Odor decoding method based on olfactory EEG signals

Abstract

This invention discloses an odor decoding method based on olfactory electroencephalogram (EEG) signals, comprising the following steps: inputting a training set into a random forest model for training to obtain a trained random forest model; inputting the sample to be tested into the trained random forest model and outputting the classification result; each sample is a dimension-reduced feature matrix corresponding to a target characteristic gas. The method for obtaining the dimension-reduced feature matrix includes: depriving a rat model of water for 12-15 hours; stimulating the rat model with a gas containing a target characteristic gas; and synchronously acquiring the local field potential signal of the rat model based on an array electrode within 5 seconds after stimulation; and obtaining the dimension-reduced feature matrix based on the local field potential signal. The odor decoding method based on olfactory EEG signals of this invention has the advantages of high specificity, high sensitivity, strong real-time performance, and portability, with a response time ≤1s and a decoding accuracy of up to 94%.

Description

Technical Field

[0001] This invention belongs to the field of hazardous chemical detection technology, specifically relating to an odor decoding method based on olfactory electroencephalogram (EEG) signals. Background Technology

[0002] In recent years, explosions have occurred frequently. Most flammable and explosive hazardous chemicals are highly volatile, and the characteristic gases they release are key monitoring indicators for explosion risk early warning. Existing gas detection methods are mainly divided into manual sensory detection and physicochemical instrument detection (including traditional electronic noses). Compared with bio-electronic noses, the above methods generally suffer from problems such as susceptibility to environmental interference, high cost, insufficient portability and flexibility, and difficulty in achieving the sensitivity and specificity of natural biological olfaction.

[0003] Bioelectronic noses utilize olfactory receptors, olfactory cells, olfactory tissues, or live animals as sensitive sensing elements, exhibiting excellent sensitivity and specificity. The rat olfactory system demonstrates outstanding performance, with liquid-phase odor molecule recognition sensitivity reaching the pM~nM level. It can distinguish a vast number of odor molecules and integrate multi-component odors to form complex olfaction, making it an ideal model for developing a live bioelectronic nose. The rat olfactory pathway involves the olfactory epithelium converting odor signals into electrochemical signals and transmitting them to the olfactory bulb. After encoding by the olfactory bulb, these signals are transmitted to the olfactory cortex via two types of projection neurons: mitral cells and plexiform cells, ultimately forming olfactory perception. Brain-computer interfaces can collect olfactory-evoked brain neural electrical signals, completing the conversion of neural signals into odor information. Currently, there is no integrated technology combining olfactory receptor engineering, brain-computer signal acquisition, in vivo synchronous regulation, and rapid neural signal decoding applied in the field of hazardous chemical gas detection. Summary of the Invention

[0004] In view of the shortcomings of the existing technology, the purpose of this invention is to provide an odor decoding method based on olfactory electroencephalogram (EEG) signals.

[0005] The objective of this invention is achieved through the following technical solution.

[0006] A method for decoding odors based on olfactory electroencephalogram (EEG) signals includes the following steps:

[0007] Step 1: Combine multiple samples into a training set and input them into the random forest model for training to obtain a trained random forest model. Each sample is a dimensionality-reduced feature matrix corresponding to a class of target feature gases. The samples in the training set correspond to a total of M classes of target feature gases.

[0008] Step 2: Input the sample to be tested into the trained random forest model, and the trained random forest model outputs the classification result;

[0009] The method for obtaining the dimension-reduced feature matrix includes: subjecting the rat model to water deprivation for 12-15 hours, stimulating the rat model with a gas containing a target characteristic gas, and synchronously acquiring the local field potential signal of the rat model based on an array electrode within 5 seconds after stimulation. The array electrode contains C acquisition channels, where C is greater than 4. The power spectral density of the local field potential signal in the β band is extracted, and the power spectral density of the β band is constructed into a power spectral density matrix with dimension C×F, where F is the number of frequency sampling points in the β band. The PCA algorithm is used to reduce the dimension of the power spectral density matrix to 4×F to obtain the dimension-reduced feature matrix.

