An Automatic Parkinson's Disease Recognition Method Based on Hybrid Handwriting Features
By collecting and processing handwritten image data from Parkinson's patients, and utilizing pixel fusion and feature-level fusion techniques, combined with convolutional neural networks and support vector machines, the efficiency and accuracy issues in Parkinson's disease diagnosis in existing technologies have been resolved, achieving more efficient early diagnosis.
Patent Information
- Application Number
- CN202410334228.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-03-22
- Publication Date
- 2026-11-17
- Estimated Expiration
- 2044-03-22
AI Technical Summary
Existing technologies are poor in terms of efficiency, accuracy, and adaptability in the diagnosis of Parkinson's disease, and diagnostic methods based on handwritten images fail to fully utilize the cross-information between handwritten trajectories and templates, resulting in diagnostic inconsistencies and high costs.
Spiral and meandering handwritten image data of Parkinson's patients were collected. After image enhancement and augmentation preprocessing, a fused image was generated by pixel fusion and input into a pre-trained convolutional neural network to extract feature vectors. The feature vectors were then combined with a support vector machine for classification.
It improves the accuracy, specificity, and sensitivity of early diagnosis of Parkinson's disease, assisting clinicians in diagnosing Parkinson's disease more accurately.
Smart Images

Figure CN118116083B_ABST
Abstract
Description
Technical Field
[0001] This disclosure relates to the field of image recognition technology, and more particularly to an automatic identification method for Parkinson's disease based on mixed handwriting features. Background Technology
[0002] Parkinson's disease is a chronic, progressive neurological disorder that typically affects middle-aged and elderly individuals. It is primarily caused by the gradual death or damage of neurons in the brain responsible for controlling movement. Typical symptoms include muscle rigidity, tremors, bradykinesia, and postural instability. Besides affecting motor function, Parkinson's disease can also lead to other symptoms such as cognitive decline, mood disorders, and sleep disturbances. Currently, there is no cure for Parkinson's disease, but symptoms can be alleviated and the patient's quality of life improved through medication, physical therapy, and surgery. In the clinical diagnosis of Parkinson's disease, the most common method is a physician's clinical assessment, including history taking and a neurological physical examination. By observing and assessing the patient's symptoms, combined with information such as medical history, the physician can make a preliminary diagnosis and assessment. However, clinical assessments are often influenced by the physician's subjective judgment; different physicians may have different observations and assessments of the patient's symptoms, which can lead to inconsistencies in diagnosis.
[0003] Most existing technologies diagnose Parkinson's disease by collecting body data or electroencephalograms (EEGs) of patients during specific movements. This method requires specialized equipment and procedures, is invasive, has limited application scenarios, and is costly, making it unsuitable for routine diagnosis of Parkinson's patients. Furthermore, most current methods for diagnosing Parkinson's disease based on handwritten images only extract and fuse features to compensate for information gaps between different handwritten images. They do not utilize the information overlapping between the handwriting trajectory and the handwriting template, and cannot fully extract the visual features of handwritten images to achieve a good effect in assisting clinicians in diagnosis.
[0004] It is evident that there is an urgent need for an automatic Parkinson's disease identification method based on mixed handwriting features that offers high efficiency, accuracy, and adaptability. Summary of the Invention
[0005] In view of this, the present disclosure provides an automatic identification method for Parkinson's disease based on mixed handwriting features, which at least partially solves the problems of poor evaluation efficiency, accuracy and adaptability in the prior art.
[0006] This disclosure provides an automatic Parkinson's disease identification method based on hybrid handwriting features, including:
[0007] Step 1: Collect handwritten image data from Parkinson's disease patients, including spiral handwritten images and meandering handwritten images;
[0008] Step 2: Perform data preprocessing on the spiral handwritten image and the meandering handwritten image, wherein the preprocessing includes image enhancement and image augmentation;
[0009] Step 3: The preprocessed spiral handwritten image and the meandering handwritten image are fused at the pixel level using a pixel fusion method to generate a fused image;
[0010] Step 4: Input the spiral handwritten image and the fused image into a pre-trained convolutional neural network, and extract the feature vectors of the flattening layer as the spiral handwritten image feature F1 and the fused image feature F2;
[0011] Step 5: Use the feature vector concatenation method to perform feature-level fusion of the spiral handwritten image feature F1 and the fused image feature F2 to generate fused features;
[0012] Step 6: Input the fused features into the support vector machine, classify the fused features, and obtain the diagnostic results.
