Diagnosis and treatment integrated treatment effect prediction method and system based on artificial neural network

Abstract

The invention discloses a diagnosis and treatment integrated treatment effect prediction method and system based on an artificial neural network, and the method carries out the deep learning processing of a microbubble cavitation signal through an artificial neural network model, can accurately predict the treatment effect under different ultrasonic parameter combinations, can also adjust the ultrasonic parameters in real time according to the prediction result, and improves the accuracy of diagnosis and treatment. A treatment scheme is optimized, dependence on subjective experience of doctors is reduced, and scientificity and accuracy of prediction are greatly improved; meanwhile, by dynamically controlling the ultrasonic characteristics and the microbubble characteristics, the precision and the curative effect of focused ultrasound in the treatment process are improved, and the risk of damage to surrounding normal tissues is reduced to the maximum extent. The passive cavitation detection data and the self-learning ability of the ANN are utilized, and the system can continuously accumulate treatment data and automatically update and optimize model parameters, so that personalized treatment requirements of different patients are met, customized treatment is provided for the patients, and the safety and the treatment effect of the patients are enhanced.

Description

Technical Field

[0001] The present invention relates to the field of focused ultrasound technology, and in particular to a method and system for predicting treatment effects of integrated diagnosis and treatment based on an artificial neural network. Background Art

[0002] High-intensity focused ultrasound (HIFU) is a non-invasive treatment technology that has been widely used in the ablation treatment of tumors, stones and other diseased tissues. HIFU generates high energy at the target tissue by focusing sound waves, thereby achieving local temperature increase and tissue ablation. However, HIFU will induce a variety of physical effects during the propagation process, among which cavitation effect is one of the key mechanisms for its therapeutic effect.

[0003] The cavitation effect refers to the fact that under the action of ultrasound, tiny bubbles vibrate violently or even collapse in the sound field, resulting in local high temperature, pressure changes and micro-jet formation. These cavitation phenomena can directly destroy the cell membrane of the target tissue and enhance the therapeutic effect of high-intensity focused ultrasound therapy. At the same time, the signal generated by the cavitation effect contains key information such as microbubble vibration characteristics and sound field intensity, which can provide feedback for the treatment process.

[0004] However, since the acoustic cavitation process is complex and difficult to monitor accurately, how to make full use of the acoustic cavitation signal to accurately predict the effect of high-intensity focused ultrasound therapy remains a technical problem that needs to be solved urgently. Summary of the invention

[0005] In view of this, the purpose of the present invention is to provide a method and system for predicting treatment effects based on an artificial neural network for integrated diagnosis and treatment. The method determines the treatment effect by using an artificial neural network to predict the microbubble cavitation effect.

[0006] In order to achieve the above object, the present invention provides the following technical solutions:

[0007] The method for predicting the treatment effect of integrated diagnosis and treatment based on an artificial neural network provided by the present invention comprises the following steps:

[0008] S1: Determine the auxiliary variables in the neural network model;

[0009] S2: By means of passive cavitation detection method, the system is adjusted to adjust the ultrasonic excitation of microbubble cavitation and the cavitation signal is received by the transducer;

[0010] S3: performing signal processing on the cavitation signal;

[0011] S4: Preliminary construction of neural network model based on auxiliary variables and experimental results;

[0012] S5: training neural network to predict cavitation intensity;

[0013] S6: Compare the predicted value with the true value to verify the accuracy of the model.

[0014] Furthermore, the auxiliary variables include at least one of ultrasound pulse intensity, ultrasound pulse driving sound pressure, driving frequency, pulse duration, pulse repetition frequency, microbubble concentration and ultrasound contrast agent type.

[0015] Further, the passive cavitation detection method in step S2 is performed in the following manner:

[0016] S21: Acquiring a signal to generate a pulsed sinusoidal signal with adjustable voltage amplitude, pulse duration and pulse repetition frequency;

[0017] S22: After being amplified by the power amplifier, the focused transducer is driven to emit sound waves, and the ultrasonic waves are transmitted to the microbubble solution. The microbubbles undergo steady-state cavitation or transient cavitation under the stimulation of the sound waves, and the generated sound signals are transmitted to the surroundings;

[0018] S23: the acoustic signal is received by another receiving transducer, the acoustic signal is converted into an electrical signal through the inverse piezoelectric effect, and the cavitation signal is displayed on an oscilloscope;

[0019] S24: Save the data and perform cavitation signal processing.

