Alzheimer's disease data processing method and system

Abstract

The present invention relates to the field of data processing technology, and in particular to a method and system for processing Alzheimer's disease data. The method comprises the following steps: obtaining indicator information of an object to be detected; encoding the indicator information, assigning a unique code bit to each bit of the indicator information to form a first coding sequence combination; encoding the first coding sequence combination, amplifying the number of valid characters of the first coding sequence combination by m times to form a second coding sequence combination; deconvolving the second coding sequence combination, amplifying the second coding sequence combination to form a one-dimensional convolution array; performing vertical staggered stacking on all arrays in the one-dimensional convolution array to form a two-dimensional array; inputting the two-dimensional array into a first neural network to obtain a first prediction standard threshold; inputting the two-dimensional array into a second neural network to obtain a second prediction standard threshold; and performing weighted sum calculation on the first prediction standard threshold and the second prediction standard threshold to obtain a third prediction standard threshold.

Description

Technical Field

[0001] The present invention relates to the field of data processing technology, and in particular to a method and system for processing Alzheimer's disease data. Background Art

[0002] Alzheimer's disease is a widespread neurodegenerative disease, and its impact is particularly concentrated in the elderly population. The core feature of this disease is a significant and progressive decline in cognitive function, which is manifested as a significant decline in memory, including the loss of short-term and long-term memory, as well as impairment in multiple cognitive areas such as thinking, judgment, and learning ability. In addition, patients are often accompanied by mood changes, abnormal behavior, and a gradual loss of the ability to take care of themselves in daily life, such as a decline in basic activities such as dressing, eating, and personal hygiene. Currently, with the acceleration of the aging process of the global population, the incidence of this disease is showing a rising trend, becoming an increasingly serious social public health problem. In view of this situation, in-depth research and timely intervention in Alzheimer's disease are particularly important.

[0003] Currently, mainstream Alzheimer's disease diagnostics rely primarily on clinical symptom assessment, neuropsychological testing, magnetic resonance imaging (MRI) or positron emission tomography (PET), or cerebrospinal fluid (CSF) analysis for markers. The first two methods, relying on subjective patient descriptions, offer low sensitivity, while the third and fourth methods are not only expensive but also invasive, posing significant biosafety risks.

[0004] For these reasons, blood tests are becoming increasingly popular for early screening of Alzheimer's disease. Due to their low cost and ease of use, they can be widely adopted by the general population. Based on the results of these tests, the patient can decide whether to undergo imaging methods such as MRI / PET or cerebrospinal fluid testing. However, as a new diagnostic method, blood tests still have relatively low sensitivity. Consequently, in actual use, it is very common for the prediction criteria to be too high, leading to a large number of missed tests, or for the prediction criteria to be too low, leading to subsequent over-testing.

[0005] Among them, chemiluminescence is the main method for blood testing for Alzheimer's disease. Its main principle is to use an instrument to detect the light intensity of the chemiluminescence reaction to quantitatively analyze the target substance in the sample. The standard curve of concentration-luminescence intensity based on the instrument is used to quantitatively analyze the target substance in the sample. Generally speaking, once the concentration of the target substance obtained by this method exceeds the distribution range of the normal population, it is considered that there is a risk of Alzheimer's disease. However, in actual clinical applications, it is found that this prediction standard threshold varies from person to person. If the same prediction standard threshold is used for different people, problems of missed diagnosis or misdiagnosis will occur, and the detection accuracy rate will be low. Summary of the Invention

[0006] The present invention provides an Alzheimer's disease data processing method and system, which can solve the problem in related technologies of low detection accuracy caused by the same Alzheimer's disease prediction standard threshold for different subjects to be detected.