[0010] The method for constructing a rat model includes: using rats overexpressing Olr1356 in the olfactory epithelium as modified rats; acclimating these rats for at least one week; after acclimation, withholding water for 12-15 hours while continuing normal feeding; and subjecting them to paradigm training for N consecutive days until they learn the behavioral paradigm, thus obtaining the rat model. Specifically, the modified rats were deprived of water for 12-15 hours before each day's paradigm training, and the daily training time was 1-2 hours. Paradigm training was conducted within an open-top ring-shaped structure: the modified rats were placed inside the ring, with two points on the bottom surface of the ring designated as the first and second positions, the straight-line distance between the first and second positions being 1-1.5 meters. The training consists of a cyclical reward method. A single reward method includes: allowing the modified rat to freely explore within a circular structure; when the rat approaches a first location, providing air stimulation for 5-10 seconds at that location; then providing 1-2 mL of water as the first reward at the first location; then allowing the rat to continue exploring freely within the circular structure; and when the rat approaches a second location, providing 1-2 mL of water as the second reward at the second location. If the rat stays at the first location for less than 5 seconds or does not approach the second location after the first reward, the reward method is considered a failure; otherwise, the reward method is considered a success.

[0011] Let X be the success rate of the reward method in the daily paradigm training. If X is greater than 85% for three consecutive days of paradigm training, then the modified rats are judged to have learned the behavioral paradigm.

[0012] In the above technical solution, a rat model is stimulated within an annular structure using a gas containing a target characteristic gas. The stimulation method includes: placing the rat model within the annular structure and allowing it to explore freely; when the rat model approaches a first position, providing the rat model with a gas containing the target characteristic gas for 2-5 seconds at the first position, followed by providing the rat model with clean air for 3-5 seconds to remove odor residue; then providing the rat model with 1-2 mL of water as reward A at the first position; and then allowing the rat model to continue exploring freely within the annular structure; when the rat model approaches a second position, providing the rat model with 1-2 mL of water as reward B at the second position; if the rat model stays at the first position for less than 5 seconds or does not approach the second position after reward A, the stimulation of the rat model is considered a failure.

[0013] In the above technical solution, the annular component is a cuboid with an open top surface. A first through hole and a third through hole are formed on one side wall of the annular component, and a second through hole is formed on the side wall of the annular component opposite to the first through hole. The first through hole and the third through hole are close to the first position, and the second through hole is close to the second position. A first pipe is connected to the outside of the first through hole for supplying gas to the modified rat / rat model at the first position. A third pipe is connected to the outside of the third through hole for supplying water to the modified rat / rat model at the first position. A second pipe is connected to the outside of the second through hole for supplying water to the modified rat / rat model at the second position.

[0014] In the above technical solution, the array electrode is a microelectrode array.

[0015] In the above technical solution, the microelectrode array is implanted in the olfactory bulb of the right brain of the rat model.

[0016] The beneficial effects of this invention are as follows:

[0017] This invention is the first to employ bioengineering modification combined with brain-computer interface technology to achieve odor classification and recognition: by using adeno-associated virus transfection technology to overexpress the specific olfactory receptor Olr3 in the olfactory mucosa of rats, the olfactory nerve perception system is engineered to achieve specific enhancement. Based on the olfactory perception brain-computer interface, animal behavior training is conducted on rats to train them to form a stable behavioral pattern, effectively enhancing the modified rats' ability to recognize and orient themselves to target odors, reducing EEG interference caused by irrelevant actions, emotions, and attention fluctuations in the modified rats, improving the data stability and signal-to-noise ratio of the input EEG signals for decoding analysis, and thus improving the classification accuracy and reliability of odor decoding.

[0018] The odor decoding method based on olfactory EEG signals of this invention has the advantages of high specificity, high sensitivity, strong real-time performance, and portability. The response time is ≤1s and the decoding accuracy can reach up to 94%. It can be used for rapid detection of hazardous chemicals in scenarios such as public safety and chemical production, and has high practical value. Attached Figure Description

[0019] Figure 1 This is a schematic diagram of the ring-shaped component.

[0020] Figure 2 This is a schematic diagram of the gas cylinder structure;

[0021] Figure 3 The average trajectory length for a single trajectory test method modified for rats and rat models;

[0022] Figure 4 The movement trajectories of a single modified rat and a single rat model in the annular structure;

[0023] Figure 5 The accuracy of the prediction sets for Example 6 and Comparative Example 1 when T takes different values;

[0024] Figure 6 To predict the accuracy of the prediction results for samples corresponding to nitrotoluene, dinitrotoluene, and trinitrotoluene in the prediction set, where (a) represents nitrotoluene, (b) represents trinitrotoluene, and (c) represents dinitrotoluene.

