PDVH behavior discrimination method and system based on machine learning

Abstract

The invention belongs to the technical field of biological medicine, and discloses a PDVH behavior discrimination method and system based on machine learning, and the method comprises the steps: constructing a PD mouse model and a PDVH mouse model, and respectively obtaining mouse behavior sequences of the PD mouse model and the PDVH mouse model in a behavioral test; key behavior modes related to illusion and a transformation rule between the behavior modes are extracted from the mouse behavior sequence; and inputting the key behavior mode and the transition rule between the behavior modes into a PDVH discrimination model constructed based on machine learning to train and discriminate whether a PDVH behavior occurs. According to the method, the visual illusion-related behavior pattern of the model mouse with the visual illusion is identified and verified through multi-dimensional monitoring, the occurrence of the illusion-related behavior is accurately diagnosed, and possibility is provided for drug screening, drug efficacy evaluation and mechanism research of the Parkinson's visual illusion.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of biomedicine, in particular to a PDVH behavior discrimination method and system based on machine learning. BACKGROUND

[0002] PD (Parkinson's Disease) is a common neurodegenerative disease, mainly manifested as motor symptoms such as tremor, rigidity and bradykinesia. However, as the disease progresses, about 20%-40% of Parkinson's disease patients will develop visual hallucinations PDVH (Parkinson's Disease-related Visual Hallucinations), which significantly affects the quality of life and disease management of patients. The existing PDVH animal model has not been able to fully simulate the hallucination symptoms of patients, and lacks accurate behavioral function detection methods.

[0003] In the existing animal model, the common hallucination model mainly relies on the application of hallucinogens. Hallucinogens such as LSD (Lysergic Acid Diethylamide), DMT (Dimethyltryptamine) are widely used to study hallucination symptoms of mental disorders such as schizophrenia and depression. However, these models are still limited in simulating visual hallucinations and related behavior characteristics in the context of Parkinson's disease. Therefore, how to construct a mouse model that can truly simulate Parkinson's visual hallucinations and accurately assess the process of hallucination through behavioral methods is still a problem to be solved.

[0004] Trihexyphenidyl hydrochloride, as an anticholinergic drug, is commonly used in clinical treatment of motor symptoms of Parkinson's disease. Studies have shown that when a higher dose of trihexyphenidyl hydrochloride is used, it can induce symptoms similar to visual hallucinations, making trihexyphenidyl hydrochloride an ideal drug for studying PDVH models. However, a complete PDVH mouse model induced by trihexyphenidyl hydrochloride has not been developed in existing research, and there is a lack of accurate assessment methods for hallucination behavior. Therefore, it is speculated that under the premise of restoring Parkinson's drug treatment as much as possible, trihexyphenidyl hydrochloride may induce behavior phenotypes highly related to Parkinson's hallucinations.

[0005] Trihexyphenidyl hydrochloride is an anticholinergic drug commonly used in clinical treatment to improve tremor and muscle rigidity and other motor symptoms in patients with Parkinson's disease. Previous studies have reported that at higher doses, this drug can induce central nervous system reactions including abnormal visual perception. However, the above studies have focused on single behavior changes or non-specific psychiatric symptoms, and there is still a lack of systematic verification of whether a stable, repeatable and Parkinson's disease characteristic background visual hallucination related behavior phenotype can be formed.

[0006] Further, the main reason why a systematic PDVH mouse model based on benzhexol hydrochloride intervention in the background of Parkinson's disease has not been established in the prior art is that, on the one hand, rodents cannot directly reflect hallucination experience through subjective reports, and lack objective and quantifiable behavioral evaluation indicators; on the other hand, the Parkinson's model itself has obvious movement disorders and behavioral laterality, and traditional behavioral parameters are difficult to distinguish non-specific movement changes from potential hallucination-related behaviors. Therefore, there is still a lack of a technical solution that can accurately identify and evaluate the visual hallucination-related behavior phenotype in the background of Parkinson's disease. SUMMARY

[0007] To solve the problems in the prior art, the present application provides a PDVH behavior discrimination method and system based on machine learning, which can simulate the background of Parkinson's disease drug treatment as much as possible, and through benzhexol hydrochloride intervention combined with specific behavior recognition and evaluation strategies, a PDVH animal model with stable behavior phenotype is established, thereby realizing the objective and quantitative evaluation of visual hallucination-related behavior.

[0008] The present application uses the method of benzhexol hydrochloride to induce a Parkinson's disease visual hallucination (PDVH) mouse model, and combines a multi-dimensional behavior monitoring system and a time series analysis method to perform accurate behavior analysis.

[0009] The present application adopts the following technical solutions.

[0010] The first aspect of the present application provides a PDVH behavior discrimination method based on machine learning, comprising:

[0011] Constructing a PD mouse model and a PDVH mouse model, and obtaining mouse behavior sequences of the PD mouse model and the PDVH mouse model in behavioral tests, respectively;

[0012] Extracting key behavior patterns related to hallucinations and transition rules between behavior patterns from the mouse behavior sequences;

[0013] Inputting the key behavior patterns and the transition rules between behavior patterns into a PDVH discrimination model constructed based on machine learning to train and discriminate whether PDVH behavior occurs.

[0014] Optionally, constructing the PDVH mouse model comprises: using benzhexol hydrochloride as an inducing agent after the PD mouse model is stable, using a dosage of 3-5 mg / kg / day, using an oral administration method, and continuously administering for 7-21 days to construct the PDVH mouse model.

[0015] Optionally, the transition rule between behavior patterns is extracted from the mouse behavior sequence, including:

[0016] A behavior state set is constructed, including movement, exploration, nursing and defense immobility;

[0017] The transition relationship between adjacent behaviors is extracted from the mouse behavior sequence, and the number of transitions between all behaviors is counted;

[0018] A first-order behavior transition probability matrix is constructed based on the number of transitions;

[0019] A first-order transition entropy is calculated based on the first-order behavior transition probability matrix to represent the transition rule between behavior patterns.

[0020] Optionally, the first-order behavior transition probability matrix is constructed based on the number of transitions, including:

[0021] The total number of transitions of the starting behavior is the denominator, and the single transition number of the starting behavior to another behavior is the numerator, to calculate the first-order behavior transition probability matrix of any two behaviors, wherein the starting behavior and the other behavior are behaviors in the behavior state set.

[0022] Optionally, the first-order transition entropy is calculated based on the first-order behavior transition probability matrix according to the following formula:

[0023] ,

[0024] Wherein, represents the first-order transition entropy, represents the starting behavior state, represents the starting behavior state of the mouse, the subsequent behavior state that occurs afterward, represents the starting behavior, the probability of occurrence, represents the conditional probability of transition from behavior state to behavior state , and is obtained by dividing the number of transitions from behavior state to behavior state by the total number of transitions of behavior state , represents the conditional probability of transition from behavior state to behavior state .

[0025] Optionally, the kinematic test includes mine site test, pole climbing test, grip strength test, and rotarod motor function test.

[0026] Optionally, the method further comprises deleting the mouse behavior sequence corresponding to the simple movement disorder of the PD mouse model after obtaining the mouse behavior sequence.