[0007] In order to solve the technical problems in the background technology, a first aspect of the present invention provides a method for processing Alzheimer's disease data, the method comprising the following steps:

[0008] Obtain indicator information of the object to be detected;

[0009] Encoding the indicator information, assigning a unique code bit to each bit of data in the indicator information to form a first coding sequence combination;

[0010] Encoding the first coding sequence combination, and multiplying the number of valid characters of the first coding sequence combination by m times to form a second coding sequence combination, where m is a positive integer greater than or equal to 2;

[0011] performing deconvolution on the second coding sequence combination, and amplifying the second coding sequence combination into a one-dimensional convolution array;

[0012] Performing vertical staggered stacking on all arrays in the one-dimensional convolution array to form a two-dimensional array;

[0013] Inputting the two-dimensional array into a first neural network to obtain a first prediction standard threshold; the first neural network is a convolutional neural network;

[0014] Inputting the two-dimensional array into a second neural network to obtain a second prediction standard threshold; the second neural network is a cutting neural network;

[0015] A third prediction standard threshold is obtained by performing a weighted sum calculation on the first prediction standard threshold and the second prediction standard threshold.

[0016] Optionally, the first encoding sequence combination includes escape sequences corresponding to each bit of data, and each escape sequence includes multiple characters;

[0017] The step of encoding the first coding sequence combination and increasing the number of valid characters of the first coding sequence combination by m times to form a second coding sequence combination, where m is a positive integer greater than or equal to 2, comprises:

[0018] Traversing each escape sequence in the first encoding sequence combination;

[0019] Extract the last m bits of each escape sequence as the escape extraction sequence;

[0020] Each character of each m-bit escape extraction sequence is encoded to increase the number of valid characters of the first encoding sequence combination by m times to form a second encoding sequence combination, where m is a positive integer greater than or equal to 2.

[0021] Optionally, the step of performing deconvolution on the second coding sequence combination and amplifying the second coding sequence combination into a one-dimensional convolution array includes:

[0022] A deconvolution operation is performed on the second coding sequence combination using a 3*2 convolution kernel, with a step size of 1 each time. After the deconvolution operation, the total data length of the second coding sequence group is expanded to form a one-dimensional convolution array.

[0023] Optionally, after the step of deconvolving the second coding sequence combination and amplifying the second coding sequence combination into a one-dimensional convolution array is completed, and before the step of vertically stacking all arrays in the one-dimensional convolution array to form a two-dimensional array is performed, the following steps are further performed:

[0024] Adjust the one-dimensional convolution array X to obtain a one-dimensional convolution adjustment array , where the one-dimensional convolution array X includes , the total data length of the one-dimensional convolution array X is n;

[0025] By formula , get the one-dimensional convolution adjustment array ;

[0026] One-dimensional convolution resize array include ;in, Represents a one-dimensional convolution adjustment array The kth value in , j represents the imaginary unit, and the calculation formula of j is as follows: .

[0027] Optionally, the step of performing staggered vertical stacking on all arrays in the one-dimensional convolution array to form a two-dimensional array includes:

[0028] Determine the first row of the one-dimensional convolution array of the two-dimensional array;

[0029] All data in the first row of the one-dimensional convolution array are shifted right by one position to obtain the qth row of the one-dimensional convolution array of the two-dimensional array, where q is a positive integer greater than 1 and less than or equal to n.

[0030] Optionally, the first row of the one-dimensional convolution array includes The one-dimensional convolution array X.

[0031] Optionally, the first row of the one-dimensional convolution array includes One-dimensional convolution resized array , the two-dimensional array formed is:

[0032] .

[0033] Optionally, the step of inputting the two-dimensional array into a first neural network to obtain a first prediction standard threshold, wherein the first neural network is a convolutional neural network, comprises:

[0034] The two-dimensional array is input into a first neural network, which is sequentially arranged from the input end to the output end, including a first convolutional layer, a first activation layer, a second convolutional layer, a second activation layer, a first pooling layer, a third convolutional layer, a third activation layer, a fourth convolutional layer, a fourth activation layer, a second pooling layer, a fifth convolutional layer, a fifth activation layer, a sixth convolutional layer, a sixth activation layer, a third pooling layer and a fully connected layer, and the first prediction standard threshold is obtained by output.