[0020] Further, the signal processing of the cavitation signal in step S3 is performed in the following manner:

[0021] The collected cavitation signals are processed, the spectrum features are extracted using fast Fourier transform, and the time-frequency domain features are extracted using filters to determine the key features of microbubble cavitation behavior related to the therapeutic effect; the extracted key features include subharmonic and superharmonic components; based on the processed data, the cavitation behavior characteristics related to the therapeutic effect are identified.

[0022] Further, the signal processing of the cavitation signal in step S3 is performed in the following manner:

[0023] Collect the cavitation signal f(t) (t=0, 1, ···, N-1) collected within a pulse output time, where N represents the number of sampling points within a sampling period T;

[0024] Perform fast Fourier transform FFT on the cavitation signal to get F(f):

[0025]

[0026] Wherein, F(f) represents the Fourier transform frequency domain characteristic signal of the cavitation signal;

[0027] The amplitudes of superharmonics and higher harmonics are obtained according to the obtained frequency domain characteristic signals, and the superharmonics and higher harmonics are extracted through a bandpass comb peak filter; used to characterize the steady-state cavitation dose SCD;

[0028] Subharmonics, fundamental waves, superharmonics and higher harmonics are filtered out by a bandpass notch filter to obtain x'(t), and then according to the formula:

[0029]

[0030] The broadband noise RMS value is calculated to characterize the inertial cavitation dose ICD; where

[0031] Where, t represents the tth pulse;

[0032] U RMS represents the RMS value of broadband noise;

[0033] x'(i) represents the signal after comb filtering to remove subharmonics, fundamental waves, superharmonics and higher harmonics;

[0034] i represents the i-th data point;

[0035] F(f) represents the result after FFT of the original signal x(t);

[0036] x(t) represents the original signal.

[0037] Further, the neural network model in step S4 is constructed in the following manner;

[0038] Determine an input layer, the input layer includes a variety of parameters related to the ultrasonic cavitation process, and combine the parameters in the following manner as a characteristic input item X of the model: X = [Ps, f, PD, PRF, N, UCA]; wherein, the ultrasonic pulse driving sound pressure Ps, the driving frequency f, the pulse duration PD, the pulse repetition frequency PRF, the microbubble concentration N, and the microbubble type UCA;

[0039] The output variable Y is determined by the following combination: Y = [SCD, ICD]; where SCD and ICD are important indicators of the intensity of reaction cavitation;

[0040] Determine the mapping relationship between X and Y as: FX = Y; where F is a neural network model;

[0041] Construct a neural network model consisting of an input layer, a hidden layer, and an output layer as follows:

[0042] The input layer has M input neurons, and the output layer has N output neurons;

[0043] The input layer contains parameters x1,x2,...,xm Corresponding to the M input parameters, the parameters of the output layer are y1,y2,...,y n Corresponding to N output parameters; l,k,p represents the weight value between the pth input parameter in the input layer and the kth neuron in the lth layer; the output of the kth neuron in the lth layer is expressed as:

[0044]

[0045] in, is the output of the kth neuron in the lth layer;

[0046] is the activation function;

[0047] The pth input parameter is the input to the kth neuron in the lth layer.

[0048] Further, the neural network is trained and the cavitation intensity is predicted in the following manner in step S5:

[0049] Construct an initial data set and obtain model training samples through data enhancement;

[0050] Constructing a training set and a validation set, wherein the training set is used to train the ANN to obtain a prediction model, and the validation set is used to verify the accuracy of the prediction model;

[0051] The data were normalized according to the following formula:

[0052]

[0053] Among them, x,y and χ,y represent the data before and after normalization respectively, x min ,y min ,x max and max are the minimum and maximum values ​​in the sample data respectively;

[0054] After multiplying the weight ratio and adding the offset, an activation function is used. The activation function uses the Sigmoid function. The formula is as follows:

[0055]

[0056] The back propagation algorithm is used for training, and the data in the training set is used to train the network. During the training process, the connection weights between neurons are updated by minimizing the error back propagation. The formula is as follows:

[0057]

[0058] in,

[0059] s is the number of training times;

[0060] η is the training efficiency;

[0061] δ l,p is the learning error, which is determined by the following formula:

[0062]

[0063] is the output of the pth neuron in the lth layer;

[0064] K is the number of nodes in the hidden layer.