[0027] Optionally, the key behavior patterns related to hallucination are extracted from the mouse behavior sequence, including:

[0028] Free activity data of the control group, the PD mouse model and the PDVH mouse model are collected, and the free activity data are unsupervisedly clustered into a plurality of basic behavior types;

[0029] The difference in the proportion of the basic behavior types in the behavior of the control group and the PD mouse model is compared, and the corresponding basic behavior type is selected as the first intermediate behavior type when the difference is less than a difference threshold;

[0030] The proportion of the first intermediate behavior type and the motor function index are correlated, and the behavior type related to the motor function of Parkinson is selected;

[0031] The difference in the proportion of the basic behavior types in the behavior of the PD mouse model and the PDVH mouse model is compared, and the basic behavior type with a significant difference is selected as the second intermediate behavior type;

[0032] The second intermediate behavior type is deleted, and the behavior type related to the motor function of Parkinson is selected as the key behavior pattern related to hallucination.

[0033] Optionally, the method further comprises: segmenting the mouse behavior sequence according to an optimal time window, drawing a behavior time curve of the proportion of the duration of the appearance of the key behavior pattern with respect to time, and the behavior time curve is used to quantify the degree of hallucination; wherein the optimal time window is determined according to the similarity between the behavior time curves.

[0034] The second aspect of the present application provides a PDVH behavior discrimination system based on machine learning, which is used to realize the above-mentioned PDVH behavior discrimination method based on machine learning, comprising:

[0035] A construction module is used to construct a PD mouse model and a PDVH mouse model, and obtain the mouse behavior sequence of the PD mouse model and the PDVH mouse model in the behavior test, respectively;

[0036] An extraction module is used to extract the key behavior patterns related to hallucination and the transition rules between the behavior patterns from the mouse behavior sequence;

[0037] A discrimination module is used to input the key behavior patterns and the transition rules between the behavior patterns into a PDVH discrimination model based on machine learning for training and discrimination of whether the PDVH behavior occurs.

[0038] The third aspect of the present application provides an electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the computer program, when loaded into the processor, implements the above-mentioned PDVH behavior discrimination method based on machine learning.

[0039] The fourth aspect of the present application provides a computer readable storage medium, wherein the computer readable storage medium stores a computer program, and the computer program, when executed by a processor, implements the above-mentioned PDVH behavior discrimination method based on machine learning.

[0040] Compared with the prior art, the present application has at least the following beneficial effects:

[0041] The present application constructs a mouse model of Parkinson visual hallucination induced by benztropine hydrochloride, and trains a PDVH discrimination model based on machine learning by extracting key behavior patterns related to hallucination and the transition rules between behavior patterns, so that the discrimination model can capture the occurrence timing and transition rules of hallucination behavior, improve the accuracy and efficiency of behavior evaluation of the discrimination model, provide a more powerful tool for analyzing behavior characteristics related to hallucination, and provide strong support for drug screening, efficacy evaluation and mechanism research of Parkinson visual hallucination. BRIEF DESCRIPTION OF DRAWINGS

[0042] Figure 1 It is a schematic diagram of a PDVH behavior discrimination method;

[0043] Figure 2 It is a schematic diagram of identification and definition of Parkinson motor dysfunction related behavior patterns;

[0044] Figure 3 It is a schematic diagram of identification and definition of visual hallucination related behavior patterns,

[0045] Figure 4 It is a schematic diagram of a mouse behavior transition map;

[0046] Figure 5 It is a schematic diagram of prediction effect of Parkinson hallucination based on behavior characteristics and transition rules;

[0047] Figure 6 It is a schematic diagram of time course evaluation of hallucination state;

[0048] Figure 7 It is a schematic diagram of a behavior analysis system module for evaluating Parkinson hallucination. DETAILED DESCRIPTION

[0049] In order to further illustrate the present application, a series of examples are given below. These examples can enable a person skilled in the art to fully understand the present application, but should not be considered as limiting the scope of the present application.

[0050] The application provides a PDVH behavior discrimination method based on machine learning, comprising:

[0051] constructing a PD mouse model and a PDVH mouse model, and acquiring mouse behavior sequences of the PD mouse model and the PDVH mouse model in a behavior test, respectively;

[0052] extracting key behavior patterns related to hallucinations and transition rules between the behavior patterns from the mouse behavior sequences;

[0053] inputting the key behavior patterns and the transition rules between the behavior patterns into a PDVH discrimination model based on machine learning to train and discriminate whether PDVH behavior occurs.

[0054] Optionally, constructing the PDVH mouse model comprises: using trihexyphenidyl hydrochloride as an inducing agent to construct the PDVH mouse model after the PD mouse model is stable, in a dose of 3-5 mg / kg / day and an oral administration mode, and continuously administering for 7-21 days.

[0055] Optionally, extracting the transition rules between the behavior patterns from the mouse behavior sequences comprises:

[0056] constructing a behavior state set, wherein the behavior state set comprises movement, exploration, nursing and defense immobility;

[0057] extracting a conversion relationship of adjacent behaviors from the mouse behavior sequences, and counting transition times between all behaviors;

[0058] constructing a first-order behavior transition probability matrix based on the transition times;

[0059] calculating a first-order transition entropy based on the first-order behavior transition probability matrix to represent the transition rules between the behavior patterns.

[0060] Optionally, constructing the first-order behavior transition probability matrix based on the transition times comprises:

[0061] calculating a first-order behavior transition probability matrix of any two behaviors by taking a total transition time of a starting behavior as a denominator and taking a single transition time of the starting behavior to another behavior as a numerator, wherein the starting behavior and the another behavior are behaviors in the behavior state set.

[0062] Optionally, the first-order transition entropy is calculated based on the first-order behavior transition probability matrix according to the following formula:

[0063] ,

[0064] wherein, represents the first-order transition entropy, represents a starting behavior state, representing the initial behavior state of the mouse a subsequent behavior state that occurs thereafter, representing the initial behavior the probability of occurrence, representing the conditional probability of transitioning from a behavior state to a behavior state and is obtained by dividing the number of transitions of a behavior state to a behavior state by the total number of transitions of a behavior state , and representing the conditional probability of transitioning from a behavior state to a behavior state .

[0065] Optionally, the kinematic test includes a mine site test, a pole climbing test, a grip strength test, and a rotarod motor function test.

[0066] Optionally, the method further comprises deleting the mouse behavior sequence corresponding to the simple motor disorder of the PD mouse model after obtaining the mouse behavior sequence.

[0067] Extracting the key behavior pattern related to hallucination from the mouse behavior sequence includes:

[0068] Collecting free activity data of the control group, the PD mouse model, and the PDVH mouse model, and performing unsupervised clustering on the free activity data to divide the free activity data into a plurality of basic behavior types;

[0069] Comparing the difference in behavior proportion of the basic behavior types in the control group and the PD mouse model, and selecting the corresponding basic behavior type as the first intermediate behavior type when the difference is less than a difference threshold;

[0070] Performing correlation analysis on the proportion of the first intermediate behavior type and the motor function index, and selecting a behavior type related to the motor function of Parkinson's disease;

[0071] Comparing the difference in behavior proportion of the basic behavior types in the PD mouse model and the PDVH mouse model, and selecting the corresponding basic behavior type as the second intermediate behavior type when the difference is significant;

[0072] Deleting the behavior type related to the motor function of Parkinson's disease from the second intermediate behavior type as the key behavior pattern related to hallucination.