[0035] Optionally, the step of inputting the two-dimensional array into a second neural network to obtain a second prediction standard threshold value; wherein the second neural network is a step of cutting a neural network, comprises:

[0036] In a segmentation layer of the second neural network, the two-dimensional array is segmented into a plurality of two-dimensional sub-arrays of equal size according to a target size;

[0037] Each two-dimensional sub-array is fed into the first fully connected layer of the second neural network, and the dimension of each two-dimensional sub-array is scaled;

[0038] Record the position information of the data blocks output by each two-dimensional sub-array from the first fully connected layer;

[0039] Based on the position information of the data blocks output from the first fully connected layer of each two-dimensional sub-array, all the data blocks output from the first fully connected layer are superimposed to form a data block superposition array;

[0040] Inputting the data block stacking array into the multi-head attention layer of the second neural network, and adding the position information of the data blocks to obtain the data block padded stacking array;

[0041] The padded stacked array of data blocks is sequentially input into the second fully connected layer of the second neural network, the third fully connected layer of the second neural network, and the output layer of the second neural network, and the second prediction standard threshold is output from the output layer of the second neural network.

[0042] Optionally, the third prediction standard threshold is calculated by a weighted formula, wherein the weighted formula is:

[0043] ;

[0044] in is the first prediction standard threshold, is the second prediction standard threshold, a and b are weight coefficients, and a and b are adjusted according to the training results.

[0045] In order to solve the technical problems in the background technology, the second aspect of the present invention provides an Alzheimer's disease data processing system, which includes a processor and a memory, wherein the memory stores at least one instruction or program, and the instruction or program is loaded and executed by the processor to implement the Alzheimer's disease data processing method as described in the first aspect of the present invention.

[0046] The Alzheimer's disease data processing method and system provided by the present invention can calculate the Alzheimer's disease prediction standard threshold corresponding to the indicator information of the object to be detected, that is, the third prediction standard threshold, and realize personalized setting of the standard threshold for Alzheimer's disease prediction to improve the detection accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] In order to more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the specific embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0048] Figure 1 A flow chart of a method for processing Alzheimer's disease data according to an embodiment of the present invention is shown;

[0049] Figure 2 A flow chart showing step S6 is shown;

[0050] Figure 3 A flow chart of step S7 is shown. DETAILED DESCRIPTION

[0051] The following is a clear and complete description of the technical solutions of the present invention in conjunction with the accompanying drawings. Obviously, the embodiments described are only some of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making any creative efforts are within the scope of protection of the present invention.

[0052] In addition, the technical features involved in the different embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0053] Figure 1 A flow chart of an Alzheimer's disease data processing method provided by an embodiment of the present invention is shown. It should be noted that each step of the Alzheimer's disease data processing method of the present invention is completely executed by a computer.

[0054] from Figure 1 As can be seen from FIG, the Alzheimer's disease data processing method includes the following steps:

[0055] Step S1: Acquire index information of an object to be detected, wherein the index information includes multiple bits of text data and multiple bits of numerical data.

[0056] The text data items include gender, family medical history, and genotype screening status, while the numerical data items include age, various scores of the Mental State Examination (MMSE), and various scores of the Mental Cognitive Assessment (MoCA).

[0057] The text data and numerical data of the above multiple items may be stored with spaces between different items.

[0058] For example, the indicator information of the object A to be detected includes:

[0059] Gender: Male; Family history: None within three generations; None outside three generations; Age: 77; MMSE: 111111111110111111111111111111 MoCA: 101111111111111 1111101111111111

[0060] Step S2: Encode the indicator information, and assign a unique code bit to each bit of data in the indicator information to form a first coding sequence combination.

[0061] The first coding sequence combination includes escape sequences corresponding to each bit of data, and each escape sequence includes multiple characters.

[0062] The process of assigning a unique code bit to each bit of data in the indicator information to form a first coding sequence combination enables the encoding of the indicator information in step S2 to convert the text data and numerical data in the indicator information to unify different data types in the indicator information.

[0063] Exemplarily, the indicator information may be encoded using Unicode encoding.

[0064] To improve the validity of the data, the space information in the indicator information is deleted before step S2, or the escape sequence corresponding to the space information in the indicator information is deleted after step S2 to obtain a first coding sequence combination that does not include space information.

