Alzheimer's disease clock drawing experiment automatic evaluation method and system based on non-rigid registration

Abstract

The invention provides a senile dementia clock drawing experiment automatic evaluation method and system based on non-rigid registration, and belongs to the technical field of artificial intelligence, and the method comprises the steps: S1, obtaining clock drawing experiment data of a testee, and carrying out the preprocessing of the obtained clock drawing experiment data, and obtaining a first point cloud image; s2, performing non-rigid registration on the first point cloud image and a preset standard template to obtain a second point cloud image after non-rigid registration; s3, performing clock drawing experiment analysis on the second point cloud image to obtain an analysis result; and S4, scoring the analysis result according to a preset scoring standard to obtain a total score of the senile dementia scale of the testee. The non-rigid registration method based on machine vision has the advantages that automatic and intelligent evaluation of the senile dementia scale clock drawing experiment is achieved, evaluation efficiency and accuracy are improved, a more effective auxiliary means is provided for diagnosis and treatment of senile dementia patients, and the non-rigid registration method based on machine vision has wide application prospects.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to an automatic evaluation method and system for an Alzheimer's disease clock drawing experiment based on non-rigid registration. Background Art

[0002] The Clock Drawing Test (CDT) is a widely used neuropsychological assessment tool, mainly used to assess cognitive function, especially in the diagnosis of early dementia (such as Alzheimer's disease). The test method is to ask the subject to draw a clock on paper, fill in the time numbers on the clock, and then point the hour hand and minute hand to the specified time respectively. It is used to check the subject's comprehension, action execution ability, digital knowledge, planning, abstract thinking ability, graphic reconstruction ability, attention, visual memory, visual-spatial ability, etc.

[0003] Currently, CDT assessment is mainly done manually by clinicians, who score based on the accuracy of the clock, the arrangement of the numbers, the position of the hands, and the overall neatness and coordination. However, this manual assessment method is not only time-consuming and laborious, but also easily affected by the subjective judgment of the assessor, resulting in possible differences in the consistency and accuracy of the assessment results. Summary of the invention

[0004] In order to solve the above technical problems, the present invention provides an automatic evaluation method and system for the Alzheimer's disease clock drawing experiment based on non-rigid registration, so that patients can be tested at any time more conveniently, and further optimizes the automatic evaluation system's understanding and analysis of the clock deformation characteristics, making the evaluation results more quantitative and objective.

[0005] The technical problem solved by the present invention can be achieved by adopting the following technical solutions:

[0006] An automatic evaluation method for the Alzheimer's clock drawing experiment based on non-rigid registration, comprising:

[0007] Step S1, obtaining clock drawing experimental data of a subject, and preprocessing the obtained clock drawing experimental data to obtain a first point cloud image;

[0008] Step S2, performing non-rigid registration on the first point cloud image and a preset standard template to obtain a second point cloud image after non-rigid registration;

[0009] Step S3, performing a clock drawing experiment analysis on the second point cloud image to obtain an analysis result;

[0010] Step S4, scoring the analysis results according to a preset scoring standard to obtain the total score of the Alzheimer's disease rating scale of the subject.

[0011] Preferably, in step S1, after obtaining the clock drawing experiment data of the subject and before preprocessing the obtained clock drawing experiment data, the method further includes:

[0012] The obtained clock drawing experimental data is preliminarily preprocessed; the preliminarily preprocessing includes:

[0013] Performing size standardization on the clock drawing experiment data; and / or

[0014] grayscale the clock drawing experiment data; and / or

[0015] Performing filtering processing on the clock drawing experiment data; and / or

[0016] The clock drawing experimental data is subjected to edge enhancement processing.

[0017] Preferably, the preprocessing of the acquired clock drawing experiment data comprises:

[0018] Using the target detection and segmentation model to identify the clock drawing experimental data and extract feature data, the feature data including label information and position information of the dial, numbers, hour hand and minute hand;

[0019] According to the position information in the feature data, the dial, numbers, hour hand and minute hand are respectively mapped into three-dimensional space to obtain the first point cloud image.

[0020] Preferably, the step S2 comprises:

[0021] Using the first point cloud image as the source point cloud and the standard template as the target point cloud;

[0022] Based on a preset point cloud registration algorithm, a Gaussian mixture model is used to fit the source point cloud, the target point cloud is used as the data points generated by downsampling the Gaussian mixture model, and the Gaussian mixture model parameters are iteratively adjusted to obtain a non-rigid registration result between the source point cloud and the target point cloud, wherein the non-rigid registration result includes the second point cloud image.

[0023] Preferably, the preset point cloud registration algorithm includes a consistent point drift algorithm.