[0025] Wherein, 1: ring-shaped component, 1-1: first through hole, 1-2: second through hole, 1-3: third through hole, 2: gas cylinder, 2-1: air inlet pipe, 2-2: air outlet pipe. Detailed Implementation

[0026] The technical solution of the present invention will be described in detail below with reference to the accompanying drawings and embodiments.

[0027] Example 1

[0028] A method for decoding odors based on olfactory electroencephalogram (EEG) signals includes the following steps:

[0029] Step 1: Multiple samples are combined into a training set and input into a random forest model (Breiman, L. (2001). Random Forests. Machine Learning, 45(1), 5-32.) for training (the target feature gas type corresponding to each sample in the training set is taken as the ground truth) to obtain a trained random forest model, where each sample is a dimensionality-reduced feature matrix corresponding to a type of target feature gas, and the samples in the training set correspond to a total of M types of target feature gases;

[0030] Step 2: Input the sample to be tested into the trained random forest model. The trained random forest model outputs the classification result (the classification result is an array of dimension 1*M, which consists of M probability values. The target feature gas category corresponding to the largest probability value in the classification result is used as the prediction result for the sample to be tested).

[0031] The method for obtaining the dimension-reduced feature matrix includes: depriving the rat model of water for 12-15 hours, stimulating the rat model with a gas containing a target characteristic gas, and synchronously acquiring the local field potential signal (LFP signal, where the horizontal axis of the LFP signal is time and the vertical axis is amplitude, and the acquisition frequency of the local field potential signal is 20kHz) based on the array electrode within 5 seconds after stimulation. The array electrode contains C acquisition channels, where C is greater than 4 (in this embodiment, C=18). The power spectral density of the local field potential signal in the β band is extracted. The frequency band ranges from 15 to 30 Hz. The formula for calculating the power spectral density is: Redwan SM, Uddin MP, Ulhaq A, et al. Power spectral density-based restingstate EEG classification of first-episode psychosis [J]. Scientific Reports, 2024, 14(1): 1-12. The power spectral density of the β band is constructed into a power spectral density matrix of dimension C×F (each element in the power spectral density matrix is ​​a power spectral density value), where F is the number of frequency sampling points in the β band. The power spectral density matrix is ​​reduced to 4×F using the PCA algorithm (for details of the PCA algorithm, see: Pearson, K. (1901). On Lines and Planes of Closest Fit to Systems of Points in Space. Philosophical Magazine Series 6, 2(11), 559-572), and the dimension-reduced feature matrix is ​​obtained.

[0032] The method for constructing a rat model includes: using rats overexpressing Olr1356 in the olfactory epithelium as modified rats, acclimating them for one week (during which the modified rats are petted daily to reduce their stress response), depriving them of water for 12 hours after the acclimation period, and feeding them normally (to enhance their motivation for water rewards), and then subjecting them to paradigm training for N consecutive days until they learn the behavioral paradigm (in this embodiment, N=8) to obtain the rat model. Before each paradigm training session, the modified rats are deprived of water for 12 hours. The daily paradigm training lasts for one hour and is conducted within an open-top ring-shaped component 1: the modified rats are placed within the ring-shaped component 1, with two points on the bottom surface of the ring-shaped component 1 designated as the first and second positions, and the straight-line distance between the first and second positions being 1 meter. The paradigm training consists of cyclical reward methods. A single reward method includes: allowing the modified rat to freely explore within a ring-shaped component 1; when the modified rat approaches a first position, providing air stimulation to the modified rat at the first position for 5 seconds; then providing the modified rat with 1.2 mL of water at the first position as the first reward; then allowing the modified rat to continue freely exploring within the ring-shaped component 1; when the modified rat approaches a second position, providing the modified rat with 1.2 mL of water at the second position as the second reward; if the modified rat stays at the first position for less than 5 seconds or does not approach the second position after the first reward, the reward method is considered a failure (if a single reward method fails, the next reward method will be started again after 10 seconds); otherwise, the reward method is considered a success.

[0033] Let X be the success rate of the reward method in daily paradigm training. If X is greater than 85% for three consecutive days of paradigm training, then the modified rat is considered to have learned the behavioral paradigm.