[0013] According to a specific implementation of an embodiment of this disclosure, step 2 specifically includes:
[0014] Step 2.1: Perform a linear transformation on the value channels of the spiral handwritten image and the meandering handwritten image to increase the difference in pixel values and make the image clearer;
[0015] Step 2.2 involves increasing the brightness of the spiral and meandering handwritten images, flipping the images, adding edge filtering, and inverting the image colors. The formula for increasing the brightness is as follows:
[0016] I enhanced =α·I original +β
[0017] Among them, I enhanced I represents the enhanced image. original This represents the original image, where α controls the magnification factor and β controls the amount of motion.
[0018] The formula for calculating the flipped image is as follows:
[0019] R'(x,y)=R(width-x-1,y)
[0020] G'(x,y)=G(width-x-1,y)
[0021] B'(x,y)=B(width-x-1,y)
[0022] The original image is I. original = (R, G, B), the horizontally flipped image is I flipped= (R′,G′,B′), where (x,y) represents the pixel coordinates in the image, and width is the width of the image;
[0023] The formula for adding edge filtering to the image is as follows:
[0024]
[0025] Wherein, the horizontal edge intensity and the vertical edge intensity are the convolution results in the corresponding directions;
[0026] The technical formula for flipping colors is as follows:
[0027] I inverted (x,y)=255-I original (x,y)
[0028] Among them, I inverted This represents the flipped image, where (x,y) represents the pixel coordinates in the image.
[0029] According to a specific implementation of an embodiment of this disclosure, step 3 specifically includes:
[0030] Step 3.1: Construct the Gaussian pyramids corresponding to the preprocessed spiral handwritten image and the meandering handwritten image;
[0031] Step 3.2: Construct the Pyramid of Laplace based on the Pyramid of Gauss;
[0032] Step 3.3: Merge and reconstruct the Pyramid of Laplace to obtain a merged image.
[0033] According to a specific implementation of an embodiment of this disclosure, step 3.1 specifically includes:
[0034] A Gaussian pyramid is constructed by repeatedly downsampling the preprocessed spiral and meandering handwritten images. Each downsampling reduces the image size by half, using the following formula:
[0035] G i+1 =pyrDown(G i )
[0036] Among them, G i Let represent the image of the i-th Gaussian pyramid, and pyrDown(·) is the downsampling operation.
[0037] According to a specific implementation of an embodiment of this disclosure, step 3.2 specifically includes:
[0038] Starting from the top of the Gaussian pyramid, the Laplace pyramid is constructed through upsampling and differencing operations, with the following formula:
[0039] Li =G i -pyrUp(G i+1 )
[0040] Among them, L i Let represent the Laplacian pyramid image of the i-th layer, and pyrUp(·) is the upsampling operation.
[0041] According to a specific implementation of an embodiment of this disclosure, step 3.3 specifically includes:
[0042] Step 3.3.1: Perform a weighted fusion operation on each level of the Laplace pyramid, using the following formula:
[0043] B i =μ×L 1,i +(1-μ)×L 2,i
[0044] Among them, B i L represents the image after fusion at the i-th layer. 1,i and L 2,i Let represent the Laplacian pyramid images of the two input images at the i-th layer, and μ represent the image weights;
[0045] Step 3.3.2 involves upsampling and overlaying the fused Laplacian pyramid to reconstruct the fused image, using the following formula:
[0046] R i =pyrUp(B i )+B i-1
[0047] Among them, R i B represents the reconstructed image of layer i. i-1 It is the fused image from the previous layer, R0 = B0.
[0048] According to a specific implementation of this disclosure, the expression for the fusion feature is:
[0049]
[0050] Where F represents the fusion feature, This represents the concatenation operation of feature vectors.