[0024] Preferably, the iterative adjustment of Gaussian mixture model parameters further comprises:

[0025] The Gaussian mixture model parameters are regularized based on the regenerated Hilbert space.

[0026] Preferably, the step S3 comprises:

[0027] Calculating the Euclidean distance between two adjacent points in the dial contour point cloud in the second point cloud image, and comparing the maximum value of the Euclidean distance with a first preset threshold to obtain a first comparison result, and determining the contour completeness detection result based on the first comparison result; and / or

[0028] Calculating the Hausdorff distance between the digital point cloud in the second point cloud image and the corresponding number in the preset standard template, and comparing the Hausdorff distance with a second preset threshold to obtain a second comparison result, and determining the digital correctness detection result based on the second comparison result; and / or

[0029] Calculating the Euclidean distance between the centroid of the digital point cloud in the second point cloud image and the corresponding digital centroid in the preset standard template, and comparing the Euclidean distance with a third preset threshold to obtain a third comparison result, and determining the digital sorting detection result based on the third comparison result; and / or

[0030] Check whether the pointer is complete and whether the pointer is correctly placed to obtain the pointer correctness test result;

[0031] The analysis result includes the contour integrity detection result, and / or the digital correctness detection result, and / or the digital sorting detection result, and / or the pointer correctness detection result.

[0032] Preferably, the detection of whether the pointer is correctly placed is divided into the detection of the correctness of the hour hand and the correctness of the minute hand; wherein,

[0033] The steps of detecting the correctness of the hour hand include:

[0034] To set the hour hand, you need to point to number N, where N = 1, 2, 3, ..., 12;

[0035] Establish three three-dimensional coordinate systems. The first coordinate system uses the straight line connecting the number N and the center of the dial as the X-axis. The second coordinate system uses the straight line connecting the number N+1 and the center of the dial as the X-axis. The third coordinate system uses the straight line connecting the centroid of the hour hand point cloud and the center of the dial as the X-axis. The three three-dimensional coordinate systems all use the vertical direction of the center of the dial as the Z-axis. The Y-axes of the three three-dimensional coordinate systems are generated based on the right-hand rule.

[0036] Perform matrix transformation on the first coordinate system and the second coordinate system respectively toward the third coordinate system. If the rotation angles are one positive and one negative, the position of the hour hand is correct; otherwise, the position of the hour hand is wrong.

[0037] The step of detecting the correctness of the minute hand comprises:

[0038] To set the minute hand, point to the number M, where M = 0, 1, 2, 3, ..., 59;

[0039] Determine a first vector formed by the center of the dial and the mass point of the number M, and a second vector formed by the center of the dial and the mass center of the minute hand point cloud, and calculate the angle between the first vector and the second vector;

[0040] The included angle is compared with a preset angle threshold. If the included angle exceeds the preset angle threshold, the minute hand position is wrong; otherwise, the minute hand position is correct.

[0041] Preferably, the step S4 further includes:

[0042] Step S5, displaying the total score of the Alzheimer's disease rating scale of the subject and the cognitive level corresponding to the total score.

[0043] The present invention also provides an automatic evaluation system for the Alzheimer's disease clock drawing experiment based on non-rigid registration, which is used to implement the automatic evaluation method for the Alzheimer's disease clock drawing experiment based on non-rigid registration, comprising:

[0044] The acquisition module is used to obtain the clock drawing experiment data of the subjects;

[0045] A preprocessing module, connected to the acquisition module, for preprocessing the acquired clock drawing experiment data to obtain a first point cloud image;

[0046] A registration module, connected to the preprocessing module, configured to perform non-rigid registration on the first point cloud image and a preset standard template to obtain a second point cloud image after non-rigid registration;

[0047] An analysis module, connected to the registration module, configured to perform a clock drawing experiment analysis on the second point cloud image to obtain an analysis result;

[0048] The scoring module is connected to the analysis module and is used to score the analysis results according to a preset scoring standard to obtain the total score of the Alzheimer's disease scale of the subject.

[0049] The advantages or beneficial effects of the technical solution of the present invention are:

[0050] The present invention realizes the automated and intelligent evaluation of the clock drawing experiment of the Alzheimer's scale based on the non-rigid registration method of machine vision, improves the evaluation efficiency and accuracy, provides a more effective auxiliary means for the diagnosis and treatment of patients with Alzheimer's disease, and has broad application prospects. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1 It is a flow chart of an automatic evaluation method of the Alzheimer's disease clock drawing experiment based on non-rigid registration in a preferred embodiment of the present invention;

[0052] Figure 2It is a flow chart of an automatic evaluation method of the Alzheimer's disease clock drawing experiment based on non-rigid registration in a preferred embodiment of the present invention;

[0053] Figure 3 This is a structural block diagram of an automatic evaluation system for the Alzheimer's disease clock drawing experiment based on non-rigid registration in a preferred embodiment of the present invention. DETAILED DESCRIPTION

[0054] 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.

