A cognitive ability intelligent classification method and system based on YOLO algorithm
Through the intelligent classification method of cognitive ability based on YOLO algorithm, the image of the subjects copied is analyzed using feature extraction and machine learning models, which solves the problems of long time and great influence of subjective factors in the existing screening method, and achieves rapid and accurate Alzheimer's disease screening.
Patent Information
- Application Number
- CN202111535447.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2021-12-15
- Publication Date
- 2024-11-26
- Estimated Expiration
- 2041-12-15
AI Technical Summary
Existing Alzheimer's screening methods such as the MMSE scale require a long time to complete, rely on doctor's experience, and the results are greatly affected by subjective factors, and are not suitable for large-scale screening.
Using the cognitive ability intelligent classification method based on YOLO algorithm, the subjects copied images of standard graphics are obtained, and the feature extraction model and machine learning model are used to label and classify feature regions and feature points to construct initial distance sequences and feature sequences. , perform Fourier transform and finally determine the type of cognitive ability of the subject.
It realizes rapid and accurate cognitive classification of the subjects, and is suitable for large-scale preliminary screening of Alzheimer's disease, reducing dependence on doctor's experience and improving the objectivity of the results.
Smart Images

Figure CN114187283B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of computer medical image analysis, and in particular to a cognitive ability intelligent classification method and system based on the YOLO algorithm. Background Art
[0002] Alzheimer's disease (AD) is a degenerative disease of the central nervous system with an insidious onset and a slow course, which is more common in people over 65 years old. Since the early symptoms of AD are relatively hidden and difficult to detect directly, they are usually diagnosed in the middle and late stages of the best intervention period, so early detection and early prevention are necessary. At present, the clinical screening of AD mainly relies on the Minimum Cognitive Evaluation Scale (MMSE) compiled by Folstein in the United States in 1975. The scale includes 30 items such as orientation, attention, language, memory, and visual-spatial skills, each with 1 point, for a total of 30 points. The dementia cutoff value varies according to the level of education. Due to its advantages such as simple operation, ease of operation, and ideal validity, it is suitable for the examination of clinical cognitive impairment and the screening of dementia in the community population. However, the MMSE scale still faces the following problems in clinical use: (1) The evaluation process requires a one-on-one doctor-patient relationship, which takes a long time (usually more than 30 minutes) and is not suitable for large-scale population screening; (2) The evaluation doctor must have rich experience, especially the pentagon replication test sub-item in the scale. The evaluation results are greatly affected by subjective factors, and different evaluation doctors often give different evaluation results. Therefore, there is an urgent need for a method and system that can accurately, intelligently and quickly classify the cognitive abilities of the subjects. Summary of the invention
[0003] The purpose of the present invention is to provide a method and system for intelligent classification of cognitive abilities based on the YOLO algorithm, which can achieve large-scale preliminary screening of Alzheimer's disease by classifying cognitive abilities.
[0004] To achieve the above object, the present invention provides the following solutions:
[0005] A cognitive ability intelligent classification method based on the YOLO algorithm, comprising:
[0006] The image of the test pattern is obtained as the test image; the test pattern is obtained after the testee copies the standard pattern; the standard pattern includes a plurality of overlapping polygons;
[0007] The test image is input into a feature extraction model, and the feature area and all feature points in the test image are annotated to obtain an annotated test image; the feature extraction model is obtained by training the YOLO neural network based on the historical test image;
[0008] The cognitive ability type of the subject is determined based on the labeled test image.
[0009] Optionally, the standard shape includes two overlapping pentagons; and the overlapping portion of the two pentagons is a rhombus.
[0010] Optionally, before acquiring the image of the test pattern as the test image, the method further includes:
[0011] Obtain multiple historical test images;
[0012] Using a rectangular frame to mark the characteristic area and all characteristic points in each of the historical test images, to obtain a marked historical test image; the characteristic area is the overlapping part; the characteristic points are the vertices of the polygon;
[0013] The annotated historical test image is used as input and the rectangular frame is used as output, and the YOLO neural network is trained to obtain the feature extraction model.