[0034] Example 2

[0035] An odor decoding method based on olfactory EEG signals, in addition to the method described in Example 1, involves stimulating a rat model within an annular component 1 with a gas containing a target characteristic gas. The stimulation method includes: placing the rat model within the annular component 1 and allowing it to explore freely; when the rat model approaches a first position, providing the rat model with a gas containing the target characteristic gas for 5 seconds at the first position, followed by providing the rat model with clean air for 3 seconds to remove odor residue; then providing the rat model with 1.2 mL of water as reward A at the first position; and then allowing the rat model to continue exploring freely within the annular component 1; when the rat model approaches a second position, providing the rat model with 1.2 mL of water as reward B at the second position; if the rat model stays at the first position for less than 5 seconds or does not approach the second position after reward A, the stimulation of the rat model is considered a failure (if stimulating the rat model with a gas containing the target characteristic gas fails, the next stimulation of the rat model with a gas containing the target characteristic gas is initiated after 10 seconds).

[0036] Example 3

[0037] An odor decoding method based on olfactory EEG signals, in addition to the method described in Example 2, includes an annular component 1 as follows: Figure 1 As shown, the annular component 1 is a cuboid with an open top (the base of the cuboid is a square with a side length of 100cm, and the height of the cuboid is 50cm). An activity area is formed inside the annular component 1, which allows the modified rat / rat model to move and sniff freely. A first through hole 1-1 (square, 3cm×3cm) and a third through hole 1-3 (square, 3cm×3cm) are formed on one side wall of the annular component 1. A second through hole 1-2 (square, 3cm×3cm) is formed on the side wall of the annular component 1 opposite to the first through hole. The first and third through holes are close to the first position ( (Not shown in the figure), the second through hole is close to the second position (not shown in the figure), the first through hole is connected to the first pipe (not shown in the figure) for supplying gas (air or gas containing the target characteristic gas) to the modified rat / rat model at the first position, the third through hole is connected to the third pipe (not shown in the figure) for supplying water to the modified rat / rat model at the first position, the second through hole is connected to the second pipe for supplying water to the modified rat / rat model at the second position, and the center of the first through hole, the center of the second through hole and the center of the third through hole are 5cm away from the bottom surface of the annular member 1 respectively;

[0038] The array electrode is a microelectrode array (manufacturer: Microprobes for Life Science, model: CustomMWA). The microelectrode array is a 21-channel tetrade electrode array, of which 18 are recording electrodes for data acquisition, and the remaining 3 are reference electrodes. Each tetrade electrode is made of four platinum-iridium alloy microfilaments (composed of 90 wt% Pt and 10 wt% Ir), and each tetrade electrode has a diameter of 34 mm. Furthermore, the Tetrode electrode is coated with a polyimide insulating coating, and its tip is platinum-plated, reducing the impedance to 0.3. the following.

[0039] The microelectrode array was implanted in the right olfactory bulb of the rat model's brain (locating to: 8.0 mm posterior to the anterior fontanelle and 1.0 mm lateral to the right of the midline). The initial implantation depth of the microelectrode array was 312 mm. To ensure the stability and reliability of the acquired local field potential signals, the implantation depth of the microelectrode array needs to be adjusted. After adjustment, the microelectrode array is implanted in the rat model and can be used for subsequent acquisition of local field potential signals. The adjustment steps are as follows: After implanting the microelectrode array into the right olfactory bulb of the rat model, the microelectrode array is inserted daily at a preset step size of 31.25 mm. The implantation depth of the microelectrode array was lowered, and the Spike signal acquired by the microelectrode array was recorded. When the target neuron's electrical activity signal characteristics (i.e., Spike discharge) were detected in the Spike signal, the implantation depth of the microelectrode array was no longer lowered, and the implantation of the microelectrode array in the rat model was completed.

[0040] Example 4

[0041] An odor decoding method based on olfactory EEG signals, in Example 3, the process of obtaining the modified rats includes: anesthetizing the rats (male adult Long-Evans rats, weighing approximately 400-600g and aged 6-10 months), placing the rats horizontally with their abdomens facing upwards, and diluting the recombinant virus to the target titer using physiological saline (the target titer is...). (vg / ml), drop the solution into the olfactory epithelium on both sides of the rat's nasal cavity using a pipette. The diluted recombinant virus was left to stand for 15 minutes to allow it to fully contact the olfactory epithelium of the rats and complete the delivery of the recombinant virus. The rats were then fed for 21 days to obtain the modified rats. The recombinant virus was pAAV-CMV-EGFP-2A-Olr1356-3FLAG, which was purchased from Shanghai Heyuan Biotechnology (Group) Co., Ltd.