[0055] It should be noted that, in the absence of conflict, the embodiments of the present invention and the features in the embodiments may be combined with each other.

[0056] The present invention will be further described below in conjunction with the accompanying drawings and specific embodiments, but they are not intended to limit the present invention.

[0057] See also Figure 1 In a preferred embodiment of the present invention, based on the above problems existing in the prior art, a method for automatically evaluating a non-rigid registration method based on machine vision in an Alzheimer's disease clock drawing experiment is provided, comprising:

[0058] Step S1, obtaining clock drawing experimental data of a subject, and preprocessing the obtained clock drawing experimental data to obtain a first point cloud image;

[0059] The clock drawing test (CDT) is a test in which subjects draw a clock according to the drawing requirements to assess the Alzheimer's disease scale.

[0060] First, the drawing requirements of the clock drawing experiment are clarified. The drawing requirements can be given on-site by a professional doctor according to the purpose of the assessment and the subject's condition, or they can be randomly generated by the system. The drawing requirements at least include specifying the time that the subject needs to draw, that is, the specific time that the hour hand and minute hand need to point to. Exemplarily, it is expressed in the format of XX:XX. At the same time, the completion time of the subject can also be limited, for example, to 10 minutes.

[0061] Next, the subject starts drawing according to the given drawing requirements. The drawing process can be done on traditional paper, and the subject uses a pencil, pen or other writing tool to complete the drawing of the clock. After the drawing is completed, the drawn clock is photographed by a high-definition camera or scanner and the paper image is converted into a digital image format. Alternatively, specialized digital devices such as digital drawing tablets or touch screens can be used. These devices are usually equipped with pressure sensing and position tracking functions, which can record the subject's pen stroke trajectory, pressure changes and other information in real time. The subject only needs to draw the clock on the device according to the drawing requirements, and the system will automatically capture and save the entire drawing process, including the hand-drawn clock image, drawing time and any necessary interactive information.

[0062] Once the clock drawing experiment data of the subjects are obtained, subsequent preprocessing can be performed. The preprocessing includes but is not limited to grayscale conversion, size standardization, denoising, edge enhancement and other preliminary preprocessing operations, and then segmentation recognition and feature extraction based on machine learning technology (such as the Mark-RCNN model), including the labels and positions of the dial, numbers, hour hand and minute hand, and the extracted feature image is converted into a point cloud image.

[0063] The image in the clock drawing experiment data is a two-dimensional image, and the first point cloud image is a three-dimensional image. The Z-axis coordinate of the first point cloud image is 0.

[0064] Step S2, performing non-rigid registration on the first point cloud image and a preset standard template to obtain a second point cloud image after non-rigid registration;

[0065] Specifically, the first point cloud image (denoted as point set A) is non-rigidly registered with the preset standard template (denoted as point set B) to generate a transformed second point cloud image (namely the registered point set A') that is more aligned with the standard template. Non-rigid registration allows point set A to undergo local deformation to adapt to the shape and features of point set B while maintaining the overall structural similarity.

[0066] Step S3, performing a clock drawing experiment analysis on the second point cloud image to obtain an analysis result;

[0067] Specifically, a clock drawing experiment analysis is carried out on the point cloud image after non-rigid registration, including contour integrity detection, number correctness detection, number sorting detection, and pointer correctness detection; among them, the contour integrity detection verifies whether the dial is closed, the number correctness detection verifies whether the 12 numbers are arranged correctly in order, and the pointer correctness detection verifies whether the pointer is complete and placed in the correct position.

[0068] Step S4, scoring the analysis results according to a preset scoring standard to obtain the total score of the subject's Alzheimer's disease rating scale.

[0069] Specifically, according to the scoring criteria, the scale score is automatically calculated for the analysis results. The application prospects of the present invention are broad, and it can be applied to the automatic evaluation system of the clock drawing experiment of the Alzheimer's scale to improve the evaluation efficiency and accuracy. Through the implementation of this patent, automated and intelligent Alzheimer's evaluation can be realized, providing a more effective auxiliary means for the diagnosis and treatment of patients with Alzheimer's disease.

[0070] As a preferred embodiment, in step S1, after obtaining the clock drawing experiment data of the subject and before preprocessing the obtained clock drawing experiment data, the method further includes:

[0071] The obtained clock drawing experimental data is preprocessed; the preprocessing includes:

[0072] Standardize the clock drawing experiment data; and / or

[0073] Grayscale the clock drawing experiment data; and / or

[0074] Filtering the clock drawing experiment data; and / or

[0075] Perform edge enhancement on the clock drawing experiment data.