[0014] Optionally, determining the cognitive ability type of the subject according to the labeled test image specifically includes:
[0015] Determine whether the overlapping portion exists in the plurality of portions annotated in the annotated test image, and obtain a first determination result;
[0016] If the first judgment result is no, determining that the cognitive ability type of the subject is unqualified;
[0017] If the first judgment result is yes, constructing an initial distance sequence according to the feature area and the feature point;
[0018] The cognitive ability type of the subject is determined according to the initial distance sequence.
[0019] Optionally, constructing an initial distance sequence according to the feature area and the feature point specifically includes:
[0020] Determine the coordinates of the midpoint of the rectangular box marking the feature point as the coordinates of the corresponding feature point;
[0021] Determine the midpoint coordinates of the rectangular frame marking the feature area as the midpoint coordinates of the corresponding feature area;
[0022] According to the coordinates of the feature points and the coordinates of the midpoint of the feature area, the distance between each feature point and the midpoint of the feature area is calculated to obtain the initial distance sequence.
[0023] Optionally, determining the cognitive ability type of the subject according to the initial distance sequence specifically includes:
[0024] Eliminate outliers in the initial distance sequence to obtain a standard distance sequence;
[0025] Determine whether the number of elements in the standard distance sequence is within a preset number range, and obtain a second determination result;
[0026] If the second judgment result is no, determining that the cognitive ability type of the subject is unqualified;
[0027] If the second judgment result is yes, constructing a feature sequence of the test pattern;
[0028] The feature sequence is input into a cognitive ability classification model to determine the cognitive ability type of the subject; the cognitive ability classification model is obtained by training an initial classification model using a feature sequence set.
[0029] Optionally, the initial classification model is a decision tree, a random forest, a support vector machine or a K nearest neighbor model.
[0030] Optionally, the step of constructing the feature sequence of the test pattern specifically includes:
[0031] Taking the midpoint of the feature area as the center of the circle and taking the maximum value in the standard distance sequence and half of the maximum value in the standard distance sequence as the radius, respectively, to obtain concentric circles;
[0032] Dividing the concentric circles by a straight line group to obtain a plurality of local circle regions;
[0033] Construct an initial feature sequence according to the number of test pattern corners contained in each local circular area;
[0034] Performing Fourier transformation on the initial feature sequence to obtain an initial feature sequence after Fourier transformation;
[0035] The modulus value of the initial feature sequence after the Fourier transformation is determined as the feature sequence of the test pattern.
[0036] Optionally, before acquiring the image of the test pattern as the test image, the method further includes:
[0037] The feature sequence set is used as input, and the cognitive ability types of the subjects corresponding to the feature sequences in the feature sequence set are used as output, and the initial classification model is trained to obtain the cognitive ability classification model.
[0038] A cognitive ability intelligent classification system based on the YOLO algorithm, including:
[0039] A test image acquisition module is used to acquire an image of a test pattern as a test image; the test pattern is obtained by the testee copying a standard pattern; the standard pattern includes a plurality of overlapping polygons;
[0040] A labeling module is used to input the test image into a feature extraction model, label the feature area and all feature points in the test pattern, and obtain a labeled test image; the feature extraction model is obtained by training the YOLO neural network based on historical test images;
[0041] The cognitive ability type determination module is used to determine the cognitive ability type of the subject based on the labeled test image.
[0042] According to the specific embodiments provided by the present invention, the present invention discloses the following technical effects:
[0043] The present invention provides a method and system for intelligent classification of cognitive ability based on the YOLO algorithm, the method comprising: obtaining an image of a test pattern as a test image; the test pattern is obtained after the testee copies a standard pattern; the standard pattern includes a plurality of overlapping polygons; the test image is input into a feature extraction model, and the feature area and all feature points in the test pattern are annotated to obtain an annotated test image; the feature extraction model is obtained after training a YOLO neural network based on historical test images; and the type of cognitive ability of the testee is determined based on the annotated test image. The present invention annotates the feature areas and feature points of the test image through the trained YOLO neural network, and then classifies the cognitive ability, and the classification results are used for large-scale preliminary screening of Alzheimer's disease. BRIEF DESCRIPTION OF THE DRAWINGS
[0044] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.