[0042] Example 5

[0043] Based on Example 4, 4800 dimensionality-reduced feature matrices were obtained as a dataset: Four rat models were prepared, and each rat model obtained multiple dimensionality-reduced feature matrices daily according to the method for obtaining dimensionality-reduced feature matrices in Example 1. Each rat model underwent stimulation for a total of 2-3 hours daily. The dimensionality-reduced feature matrices corresponded to the local field potential signal at T seconds after stimulation. After every 5 stimulation cycles, the rat models rested for 5 minutes to reduce the impact of fatigue on the experimental results (the experimental environment needed to be kept quiet throughout the process). All dimensionality-reduced feature matrices obtained daily by each rat model corresponded to M-type target characteristic gases. The stimulation order of the M-type target characteristic gases was randomized.

[0044] Prepare a gaseous medium containing a target characteristic gas. The gaseous medium can be pure water vapor or a mixed vapor. When it is a mixed vapor, it is a mixture of water vapor (carrier gas) and the vapor of a dopant. The dopant is nitrotoluene (NT), dinitrotoluene (DNT), trinitrotoluene (TNT), citral, or octanoic acid. That is, the target characteristic gas is nitrotoluene (NT) vapor, dinitrotoluene (DNT) vapor, trinitrotoluene (TNT) vapor, citral vapor, octanoic acid vapor, or water vapor. Nitrotoluene (NT) vapor, dinitrotoluene (DNT) vapor, and trinitrotoluene (TNT) vapor correspond to hazardous chemicals, while citral vapor, octanoic acid vapor, and water vapor correspond to interfering odors. M=6.

[0045] The specific settings are as follows: Figure 2 The gas cylinder shown is as follows: Gas cylinder 2 contains liquid and is a sealed cylinder. The top end of an inlet pipe 2-1 is located outside the cylinder, and the bottom end of the inlet pipe extends into the liquid inside the cylinder. The top end of an outlet pipe 2-2 is located outside the cylinder, and the top end of the outlet pipe 2-2 is connected to the first pipe. The bottom end of the outlet pipe extends into the cylinder and is located on the surface of the liquid inside. The liquid inside the cylinder is pure water or a mixture. When it is a mixture, the mixture consists of water and a dopant, such as nitrotoluene (NT), dinitrotoluene (DNT), trinitrotoluene (TNT), citral, or octanoic acid. The concentration of the dopant in the mixture is X mol / L. Air is introduced into the cylinder through the inlet pipe at a flow rate of 5 mL / s. After air is introduced through the inlet pipe, the air bubbles through the liquid inside the cylinder, and a large amount of water vapor and the vapor of the dopant are discharged from the outlet pipe and enter the first pipe. X = , , or Each rat model was stimulated at least 50 times under each X and each type of target characteristic gas, so that the rat model could obtain 50 dimension-reduced feature matrices under each X and each type of target characteristic gas (the dimension-reduced feature matrix corresponds to the local field potential signal at T s after stimulation).

[0046] The accuracy of prediction results based on local field potential signals at different times after stimulation was investigated, with T = 0.2, 0.4, 0.6, 0.8, 1, 1.2, 1.4, 1.6, 1.8 or 2s. Each selected value of parameter T corresponds to a dataset containing 4800 samples; that is, under a single T condition, a total of 4800 dimensionality-reduced feature matrices were obtained from the four rat models, which constitute the dataset corresponding to that T.

[0047] Example 6

[0048] An odor decoding method based on olfactory EEG signals, building upon Example 5, uses 70% of the dataset from Example 5 as the training set and the remaining 30% as the prediction set when a value for T is selected. The training set is divided into five parts, and 5-fold cross-validation is used to optimize the number of decision trees. The maximum growth depth of the decision trees is not artificially limited, allowing them to grow naturally to their inherent convergence state, resulting in a trained random forest model. The prediction set is input into the trained random forest model, which outputs classification results. The accuracy of the prediction set is then determined based on the classification results. In this odor decoding method, the classification result is not controlled by X but is determined solely by the target feature gas category.