[0076] Specifically, in this embodiment, for the acquired clock drawing experiment data, a series of preliminary preprocessing steps are first performed to ensure the consistency of the data and the effectiveness of the analysis.

[0077] The preprocessing steps include but are not limited to: normalizing the image size, adjusting all images to a preset uniform size for subsequent uniform processing; then, graying the standardized images, converting color images into grayscale images, reducing the amount of data and simplifying the analysis process; then, filtering the grayscale images, using appropriate filtering algorithms (such as Gaussian filtering, median filtering, etc.) to remove noise in the image and improve image quality; finally, edge enhancement processing is performed on the filtered images, and edge information in the image is enhanced through edge detection algorithms (such as Canny edge detector) to more clearly identify and analyze key features in the clock drawing experiment.

[0078] As a preferred implementation, preprocessing the acquired clock drawing experiment data includes:

[0079] Use the target detection and segmentation model to identify the clock drawing experiment data and extract feature data, including the label information and position information of the dial, numbers, hour hand and minute hand;

[0080] According to the position information in the feature data, the dial, numbers, hour hand and minute hand are respectively mapped into the three-dimensional space to obtain the first point cloud image.

[0081] Furthermore, the target detection and segmentation model is a convolutional neural network model. The target detection and segmentation model may adopt a Mark-RCNN model.

[0082] Specifically, in this embodiment, after completing the preliminary preprocessing operation on the clock drawing experiment data, a target detection and segmentation model based on a convolutional neural network, such as a Mark-RCNN model, is used to perform an in-depth analysis on the preprocessed image.

[0083] The Mark-RCNN model can accurately identify and extract key feature data in the clock drawing experiment, including the label information of the dial, numbers, hour hand and minute hand, and their position information in the image. Subsequently, using this position information, the identified elements are mapped from the two-dimensional image space to the three-dimensional space. Specifically, each element (dial, numbers, hour hand, minute hand) is set to 0 (assuming that the image plane is the XY plane) according to the position of its pixel coordinates on the two-dimensional image, thereby converting them into corresponding three-dimensional point cloud coordinates to generate the first point cloud image.

[0084] As a preferred embodiment, step S2 includes:

[0085] The first point cloud image is used as the source point cloud and the standard template is used as the target point cloud;

[0086] Based on the preset point cloud registration algorithm, the source point cloud is fitted using a Gaussian mixture model, and the target point cloud is used as the data points generated by downsampling the Gaussian mixture model. The parameters of the Gaussian mixture model are iteratively adjusted to obtain a non-rigid registration result between the source point cloud and the target point cloud. The non-rigid registration result includes a second point cloud image.

[0087] As a preferred implementation, the preset point cloud registration algorithm includes a Coherent Point Drift (CPD) algorithm.

[0088] In the Open3D environment, the CPD algorithm is a classic point cloud registration method that is suitable for handling non-rigid (non-rigid) registration problems of unordered point clouds. The CPD algorithm achieves registration by building a probabilistic relationship model between point clouds and seeking to maximize the consistency between corresponding point pairs. Specifically, the algorithm will find a probabilistic transfer matrix and a transformation matrix to map the points of the source point cloud (source) to the target point cloud (target), while considering the consistency and probability distribution characteristics between the points, where the corresponding position of each source point on the target point cloud follows a Gaussian distribution.

[0089] Specifically, in this embodiment, the first point cloud image is set as the source point cloud, and the standard template is used as the target point cloud; a preset point cloud registration algorithm, such as the CPD point cloud registration algorithm, is used to perform the non-rigid registration process. Specifically, the source point cloud is fitted using a Gaussian mixture model (GMM), and the target point cloud is regarded as the data points generated by GMM downsampling; the parameters of the GMM are adjusted iteratively to maximize the consistency between the source point cloud and the target point cloud, and finally a point cloud image after non-rigid registration of the source point cloud and the target point cloud is obtained, that is, the second point cloud image.

[0090] GMM parameters include: λ and β. λ is used to balance the weight between the goodness of maximum likelihood fitting and regularization; β controls the strength of the smoothing effect. By continuously adjusting λ and β, it is possible to achieve accurate alignment between the source point cloud and the target point cloud while maintaining the overall structure of the point cloud.

[0091] The larger the β, the stronger the rigid change and the weaker the non-rigid change.

[0092] As a preferred implementation, the iterative adjustment of Gaussian mixture model parameters further includes:

[0093] The parameters of the Gaussian mixture model are regularized based on the Reproducing kernel Hilbert space (RKHS).