[0073] Optionally, the method further comprises segmenting the mouse behavior sequence according to an optimal time window, drawing a behavior time curve of the proportion of the duration of the occurrence of the key behavior pattern with respect to time, and using the behavior time curve to quantify the degree of hallucination; wherein the optimal time window is determined according to the similarity between the behavior time curves.

[0074] The following gives specific examples.

[0075] Example 1 Construction of PDVH mouse model based on benztropine induction

[0076] Experimental subjects and grouping: 8-week-old adult male C57BL / 6J mice were selected, weighing about 20-25 g, and free to eat and drink water. The mice were randomly divided into blank control group, PD group, and PDVH group.

[0077] (1) Blank control group (Control group): The mice in this group were not subjected to 6-OHDA injury modeling, but only underwent sham operation treatment, and an equal volume of normal saline was injected. This group was used to provide normal behavioral baseline data.

[0078] (2) PD model group (PD group): This group of mice established a unilateral 6-OHDA injury model. According to the established stereotactic method, 6-hydroxydopamine (6-OHDA) was injected into the target brain area to construct a Parkinson's disease motor impairment model. After the operation, the recovery period (such as 2-3 weeks, as actual) was given, during which no benztropine was administered. This group was used to evaluate the basic influence of Parkinson's pathology on behavior and served as an important control for distinguishing between movement disorder behavior and hallucination-related behavior.

[0079] (3) PDVH model group (PD + benztropine group): This group of mice first established a unilateral 6-OHDA Parkinson's model, and then administered benztropine hydrochloride after the model was stable. The benztropine administration regimen was 3 mg / kg / day, and the gavage was administered continuously for 14 days. This group was used to induce visual hallucination-related behavior phenotypes in the context of Parkinson's pathology.

[0080] Experimental method: PD model construction: In the groups that needed PD modeling, unilateral intracerebral stereotactic injection of 6-OHDA was used to establish striatal-nigral pathway injury; a 4-week recovery and stabilization period was set after induction, followed by basic motor function evaluation:

[0081] Open field test: used to evaluate the mouse's spontaneous activity, exploration behavior, and other indicators. PD mice often exhibit decreased total activity and reduced movement speed due to motor slowing and reduced activity, and may also have changes in exploration motivation; changes in kinematic parameters in the open field can reflect motor dysfunction and related behavior changes. After 4 weeks of PD modeling, the mice were placed in an open field behavioral apparatus for testing. The apparatus included a closed square box, which was equipped with a top camera recording system and behavior analysis software for automatically tracking mouse displacement trajectories and outputting behavior parameters. During testing, the mice were gently placed in the center of the open field, and the timing was started and the mouse's spontaneous activity was continuously recorded for 5 minutes. The behavior analysis software was BehaviorAtlas 3D-AI.

[0082] The behavior analysis software automatically calculates and outputs the following indicators: 1) total movement distance and / or average speed: reflecting the overall level of spontaneous movement; 2) trajectory heat map and edge preference: used to assist in presenting the activity distribution characteristics.

[0083] Rope climbing test: used to test and evaluate the motor coordination, motor initiation ability and descending motor control ability of mice. PD mice show motor slowing, decreased coordination and difficulty in starting due to impaired nigrostriatal pathway. By measuring the changes in rope climbing-related indicators, the degree of motor dysfunction can be reflected. After 4 weeks of PD modeling, the mice were tested using a rope climbing test device. The device includes a vertical round rod with non-slip texture on the surface, and a platform or ball head is provided at the top of the vertical round rod for placing the mice. The bottom of the round rod is connected to a soft pad to prevent injury from falling. During the test, the mice were placed head-up on the platform / ball head at the top of the rod, allowing them to grip the round rod with their limbs. After starting the timer, the time parameters for the mice to complete the following processes were recorded: 1) T-turn: the time for the mouse to turn from head-up to head-down; 2) T-total: the total time for the mouse to turn from the starting point, climb to the bottom of the rod and touch the ground with its limbs / reach the base.

[0084] Grip strength test: forelimb grip strength test is used to test and evaluate forelimb neuromuscular function. PD mice have abnormal nigrostriatal pathways, resulting in abnormal grip strength. By testing the changes in grip strength, the condition of motor dysfunction can be reflected. After 4 weeks of PD modeling, the mice were tested for grip strength using a grip strength measurement system. When the mouse's forelimbs grasp the grip rod on the grid, gently pull the tail backward, and the maximum grip strength is reached when the mouse's forelimbs are released. The biofunction signal system software records the data. Each mouse was tested 3 times, and the mean value was taken for statistical analysis.

[0085] Rotarod motor function test: the fatigue rotating rod test is commonly used to evaluate the coordination and balance of animal movement. During the behavior test, in order to select mice that can adapt to the fatigue rotating rod movement under the same conditions, all mice were pre-trained on the fatigue rotating rod for 2 days before modeling. The animals were placed on the stationary rotating rod for 2 minutes, then trained twice at 5 rmp (Revolutions Per Minute) and 40 rmp, each for 5 minutes. If the mouse falls during training, it is placed back on the rotating rod to continue training until the end. After 4 weeks of modeling, the mice were placed on the rotating rod, and the speed was adjusted from 5 rmp to 40 rmp. The time of the first fall of the mouse from the rotating rod was recorded, and the data was analyzed.