[0065] Step S3: Encode the first coding sequence combination, and multiply the number of valid characters of the first coding sequence combination by m times to form a second coding sequence combination, where m is a positive integer greater than or equal to 2.

[0066] Exemplarily, step S3 may be implemented by the following steps:

[0067] Step S31: traverse each escape sequence in the first encoding sequence combination.

[0068] Step S32: extracting the last m bits of each escape sequence as an escape extraction sequence.

[0069] Step S33: Encode each character of the m-bit escape extraction sequence to increase the number of valid characters of the first encoding sequence combination by m times to form a second encoding sequence combination, where m is a positive integer greater than or equal to 2.

[0070] For example, taking the escape sequence \u6027 in the first code sequence combination as an example, the last four digits of the escape sequence \u6027 are extracted as the escape extraction sequence. Each digit in the escape extraction sequence 6027 is then encoded to obtain the second code sequence [36, 30, 32, 37], thereby quadrupling the data length of the escape sequence \u6027. If the total data length in the first code sequence combination is 87, the total data length of the second code sequence combination obtained after quadrupling the data length is 348.

[0071] In this embodiment, the effective data length of the first coding sequence combination is increased by m times to form a second coding sequence combination to enhance the data features in the indicator information, laying the foundation for improving the accuracy of subsequent data processing. Optionally, the first coding sequence combination can be encoded using HEX.

[0072] Step S4: performing deconvolution on the second coding sequence combination, and amplifying the second coding sequence combination into a one-dimensional convolution array.

[0073] In step S4, deconvolution is performed on the second coding sequence combination to amplify the second coding sequence combination into a one-dimensional convolution array, so as to further enhance the data features in the indicator information.

[0074] Exemplarily, a small convolution kernel may be used to perform deconvolution on the second coding sequence combination. For example, a 3*2 convolution kernel may be used to perform deconvolution on the second coding sequence combination, with a step size of 1 each time. After the deconvolution operation, the total data length of the second coding sequence group is expanded to form a one-dimensional convolution array. For example, when the total data length of the second coding sequence group is 348, a 3*2 convolution kernel may be used to perform deconvolution on the second coding sequence combination, with a step size of 1 each time. After the deconvolution operation, the total data length of the second coding sequence group is expanded to form a one-dimensional convolution array with a total data length of 512.

[0075] In order to enhance the correlation between different project data and enable subsequent data processing to effectively extract the correlation features between different project data, after step S4 is completed, the step of adjusting the obtained one-dimensional convolution array X to obtain a one-dimensional convolution adjustment array can be performed, wherein the one-dimensional convolution array X includes , the total data length of the one-dimensional convolution array X is n. This step includes: , get the one-dimensional convolution adjustment array . One-dimensional convolution adjustment array include Among them, Represents a one-dimensional convolution adjustment array The kth value in , j represents the imaginary unit, and the calculation formula of j is: .

[0076] It should be noted that the one-dimensional convolution adjustment array The same as the total data length of the one-dimensional convolution array X.

[0077] Step S5: performing vertical staggered stacking on all arrays in the one-dimensional convolution array to form a two-dimensional array.

[0078] Step S5 can be implemented by the following steps:

[0079] Step S51: Determine the first row of the one-dimensional convolution array of the two-dimensional array.

[0080] The first row of the one-dimensional convolution array may be the one in step S4 including The one-dimensional convolution array X can also include One-dimensional convolution resized array .

[0081] Resize array with 1D convolution Taking the first row of one-dimensional convolution array as an example, the first row of one-dimensional convolution array includes .

[0082] Step S52: Shift all the data in the first row of the one-dimensional convolution array one position to the right in sequence to obtain the qth row of the two-dimensional array, where q is a positive integer greater than 1 and less than or equal to n; the resulting two-dimensional array is .

[0083] For example, if we adjust the array by one-dimensional convolution Taking the first row of one-dimensional convolution array as an example, all the data in the first row of one-dimensional convolution array are shifted right by one position to obtain the second row of one-dimensional convolution adjustment array. The second row of one-dimensional convolution adjustment array is The last bit of the first row of the one-dimensional convolution array is X cn Prepend to the first bit in the second row of the one-dimensional convolution adjustment array.