[0065] Further, the verification model in step S6 is performed in the following manner:

[0066] Use the validation set to evaluate the prediction performance of the model, adjust the model structure according to the validation results, and adjust the number of neural network layers, the number of neurons in each layer, the learning rate, etc. according to the validation results to ensure the generalization ability, efficiency and accuracy of the model;

[0067] The SCD and ICD after passive cavitation detection signal processing and analysis were compared with the SCD and ICD predicted by the neural network model. The loss function introduced relative error and root mean square error as evaluation indicators to verify the accuracy of the method. The relative error calculation formula is as follows:

[0068]

[0069] The root mean square error calculation formula is as follows:

[0070]

[0071] in, Represents the calculation result of the g-th sample;

[0072] Represents the experimental results of the g-th sample.

[0073] The diagnosis and treatment integrated treatment effect prediction system based on artificial neural network provided by the present invention includes a signal generator, a power amplifier, a focusing transducer, a receiving transducer, an oscilloscope, and a signal processing system;

[0074] The signal generator is used to generate a pulsed sinusoidal signal;

[0075] The power amplifier, the pulsed sinusoidal signal is input into the transmitting transducer after passing through the power amplifier;

[0076] The focusing transducer is used to generate and transmit sound waves to the microbubble solution;

[0077] The receiving transducer is used to receive the cavitation signal of the sound wave in the microbubble cavitation area;

[0078] The oscilloscope is used to display the cavitation signal;

[0079] The signal processing system is used to receive the cavitation signal and perform analysis and processing; the signal processing system is performed in the following manner:

[0080] The collected cavitation signals are processed, the spectral features are extracted using fast Fourier transform, and the time-frequency domain features are extracted using filters to determine the key information related to the microbubble cavitation behavior and the therapeutic effect; the extracted key information includes subharmonics, superharmonics, and broadband noise; based on the processed data, the cavitation behavior characteristics related to the therapeutic effect are identified.

[0081] Furthermore, the training samples of the neural network model are obtained by data enhancement.

[0082] The beneficial effects of the present invention are:

[0083] The present invention uses an artificial neural network (ANN) model to perform deep learning processing on microbubble cavitation signals, which can accurately predict the treatment effects under different ultrasound parameter combinations. It can also adjust the ultrasound parameters in real time according to the prediction results, optimize the treatment plan, reduce the dependence on the doctor's subjective experience, and greatly improve the scientificity and accuracy of the prediction; at the same time, by dynamically controlling the sound field intensity and microbubble dynamic behavior, the accuracy and efficacy of focused ultrasound in the treatment process are improved, and the risk of damage to surrounding normal tissues is minimized.

[0084] The present invention utilizes cavitation signal data and the self-learning ability of ANN. The system can continuously accumulate treatment data, automatically update and optimize model parameters, so as to adapt to the personalized treatment needs of different patients, provide customized treatment for patients, and enhance patient safety and treatment effect. Through direct prediction by the model, the disadvantage of not predicting the microbubble cavitation effect to judge the treatment effect is overcome, which greatly simplifies the evaluation process of the treatment effect, shortens the operation time of clinical application, and improves the diagnosis and treatment efficiency. The use of acoustic cavitation signals to predict the treatment effect of microbubble-mediated ultrasound therapy has important application prospects. This method is expected to improve the accuracy of treatment effect evaluation and provide scientific decision support for clinical practice.

[0085] Other advantages, objectives and features of the present invention will be described in the following description to some extent, and to some extent, will be obvious to those skilled in the art based on the following examination and study, or can be taught from the practice of the present invention. The objectives and other advantages of the present invention can be realized and obtained through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0086] In order to make the purpose, technical solution and beneficial effects of the present invention clearer, the present invention provides the following drawings for explanation.