[0086] Behavioral results show that: Figure 1Fig. 2 shows the results of immunostaining of TH-positive neurons in the SN of the mice, wherein a indicates the schematic diagram of the time points of mouse modeling and behavioral detection, b indicates the movement trajectory of the mice in the blank control group and the PD group in the open field test, c indicates the statistics of the movement distance and speed of the mice in the blank control group and the PD group in the open field test, d indicates the falling time of the mice in the blank control group and the PD group in the rotarod test, e indicates the climbing time and falling time of the mice in the blank control group and the PD group in the pole climbing test, f indicates the grip strength of the mice in the blank control group and the PD group, g indicates the results of immunostaining of TH-positive neurons, wherein i indicates the non-injured side, ii indicates the injured side, h indicates the statistics of the number of TH-positive neurons on the non-injured side and the injured side of the PD group, i indicates the statistics of the number of head tremors based on benztropine for two consecutive weeks, j indicates the number of head tremors within ten minutes after the seventh day of administration; the data is represented by mean ± standard error; it can be seen that, compared with the blank control group, the 6-OHDA-treated mice showed significantly reduced activity speed and distance in the open field, shorter rotarod descending latency, longer turning and pole descending time in the rotarod test, and weaker grip strength. As shown in Fig. 2e, the mice in the PD group showed significantly longer pole descending time and turning time than the mice in the blank control group, indicating that the mice in the PD group had difficulty in climbing the pole and turning around the pole. As shown in Fig. 2f, the mice in the PD group showed significantly weaker grip strength than the mice in the blank control group, indicating that the mice in the PD group had difficulty in holding the pole. As shown in Fig. 2g, the number of TH-positive neurons in the SN of the mice in the PD group was significantly lower than that in the blank control group, indicating that the 6-OHDA treatment caused a significant loss of dopaminergic neurons in the SN of the mice. As shown in Fig. 2h, the number of TH-positive neurons in the non-injured side of the mice in the PD group was significantly lower than that in the blank control group, indicating that the 6-OHDA treatment caused a significant loss of dopaminergic neurons in the non-injured side of the mice. As shown in Fig. 2i, the number of head tremors based on benztropine for two consecutive weeks was significantly higher than that in the blank control group, indicating that the mice in the PD group had a higher incidence of head tremors. As shown in Fig. 2j, the number of head tremors within ten minutes after the seventh day of administration was significantly higher than that in the blank control group, indicating that the mice in the PD group had a higher incidence of head tremors. The above results demonstrate the successful construction of a unilateral injury Parkinson's mouse model. Figure 1 The results of immunostaining shown in Fig. 2g and h show that there is a significant reduction in dopaminergic neurons in the SN. Specifically, the number of tyrosine hydroxylase-positive (TH+) cells on the 6-OHDA-injured side is significantly lower than that on the contralateral cerebral hemisphere. In contrast, the blank control group CTRL mice show symmetrical TH+ cell distribution on both sides, indicating that 6-OHDA has specific neurotoxicity to dopaminergic neurons. The above results demonstrate the successful construction of a unilateral injury Parkinson's mouse model.

[0087] Hallucination-related state induction and verification: After the PD state is stable, benztropine is administered for 14 consecutive days (3 mg / kg / d, gavage), and the number of head tremors is recorded on days 1, 3, 5, 7, and 14 of administration, combined with Figure 1 As shown in Fig. 2i and j, the results show that the mice can be induced to exhibit significant head tremor-like behavior when administered for 7 consecutive days, and can be stable and sustained for 14 days with administration, which is similar to the head tremor induced by classic hallucinogens.

[0088] Example 2 Identification and verification of visual hallucination-related behavior patterns

[0089] Step 1: The BehaviorAtlas 3D-AI animal behavior analysis system was used to collect the free activity data of the mice in the blank control group, the PD group, and the PDVH group, and the free activity data was unsupervised clustered into multiple basic behavior types.

[0090] Specifically, the data of the mice's 20-minute task-free free activity was recorded additionally for subsequent analysis.

[0091] Step 2: compare the difference of the proportion of the basic behavior type in the control group and the PD group, and select the corresponding basic behavior type as the first intermediate behavior type when the difference is less than the difference threshold.

[0092] It should be noted that the difference calculation described in the present application can be calculated by statistical methods, for example, the p value obtained based on t test, when the p value is less than 0.05, it is considered that there is a significant difference, and the difference threshold can also be set by the person skilled in the art according to the application, and the asterisks in the drawings of the present application represent different degrees of statistical difference, and the more asterisks, the greater the statistical difference.

[0093] Step 3: correlation analysis of the proportion of the first intermediate behavior type and the motor function index is performed to select the behavior type related to the motor function of Parkinson.

[0094] Specifically, the Pearson correlation coefficient is used to calculate the correlation between the proportion of each behavior type and the above-mentioned motor function index. The specific steps include: calculating the behavior proportion of each behavior type for each experimental animal; collecting the corresponding traditional motor behavior index of the same animal; constructing the behavior proportion data matrix and the motor function index data matrix; calculating the correlation coefficient between the two; according to the absolute value of the correlation coefficient and the statistical significance p<0.05, the behavior type significantly related to the motor function is screened out.

[0095] In the present application, when the behavior proportion of a behavior type is significantly correlated with at least one motor function index, it is considered that the behavior type is related to the change of the motor function of Parkinson, and it is determined as a motor-related behavior for subsequent analysis. Since the most hallucination-related behavior needs to be found among all the behaviors that have differences after administration, behaviors that are definitely related to movement need to be excluded first.

[0096] Step 4: compare the difference of the proportion of the basic behavior type in the PD group and the PDVH group, and select the corresponding basic behavior type with significant difference as the second intermediate behavior type.

[0097] It can be understood that the comparison before is the pathological change caused by PD Parkinson itself, and here is what kind of behavior difference will be caused by the administration of benhexol on the basis of PD.

[0098] Step 5: delete the second intermediate behavior type and select M1, M34 as the key behavior pattern related to hallucination.

[0099] Forty different behavior types are obtained by using the behavior module of the BehaviorAtlas 3D-AI animal behavior analysis system.

[0100] 1) Behavior pattern screening and analysis.

[0101] As Figure 2 shown, Figure 2 a indicates the comparison of the proportion of behavior types between the control group and the PD group mice, the dotted box indicates the behavior type with statistically significant difference, One-way ANOVA, * p < 0.05, ** p < 0.01, **** p < 0.0001 compared with the control group; b indicates the clustering diagram of the spontaneous behavior module with difference, c-f indicates the difference analysis of the proportion of single movement behavior between the PD group and the control group, the data is expressed as mean ± standard error. * p < 0.05, ** p < 0.01, **** p < 0.0001 compared with the control group, g indicates the two-dimensional embedding analysis result of the behavior proportion data based on t-SNE algorithm. The support vector machine decision boundary is constructed using Gaussian kernel function; h indicates the correlation between the proportion of climbing behavior and standing behavior and the movement speed; i indicates the correlation between the proportion of left turning behavior and the latency of rotarod experiment, n = 12 in each group; The correlation coefficient (r) and the significance level (p value) are marked in the figure.

[0102] As Figure 3 shown, Figure 3 a indicates the use of a Venn diagram to screen the behavior numbers that change regularly in the three groups, b, c indicates the behavior label annotated by behavior characteristics, including movement or non-movement related behavior, d indicates the proportion of the duration of all screened behaviors to the total duration of all behaviors, the comparison between groups uses one-way ANOVA, and the difference is significant compared with the control group; and the difference is significant compared with the PD group * p < 0.05, ** p < 0.01, *** p < 0.001, e indicates the change characteristics of the moving speed of the head skeleton points (nose, left and right ears) with time during the M1 behavior duration, the gaze is characterized by the head point maintaining a low speed for a long time, f indicates the change characteristics of the moving speed of the head skeleton points (nose, left and right ears) with time during the M34 behavior duration, head tremor is characterized by a significant increase in the speed of the nose point, g indicates the schematic diagram of the eye tracking experiment of mice, h indicates the pupil movement distribution diagram of mice in the PD group and the PDVH group, i indicates the difference statistics of the pupil point dispersion of mice in the PD group and the PDVH group, j indicates the proportion of mice in the three groups in the M1 behavior mode after treatment with pimavanserin, k indicates the proportion of mice in the three groups in the M34 behavior mode after treatment with pimavanserin, the comparison between groups uses one-way ANOVA.