[0045] Figure 1 Flow chart of the cognitive ability intelligent classification method based on the YOLO algorithm in an embodiment of the present invention;
[0046] Figure 2 It is a schematic diagram of a standard graph in an embodiment of the present invention;
[0047] Figure 3 A schematic diagram of a test image in an embodiment of the present invention;
[0048] Figure 4 A schematic diagram of a labeled test image in an embodiment of the present invention;
[0049] Figure 5 Schematic diagram of local features of a test image in an embodiment of the present invention;
[0050] Figure 6 This is a flow chart of a community rapid screening method for Alzheimer's disease for target detection in an embodiment of the present invention;
[0051] Figure 7 This is a flow chart of the cognitive ability classification model training method in an embodiment of the present invention. DETAILED DESCRIPTION
[0052] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0053] The purpose of the present invention is to provide a method and system for intelligent classification of cognitive abilities based on the YOLO algorithm, which can achieve large-scale preliminary screening of Alzheimer's disease by classifying cognitive abilities.
[0054] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.
[0055] like Figure 1 and 6 The present invention provides a cognitive ability intelligent classification method based on the YOLO algorithm, comprising:
[0056] Step 101: Acquire an image of a test pattern as a test image; the test pattern is obtained by the testee copying a standard pattern; the standard pattern includes a plurality of overlapping polygons; the schematic diagram of the standard pattern is as follows Figure 2 ; Test image schematic diagram as shown Figure 3 .
[0057] Step 102: Input the test image into the feature extraction model, annotate the feature area and all feature points in the test image, and obtain an annotated test image; the feature extraction model is obtained by training the YOLO neural network based on the historical test image; the schematic diagram of the annotated test image is as follows Figure 4 .
[0058] Step 103: Determine the cognitive ability type of the subject based on the labeled test image.
[0059] Specifically, the standard shape includes two overlapping pentagons; and the overlapping portion of the two pentagons is a rhombus.
[0060] Before step 101, the method further includes:
[0061] Obtain multiple historical test images;
[0062] The characteristic regions and all characteristic points in each historical test image are annotated using a rectangular frame to obtain an annotated historical test image; the characteristic regions are the overlapping parts; the characteristic points are the vertices of the polygons;
[0063] With the annotated historical test image as input and the rectangular box as output, the YOLO neural network is trained to obtain the feature extraction model.
[0064] Step 103 specifically includes:
[0065] Determine whether there is an overlapping portion among the multiple portions annotated in the annotated test image, and obtain a first determination result;
[0066] If the first judgment result is negative, the cognitive ability type of the subject is determined to be unqualified;
[0067] If the first judgment result is yes, construct an initial distance sequence according to the feature area and the feature point;
[0068] According to the initial distance sequence, the cognitive ability type of the subject is determined.
[0069] Among them, the initial distance sequence is constructed according to the feature area and feature points, specifically including:
[0070] Determine the coordinates of the midpoint of the rectangular box marking the feature point as the coordinates of the corresponding feature point;
[0071] Determine the midpoint coordinates of the rectangular frame marking the feature area as the midpoint coordinates of the corresponding feature area;
[0072] According to the coordinates of the feature points and the coordinates of the midpoint of the feature area, the distance between each feature point and the midpoint of the feature area is calculated to obtain an initial distance sequence.
[0073] Specifically, according to the initial distance sequence, the cognitive ability type of the subject is determined, including:
[0074] Eliminate outliers in the initial distance sequence to obtain a standard distance sequence;
[0075] Determine whether the number of elements in the standard distance sequence is within a preset number range, and obtain a second determination result;
[0076] If the second judgment result is negative, the cognitive ability type of the subject is determined to be unqualified;
[0077] If the second judgment result is yes, constructing a feature sequence of the test pattern;
[0078] The feature sequence is input into the cognitive ability classification model to determine the cognitive ability type of the subject; the cognitive ability classification model is obtained by training the initial classification model using the feature sequence set.