[0087] Figure 1 This is a flow chart of the method for predicting treatment effect of integrated diagnosis and treatment based on artificial neural network.

[0088] Figure 2 This is the flow chart of the ultrasonically stimulated microbubble cavitation detection experiment.

[0089] Figure 3 This is the flow chart of passive cavitation detection signal processing.

[0090] Figure 4 This is the structure diagram of the neural network model. DETAILED DESCRIPTION

[0091] The present invention is further described below in conjunction with the accompanying drawings so that those skilled in the art can better understand the present invention and implement it.

[0092] like Figure 1 As shown, the method for predicting the treatment effect of integrated diagnosis and treatment based on artificial neural network provided in this embodiment includes the following steps:

[0093] S1: Determine auxiliary variables in the neural network model, wherein the auxiliary variables include: ultrasound pulse intensity, driving frequency, pulse duration, pulse repetition frequency, microbubble concentration, and ultrasound contrast agent type;

[0094] S2: By means of passive cavitation detection method, the system is adjusted to adjust the ultrasonic excitation of microbubble cavitation and the cavitation signal is received by the transducer;

[0095] S3: Process the cavitation signal on the PC side;

[0096] S4: Preliminary construction of neural network model based on auxiliary variables and experimental results;

[0097] S5: training neural network to predict cavitation intensity;

[0098] S6: Compare the predicted value with the true value to verify the accuracy of the model;

[0099] The specific implementation process of each step is described in detail below.

[0100] S1: Determine auxiliary variables in the neural network model, wherein the auxiliary variables include: ultrasonic pulse driving sound pressure, driving frequency, pulse duration, pulse repetition frequency, microbubble concentration and microbubble type;

[0101] The prediction effect of the neural network depends largely on the selection of auxiliary variables. These parameters can fully reflect the key factors that affect the experimental results of passive cavitation detection. To this end, I selected parameters related to ultrasonic pulse characteristics and microbubble characteristics to ensure that the input layer can cover the physical and chemical variables directly related to the cavitation effect. The combination of these parameters can fully capture the changes in microbubble dynamics and cavitation signals under experimental conditions, providing a solid foundation for the prediction results.

[0102] In this embodiment, the selection of auxiliary variables in the neural network model specifically includes:

[0103] First, in the process of passive cavitation detection, the ultrasonic pulse intensity directly determines the acoustic radiation force and pressure amplitude acting on the microbubbles, and is an important parameter affecting the microbubble oscillation behavior and cavitation effect. Higher pulse intensity may lead to the generation of inertial cavitation, while lower intensity may only induce steady-state cavitation. Therefore, the ultrasonic pulse driving sound pressure and frequency are selected as auxiliary variables for the establishment of the neural network prediction model.

[0104] Secondly, the pulse duration determines the duration of ultrasound action, affecting the stability and cumulative effect of microbubble oscillation. Long pulses may cause microbubbles to rupture and asymmetric oscillation. The pulse repetition frequency affects the oscillation state of microbubbles under continuous excitation and the cumulative degree of cavitation effect. At high repetition frequencies, microbubbles may not fully recover to their initial state. Therefore, the ultrasound pulse duration and repetition frequency are selected as auxiliary variables for the establishment of the neural network prediction model.

[0105] Finally, the concentration of microbubbles directly affects the intensity and characteristics of the cavitation signal. Higher concentrations may lead to enhanced interactions between microbubbles, thereby affecting the intensity of the cavitation signal. Different types of microbubbles have different inert gas and shell properties, which significantly affect the microbubble oscillation behavior and the performance of the cavitation signal. Therefore, microbubble concentration and microbubble type are selected as auxiliary variables for the establishment of the neural network prediction model.

[0106] In summary, based on the correlation between ultrasonic pulse characteristics and microbubble properties and cavitation signal intensity, ultrasonic pulse driving sound pressure, driving frequency, pulse duration, pulse repetition frequency, microbubble concentration and microbubble type were selected as auxiliary variables of the neural network prediction model.