[0103] By comparing the control group and the PD group, as Figure 2As shown in a-g, first, it was found that there were 6 behaviors that showed significant differences, and this difference could significantly distinguish the two groups of mice, which indicated that the changes in spontaneous behavior might be related to the motor function or non-motor function of Parkinson. Further correlation analysis was performed between the proportion of spontaneous behavior and the behavior data of the mice, as shown in Figure 2 As shown in h, the climbing and lifting behaviors were positively correlated with the movement speed in the open field, as shown in Figure 2 As shown in i, the left turning behavior was negatively correlated with the time of falling off the rotating rod. This indicated that part of the behaviors were related to the motor function of Parkinson. By comparing the PD group and the PDVH group, 8 behaviors (M1, M2, M7, M8, M19, M34, M38, M39) showed significant differences, indicating that they were related to non-motor state specificity, as shown in Figure 3 As shown in b-c, M1 represents arching with staring, M2 represents arching, M7 and M8 represent classic grooming, M19 represents left grooming, and M34 represents arching with head tremor, and M38 represents trotting. By excluding the behaviors that already had differences between the control group and the PD group, the present disclosure selected two key behavior patterns, M1 and M34. Both M1 and M34 showed similar bending postures, accompanied by head twitching and staring-related behaviors. These characteristics made M1 and M34 be determined as arching states related to hallucinations, as shown in Figure 3 As shown in d, e represents the change characteristics of the movement speed of the head skeleton points (nose, left and right ears) over time during the duration of the M1 behavior, and staring is manifested as the head points continuously maintaining a low speed, and f represents the change characteristics of the movement speed of the head skeleton points (nose, left and right ears) over time during the duration of the M34 behavior, and head tremor is manifested as a significant increase in the speed of the nose point.

[0104] It can be understood that when the control group and the PD group were compared for the first time, the PDVH group was not added, and the behaviors obtained here were mainly differences in grooming, left turning, climbing, and arching, and these four were used in the exclusion. In the subsequent behavior screening, the first to be excluded were climbing and left turning, which were behaviors related to motor function, and if the PD changed compared with the control, the PDVH did not change, which indicated that this might be the effect of the drug in treating PD, and also needed to be excluded, and finally the behaviors that changed compared with the control, the PD group did not change (Parkinson itself did not affect), and the PDVH group changed (the comprehensive effect after administration) were selected as M1, M2, M19, M34, and M38. M2 and M38 are very common behaviors, and M19 is a reduced behavior, none of which meet the characteristics of hallucinations. However, M1 and M34 are potential candidate behaviors.

[0105] 2) Gazing behavior was identified using a head-mounted eye tracker. This device was designed to minimize physical interference. It consisted only of the necessary imaging components: a miniature CMOS camera mounted on a skull support, which was securely fixed to the exposed skull of the mouse. Figure 3 As shown in g. The camera is controlled and image processed using LabVIEW software, recording eye movement data in real time. The gaze behavior in mice is characterized by prolonged fixation of the eyeballs on a fixed target, accompanied by low eye movement speed and stable pupil position. This phenomenon was significantly increased after benztropine-induced modeling, as shown in... Figure 3 As shown in h and i, the probability of gazing behavior is higher, and the dispersion of pupil movement is significantly reduced, i.e., gazing.

[0106] Eye movement speed below the speed threshold and pupil position change less than the change threshold indicate low eye movement speed and stable pupil position.

[0107] The identification of the staring behavior demonstrated that it was a specific behavior related to visual hallucinations induced by benztropine in the context of Parkinson's pathology, rather than a simple motor disorder or non-specific behavioral change.

[0108] 3) Validation of behavioral patterns: Mice in the PDVH group were subcutaneously injected with pimovaserin 3 mg / kg / day for days 12 to 14, and behavioral tests were performed 30 minutes after the last injection. Pimovaserin is a 5-HT2A receptor inverse agonist and has been shown to have clinical efficacy in treating PD psychosis. Behavioral analysis results showed that pimovaserin treatment significantly reduced the duration of gaze and head tics, such as... Figure 3 As shown in j and k. The above results demonstrate that benztropine-induced behavioral changes can serve as a reliable marker for PDVH discrimination.

[0109] Example 3: Plotting behavioral transition maps and predicting behavioral dynamics in PDVH mice

[0110] By analyzing the behavioral transitions between the PDVH group and the PD group, a behavioral transition atlas was constructed, revealing the patterned differences in behavioral dynamics between the two groups. Through behavioral transition probability matrices, Markov models, and principal component analysis, the potential patterns and stability of the behavioral structure in PDVH mice were identified.

[0111] like Figure 4 As shown, Figure 4In the table, 'a' represents manual classification and annotation of all behaviors based on shared skeletal features and behavioral trends; 'b' represents the transition probability between various behaviors; 'c' represents the transition probability between the four common main behavioral types in mice; 'd' represents the statistical difference in transition among the above four behaviors, with significant differences compared to the PD group: *p<0.05, **p<0.01, ***p<0.001; 'e' represents the entropy rate statistics of first-order transition entropy, with significant differences compared to the PD group: ***p<0.001. Data are expressed as mean ± SEM. Unpaired t-tests were used for inter-group comparisons. f represents the volcano plot analysis of the difference in behavioral transfer probabilities between the PD group and the PCVVH group, with the horizontal axis representing the logarithmic change in behavioral transfer probability and the vertical axis representing the significance of the difference. g represents a schematic diagram of PCA dimensionality reduction analysis of the real behavioral sequences and the behavioral sequences simulated by the first-order Markov model based on 20 behavioral transfers with significant differences between the two groups. h represents the use of JSD (Jensen–Shannon Divergence, JS dispersion) to quantitatively compare the similarity and difference between the simulated behavioral sequences and the real behavioral sequences in the distribution of behavioral transfer probability, and to comprehensively evaluate the fitting effect of the first-order Markov model on the structure of the two groups of behavioral sequences.

[0112] Mapping behavioral transitions and predicting behavioral dynamics in PDVH mice specifically includes:

[0113] 1) Define behaviors and behavior transfers, and construct a set of behavioral states: Clearly classify and define the behavioral types of experimental animals, determine the set of behavioral states and the transfer relationships between behaviors, and construct a system of behavior transfer events. Refer to the Stanford University Mouse Behavioral Atlas Index, such as... Figure 4 As shown in Figure a, spontaneous behaviors of mice are divided into four categories, totaling ten different behaviors: movement, exploration, nursing, and defensive immobility. Movement includes, but is not limited to, jumping, turning left, turning right, trotting, and walking; exploration includes, but is not limited to, sniffing, climbing, and raising the body; nursing includes, but is not limited to, grooming; and defensive immobility includes, but is not limited to, arching the back.

[0114] 2) Construction of behavior transfer network: The transition relationship between adjacent behaviors was extracted from the continuous behavioral time sequence of mice in the PD group and PDVH group. A first-order behavior transfer probability matrix was constructed based on the transition relationship, and the first-order transfer entropy was calculated based on the first-order behavior transfer probability.