[0094] Specifically, in this embodiment, in the process of fitting the source point cloud using the Gaussian mixture model GMM, in order to enhance the smoothness of the transformation and prevent overfitting, the regenerated Hilbert space (RKHS) is introduced to regularize the parameters of the GMM. Specifically, by integrating the regularization mechanism of RKHS within the framework of GMM, we can effectively constrain the non-smooth phenomenon in the transformation process, ensure that while iteratively adjusting the GMM parameters to optimize the registration effect between the source point cloud and the target point cloud, the continuity and stability of the transformation are maintained, thereby obtaining an accurate and smooth non-rigid registration result.

[0095] The clock drawing experiment analysis includes contour integrity detection, number correctness detection, number sorting detection, and pointer correctness detection.

[0096] As a preferred embodiment, the contour integrity test verifies whether the dial is closed, including:

[0097] The Euclidean distance between two adjacent points in the dial contour point cloud in the second point cloud image is calculated, and the maximum value of the Euclidean distance is compared with a first preset threshold to obtain a first comparison result, and the contour completeness detection result is determined based on the first comparison result.

[0098] Specifically, in this embodiment, the KD tree (K-dimensional tree) principle is used to traverse each point in the dial outline point cloud in the second point cloud image. For each point, the KD tree is used to search for its nearest K neighbors (usually K is 2, i.e., two directly adjacent points), and the Euclidean distances between these adjacent point pairs are calculated; all calculated Euclidean distances are sorted, and the maximum value is extracted; and the maximum value is compared with a preset first preset threshold:

[0099] If the maximum value is greater than the first preset threshold, it indicates that there is a large gap in the dial contour, that is, the dial is not closed, and the contour integrity test result is unqualified;

[0100] On the contrary, if the maximum value is less than or equal to the first preset threshold, it is judged that the dial contour is well closed and the contour integrity detection result is qualified.

[0101] As a preferred embodiment, the number correctness detection verifies that the 12 numbers are correct, including:

[0102] Calculate the Hausdorff distance between the digital point cloud in the second point cloud image and the corresponding number in the preset standard template, compare the Hausdorff distance with a second preset threshold, obtain a second comparison result, and determine the digital correctness detection result based on the second comparison result.

[0103] Specifically, in this embodiment, the digital point cloud in the registered second point cloud image is non-rigidly measured with the corresponding numbers in the standard template according to the classification labels of Mask-RCNN. Specifically:

[0104] For each digital point cloud in the second point cloud image, find the corresponding standard digital point cloud in the preset standard template according to its classification label; calculate the Hausdorff distance between the two point clouds as a similarity measurement indicator, and compare the calculated Hausdorff distance value with the set second preset threshold:

[0105] When the Hausdorff distance value is greater than the second preset threshold, it indicates that there is a significant difference between the digital point cloud and the corresponding number in the standard template, and it is determined that there is an error in the number, and the digital correctness test result is unqualified;

[0106] When the Hausdorff distance value is less than or equal to the second preset threshold, the number is determined to be correct and the number correctness detection result is qualified.

[0107] As a preferred embodiment, the number order detection verifies whether the 12 numbers are correctly arranged in order, including:

[0108] Calculate the Euclidean distance between the centroid of the digital point cloud in the second point cloud image and the corresponding digital centroid in the preset standard template, and compare the Euclidean distance with a third preset threshold to obtain a third comparison result, and determine the digital sorting detection result based on the third comparison result.

[0109] Specifically, in this embodiment, for each number, the centroid of the digital point cloud in the second point cloud image is determined, recorded as the first centroid; the centroid of the corresponding number in the standard template is determined, recorded as the second centroid; the Euclidean distance between the first centroid and the second centroid is calculated, and the calculated Euclidean distance of the centroid is compared with the corresponding preset third preset threshold:

[0110] If the Euclidean distance between the centroid of any digital point cloud in the second point cloud image and the number in its corresponding standard template is greater than a third preset threshold, it is determined that there is a deviation in the digital ordering at this position, and the overall digital ordering detection result is incorrect;

[0111] Only when the Euclidean distances of the centroids of all corresponding digital point clouds are less than or equal to the third preset threshold, it is determined that the 12 numbers are correctly arranged in order and the digital sorting detection result is correct.

[0112] As a preferred embodiment, the pointer correctness detection verifies whether the pointer is complete and placed in the correct position, including:

[0113] Check whether the pointer is complete and whether the pointer is placed correctly to obtain the pointer correctness test result.

[0114] Specifically, in this embodiment, the pointer correctness detection is divided into two steps: detecting whether the pointer is complete and detecting whether the pointer is correctly placed.