[0079] Among them, the initial classification model is a decision tree, random forest, support vector machine or K nearest neighbor model.
[0080] like Figure 5 , construct the feature sequence of the test pattern, specifically including;
[0081] Take the midpoint of the feature area as the center, and draw circles with the maximum value in the standard distance sequence and half of the maximum value in the standard distance sequence as the radius to obtain concentric circles;
[0082] The concentric circles are divided by the straight line group to obtain multiple local circle areas;
[0083] Construct an initial feature sequence according to the number of test pattern corners contained in each local circular area;
[0084] Performing Fourier transform on the initial feature sequence to obtain the initial feature sequence after Fourier transform;
[0085] The modulus value of the initial feature sequence after Fourier transformation is determined as the feature sequence of the test pattern.
[0086] like Figure 7 , before acquiring the image of the test pattern as the test image, it also includes:
[0087] With the feature sequence set as input and the cognitive ability types of the subjects corresponding to the feature sequences in the feature sequence set as output, the initial classification model is trained to obtain a cognitive ability classification model.
[0088] like Figure 2 As shown in the figure, two intersecting pentagons are drawn on a piece of white paper, and the test subject is required to draw them accurately. Scoring criteria: The pentagon must have 5 clear angles and 5 sides. At the same time, the intersection of the two pentagons forms a diamond shape, and the jitter of the lines and the shape can be ignored.
[0089] In order to achieve end-to-end rapid detection, the steps of the present invention are as follows:
[0090] 1. Use the YOLO neural network to detect the corners of the pentagon replication result. It should be noted that ideally there are ten corners, but in some cases, due to some reasons such as the corners are not closed, they cannot be detected or the number of detected corners is not equal to 10 due to interference from the connection line. All the coordinates of the target detection belonging to the corner category are saved and a sequence M = [(X1, Y1), (X2, Y2).... (X n ,Y n ))]Where n is the number of target detection angles.
[0091] 2. It is necessary to detect the position of the rhombus obtained by the intersection of two pentagons, and take the center of the rhombus as the center of the entire image. Its coordinates are marked as (X0, Y0). If the rhombus cannot be detected, that is, the intersection of the two pentagons cannot be found, it is considered that the person's MMSE structural ability is unqualified.
[0092] 3. In order to solve the interference of abnormal corners detected, the statistical Grubbs's law is used. Specifically, for the coordinates of each corner (X i ,Y i ), calculate (X i ,Y i ) and (X0, Y0), construct a distance sequence D = [d1, d2…d n ], where d i Indicates (X i ,Y i ) and (X0, Y0), where Euclidean distance is used for calculation. Finally, based on Grubbs' rule, the null hypothesis and alternative hypothesis are established:
[0093] Hypothesis 1 (H0) H0: There are no outliers in D.
[0094] Hypothesis 2 (H1): There is a new outlier in D.
[0095] For each D, if there is an outlier, delete it and get a new D, until there is no outlier in D, and the cycle ends. Let the number of elements in D be c, if c>=12 or c<=8, it is considered that the person's MMSE structural ability is unqualified.
[0096] 4. With (X0, Y0) as the center, Max{D i} as the radius, make a circle; at the same time In addition, two 45° diagonal lines, horizontal lines and vertical lines are used to divide the circle into 16 areas according to the distance from the center of the circle from far to near and the angle with the center of the circle from small to large, namely Z1, Z2.....Z 16 Then, calculate the number of corners in each region and save the number into a sequence, recorded as
[0097] To make N rotationally invariant, N is subjected to discrete Fourier transform, and then the complex value is modulo to obtain N'.
[0098] 5. Select s images from the positive samples and construct a positive sample set, denoted as It should be noted that the s pictures here need to be slightly different.