[0115] Further analysis of the annotated behavioral maps yielded the first-order behavioral transition probability matrices for the PD and PDVH groups of mice, such as... Figure 4 As shown in Figure b, the transfer differences between the four types of biological behaviors are as follows: Figure 4 As shown in Figure c, statistical results indicate that the probability of various behaviors shifting to nurturing behaviors decreased significantly, while exploratory and defensive behaviors increasingly transitioned to defensive immobility. Figure 4The first-order transition entropy is used to quantify the randomness and regularity of behavior transition. The lower the entropy value, the stronger the regularity and predictability of behavior transition; the higher the entropy value, the stronger the randomness of behavior transition.

[0116] It should be noted that the behavior is defined and its meaning is annotated according to the skeleton and video segment, with reference to the behavior atlas index of Stanford University.

[0117] The first-order transition entropy is calculated based on the first-order behavior transition probability matrix according to the following formula:

[0118]

[0119] wherein, denotes the first-order transition entropy, denotes the starting behavior state, denotes the starting behavior state of the mouse, the subsequent behavior state that occurs thereafter, denotes the starting behavior, the probability of occurrence, denotes the conditional probability of transition from the behavior state to the behavior state , and is obtained by dividing the number of transitions from the behavior state to the behavior state by the total number of transitions of the behavior state , denotes the conditional probability of transition from the behavior state to the behavior state .

[0120] As shown in the results in e in the detailed description, Figure 4 the first-order transition entropy of the PDVH mice is lower, showing stronger regularity and predictability, while the behavior transition of the PD group mice shows higher randomness.

[0121] Specifically, the transition relationship of adjacent behaviors is extracted from the continuous behavior time sequence of the PD group mice and the PDVH mice, the number of transitions between all behaviors is counted, and the first-order behavior transition probability matrix is constructed based on the number of transitions. The total number of transitions of a certain behavior as the denominator, and the number of single transitions of the behavior to another behavior as the numerator, the transition probability between any two behaviors is calculated.

[0122] 3) Construction of Markov model and behavior prediction based on state set and state transition matrix: independent first-order Markov behavior sequence models are constructed for the PD group and the PDVH group. First, based on the real behavior data of each group, the first-order behavior transition probability matrix specific to each group is calculated. Subsequently, 20 key behavior transitions with statistically significant differences are screened out through a volcano plot, such as Figure 4 ​As shown in f. Next, using independent Markov models, 16 simulated behavioral sequences with the same number of real samples were generated, ensuring each group contained 16 real sequences and 16 simulated sequences. PCA dimensionality reduction analysis was then performed on the real and simulated sequences using the selected significantly different behavioral transitions as features. (See figure.) Figure 4 As shown in Figure g, the simulated sequences of PDVH mice clustered well with the real data, while the simulated sequences of PD mice were significantly separated from the real data, suggesting that the model can effectively capture stable behavioral differences between the two groups.

[0123] 4) Evaluating the similarity between the model and real data based on JSD. The accuracy of the model was further verified by calculating the similarity of the distributions of the simulated and real sequences using JSD values. A low JSD value indicates that the model can accurately capture the behavioral structural differences between the PDVH and PD groups. For example... Figure 5 As shown in Figure h, the difference in distribution between the Markov model-generated data and the real behavioral data was quantitatively evaluated by calculating the JSD to further verify the model's ability to capture behavioral structural features. Sixteen mice from each of the PDVH and non-PDVH groups were included, and the distributions of real behavior and model-generated simulated behavior were constructed separately. The results showed that the overall JSD values ​​between the simulated and real data were low in both groups, indicating that the model can reconstruct real behavioral patterns well. Furthermore, in the joint distribution of real and simulated data, the PDVH and non-PDVH groups exhibited clearly separable structural boundaries, indicating that the model can not only effectively capture behavioral features within each group but also retain key differences between the two groups.

[0124] Example 4: Diagnosing Hallucinations Based on Characteristic Behaviors and Transition Patterns

[0125] Whether multidimensional analysis based on behavioral change patterns and characteristic behaviors is superior to single-behavioral parameter assessment in predicting hallucinations is investigated. Through ROC analysis, the Youden index, and model evaluation, it is demonstrated that comprehensive behavioral change characteristics can significantly improve the accuracy of hallucination prediction. This includes the following:

[0126] 1) Behavioral Feature Extraction and Model Building: Behavioral features related to hallucination occurrence were extracted, including the proportion of characteristic behaviors and behavioral transition features (the probability of transitions between hallucination-related states). Based on these features, a multi-dimensional prediction model was built, combining behavioral transition patterns with HHS (hallucination-related hunching states). The single-parameter evaluation model only used individual behavioral parameters (e.g., separate evaluations of M1 and M34).

[0127] 2) Model training and ROC analysis: The performance of the models was evaluated using ROC (Receiver Operating Characteristic) analysis. The AUC (Area Under Curve) values of different models were compared.

[0128] Figure 5 a-c represent the accuracy of different training models in distinguishing PD and PDVH groups of mice, where a represents the AUC comparison of integrated features (behavioral transition + M1 + M34), single behavioral transition features, and M1 + M34 combination; b represents the AUC comparison of M1 + M34 combination and single M1, M34 features; c represents the AUC comparison of single behavioral transition, M1, M34 features; d represents the Youden index analysis under different diagnostic thresholds for each behavioral feature, with higher values indicating better diagnostic accuracy; e represents the sensitivity and specificity quantification results of each core behavioral feature at the optimal threshold, which directly reflects the diagnostic accuracy of the model.

[0129] Figure 5 a-c results show that the classification accuracy of the integrated model in predicting hallucinations is significantly better than that of the single parameter model, with an AUC value of 0.96, while the AUC values of the single parameter model are: AUC M1 =0.90, AUC M34 =0.93, AUCtransition=0.92, which indicates that individual behavioral indicators cannot fully capture the complexity of hallucinations, and the integrated analysis of transition patterns can more effectively predict the occurrence of hallucinations.