[0049] Comparative Example 1

[0050] An odor decoding method based on olfactory EEG signals is basically the same as the odor decoding method based on olfactory EEG signals in Example 6, except that the modified rats in this comparative example are replaced with Long-Evans rats.

[0051] The accuracy of the prediction sets of Example 6 and Comparative Example 1 when T takes different values ​​is as follows: Figure 5 As shown, the accuracy of Example 6 can reach up to 80.0±5.2% (the accuracy of a single rat model can reach up to 94.0%); the accuracy of Comparative Example 1 can reach up to 65.0±4.6% (the accuracy of a single rat model can reach up to 78.0%). The accuracy of odor classification in Example 6 is significantly higher than that in Comparative Example 1.

[0052] When T=2s, samples corresponding to nitrotoluene (NT), dinitrotoluene (DNT), or trinitrotoluene (TNT) in the prediction set are selected (i.e., the target characteristic gas is nitrotoluene (NT), dinitrotoluene (DNT), or trinitrotoluene (TNT)), and the accuracy of the prediction results at different X values ​​is calculated. The accuracy of the prediction results for samples corresponding to nitrotoluene (NT) in Example 6 and Comparative Example 1 is as follows: Figure 6 As shown in (a), the accuracy of the prediction results for the samples corresponding to trinitrotoluene (TNT) in the prediction set of Example 6 and Comparative Example 1 is as follows: Figure 6As shown in (b), the accuracy of the prediction results for the samples corresponding to dinitrotoluene (DNT) in the prediction set of Example 6 and Comparative Example 1 is as follows: Figure 6 As shown in (c), it can be seen that when the value of X is in the range of mol / L to At concentrations within mol / L, the average accuracy of odor classification in Example 6 was higher than that in Comparative Example 1.

[0053] Rats were deprived of water for 12 hours but fed normally. They were then allowed to cycle through an annular structure 1 20 times using the following trajectory test method. Each trajectory test consisted of the following steps: the rat was allowed to explore freely within the annular structure 1; when the rat approached the first position, 1.2 mL of water was provided through the third opening; the rat continued to explore freely within the annular structure 1; when the rat approached the second position, 1.2 mL of water was provided through the second opening; the rat continued to explore freely within the annular structure 1 again; when the rat approached the first position again, 1.2 mL of water was provided through the third opening, ending the single trajectory test. Rats rested for 2 minutes after every 10 trajectory test cycles. During each trajectory testing, a high-definition camera was installed above the annular component 1 to continuously capture video of the rat's movement within the annular component 1. A light-emitting marker (LED light) was placed on the rat's head. By identifying the position of the marker in the continuous video frames, the rat's real-time position coordinates within the annular component 1 were extracted. Based on these real-time position coordinates, the rat's trajectory was reconstructed. The rat was either a modified rat or a rat model. The average trajectory length of a single trajectory test using the rat model is as follows: Figure 3 As shown in the "After Learning the Behavioral Paradigm", the average trajectory length of the modified single trajectory test method for rats is as follows: Figure 3 As shown in the "Learned Behavioral Paradigm". According to Figure 3 It can be seen that the average trajectory length of the modified rats before learning the behavioral paradigm was 403±60.9cm, while the average trajectory length of the rat model after learning the behavioral paradigm was 254±15.8cm, indicating that the rats' activities were more purposeful after learning the behavioral paradigm.

[0054] Taking a single rat as an example, the modified rat before learning the behavioral paradigm exhibited strong randomness in its movement trajectory within the circular component 1, such as... Figure 4 As shown in (a), the rats' roaming paths were scattered, with frequent turns and unstable movement directions, showing no obvious tendency to approach the target odor source area, indicating that they had not yet established a stable behavioral pattern related to the target odor. After learning the behavioral paradigm, the rat model's behavioral trajectory changed significantly; its movement path gradually shifted from a random distribution to directional movement towards the target odor source area, showing a stronger tendency towards the target odor source area, such as... Figure 4As shown in (b), the pathways to the target odor source area were more concentrated, the ineffective walking distance was reduced, and the dwelling and approach behaviors towards the target odor and reward areas increased. These results indicate that the behavioral paradigm can effectively enhance the modified rats' ability to recognize and orient themselves to target odors, transforming them from an initial random exploration state to a stable, goal-oriented behavioral pattern; at the same time, it can effectively improve the modified rats' ability to track odor sources using olfactory cues.