[0051] According to a specific implementation of an embodiment of this disclosure, step 6 specifically includes:
[0052] The fused features are divided into training and test sets by leaving one-out-of-one-out. The feature vectors and labels of the training set are input into a support vector machine to train a classifier, resulting in a classifier model. The trained classifier model is used to classify new samples, and predictions are made based on the output of the decision function. The class with the highest probability value is taken as the final diagnosis result.
[0053] According to a specific implementation of this disclosure, the expression of the decision function is as follows:
[0054]
[0055] Where x is the input feature vector, y i For the corresponding category label, θ i For the Lagrange multipliers of the support vector machine, x i For support vectors, <·> represents the inner product of feature vectors, and z is the bias term.
[0056] According to a specific implementation of an embodiment of this disclosure, step 6 further includes:
[0057] An optimization problem is constructed based on the decision function to ensure that the classification boundary can separate samples of different classes and maximize the margin. The expression for the optimization problem is:
[0058]
[0059]
[0060] Where C is the regularization parameter, used to balance the size of the margin and the number of misclassified samples, j represents the index of the sample, and n represents the number of samples.
[0061] The automatic Parkinson's disease identification scheme based on hybrid handwriting features in this embodiment includes: Step 1, collecting handwritten image data of Parkinson's disease patients, wherein the handwritten image data includes spiral handwriting images and meandering handwriting images; Step 2, performing data preprocessing on the spiral handwriting images and meandering handwriting images, wherein the preprocessing includes image enhancement and image augmentation; Step 3, using a pixel fusion method to perform pixel-level fusion of the preprocessed spiral handwriting images and meandering handwriting images to generate a fused image; Step 4, inputting the spiral handwriting images and fused images into a pre-trained convolutional neural network, and extracting the feature vectors of the flattened layer as spiral handwriting image feature F1 and fused image feature F2; Step 5, using a feature vector concatenation method to perform feature-level fusion of spiral handwriting image feature F1 and fused image feature F2 to generate fused features; Step 6, inputting the fused features into a support vector machine, classifying the fused features, and obtaining a diagnostic result.
[0062] The beneficial effects of this disclosure are as follows: The scheme of this disclosure collects spiral and meandering handwriting images of Parkinson's patients. Image enhancement and image augmentation are applied to the spiral and meandering handwriting images for preprocessing. Pixel fusion is used to perform pixel-level fusion of the preprocessed spiral and meandering handwriting images to generate a fused image. The spiral and fused images are input into a pre-trained convolutional neural network, and the feature vectors of the flattened layers are extracted as spiral handwriting image feature F1 and fused image feature F2. Feature vector concatenation is used to perform feature-level fusion of spiral handwriting image feature F1 and fused image feature F2 to generate a fused feature. This hybrid fusion method yields more comprehensive and richer features, amplifying the differences between patients and healthy individuals. The fused features are input into a support vector machine for classification. Experimental results show that for the early diagnosis of Parkinson's disease, the accuracy, specificity, and sensitivity of this invention are improved compared to existing technologies. Clinicians can diagnose Parkinson's patients more accurately based on the classification results combined with subjective judgment. Attached Figure Description
[0063] To more clearly illustrate the technical solutions of the embodiments of this disclosure, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this disclosure. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0064] Figure 1 A flowchart illustrating an automatic Parkinson's disease identification method based on hybrid handwriting features provided in this disclosure embodiment;
[0065] Figure 2 This is a schematic diagram illustrating the specific implementation architecture of an automatic Parkinson's disease identification method based on hybrid handwriting features, provided in an embodiment of this disclosure. Detailed Implementation
[0066] The embodiments of this disclosure will now be described in detail with reference to the accompanying drawings.
[0067] The following specific examples illustrate the implementation of this disclosure. Those skilled in the art can easily understand other advantages and effects of this disclosure from the content disclosed in this specification. Obviously, the described embodiments are only a part of the embodiments of this disclosure, and not all of them. This disclosure can also be implemented or applied through other different specific embodiments, and the details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of this disclosure. It should be noted that, in the absence of conflict, the following embodiments and features in the embodiments can be combined with each other. Based on the embodiments in this disclosure, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this disclosure.