[0130] 3) Youden index evaluation and sensitivity improvement: To further evaluate the clinical diagnostic effect of the model, the Youden index was calculated, which is widely used in biomedical research to determine the optimal diagnostic threshold. The results show that the Youden index of the integrated model is 0.875, which is significantly higher than that of all other single variable models. This enhanced discrimination ability is mainly due to the improvement of the sensitivity of the model in detecting hallucination-related behavioral states, while maintaining comparable specificity to other models, as shown in d. Figure 5

[0131] 4) Transition rule and sensitivity analysis: Through sensitivity analysis of the model, it was found that the introduction of behavioral transition patterns significantly improved the detection sensitivity of hallucination diagnosis. The model successfully captured more complex behavioral dynamics that were often ignored in single behavioral measurements. The complexity of mouse behavior was preserved, and the diagnostic ability was significantly improved, as shown in e. Figure 6

[0132] Example 5 Time course evaluation of hallucination state​​

[0133] Through time-course analysis, an appropriate time window was selected to measure the degree of hallucination, capture the temporal changes in hallucination-related behavioral states, and compare the temporal patterns of hallucination state changes between PDVH mice and PD mice. Specifically, this includes the following:

[0134] 1) Temporal distribution analysis of behavioral characteristics: Figure 6 In the figure, a and b represent the longest duration and frequency of M1 and M34 behaviors in the PD and PDVH groups, respectively. Data are expressed as mean ± standard error. Unpaired t-tests were used for statistical analysis between groups. The sample size in each group was n=16. Significant differences were found compared to the PD group (***p<0.001, ****p<0.0001). d represents the smoothed trajectory curves showing the proportion of M1 and M34 behaviors in the PD and PDVH groups over time windows. The shaded area represents the 95% confidence interval. Significant differences were found compared to the PD group (*p<0.05, **p<0.001, ***p<0.0001). e represents the Fréchet distance heatmap of the proportion of M1 and M34 behaviors in the PD and PDVH groups under different time window divisions. The values ​​in the cells are the calculated distance values. Figure 6 From a and b, we know that the average longest duration of a single occurrence of M1 and M34 is 10.64±2.19 seconds and 7.17±1.14 seconds, respectively, and the frequency of occurrence per minute for the two behaviors is M1: 2.47±0.22 and M34: 4.09±0.18, respectively.

[0135] 2) Time Window Selection and Optimization: Based on the behavioral characteristic time scale, this disclosure identifies three potentially suitable time windows: 5s, 10s, and 30s. The impact of different time window lengths on the behavioral sequence is further evaluated. By comparing the similarity of two sets of behavioral time curves under different time windows, 10s is determined to be the optimal time window. Figure 6 As shown in Figure e, comparing the impact of different time windows on the behavior time curve, a time window that is too short (5s) fragments continuous behavior and does not effectively improve the discrimination effect. A time window that is too long (30s) blurs key behavioral transitions, leading to a decrease in discrimination effect. It is understandable that the behavior time curve is constructed by connecting the proportions of M1 and M34 occurrences within each time window in chronological order. For example... Figure 7 As shown in d, the analysis of this time window shows that the differences in HHS mainly occur in the first 1-15 time windows, while they tend to be random in subsequent time windows.

[0136] 3) Behavioral sequence time series clustering analysis: To further verify the effectiveness of the time window, the M1 and M34 behavioral duration curves of the PD and PDVH groups of mice were subjected to hierarchical clustering, and the time series characteristics of multiple mice were unsupervised classified according to the similarity between the behavioral time curves, and the similarity of the behavioral trajectories was analyzed. The results showed that the HHS of the PDVH group had a time structure of first increasing and then decreasing over time, and the PD group lacked this structure.

[0137] Example 6 Multidimensional behavioral analysis system for behavioral evaluation of PDVH Parkinson hallucination mice

[0138] As shown in Figure 7 , this embodiment constructs a multidimensional behavioral analysis system for Parkinson's disease hallucination model mice, relying on high-precision behavior collection, AI feature recognition and time process analysis, to realize the quantitative analysis, feature mining and dynamic prediction of hallucination-related behaviors. The system as a whole is divided into three functional modules of behavior collection and quantification, core feature recognition, and hallucination state time process evaluation, and the specific implementation is as follows:

[0139] Step 1: Obtain mouse behavior sequence data, construct a behavior quantification dataset of PDVH group mice and PD group mice, and the behavior quantification data in the behavior quantification dataset include mouse body key point position, behavior trajectory, behavior pattern, and behavior statistical indicators.

[0140] Among them, the behaviors of mice are clustered into different behavior patterns based on the similarity of mouse body key point positions, and the behavior statistical indicators are obtained by analyzing the behavior trajectory, including the number of occurrences, duration, proportion of behavior duration to total behavior time, and behavior transition data. Behavior transition data includes transition probability, transition frequency and transition time series regularity between behavior patterns.

[0141] The BehaviorAtlas 3D-AI animal behavior analysis system is used to collect and cluster the behaviors of PDVH group mice and PD group mice in multiple dimensions and high precision, and output the unsupervised clustered behavior quantification dataset. The BehaviorAtlas 3D-AI animal behavior analysis system is used to collect mouse behavior from multiple perspectives, reconstruct three-dimensional posture, and perform unsupervised segmentation and clustering to obtain behavior modules that can be used for statistics, including: multi-perspective video collection and three-dimensional posture reconstruction.

[0142] The mice are placed in a 30 cm diameter open circle acrylic field with a white featureless bottom surface for free activity, and a four-camera synchronous acquisition device is used for video recording to obtain multi-view synchronous video data. The video acquisition parameters are 60 fps, resolution 848x480, and each mouse is continuously recorded for 20 minutes. A standard chessboard is used for camera calibration before the experiment, which is used for subsequent three-dimensional reconstruction. A deep learning key point tracking method is used to identify 16 body key points in the four-way video, including: nose, left and right ears, neck, back, left and right forelimbs, left and right hind limbs, left and right forepaws, left and right hind paws, tail root / tail middle / tail tip. The 2D key point coordinates of each view are combined with the camera calibration parameters to perform three-dimensional reconstruction to obtain the positions of each body key point of the mouse. According to the similarity of the positions of each body key point, different behaviors are clustered to obtain the behavior pattern of the mouse. The behavior quantification dataset includes: multi-view synchronous video data, camera calibration parameters, body key point 2D / 3D coordinate trajectories, and behavior segments, behavior patterns, behavior labels and behavior statistical indicators obtained based on the trajectories, including but not limited to occurrence frequency, duration, proportion, transfer characteristics, such as Figure 7 as shown in a of FIG. 12.

[0143] Step 2: Annotate and analyze the behavior quantification data to obtain key behaviors and transition characteristics, and predict hallucination occurrence behaviors according to the occurrence and transition rules of key behaviors.

[0144] Step 2.1: From the nearly 40 types of behavior patterns that can be analyzed by the system, taking the PD group as the control and the PDVH group as the experimental group, through multidimensional cross-analysis and statistical test, behaviors related to motor function are excluded, and key behavior characteristics M1, M34 that are significantly different between groups and highly related to hallucination phenotype are selected.

[0145] Step 2.2: Real-time eye movement data and mouse head tremor degree obtained to assist in accurate definition of core behaviors. By introducing a gaze behavior detection module, real-time monitoring of mouse pupil activity is performed, and head tremor and gaze degree are reflected by combining head skeleton point change characteristics, thereby realizing identification and definition of head tremor and gaze behaviors.

[0146] Step 2.3: Behavior transition map construction and dynamic prediction: based on the conversion rule of core behaviors, a PDVH mouse behavior transition map is constructed, and hallucination occurrence is predicted according to the Markov model, and the identification and diagnosis efficiency of the animal model hallucination phenotype can be improved by combining the behavior transition rule, such as Figure 7 as shown in b of FIG. 12.