[0055] The present invention has been described above by way of example. It should be noted that any simple modifications, alterations or other equivalent substitutions that can be made by those skilled in the art without creative effort without departing from the core of the present invention fall within the protection scope of the present invention.

Claims

1. A method for decoding odors based on olfactory electroencephalogram (EEG) signals, characterized in that, Includes the following steps: Step 1: Combine multiple samples into a training set and input them into the random forest model for training to obtain a trained random forest model. Each sample is a dimensionality-reduced feature matrix corresponding to a class of target feature gases. The samples in the training set correspond to a total of M classes of target feature gases. Step 2: Input the sample to be tested into the trained random forest model, and the trained random forest model outputs the classification result; The method for obtaining the dimension-reduced feature matrix includes: subjecting the rat model to water deprivation for 12-15 hours, stimulating the rat model with a gas containing a target characteristic gas, and synchronously acquiring the local field potential signal of the rat model based on an array electrode within 5 seconds after stimulation. The array electrode contains C acquisition channels, where C is greater than 4. The power spectral density of the local field potential signal in the β band is extracted, and the power spectral density of the β band is constructed into a power spectral density matrix with dimension C×F, where F is the number of frequency sampling points in the β band. The PCA algorithm is used to reduce the dimension of the power spectral density matrix to 4×F to obtain the dimension-reduced feature matrix. The method for constructing a rat model includes: using rats overexpressing Olr1356 in the olfactory epithelium as modified rats; acclimating these rats for at least one week; after acclimation, withholding water for 12-15 hours while continuing normal feeding; and subjecting them to paradigm training for N consecutive days until they learn the behavioral paradigm, thus obtaining the rat model. Specifically, the modified rats were deprived of water for 12-15 hours before each day's paradigm training, and the daily training time was 1-2 hours. Paradigm training was conducted within an open-top ring-shaped structure: the modified rats were placed inside the ring, with two points on the bottom surface of the ring designated as the first and second positions, the straight-line distance between the first and second positions being 1-1.5 meters. The training consists of a cyclical reward method. A single reward method includes: allowing the modified rat to freely explore within a circular structure; when the rat approaches a first location, providing air stimulation for 5-10 seconds at that location; then providing 1-2 mL of water as the first reward at the first location; then allowing the rat to continue exploring freely within the circular structure; and when the rat approaches a second location, providing 1-2 mL of water as the second reward at the second location. If the rat stays at the first location for less than 5 seconds or does not approach the second location after the first reward, the reward method is considered a failure; otherwise, the reward method is considered a success. Let X be the success rate of the reward method in the daily paradigm training. If X is greater than 85% for three consecutive days of paradigm training, then the modified rats are judged to have learned the behavioral paradigm.

2. The odor decoding method based on olfactory EEG signals according to claim 1, characterized in that, The rat model was stimulated within a ring-shaped structure using a gas containing a target characteristic gas. The stimulation method included: placing the rat model within the ring-shaped structure and allowing it to explore freely; when the rat model approached a first position, providing the rat model with the target characteristic gas for 2-5 seconds at the first position, followed by providing the rat model with clean air for 3-5 seconds; then providing the rat model with 1-2 mL of water as reward A at the first position; and then allowing the rat model to continue exploring freely within the ring-shaped structure; when the rat model approached a second position, providing the rat model with 1-2 mL of water as reward B at the second position; if the rat model stayed at the first position for less than 5 seconds or did not approach the second position after reward A, the stimulation of the rat model was considered a failure.

3. The odor decoding method based on olfactory EEG signals according to claim 2, characterized in that, The annular component is a cuboid with an open top surface. A first through hole and a third through hole are formed on one side wall of the annular component, and a second through hole is formed on the side wall opposite to the first through hole. The first and third through holes are close to a first position, and the second through hole is close to a second position. A first pipe is connected to the outside of the first through hole for supplying gas to the modified rat / rat model in the first position. A third pipe is connected to the outside of the third through hole for supplying water to the modified rat / rat model in the first position. A second pipe is connected to the outside of the second through hole for supplying water to the modified rat / rat model in the second position.

4. The odor decoding method based on olfactory EEG signals according to claim 3, characterized in that, The array electrode is a microelectrode array.

5. The odor decoding method based on olfactory EEG signals according to claim 4, characterized in that, The microelectrode array was implanted in the right olfactory bulb of the rat model's brain.

Images