[0068] It should be noted that various aspects of embodiments within the scope of the appended claims are described below. It will be apparent that the aspects described herein can be embodied in a wide variety of forms, and any particular structure and / or function described herein is merely illustrative. Based on this disclosure, those skilled in the art will understand that one aspect described herein can be implemented independently of any other aspect, and two or more of these aspects can be combined in various ways. For example, any number of aspects set forth herein can be used to implement the device and / or practice the method. Additionally, this device and / or method can be implemented using structures and / or functionalities other than one or more of the aspects set forth herein.
[0069] It should also be noted that the illustrations provided in the following embodiments are only schematic representations of the basic concept of this disclosure. The drawings only show the components related to this disclosure and are not drawn according to the number, shape and size of the components in actual implementation. In actual implementation, the form, quantity and proportion of each component can be arbitrarily changed, and the layout of the components may also be more complex.
[0070] Furthermore, specific details are provided in the following description to facilitate a thorough understanding of the examples. However, those skilled in the art will understand that the described aspects can be practiced without these specific details.
[0071] This disclosure provides an automatic identification method for Parkinson's disease based on hybrid handwriting features, which can be applied to the auxiliary diagnosis process in medical scenarios.
[0072] See Figure 1 This is a flowchart illustrating an automatic Parkinson's disease identification method based on hybrid handwriting features provided in an embodiment of this disclosure. Figure 1 and Figure 2 As shown, the method mainly includes the following steps:
[0073] Step 1: Collect handwritten image data from Parkinson's disease patients, including spiral handwritten images and meandering handwritten images;
[0074] In practice, patients can draw on a digital tablet with a template paper on it according to the trajectory. The drawn image is displayed in real time on the host screen connected to the digital tablet, and the drawing is saved after completion.
[0075] Step 2: Perform data preprocessing on the spiral handwritten image and the meandering handwritten image, wherein the preprocessing includes image enhancement and image augmentation;
[0076] Based on the above embodiments, step 2 specifically includes:
[0077] Step 2.1: Perform a linear transformation on the value channels of the spiral handwritten image and the meandering handwritten image to increase the difference in pixel values and make the image clearer;
[0078] Step 2.2 involves increasing the brightness of the spiral and meandering handwritten images, flipping the images, adding edge filtering, and inverting the image colors. The formula for increasing the brightness is as follows:
[0079] I enhanced =α·I original +β
[0080] Among them, I enhanced I represents the enhanced image. original This represents the original image, where α controls the magnification factor and β controls the amount of motion.
[0081] The formula for calculating the flipped image is as follows:
[0082] R'(x,y)=R(width-x-1,y)
[0083] G'(x,y)=G(width-x-1,y)
[0084] B'(x,y)=B(width-x-1,y)
[0085] The original image is I. original = (R, G, B), the horizontally flipped image is I flipped = (R′,G′,B′), where (x,y) represents the pixel coordinates in the image, and width is the width of the image;
[0086] The formula for adding edge filtering to the image is as follows:
[0087]
[0088] Wherein, the horizontal edge intensity and the vertical edge intensity are the convolution results in the corresponding directions;
[0089] The technical formula for flipping colors is as follows:
[0090] I inverted (x,y)=255-I original (x,y)
[0091] Among them, I inverted This represents the flipped image, where (x,y) represents the pixel coordinates in the image.
[0092] In practice, linear transformations can be performed on the value channels of spiral and meandering handwritten images to increase pixel value differences and make the image clearer. For handwritten images, they are converted from RGB color space to HSV color space, and the value channels of the HSV image are extracted. The convertScaleAbs function is then used to perform a linear transformation on the value channels, with the following formula:
[0093] g(x)=convertScaleAbs(f(x),λ,γ)=clip(λ·f(x)+γ,0,255) (1)
[0094] Where f(x) is the input image, g(x) is the output image, λ is the magnification factor, and γ is the offset. In this description, the magnification factor λ is set to 1.5 and the offset γ is set to 0.