[0147] Step 3: The hallucination state time is evaluated to quantify the intensity, distribution and evolution law of hallucination-related behaviors in the time dimension. By selecting an optimal time window, the time bin statistics of typical hallucination-related behaviors are performed to finely capture the time sequence fluctuation characteristics of the behavior state, and the differences in the behavior time mode between the PDVH group and the PD group are compared to realize the quantitative evaluation of the hallucination state over time. It can also reflect the time sequence characteristics of drug action, such as ​ as shown in c in the middle.

[0148] The application also provides a PDVH behavior discrimination system based on machine learning, which runs the above-mentioned PDVH behavior discrimination method based on machine learning. The system comprises:

[0149] A construction module is configured to construct a PD mouse model and a PDVH mouse model, and obtain mouse behavior sequences of the PD mouse model and the PDVH mouse model in a behavior test, respectively.

[0150] An extraction module is configured to extract key behavior patterns related to hallucinations and transition rules between behavior patterns from the mouse behavior sequences.

[0151] A discrimination module is configured to input the key behavior patterns and the transition rules between the behavior patterns into a PDVH discrimination model based on machine learning for training and discriminating whether PDVH behavior occurs.

[0152] As to the system in the above-mentioned embodiments, the specific manner in which each unit performs the operation has been described in detail in the embodiments related to the method, and will not be described in detail here.

[0153] The application also provides an electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the computer program, when loaded into the processor, implements the above-mentioned PDVH behavior discrimination method based on machine learning.

[0154] The application also provides a computer readable storage medium, which stores a computer program, wherein the computer program, when executed by a processor, implements the above-mentioned PDVH behavior discrimination method based on machine learning.

[0155] It should be understood that the size of the serial number of each step in the above-mentioned embodiments does not mean the order of execution, and the execution order of each process should be determined according to its function and inherent logic, and should not constitute any limitation on the implementation process of the embodiments of the application.

[0156] The present disclosure can be a system, a method, and / or a computer program product. The computer program product can include a computer readable storage medium having computer readable program instructions embodied therewith, wherein the computer readable program instructions are used to cause a processor to implement various aspects of the present disclosure.

[0157] It should be pointed out finally that the above embodiments are only used to illustrate the technical solutions of the present application but not to limit it. Although the present application has been described in detail with reference to the above embodiments, it should be understood by those skilled in the art that the specific embodiments of the present application can be modified or equivalently replaced without departing from the spirit and scope of the present application, and any modification or equivalent replacement should be covered within the protection scope of the claims of the present application.

Claims

1. A PDVH behavior discrimination method based on machine learning, characterized in that, include: PD mouse model and PDVH mouse model were constructed, and the mouse behavior sequences of PD mouse model and PDVH mouse model in behavioral tests were obtained respectively; Key behavioral patterns associated with hallucinations and the transition patterns between these patterns were extracted from mouse behavioral sequences. The key behavioral patterns and the transition patterns between them are input into a PDVH discriminant model built on machine learning for training to determine whether PDVH behavior has occurred.

2. The PDVH behavior discrimination method based on machine learning according to claim 1, characterized in that: The process of constructing a PDVH mouse model involves: after the PD mouse model has stabilized, using benztropine hydrochloride as an inducer, administering it orally at a dose of 3-5 mg / kg / day for 7-21 consecutive days to construct the PDVH mouse model.

3. The PDVH behavior discrimination method based on machine learning as described in claim 1, characterized in that: The patterns of transition between behavioral sequences extracted from mouse behavioral sequences include: Construct a set of behavioral states, which includes movement, exploration, nurturing, and defensive immobility; Extract the transition relationships between adjacent behaviors from mouse behavior sequences and count the number of transitions between all behaviors; Construct a first-order behavior transition probability matrix based on the number of transitions; First-order transition entropy is calculated based on the first-order behavior transition probability matrix to characterize the transition patterns between behavior modes.

4. The PDVH behavior discrimination method based on machine learning according to claim 3, characterized in that: Constructing a first-order behavior transition probability matrix based on the number of transitions includes: Using the total number of transitions from the initial behavior as the denominator and the number of single transitions from the initial behavior to the other behavior as the numerator, we can calculate the first-order behavior transition probability matrix for any two behaviors, where the initial behavior and the other behavior are behaviors in the set of behavior states.

5. The PDVH behavior discrimination method based on machine learning according to claim 3, characterized in that: The first-order transition entropy is calculated based on the first-order behavior transition probability matrix using the following formula: ， in, Represents the first-order transition entropy. Indicates the initial behavior state. Indicates the initial behavioral state of the mouse Subsequent behavioral states that occur afterward Indicates the initial behavior The probability of occurrence Indicates from behavioral state Transition to behavioral state The conditional probability is calculated by determining the behavioral state. Transition to behavioral state The number of transitions divided by the behavioral state The total number of transfers is obtained. Indicates from behavioral state Transition to behavioral state The conditional probability.

6. The PDVH behavior discrimination method based on machine learning according to claim 1, characterized in that: The method further includes deleting mouse behavior sequences that correspond to simple motor impairment in PD mouse models after obtaining the mouse behavior sequences.

7. The PDVH behavior discrimination method based on machine learning according to claim 6, characterized in that: Key behavioral patterns associated with hallucinations extracted from mouse behavioral sequences include: Data on free movement were collected from the control group, PD mouse model, and PDVH mouse model, and the free movement data were divided into multiple basic behavioral types by unsupervised clustering. Compare the differences in the proportion of basic behavior types in the control group and PD mouse model. When the difference is less than the difference threshold, the corresponding basic behavior type is selected as the first intermediate behavior type. Correlation analysis was conducted between the proportion of the first intermediate behavior type and motor function indicators to select the behavior types related to motor function in Parkinson's disease. By comparing the differences in the proportion of basic behavioral types in PD mouse models and PDVH mouse models, the basic behavioral types with significant differences were selected as the second intermediate behavioral types. The second intermediate behavior type was removed from the behavior types related to motor function in Parkinson's disease and identified as the key behavior patterns related to hallucination.

8. The PDVH behavior discrimination method based on machine learning according to claim 1, characterized in that: Kinematic tests include mine field tests, pole climbing tests, grip strength tests, and swivel rod motion function tests.

9. The PDVH behavior discrimination method based on machine learning according to claim 1, characterized in that: The method further includes: segmenting the mouse behavior sequence according to the optimal time window, and plotting the behavior time curve of the proportion of the occurrence of key behavior patterns under each time window as a function of time, wherein the behavior time curve is used to quantify the degree of hallucination; wherein the optimal time window is determined based on the similarity between the behavior time curves.

10. A machine learning-based PDVH behavior discrimination system, used to implement the machine learning-based PDVH behavior discrimination method as described in any one of claims 1 to 9, characterized in that, include: The module is used to build PD mouse models and PDVH mouse models, and to obtain the mouse behavior sequences of PD mouse models and PDVH mouse models in behavioral tests, respectively. The extraction module is used to extract key behavioral patterns related to hallucinations and the transition patterns between these patterns from mouse behavioral sequences. The discrimination module is used to input key behavioral patterns and the transition patterns between them into a PDVH discrimination model built on machine learning for training and to determine whether PDVH behavior has occurred.

Images