the present invention have been described, those skilled in the art can make additional changes and modifications to these embodiments once they learn the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications falling within the scope of the present invention.

Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. In this way, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalent technologies, the present invention is also intended to include these modifications and variations.

Claims

A method for identifying brain atrophy, which is characterized in that it comprises:

Use the key frame detection module to determine the key frames in the brain image sequence;

Use a key point detection module to detect key points in the key frame, and determine the first type of grading index according to the key points in the key frame;

Use an image segmentation module to segment the key frame to determine the second type of grading index;

Recognizing brain atrophy according to the first type grading index and the second type grading index.
The method according to claim 1, wherein the first type of grading index includes the largest diameter between the anterior horns, the smallest diameter between the anterior horns, the diameter between the choroid plexus of the lateral ventricle, and the outer diameter between the roof of the lateral ventricle;

The adopting the key point detection module to detect the key points in the key frame and determine the first type of classification index according to the key points in the key frame includes:

Use a key point detection module to detect the key points of the anterior angle and the key points of the lateral ventricle in the key frame;

Determine the maximum diameter between the rake angles and the minimum diameter between the rake angles according to the key points of the rake angle;

The inter-choroid plexus diameter of the lateral ventricle and the outer diameter of the parietal ventricle are determined according to the key points of the lateral ventricle.
The method of claim 1, wherein the second type of grading index includes the widest diameter of the third ventricle;

The segmentation of the key frame by the image segmentation module to determine the second type of classification index includes:

Determining the first area according to the key points in the key frame;

Binarize the first area to determine the second area;

Segmenting the second region by using an image segmentation algorithm to determine the third ventricle region;

The widest diameter of the third ventricle is determined according to the area of the third ventricle.
The method of claim 1, wherein the second type of grading index includes the maximum outer diameter of the skull and the maximum inner diameter of the skull;

The segmentation of the key frame by the image segmentation module to determine the second type of grading index includes:

Segmenting the key frame according to the CT value corresponding to the skull to determine the first boundary;

Segmenting the first boundary by using an image segmentation algorithm to determine the skull boundary;

The maximum outer diameter of the skull and the maximum inner diameter of the skull are determined according to the boundary of the skull.
The method according to any one of claims 1 to 4, wherein the identifying brain atrophy according to the first type grading index and the second type grading index comprises:

Determining a brain atrophy assessment index according to the first type grading index and the second type grading index;

The brain atrophy evaluation index is input into a brain atrophy model to identify brain atrophy.
The method of claim 1, wherein the key point detection module and the key frame detection module are convolutional neural networks.
A device for identifying brain atrophy, characterized in that it comprises:

Key frame detection module, used to detect key frames in brain image sequences;

The key point detection module is used to detect the key points in the key frame, and determine the first type grading index according to the key points in the key frame;

The image segmentation module is used to segment the key frame and determine the second type of grading index;

The recognition module is used to recognize brain atrophy according to the first type grading index and the second type grading index.
8. The device of claim 7, wherein the first type of grading index comprises the largest diameter between the anterior horns, the smallest diameter between the anterior horns, the diameter between the choroid plexus of the lateral ventricle, and the outer diameter between the parietal ventricles;

The key point detection module includes:

The first detection module is used to detect the key points of the anterior angle and the key points of the lateral ventricle in the key frame;

A first determining module, configured to determine the maximum diameter between the rake angles and the minimum diameter between the rake angles according to the key points of the rake angle;

The second determining module is used to determine the inter-choroid plexus diameter of the lateral ventricle and the outer diameter of the parietal ventricle according to the key points of the lateral ventricle.
A computer device comprising a memory, a processor, and a computer program stored in the memory and capable of running on the processor, wherein the processor executes the program when the program is executed by any one of claims 1 to 6 The steps of the method.
A computer-readable storage medium, characterized in that it stores a computer program that can be executed by a computer device, and when the program runs on a computer device, the computer device executes the method described in any one of claims 1 to 6 A step of.