[0095] The increase in brightness is achieved by enhancing the brightness by 1.5 times, which is accomplished through a linear transformation, and the formula is as follows:
[0096] I enhanced =α·I original +β (2)
[0097] Wherein, α controls the amplification factor, and β controls the displacement. In the case of increasing the brightness by 1.5 times, α = 1.5 and no displacement is required, so β = 0.
[0098] The flipped image needs to be processed for each channel separately, and the formula is as follows:
[0099] R'(x,y)=R(width-x-1,y) (3)
[0100] G'(x,y)=G(width-x-1,y) (4)
[0101] B'(x,y)=B(width-x-1,y) (5)
[0102] The original image is I. original= (R, G, B), the horizontally flipped image is I flipped = (R′,G′,B′), where (x,y) represents the pixel coordinates in the image, and width is the width of the image;
[0103] Adding edge filtering to the image is achieved by applying convolution operations. The Sobel filter can highlight the edge information of the image, and horizontal edge detection is achieved through the following convolution kernels:
[0104]
[0105] Vertical edge detection is achieved using the following convolutional kernels:
[0106]
[0107] For each pixel, apply horizontal and vertical filters separately, and calculate the edge intensity using the following formula:
[0108]
[0109] Wherein, the horizontal edge intensity and the vertical edge intensity are the convolution results in the corresponding directions;
[0110] The inverted image color is achieved by taking the complement of each pixel value in the image (255 - pixel value), and the formula is as follows:
[0111] I inverted (x,y)=255-I original (x,y) (8)
[0112] Here, (x,y) represents the pixel coordinates in the image.
[0113] Step 3: The preprocessed spiral handwritten image and the meandering handwritten image are fused at the pixel level using a pixel fusion method to generate a fused image;
[0114] Based on the above embodiments, step 3 specifically includes:
[0115] Step 3.1: Construct the Gaussian pyramids corresponding to the preprocessed spiral handwritten image and the meandering handwritten image;
[0116] Step 3.2: Construct the Pyramid of Laplace based on the Pyramid of Gauss;
[0117] Step 3.3: Merge and reconstruct the Pyramid of Laplace to obtain a merged image.
[0118] Furthermore, step 3.1 specifically includes:
[0119] A Gaussian pyramid is constructed by repeatedly downsampling the preprocessed spiral and meandering handwritten images. Each downsampling reduces the image size by half, using the following formula:
[0120] G i+1 =pyrDown(G i )
[0121] Among them, G i Let represent the image of the i-th Gaussian pyramid, and pyrDown(·) is the downsampling operation.
[0122] Furthermore, step 3.2 specifically includes:
[0123] Starting from the top of the Gaussian pyramid, the Laplace pyramid is constructed through upsampling and differencing operations, with the following formula:
[0124] L i =G i -pyrUp(G i+1 )
[0125] Among them, L i Let represent the Laplacian pyramid image of the i-th layer, and pyrUp(·) is the upsampling operation.
[0126] Furthermore, step 3.3 specifically includes:
[0127] Step 3.3.1: Perform a weighted fusion operation on each level of the Laplace pyramid, using the following formula:
[0128] B i =μ×L 1,i +(1-μ)×L 2,i
[0129] Among them, B i L represents the image after fusion at the i-th layer. 1,i and L 2,i Let represent the Laplacian pyramid images of the two input images at the i-th layer, and μ represent the image weights;
[0130] Step 3.3.2 involves upsampling and overlaying the fused Laplacian pyramid to reconstruct the fused image, using the following formula:
[0131] R i =pyrUp(B i )+B i-1
[0132] Among them, R i B represents the reconstructed image of layer i. i-1 It is the fused image from the previous layer, R0 = B0.
[0133] In practice, pixel fusion can be used to fuse the preprocessed spiral handwritten image and the meandering handwritten image at the pixel level to generate a fused image. The methods include:
[0134] The preprocessed spiral handwritten image and the meandering handwritten image are input into the pixel-level fusion module, and the Laplacian pyramid fusion is performed. The steps are as follows: constructing Gaussian pyramids of the two images, further constructing Laplacian pyramids from Gaussian pyramids, fusing the Laplacian pyramids, and reconstructing the images.