[0084] Then, all data in the second one-dimensional convolution array are shifted right by one position to obtain a third row of the one-dimensional convolution adjustment array... and this process is repeated until the n-th row of the one-dimensional convolution adjustment array is determined.

[0085] If the total data length of the first row of the one-dimensional convolution array is 512, the total data length of the two-dimensional array is expanded to 512*512.

[0086] Because the convolution kernel used in subsequent steps to perform feature extraction on the one-dimensional convolution array yields a small number of valid feature vectors, this can lead to low prediction accuracy. Therefore, this embodiment performs a vertical, offset stacking of all arrays within the one-dimensional convolution array to form a two-dimensional array. This enhances data features, amplifies the valid feature vectors for subsequent feature extraction operations, and improves prediction accuracy.

[0087] In addition, due to the influence of the convolutional neural network itself in the subsequent steps, each feature extraction can only extract features of the convolution kernel size. However, due to the correlation between different data items of the indicator information, by staggering and vertically stacking all arrays in the one-dimensional convolution array to form a two-dimensional array, the correlation between different data items can be enhanced, so that the correlation features between different data items can be effectively extracted.

[0088] Step S6: Inputting the two-dimensional array into a first neural network to obtain a first prediction standard threshold; the first neural network is a convolutional neural network.

[0089] For example, referring to Figure 2 , which illustrates a flow chart of step S6. Figure 2It can be seen that the first neural network includes a first convolutional layer, a first activation layer, a second convolutional layer, a second activation layer, a first pooling layer, a third convolutional layer, a third activation layer, a fourth convolutional layer, a fourth activation layer, a second pooling layer, a fifth convolutional layer, a fifth activation layer, a sixth convolutional layer, a sixth activation layer, a third pooling layer and a fully connected layer, which are arranged in sequence from the input end to the output end.

[0090] Among them, the first convolution layer, the second convolution layer, the third convolution layer, the fourth convolution layer, the fifth convolution layer, and the sixth convolution layer all include 9*9 convolution kernels, and the data in the two-dimensional array are traversed through the 9*9 convolution kernels with a larger receptive field of view.

[0091] Among them, the first activation layer, the second activation layer, the third activation layer, the fourth activation layer, the fifth activation layer, and the sixth activation layer all include activation functions .

[0092] Here, a is an empirical value, ranging from 0 to 1. For example, as training results are updated, the initial value of a is typically 0.7. The activation layer updates all data in the array, increasing the nonlinearity of the entire network and allowing the neural network to extract more deep structures.

[0093] Taking the two-dimensional array of size 512*512*1 obtained in the embodiment as an example, the data of the two-dimensional array first passes through the first convolution layer containing a 9*9 convolution kernel to obtain a new array of 512*512*16, and then the data of the new array is input into the first activation layer. After passing through the first activation layer, the data size is still 512*512*16, and then it is input into the second convolution layer to obtain an array of 512*512*64, and then it is input into the second activation layer to obtain an array of data size of 512*512*64, and then the data is input into the first pooling layer to obtain an array of size 256*256*128. Then after the third convolution layer, the array size remains at 256*256*128. After the third activation layer, the data size is 256*256*128. Then it is put into the fourth convolution layer, the data size is 256*256*256. Then it is put into the fourth activation layer, maintaining the data size at 256*256*256. It is put into the second pooling layer, and the data size becomes 128*128*512. Then it is put into the fifth convolution layer. Starting from the fifth convolution layer, we choose a larger stride, so after the fifth convolution layer, the data size becomes 64*64*1024. Then it is put into the fifth activation layer, the data size becomes 64*64*1024. Then it is put into the sixth convolution layer, the data size becomes 32*32*2048. Then the data is put into the sixth activation layer, the data size remains unchanged, and then the data is put into the third pooling layer, and the data size becomes 16*16*4096. The data is then fed into the fully connected layer, where the data size is compressed to 1*1*4096. Finally, the data is fed into the output layer to obtain the first prediction standard threshold.