[0115] Among them, the segmentation results of Mask-RCNN are used to check whether the pointer is complete by checking the returned label information. Specifically, the integrity of the pointer is determined by checking whether the label information is complete and meets expectations (such as whether it contains all necessary labels indicating the start, end, and middle parts of the pointer). If all relevant labels exist and are correct, the pointer integrity test is completed.

[0116] As a preferred embodiment, the detection of whether the pointer is correctly placed is divided into hour hand correctness detection and minute hand correctness detection.

[0117] The standard dial has 12 numbers, from 1 to 12, evenly distributed on the dial. The dial is divided into 60 equal parts, each representing one minute.

[0118] As a preferred embodiment, the step of detecting the correctness of the hour hand includes:

[0119] To set the hour hand, you need to point to number N, where N = 1, 2, 3, ..., 12;

[0120] Establish three three-dimensional coordinate systems. The first coordinate system uses the straight line connecting the number N and the center of the dial as the X-axis. The second coordinate system uses the straight line connecting the number N+1 and the center of the dial as the X-axis. The third coordinate system uses the straight line connecting the centroid of the hour hand point cloud and the center of the dial as the X-axis. The three three-dimensional coordinate systems all use the vertical direction of the center of the dial as the Z-axis. The Y-axes of the three three-dimensional coordinate systems are generated based on the right-hand rule.

[0121] Perform matrix transformation on the first coordinate system and the second coordinate system to the third coordinate system respectively. If the rotation angles are one positive and one negative, the hour hand position is correct; if the rotation angles are all positive or all negative, the hour hand position is wrong.

[0122] As a preferred embodiment, the step of detecting the correctness of the minute hand includes:

[0123] To set the minute hand, point to the number M, where M = 0, 1, 2, 3, ..., 59;

[0124] Determine a first vector formed by the center of the dial and the mass point of the number M, and a second vector formed by the center of the dial and the mass center of the minute hand point cloud, and calculate the angle between the first vector and the second vector;

[0125] The included angle is compared with a preset angle threshold. If the included angle exceeds the preset angle threshold, the minute hand position is wrong; otherwise, the minute hand position is correct.

[0126] The above analysis results include contour integrity detection results, digital correctness detection results, digital sorting detection results, and pointer correctness detection results.

[0127] As a preferred embodiment, the method further includes after step S4:

[0128] Step S5, displaying the total score of the Alzheimer's disease scale of the subject and the cognitive level corresponding to the total score.

[0129] Furthermore, after completing the outline completeness test, the digital correctness test, the digital order correctness test, and the pointer integrity and correctness test, each test result is scored one by one according to the preset scoring criteria.

[0130] The scoring criteria are:

[0131] If the subject can draw a closed dial, one point is awarded;

[0132] If all 12 numbers are correct and there are no missing numbers, you will get another point;

[0133] If all the numbers are placed in the correct position, one extra point will be awarded;

[0134] If the pointer is also placed in the correct position, add another point.

[0135] Furthermore, the total score of the Alzheimer's scale of the subjects was obtained according to these scoring criteria. The cognitive level of the subjects was inferred from the scoring results:

[0136] A total score of 4 indicates normal cognition;

[0137] A total score of 3 indicates mild dementia;

[0138] A total score of 2 indicates moderate dementia;

[0139] A total score of 1 indicates severe dementia.

[0140] The score and the corresponding cognitive level assessment results are then displayed on the interface for further health assessment and intervention.

[0141] The present invention also provides an automatic evaluation system for the Alzheimer's disease clock drawing experiment based on non-rigid registration, which is used to implement the automatic evaluation method for the Alzheimer's disease clock drawing experiment based on non-rigid registration, comprising:

[0142] Collection module 1, used to obtain the clock drawing experiment data of the subjects;

[0143] A preprocessing module 2, connected to the acquisition module 1, is used to preprocess the acquired clock drawing experiment data to obtain a first point cloud image;

[0144] The registration module 3 is connected to the preprocessing module 2 and is used to perform non-rigid registration on the first point cloud image and a preset standard template to obtain a second point cloud image after non-rigid registration;

[0145] The analysis module 4 is connected to the registration module 3 and is used to perform a clock drawing experiment analysis on the second point cloud image to obtain an analysis result;

[0146] The scoring module 5 is connected to the analysis module 4 and is used to score the analysis results according to a preset scoring standard to obtain the total score of the Alzheimer's disease scale of the subject.

[0147] Furthermore, it also includes a display module 6, connected to the scoring module 5, for displaying the total score of the subject's Alzheimer's scale and the cognitive level corresponding to the total score.

[0148] Specifically, the system gives specific clock drawing requirements. After the subject completes the questions, the acquisition module 1 collects the subject's images to be evaluated. The acquisition module 1 includes an acquisition device 11 and a storage device 12, which are responsible for collecting and collecting the subject's clock drawing experimental data and uploading it to the preprocessing module 2.