[0099] Here we select the positive and negative sample sets and calculate the distance between a certain sample k and T, denoted as D k ={d1,d2…d s}, calculate the sequence D k The minimum value, 25% quantile, 50% quantile, 75% quantile, and maximum value of the corresponding sample k feature T k ={t k1 , t k2 , t k3 , t k4 , t k5}. Then build the classification model Model. Note that the model can be built using models such as decision tree, random forest, SVM, and K nearest neighbor.
[0100] 6. When it is necessary to determine the type of cognitive ability, the image j made by the test subject is calculated according to the above steps to obtain N, and T is calculated j , the final prediction result is obtained based on the classification model Model trained in the previous step.
[0101] In summary, the process of the community rapid screening method for Alzheimer's disease with target detection is as follows: Figure 6 shown.
[0102] Figure 2 This is the pentagon copying result of a certain test subject. The original size of the picture is 1024*640. It should be noted that the test subject should be required to not draw other graphics in the drawing area except copying the pentagon to avoid interference.
[0103] First, the YOLO neural network is used to detect all corners of the pentagon replication result and the rhombus formed by the pentagons, and the center of the box where all the corners are located and the center of the rhombus are extracted and the coordinates are saved. The result of target detection is as follows Figure 3 shown.
[0104] Next, according to Grubbs' rule, the corners that appeared in abnormal positions in the previous step were removed, and a preliminary screening was performed based on the number of corners and whether there were rhombuses.
[0105] Secondly, according to Figure 5The steps in are used to extract local features of the image and classify the samples using a supervised machine learning model. The classification model used is the random forest model, and the parameters involved are set as follows: the number of trees in the random forest is 500, and the standard for measuring splitting is MSE.
[0106] Finally, the final judgment result is obtained and suggestions for the next step are given based on the result.
[0107] Note: The present invention has been tested, and the detection results are as follows in 200 positive samples and 200 negative samples. Figure 7 shown.
[0108] The cognitive ability intelligent classification method based on the YOLO algorithm provided by the present invention transforms the cognitive ability judgment problem into an image binary classification problem, uses target detection related algorithms to extract features, and introduces a machine learning algorithm to achieve intelligent classification and recognition, thereby obtaining the final evaluation result.
[0109] In addition, the present invention also provides a cognitive ability intelligent classification system based on the YOLO algorithm, including:
[0110] The test image acquisition module is used to acquire an image of a test pattern as a test image; the test pattern is obtained by the testee copying a standard pattern; the standard pattern includes a plurality of overlapping polygons;
[0111] The annotation module is used to input the test image into the feature extraction model, annotate the feature area and all feature points in the test image, and obtain the annotated test image; the feature extraction model is obtained by training the YOLO neural network based on the historical test image;
[0112] The cognitive ability type determination module is used to determine the cognitive ability type of the subject based on the labeled test image.
[0113] In this specification, each embodiment is described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the embodiments can be referred to each other. For the system disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part.
[0114] The principles and implementation methods of the present invention are described in this article using specific examples. The description of the above embodiments is only used to help understand the method and core idea of the present invention. At the same time, for those skilled in the art, according to the idea of the present invention, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting the present invention.
Claims
1. A cognitive ability intelligent classification method based on the YOLO algorithm, characterized in that: The method comprises: The image of the test pattern is obtained as the test image; the test pattern is obtained after the testee copies the standard pattern; the standard pattern includes a plurality of overlapping polygons; The test image is input into a feature extraction model, and the feature area and all feature points in the test image are annotated to obtain an annotated test image; the feature extraction model is obtained by training the YOLO neural network based on the historical test image; Determining the cognitive ability type of the subject according to the labeled test image; Determining the cognitive ability type of the subject according to the labeled test image specifically includes: Determine whether the overlapping portion exists in the plurality of portions annotated in the annotated test image, and obtain a first determination result; If the first judgment result is no, determining that the cognitive ability type of the subject is unqualified; If the first judgment result is yes, constructing an initial distance sequence according to the feature area and the feature point; Determining the cognitive ability type of the subject according to the initial distance sequence; The constructing of the initial distance sequence according to the feature area and the feature point specifically includes: Determine the coordinates of the midpoint of the rectangular box marking the feature point as the coordinates of the corresponding feature point; Determine the midpoint coordinates of the rectangular frame marking the feature area as the midpoint coordinates of the corresponding feature area; According to the coordinates of the feature points and the coordinates of the midpoint of the feature area, the distance between each feature point and the midpoint of the feature area is calculated to obtain the initial distance sequence.