[0094] Step S7: inputting the two-dimensional array into a second neural network to obtain a second prediction standard threshold; the second neural network is a cutting neural network.

[0095] Figure 3 A flow chart of step S7 is shown.

[0096] Step S7 can be implemented by following steps S71 to S76:

[0097] Step S71: In the segmentation layer of the second neural network, the two-dimensional array is segmented into a plurality of two-dimensional sub-arrays of equal size according to the target size.

[0098] For example, taking the two-dimensional array of size 512*512*1 obtained in the embodiment as an example, we divide the two-dimensional array into 1024 two-dimensional sub-arrays according to the target size of 16*16*1.

[0099] This embodiment divides the two-dimensional array into several two-dimensional sub-arrays of equal size to avoid the problem of large-scale data consumption caused by directly inputting the two-dimensional array into the second neural network.

[0100] Step S72: Each two-dimensional sub-array is input into the first fully connected layer of the second neural network, and the dimension of each two-dimensional sub-array is scaled.

[0101] For example, in this embodiment, 1024 two-dimensional sub-arrays are used as the maximum sequence dimension and are respectively put into the subsequent second neural network, so that the dimension of each two-dimensional sub-array can be scaled.

[0102] Step S73: Record the position information of the data blocks output from the first fully connected layer for each two-dimensional sub-array.

[0103] In order to ensure the interactivity and relevance between different data items, each two-dimensional sub-array is input into the first fully connected layer of the second neural network. After the dimension of each two-dimensional sub-array is scaled, the position information of the data blocks output by each two-dimensional sub-array from the first fully connected layer is recorded.

[0104] Exemplarily, the data blocks output from the first fully connected layer are first inserted into the same partitioned positions as the corresponding two-dimensional subarrays to form a dataset. That is, the data blocks processed by the first fully connected layer are sequentially inserted into a dataset in the order of the original two-dimensional subarray partitioning, with each row first from left to right and then from top to bottom. The corresponding position information is then added to the data blocks output from the first fully connected layer.

[0105] The position information of the even-numbered data blocks in the data set is:

[0106]

[0107] The position information of the data blocks at odd positions in the data set is:

[0108]

[0109] Wherein, 2i and 2i+1 represent the number of data blocks output from the first fully connected layer, and N represents the total number of data blocks. In this embodiment, the size of N is 1024.

[0110] At this point, each data block output from the first fully connected layer not only has its own position information, but also has relative position information.

[0111] Step S74: Based on the position information of the data blocks output from the first fully connected layer of each two-dimensional sub-array, all the data blocks output from the first fully connected layer are superimposed to form a data block superposition array.

[0112] Taking 1024 two-dimensional sub-arrays as an example, 1024 data blocks are output from the first fully connected layer. After superimposing the 1024 data blocks, a data block superposition array of size 16*16*1024 is obtained, as well as a position information set of length 1024.

[0113] Step S75: Put the data block superposition array into the multi-head attention layer of the second neural network, and add the position information of the data block to obtain the data block padded superposition array.

[0114] For example, a 16*16*1024 sized data block stacking array is put into the multi-head attention layer, and the position information of the 1024 sized data block is added to obtain an 8*8*2024 sized data block padded stacking array.

[0115] Among them, the multi-head attention layer is multiple independent attention mechanisms in parallel in the neural network, which enables the neural network to simultaneously focus on information at different positions and semantic levels in the input data, thereby enhancing the expressive ability of the neural network.

[0116] Step S76: Input the padded and superimposed array of the data blocks into the second fully connected layer of the second neural network, the third fully connected layer of the second neural network, and the output layer of the second neural network in sequence, and output the second prediction standard threshold from the output layer of the second neural network.

[0117] Exemplarily, the padded and superimposed array of data blocks of size 8*8*2024 is input into the second fully connected layer of the second neural network to obtain an array of size 2*2*8192, and then the array of size 2*2*8192 is input into the third fully connected layer to obtain an array of size 1*1*16384, and finally the array of size 1*1*16384 is input into the output layer to obtain the second prediction standard threshold.