[0149] Preprocessing module 2 specifically includes size standardization, grayscale, filtering and edge enhancement; size standardization is used to process the uploaded image and adjust the image to a uniform size; grayscale is used to convert the resized image into a grayscale image; the obtained grayscale image is filtered and denoised, and finally edge enhancement is performed to facilitate subsequent model recognition and segmentation. Mark-RCNN is then used to identify and extract the labels and positions of the dial, numbers, hour hand and minute hand; according to the pixel coordinates of the target image, the Z axis is set to 0, and the dial, numbers, hour hand and minute hand are converted into point cloud images respectively, and then uploaded to the registration module 3.

[0150] Registration module 3 uses the dial, numbers, hour hand and minute hand point clouds input by the subject as the source point cloud, and the dial, numbers, hour hand and minute hand standard templates preset in the system as the target point cloud. The CPD point cloud registration method is adopted, and the GMM model is used to fit the source point cloud. The target point cloud is regarded as the data points generated by GMM downsampling. The regenerated Hilbert space (RKHS) is introduced in the GMM to constrain the smoothness in the transformation process, by continuously adjusting λ (the trade-off between the goodness of maximum likelihood fitting and regularization) and β (the strength of the smoothing effect). The larger β is, the stronger the rigid change is and the weaker the non-rigid change is. Finally, a point cloud image after non-rigid registration of the source point cloud and the target point cloud is obtained.

[0151] The registered point cloud image is uploaded to the analysis module 4, and the analysis process includes contour completeness detection, digital correctness detection, digital sorting detection and pointer correctness detection.

[0152] The results of the analysis module 4 are summarized and transmitted to the scoring module 5. The subject's image to be evaluated is scored according to the preset standards and the results of the contour completeness test, digital correctness test, digital order correctness test and pointer integrity and correctness test in the analysis module 4. After the scoring results are obtained, the obtained scores of the image to be evaluated and the corresponding cognitive status of the subject are displayed on the electronic screen.

[0153] The division of the automatic assessment system into modules, units and sub-units in the present application is only for a better understanding of the functions involved in the technical solution of the present invention. In practice, the functions corresponding to these modules can be loaded and executed by a single or multiple hardware programs.

[0154] Those skilled in the art should be able to appreciate that the method steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of electronic hardware and software, the composition and steps of each example have been generally described in the above description according to function. Whether these functions are performed in electronic hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art may use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention.

[0155] The above are only preferred embodiments of the present invention, and are not intended to limit the implementation methods and protection scope of the present invention. Those skilled in the art should be aware that all solutions obtained by equivalent substitutions and obvious changes made using the contents of this specification and illustrations should be included in the protection scope of the present invention.

Claims

1. An automatic evaluation method for the Alzheimer's disease clock drawing experiment based on non-rigid registration, characterized in that: include: Step S1, obtaining clock drawing experimental data of a subject, and preprocessing the obtained clock drawing experimental data to obtain a first point cloud image; Step S2, performing non-rigid registration on the first point cloud image and a preset standard template to obtain a second point cloud image after non-rigid registration; Step S3, performing a clock drawing experiment analysis on the second point cloud image to obtain an analysis result; Step S4, scoring the analysis results according to a preset scoring standard to obtain the total score of the Alzheimer's disease rating scale of the subject.

2. The automatic evaluation method for the Alzheimer's disease clock drawing experiment based on non-rigid registration according to claim 1 is characterized in that: In the step S1, after obtaining the clock drawing experiment data of the subject and before preprocessing the obtained clock drawing experiment data, the step further includes: The obtained clock drawing experimental data is preliminarily preprocessed; the preliminarily preprocessing includes: Performing size standardization on the clock drawing experiment data; and / or grayscale the clock drawing experiment data; and / or Performing filtering processing on the clock drawing experiment data; and / or The clock drawing experimental data is subjected to edge enhancement processing.

3. The automatic evaluation method for the Alzheimer's disease clock drawing experiment based on non-rigid registration according to claim 1 is characterized in that: The preprocessing of the acquired clock drawing experimental data comprises: Using the target detection and segmentation model to identify the clock drawing experimental data and extract feature data, the feature data including label information and position information of the dial, numbers, hour hand and minute hand; According to the position information in the feature data, the dial, numbers, hour hand and minute hand are respectively mapped into three-dimensional space to obtain the first point cloud image.