[0135] The Gaussian pyramids constructed from the two images are built by repeatedly downsampling the images, each downsampling reducing the image size by half, as shown in the formula:
[0136] G i+1 =pyrDown(G i (9)
[0137] Among them, G i Let represent the image of the i-th Gaussian pyramid, and pyrDown(·) is the downsampling operation;
[0138] The construction of the Laplace pyramid from the Gaussian pyramid begins at the top of the Gaussian pyramid and proceeds through upsampling and differencing operations to build the Laplace pyramid. The formula is as follows:
[0139] L i =G i -pyrUp(G i+1 (10)
[0140] Among them, L i This represents the Laplacian pyramid image of the i-th layer, where pyrUp(·) is the upsampling operation;
[0141] The fusion of the Laplacian pyramid involves performing a weighted fusion operation on the Laplacian image of each layer, using the following formula:
[0142] B i =μ×L 1,i +(1-μ)×L 2,i (11)
[0143] Among them, B i L represents the image after fusion at the i-th layer. 1,i and L 2,i Let represent the Laplacian pyramid images of the two input images at the i-th layer, and μ represent the image weight, where μ can be taken as 0.5;
[0144] The reconstructed image is obtained by upsampling and overlaying the fused Laplacian pyramid, and the formula is as follows:
[0145] R i =pyrUp(B i )+B i-1 (12)
[0146] Among them, R i B represents the reconstructed image of layer i. i-1 It is the fused image from the previous layer, R0 = B0.
[0147] In the Laplace pyramid fusion described above, five pyramid layers can be set, which preserves sufficient detail without causing excessive smoothing or high computational costs.
[0148] Step 4: Input the spiral handwritten image and the fused image into a pre-trained convolutional neural network, and extract the feature vectors of the flattening layer as the spiral handwritten image feature F1 and the fused image feature F2;
[0149] In practice, the spiral handwritten image and the fused image can be cropped to conform to VGG. 16 The input size is 224×224, and it is fed into a VGG library pre-trained on the large public dataset ImageNet. 16 In this process, the one-dimensional vector output by the Flatten() layer is extracted as the spiral handwritten image feature F1 and the fused image feature F2.
[0150] The pre-trained VGG 16 Setting `include_top` to `False` will exclude fully connected layers from the output. The final output, when expanded in one dimension, will have the following formula:
[0151] F1 = Flatten(VGG) 16 (I1)) (13)
[0152] F2 = Flatten(VGG) 16 (I2)) (14)
[0153] Among them, VGG 16 (I) indicates that image I is input into the pre-trained VGG. 16 In the model, Flatten() flattens the feature map output by the convolutional neural network into a one-dimensional vector. F1 and F2 represent the feature vectors of the spiral handwritten image and the fused image, respectively.
[0154] Step 5: Use the feature vector concatenation method to perform feature-level fusion of the spiral handwritten image feature F1 and the fused image feature F2 to generate fused features;
[0155] Furthermore, the expression for the fusion feature is as follows:
[0156]
[0157] Where F represents the fusion feature, This represents the concatenation operation of feature vectors.
[0158] In practice, the spiral handwritten image features F1 and the fused image features F2 are input into the feature-level fusion module for feature vector concatenation, and the formula is as follows:
[0159]
[0160] Where F represents the fusion feature, This represents the concatenation operation of feature vectors. The features obtained through the feature-level fusion module are more comprehensive and contain richer information, which can amplify the differences between patients and healthy individuals, making them easier to distinguish.
[0161] Step 6: Input the fused features into the support vector machine, classify the fused features, and obtain the diagnostic results.
[0162] Based on the above embodiments, step 6 specifically includes:
[0163] The fused features are divided into training and test sets by leaving one-out-of-one-out. The feature vectors and labels of the training set are input into a support vector machine to train a classifier, resulting in a classifier model. The trained classifier model is used to classify new samples, and predictions are made based on the output of the decision function. The class with the highest probability value is taken as the final diagnosis result.