[0118] Step S8: performing a weighted sum calculation on the first prediction standard threshold and the second prediction standard threshold to obtain a third prediction standard threshold, wherein the third prediction standard threshold serves as the Alzheimer's disease prediction standard threshold corresponding to the indicator information of the subject to be detected.

[0119] For example, the third prediction criterion threshold value can be calculated by a weighted formula, where the weighted formula is: ;

[0120] in is the first prediction standard threshold, is the second prediction standard threshold, a and b are weight coefficients, and a and b are adjusted according to the training results.

[0121] Through the above steps, this embodiment can calculate the Alzheimer's disease prediction standard threshold corresponding to the indicator information of the object to be detected, that is, the third prediction standard threshold, and realize personalized setting of the standard threshold for Alzheimer's disease prediction to improve the detection accuracy.

[0122] After completing step S8, this embodiment can also calculate the error between the third prediction standard threshold and the true value to train the neural network. The error between the third prediction standard threshold and the true value can be calculated using the following formula:

[0123] in is the true value, is the result of calculation, N is the total number of training samples, and in the embodiment of the present invention, the value of N is 1000.

[0124] The present invention also provides an Alzheimer's disease data processing system, the Alzheimer's disease data processing system includes a processor and a memory, the memory stores at least one instruction or program, the instruction or program is loaded and executed by the processor to implement the following Figures 1 to 3 The Alzheimer's disease data processing method shown.

[0125] Obviously, the above embodiments are merely examples for clarity of explanation and are not intended to limit the implementation methods. Those skilled in the art will readily appreciate that other variations or modifications based on the above descriptions are possible. It is not necessary and impossible to enumerate all implementation methods here. Obvious variations or modifications arising therefrom remain within the scope of protection of the present invention.

Claims

1. A method for processing Alzheimer's disease data, characterized in that: The Alzheimer's disease data processing method comprises the following steps: Obtaining indicator information of the subject to be tested; the indicator information includes multiple bits of text data and multiple bits of numerical data; wherein the text data items include the gender, family medical history and genotype screening status of the subject to be tested; the numerical data items include age, various scores of mental status examination and various scores of cognitive assessment; Encoding the indicator information, assigning a unique code bit to each bit of data in the indicator information to form a first coding sequence combination; Encoding the first coding sequence combination, and multiplying the number of valid characters of the first coding sequence combination by m times to form a second coding sequence combination, where m is a positive integer greater than or equal to 2; performing deconvolution on the second coding sequence combination, and amplifying the second coding sequence combination into a one-dimensional convolution array; Performing vertical staggered stacking on all arrays in the one-dimensional convolution array to form a two-dimensional array; Inputting the two-dimensional array into a first neural network to obtain a first prediction standard threshold; the first neural network is a convolutional neural network; Inputting the two-dimensional array into a second neural network to obtain a second prediction standard threshold; the second neural network is a cutting neural network; Performing a weighted sum calculation on the first prediction standard threshold and the second prediction standard threshold to obtain a third prediction standard threshold; The third prediction standard threshold serves as the Alzheimer's disease prediction standard threshold corresponding to the indicator information of the subject to be detected.

2. The Alzheimer's disease data processing method according to claim 1, wherein: The first coding sequence combination includes escape sequences corresponding to each bit of data, and each escape sequence includes multiple characters; The step of encoding the first coding sequence combination and increasing the number of valid characters of the first coding sequence combination by m times to form a second coding sequence combination, where m is a positive integer greater than or equal to 2, comprises: Traversing each escape sequence in the first encoding sequence combination; Extract the last m bits of each escape sequence as the escape extraction sequence; Each character of each m-bit escape extraction sequence is encoded to increase the number of valid characters of the first encoding sequence combination by m times to form a second encoding sequence combination, where m is a positive integer greater than or equal to 2.