4. The automatic evaluation method for the Alzheimer's disease clock drawing experiment based on non-rigid registration according to claim 1 is characterized in that: The step S2 comprises: Using the first point cloud image as the source point cloud and the standard template as the target point cloud; Based on a preset point cloud registration algorithm, a Gaussian mixture model is used to fit the source point cloud, the target point cloud is used as the data points generated by downsampling the Gaussian mixture model, and the Gaussian mixture model parameters are iteratively adjusted to obtain a non-rigid registration result between the source point cloud and the target point cloud, wherein the non-rigid registration result includes the second point cloud image.

5. The automatic evaluation method for the Alzheimer's disease clock drawing experiment based on non-rigid registration according to claim 4 is characterized in that: The preset point cloud registration algorithm includes a consistent point drift algorithm.

6. The automatic evaluation method for the Alzheimer's disease clock drawing experiment based on non-rigid registration according to claim 4 is characterized in that: The iterative adjustment of Gaussian mixture model parameters also includes: The Gaussian mixture model parameters are regularized based on the regenerated Hilbert space.

7. The automatic evaluation method for the Alzheimer's disease clock drawing experiment based on non-rigid registration according to claim 1 is characterized in that: The step S3 comprises: Calculating the Euclidean distance between two adjacent points in the dial contour point cloud in the second point cloud image, and comparing the maximum value of the Euclidean distance with a first preset threshold to obtain a first comparison result, and determining the contour completeness detection result based on the first comparison result; and / or Calculating the Hausdorff distance between the digital point cloud in the second point cloud image and the corresponding number in the preset standard template, and comparing the Hausdorff distance with a second preset threshold to obtain a second comparison result, and determining the digital correctness detection result based on the second comparison result; and / or Calculating the Euclidean distance between the centroid of the digital point cloud in the second point cloud image and the corresponding digital centroid in the preset standard template, and comparing the Euclidean distance with a third preset threshold to obtain a third comparison result, and determining the digital sorting detection result based on the third comparison result; and / or Check whether the pointer is complete and whether the pointer is correctly placed to obtain the pointer correctness test result; The analysis result includes the contour integrity detection result, and / or the digital correctness detection result, and / or the digital sorting detection result, and / or the pointer correctness detection result.

8. The automatic evaluation method for the Alzheimer's disease clock drawing experiment based on non-rigid registration according to claim 7 is characterized in that: The detection of whether the pointer is correctly placed is divided into the detection of the correctness of the hour hand and the correctness of the minute hand; wherein, The steps of detecting the correctness of the hour hand include: To set the hour hand, you need to point to number N, where N = 1, 2, 3, ..., 12; Establish three three-dimensional coordinate systems. The first coordinate system uses the straight line connecting the number N and the center of the dial as the X-axis. The second coordinate system uses the straight line connecting the number N+1 and the center of the dial as the X-axis. The third coordinate system uses the straight line connecting the centroid of the hour hand point cloud and the center of the dial as the X-axis. The three three-dimensional coordinate systems all use the vertical direction of the center of the dial as the Z-axis. The Y-axes of the three three-dimensional coordinate systems are generated based on the right-hand rule. Perform matrix transformation on the first coordinate system and the second coordinate system respectively toward the third coordinate system. If the rotation angles are one positive and one negative, the position of the hour hand is correct; otherwise, the position of the hour hand is wrong. The step of detecting the correctness of the minute hand comprises: To set the minute hand, point to the number M, where M = 0, 1, 2, 3, ..., 59; Determine a first vector formed by the center of the dial and the mass point of the number M, and a second vector formed by the center of the dial and the mass center of the minute hand point cloud, and calculate the angle between the first vector and the second vector; The included angle is compared with a preset angle threshold. If the included angle exceeds the preset angle threshold, the minute hand position is wrong; otherwise, the minute hand position is correct.

9. The automatic evaluation method for the Alzheimer's disease clock drawing experiment based on non-rigid registration according to claim 1, characterized in that: After step S4, the following steps are also included: Step S5, displaying the total score of the Alzheimer's disease rating scale of the subject and the cognitive level corresponding to the total score.

10. An automatic evaluation system for Alzheimer's disease clock drawing experiment based on non-rigid registration, characterized in that: The method for automatically evaluating the Alzheimer's disease clock drawing experiment based on non-rigid registration according to any one of claims 1 to 9 comprises: The acquisition module is used to obtain the clock drawing experiment data of the subjects; A preprocessing module, connected to the acquisition module, for preprocessing the acquired clock drawing experiment data to obtain a first point cloud image; A registration module, connected to the preprocessing module, configured to perform non-rigid registration on the first point cloud image and a preset standard template to obtain a second point cloud image after non-rigid registration; An analysis module, connected to the registration module, configured to perform a clock drawing experiment analysis on the second point cloud image to obtain an analysis result; The scoring module is connected to the analysis module and is used to score the analysis results according to a preset scoring standard to obtain the total score of the Alzheimer's disease scale of the subject.

Images