2. The cognitive ability intelligent classification method based on the YOLO algorithm according to claim 1 is characterized in that: The standard figure includes two overlapping pentagons; and the overlapping portion of the two pentagons is a rhombus.
3. The cognitive ability intelligent classification method based on the YOLO algorithm according to claim 2 is characterized in that: Before acquiring the image of the test pattern as the test image, the method further includes: Obtain multiple historical test images; Using a rectangular frame to mark the characteristic area and all characteristic points in each of the historical test images, to obtain a marked historical test image; the characteristic area is the overlapping part; the characteristic points are the vertices of the polygon; The annotated historical test image is used as input and the rectangular frame is used as output, and the YOLO neural network is trained to obtain the feature extraction model.
4. The cognitive ability intelligent classification method based on the YOLO algorithm according to claim 1 is characterized in that: Determining the cognitive ability type of the subject according to the initial distance sequence specifically includes: Eliminate outliers in the initial distance sequence to obtain a standard distance sequence; Determine whether the number of elements in the standard distance sequence is within a preset number range, and obtain a second determination result; If the second judgment result is no, determining that the cognitive ability type of the subject is unqualified; If the second judgment result is yes, constructing a feature sequence of the test pattern; The feature sequence is input into a cognitive ability classification model to determine the cognitive ability type of the subject; the cognitive ability classification model is obtained by training an initial classification model using a feature sequence set.
5. The cognitive ability intelligent classification method based on the YOLO algorithm according to claim 4 is characterized in that: The initial classification model is a decision tree, random forest, support vector machine or K nearest neighbor model.
6. The cognitive ability intelligent classification method based on the YOLO algorithm according to claim 4 is characterized in that: The constructing of the characteristic sequence of the test pattern specifically includes: Taking the midpoint of the feature area as the center of the circle and taking the maximum value in the standard distance sequence and half of the maximum value in the standard distance sequence as the radius, respectively, to obtain concentric circles; Dividing the concentric circles by a straight line group to obtain a plurality of local circle regions; Construct an initial feature sequence according to the number of test pattern corners contained in each local circular area; Performing Fourier transformation on the initial feature sequence to obtain an initial feature sequence after Fourier transformation; The modulus value of the initial feature sequence after the Fourier transformation is determined as the feature sequence of the test pattern.
7. The cognitive ability intelligent classification method based on the YOLO algorithm according to claim 4 is characterized in that: Before acquiring the image of the test pattern as the test image, the method further includes: The feature sequence set is used as input, and the cognitive ability types of the subjects corresponding to the feature sequences in the feature sequence set are used as output, and the initial classification model is trained to obtain the cognitive ability classification model.
8. A cognitive ability intelligent classification system based on the YOLO algorithm, characterized in that: The system applies the cognitive ability intelligent classification method based on the YOLO algorithm as described in any one of claims 1 to 7, and the system includes: A test image acquisition module is used to acquire an image of a test pattern as a test image; the test pattern is obtained by the testee copying a standard pattern; the standard pattern includes a plurality of overlapping polygons; A labeling module is used to input the test image into a feature extraction model, label the feature area and all feature points in the test pattern, and obtain a labeled test image; the feature extraction model is obtained by training the YOLO neural network based on historical test images; The cognitive ability type determination module is used to determine the cognitive ability type of the subject based on the labeled test image.