3. The Alzheimer's disease data processing method according to claim 1, wherein: The step of performing deconvolution on the second coding sequence combination and amplifying the second coding sequence combination into a one-dimensional convolution array includes: A deconvolution operation is performed on the second coding sequence combination using a 3*2 convolution kernel, with a step size of 1 each time. After the deconvolution operation, the total data length of the second coding sequence group is expanded to form a one-dimensional convolution array.

4. The Alzheimer's disease data processing method according to claim 1, wherein: After the step of deconvolving the second coding sequence combination to amplify the second coding sequence combination into a one-dimensional convolution array is completed, and before the step of vertically stacking all arrays in the one-dimensional convolution array to form a two-dimensional array is performed, the following steps are further performed: Adjust the one-dimensional convolution array X to obtain a one-dimensional convolution adjustment array , where the one-dimensional convolution array X includes , the total data length of the one-dimensional convolution array X is n; By formula , get the one-dimensional convolution adjustment array ; One-dimensional convolution resize array include ; Among them, Represents a one-dimensional convolution adjustment array The kth value in , j represents the imaginary unit, and the calculation formula of j is as follows: .

5. The Alzheimer's disease data processing method according to claim 1, wherein: The step of vertically stacking all arrays in the one-dimensional convolution array to form a two-dimensional array includes: Determine the first row of the one-dimensional convolution array of the two-dimensional array; All data in the first row of the one-dimensional convolution array are shifted right by one position to obtain the qth row of the one-dimensional convolution array of the two-dimensional array, where q is a positive integer greater than 1 and less than or equal to n.

6. The Alzheimer's disease data processing method according to claim 5, characterized in that: The first row of the one-dimensional convolution array includes The one-dimensional convolution array X.

7. The Alzheimer's disease data processing method according to claim 5, wherein: The first row of the one-dimensional convolution array includes One-dimensional convolution resized array , the two-dimensional array formed is: 。 8. The Alzheimer's disease data processing method according to claim 1, wherein: The step of inputting the two-dimensional array into the first neural network to obtain the first prediction standard threshold comprises: The two-dimensional array is input into a first neural network, which is sequentially arranged from the input end to the output end, including a first convolutional layer, a first activation layer, a second convolutional layer, a second activation layer, a first pooling layer, a third convolutional layer, a third activation layer, a fourth convolutional layer, a fourth activation layer, a second pooling layer, a fifth convolutional layer, a fifth activation layer, a sixth convolutional layer, a sixth activation layer, a third pooling layer and a fully connected layer, and the first prediction standard threshold is obtained by output.

9. The Alzheimer's disease data processing method according to claim 1, wherein: The step of inputting the two-dimensional array into a second neural network to obtain a second prediction standard threshold; wherein the second neural network is a step of cutting a neural network, comprising: In a segmentation layer of the second neural network, the two-dimensional array is segmented into a plurality of two-dimensional sub-arrays of equal size according to a target size; Each two-dimensional sub-array is fed into the first fully connected layer of the second neural network, and the dimension of each two-dimensional sub-array is scaled; Record the position information of the data blocks output by each two-dimensional sub-array from the first fully connected layer; Based on the position information of the data blocks output from the first fully connected layer of each two-dimensional sub-array, all the data blocks output from the first fully connected layer are superimposed to form a data block superposition array; Inputting the data block stacking array into the multi-head attention layer of the second neural network, and adding the position information of the data blocks to obtain the data block padded stacking array; The padded stacked array of data blocks is sequentially input into the second fully connected layer of the second neural network, the third fully connected layer of the second neural network, and the output layer of the second neural network, and the second prediction standard threshold is output from the output layer of the second neural network.

10. The Alzheimer's disease data processing method according to claim 1, wherein: The third prediction standard threshold is calculated by a weighted formula, wherein the weighted formula is: ； in is the first prediction standard threshold, is the second prediction standard threshold, a and b are weight coefficients, and a and b are adjusted according to the training results.

11. An Alzheimer's disease data processing system, characterized in that: The Alzheimer's disease data processing system includes a processor and a memory, wherein the memory stores at least one instruction or program, and the instruction or program is loaded and executed by the processor to implement the Alzheimer's disease data processing method according to any one of claims 1 to 10.

Images