[0164] Furthermore, the expression for the decision function is as follows:
[0165]
[0166] Where x is the input feature vector, yi is the corresponding class label, and θ i For the Lagrange multipliers of the support vector machine, x i For support vectors, <·> represents the inner product of feature vectors, and z is the bias term.
[0167] Furthermore, step 6 also includes:
[0168] An optimization problem is constructed based on the decision function to ensure that the classification boundary can separate samples of different classes and maximize the margin. The expression for the optimization problem is:
[0169]
[0170]
[0171] Where C is the regularization parameter, used to balance the size of the margin and the number of misclassified samples, j represents the index of the sample, and n represents the number of samples.
[0172] In practice, the fused features of the dataset are divided into training and test sets by leaving one out. The feature vectors and labels of the training set are input into the support vector machine to train the classifier, and the classifier model is obtained. The trained classifier model is used to classify new samples, and prediction is made based on the output of the decision function. The class with the highest probability value is taken as the final diagnosis result.
[0173] Accuracy, sensitivity, specificity, and F1 score are selected as the metrics for evaluating the method, and their formulas are as follows:
[0174]
[0175]
[0176]
[0177]
[0178] Where TP represents true positives, TN represents true negatives, FP represents false positives, FN represents false negatives, Precision represents precision, and Recall represents recall.
[0179] The Support Vector Machine (SVM) is used as a classifier, employing a linear kernel with parameter C set to 1. In the SVM's decision function, a kernel function is used to calculate the inner product between feature vectors. For a linear kernel, the SVM's decision function can be expressed as:
[0180]
[0181] Where x is the input feature vector, yi is the corresponding class label, θi is the Lagrange multiplier of the support vector machine, and x i For support vectors, <·> denotes the inner product of feature vectors, and z is the bias term;
[0182] The goal of the support vector machine is to find the optimal decision function, ensuring that the classification boundary can separate samples of different classes and maximize the margin. This can be achieved by solving the following optimization problem:
[0183]
[0184]
[0185] Where C is the regularization parameter, used to balance the size of the margin and the number of misclassified samples.
[0186] As can be seen from the above embodiments, the present invention has good effects on some evaluation indicators and plays an effective role in the early auxiliary diagnosis of Parkinson's disease patients. Clinicians can diagnose Parkinson's disease patients more accurately by combining the classification results with subjective judgment.
[0187] This embodiment provides an automatic Parkinson's disease identification method based on hybrid handwriting features. The method collects spiral and serpentine handwriting images from Parkinson's patients, and preprocesses these images using image enhancement and image augmentation. A pixel-level fusion method is then used to generate a fused image from the preprocessed spiral and serpentine handwriting images. The spiral and fused images are then input into a pre-trained convolutional neural network, and feature vectors from the flattened layers are extracted as spiral handwriting image feature F1 and fused image feature F2. Feature vector concatenation is then used to perform feature-level fusion of spiral handwriting image feature F1 and fused image feature F2 to generate a fused feature. This hybrid fusion method yields more comprehensive and richer features, amplifying the differences between patients and healthy individuals. The fused features are then input into a support vector machine for classification. Experimental results show that this invention improves the accuracy, specificity, and sensitivity compared to existing technologies for the early diagnosis of Parkinson's disease. Clinicians diagnose Parkinson's disease patients more accurately by combining classification results with subjective judgment.
[0188] It should be understood that the various parts of this disclosure can be implemented in hardware, software, firmware, or a combination thereof.
[0189] The above description is merely a specific embodiment of this disclosure, but the scope of protection of this disclosure is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this disclosure should be included within the scope of protection of this disclosure. Therefore, the scope of protection of this disclosure should be determined by the scope of the claims.
Claims
1. An automatic identification method for Parkinson's disease based on hybrid handwriting features, characterized in that, include: Step 1: Collect handwritten image data from Parkinson's disease patients, including spiral handwritten images and meandering handwritten images; Step 2: Perform data preprocessing on the spiral handwritten image and the meandering handwritten image, wherein the preprocessing includes image enhancement and image augmentation; Step 3: The preprocessed spiral handwritten image and the meandering handwritten image are fused at the pixel level using a pixel fusion method to generate a fused image; Step 3 specifically includes: Step 3.1: Construct the Gaussian pyramids corresponding to the preprocessed spiral handwritten image and the meandering handwritten image; Step 3.2: Construct the Pyramid of Laplace based on the Pyramid of Gauss; Step 3.3: Merge and reconstruct the Pyramid of Laplace to obtain a merged image; Step 4: Input the spiral handwritten image and the fused image into a pre-trained convolutional neural network, and extract the feature vectors of the flattened layer as the features of the spiral handwritten image. and fusion of image features ; Step 5: Use the feature vector concatenation method to process the features of the spiral handwritten image. and fusion of image features Perform feature-level fusion to generate fused features; Step 6: Input the fused features into the support vector machine, classify the fused features, and obtain the diagnostic results.
2. The method according to claim 1, characterized in that... Step 2 specifically includes: Step 2.1: Perform a linear transformation on the value channels of the spiral handwritten image and the meandering handwritten image to increase the difference in pixel values and make the image clearer; Step 2.2 involves increasing the brightness of the spiral and meandering handwritten images, flipping the images, adding edge filtering, and inverting the image colors. The formula for increasing the brightness is as follows: in, This represents the enhanced image. This represents the original image, where α controls the magnification factor and β controls the amount of motion. The formula for calculating the flipped image is as follows: The original image is =(R,G,B), the horizontally flipped image is =(R′,G′,B′), where (x,y) represents the pixel coordinates in the image, and width is the width of the image; The formula for adding edge filtering to the image is as follows: Wherein, the horizontal edge intensity and the vertical edge intensity are the convolution results in the corresponding directions; The formula for calculating the color of the inverted image is as follows: in, This represents the flipped image, where (x,y) represents the pixel coordinates in the image.
3. The method according to claim 2, characterized in that... Step 3.1 specifically includes: A Gaussian pyramid is constructed by repeatedly downsampling the preprocessed spiral and meandering handwritten images. Each downsampling reduces the image size by half, using the following formula: in, This represents the image of the i-th layer of the Gaussian pyramid. It is a downsampling operation.
4. The method according to claim 3, characterized in that... Step 3.2 specifically includes: Starting from the top of the Gaussian pyramid, the Laplace pyramid is constructed through upsampling and differencing operations, with the following formula: in, This represents the image of the i-th layer of the Laplacian pyramid. It is an upsampling operation.
5. The method according to claim 4, characterized in that... Step 3.3 specifically includes: Step 3.3.1: Perform a weighted fusion operation on each level of the Laplace pyramid, using the following formula: in, This represents the image after fusion at layer i. and These represent the Laplacian pyramid images of the two input images at the i-th layer. Represents image weights; Step 3.3.2 involves upsampling and overlaying the fused Laplacian pyramid to reconstruct the fused image, using the following formula: in, This represents the reconstructed image of layer i. It is the fused image from the previous layer. .
6. The method according to claim 5, characterized in that... The expression for the fusion feature is: Where F represents the fusion feature, and ⊕ represents the concatenation operation of feature vectors.
7. The method according to claim 6, characterized in that... Step 6 specifically includes: The fused features are divided into training and test sets by leaving one-out-of-one-out. The feature vectors and labels of the training set are input into a support vector machine to train a classifier, resulting in a classifier model. The trained classifier model is used to classify new samples, and predictions are made based on the output of the decision function. The class with the highest probability value is taken as the final diagnosis result.
8. The method according to claim 7, characterized in that The expression for the decision function is: in, For the input feature vector, For the corresponding category label, For the Lagrange multipliers of support vector machines, For support vectors, Represents the inner product of eigenvectors. This is a bias term.
9. The method according to claim 8, characterized in that... Step 6 further includes: An optimization problem is constructed based on the decision function to ensure that the classification boundary can separate samples of different classes and maximize the margin. The expression for the optimization problem is: Where C is the regularization parameter, used to balance the size of the margin and the number of misclassified samples. Indicates the index of the sample. Indicates the number of samples.