[0107] S2: By means of a passive cavitation detection method, the system is adjusted to adjust the ultrasonic excitation microbubble cavitation, and the cavitation signal is received by a transducer;

[0108] like Figure 2 As shown, Figure 2 This is the experimental flow chart of passive cavitation detection. It mainly includes signal generator, power amplifier, focusing transducer, receiving transducer, oscilloscope, and signal processing system;

[0109] The signal generator is used to generate a pulsed sinusoidal signal;

[0110] The power amplifier, the pulsed sinusoidal signal is input into the transmitting transducer after passing through the power amplifier;

[0111] The focusing transducer is used to generate and transmit sound waves to the microbubble solution;

[0112] The receiving transducer is used to receive the cavitation signal of the sound wave in the microbubble cavitation area;

[0113] The oscilloscope is used to display the cavitation signal;

[0114] The signal processing system is used to receive cavitation signals and perform analysis and processing; the signal processing system is provided with a neural network model for processing and analyzing cavitation signals, and the input auxiliary variables of the neural network model include at least any one of ultrasonic pulse intensity, ultrasonic pulse driving sound pressure, driving frequency, pulse duration, pulse repetition frequency, microbubble concentration and ultrasonic contrast agent type.

[0115] The training samples of the neural network model are obtained by data enhancement. The signal processing system in this embodiment adopts a PC-side signal processing system.

[0116] S21: firstly, obtaining and using the signal to generate a pulsed sinusoidal signal with adjustable voltage amplitude, pulse duration and pulse repetition frequency;

[0117] S22: After being amplified by the power amplifier, the focused transducer is driven to emit sound waves, and the ultrasonic waves are transmitted to the microbubble solution. The microbubbles undergo steady-state cavitation or transient cavitation under the excitation of the sound waves, and the generated sound signals are transmitted to the surroundings.

[0118] S23: These signals can be received by another receiving transducer, which converts the acoustic signals into electrical signals through the inverse piezoelectric effect and displays the cavitation signals on an oscilloscope.

[0119] S24: Save the data to the PC via USB3.0 and use Matlab to process the cavitation signal.

[0120] S3: Process the cavitation signal on the PC side;

[0121] like Figure 3 As shown, Figure 3The figure is a flow chart of passive cavitation detection signal processing. The collected cavitation signals are processed using MATLAB software, and the fast Fourier transform (FFT) is applied to extract the spectrum features. The filter is used to extract the time-frequency domain features to determine the key features related to the microbubble cavitation behavior and the treatment effect; the extracted key features include subharmonics, superharmonic components, etc. Based on the processed data, the cavitation behavior features related to the treatment effect are identified.

[0122] The cavitation signal f(t) (t=0, 1, ···, N-1) (N: the number of sampling points in a sampling period T) collected within a pulse output time is first subjected to a fast Fourier transform (FFT):

[0123]

[0124] Get F(f), then find the amplitude of superharmonic and high-order harmonic according to the result, and extract superharmonic and high-order harmonic through bandpass comb peak filter to characterize steady-state cavitation dose (SCD);

[0125] Secondly, the subharmonics, fundamental wave, superharmonics and higher harmonics are filtered out by a bandpass notch filter to obtain x'(t), and then according to the formula:

[0126]

[0127] The broadband noise RMS value is calculated to characterize the inertial cavitation dose (ICD); where

[0128] t represents the tth pulse;

[0129] U RMS represents the RMS value of broadband noise;

[0130] x'(i) represents the signal after comb filtering to remove subharmonics, fundamental waves, superharmonics and higher harmonics;

[0131] i represents the i-th data point;

[0132] F(f) represents the result after FFT of the original signal x(t);

[0133] x(t) represents the original signal;

[0134] S4: Preliminary construction of neural network model based on auxiliary variables and experimental results;

[0135] A multi-layer artificial neural network (ANN) model is established, and the model adopts a multi-layer artificial neural network structure. The input layer of the artificial neural network (ANN) model provided in this embodiment includes a variety of parameters related to the ultrasonic cavitation process, and these parameters serve as feature input items of the model:

[0136] Ultrasonic pulse driving sound pressure Ps, driving frequency f, pulse duration PD, pulse repetition frequency PRF, microbubble concentration N, and microbubble type UCA are used as input variables:

[0137] X=[Ps,f,PD,PRF,N,UCA]

[0138] SCD and ICD are important indicators of reaction cavitation intensity. Taking them as output variables, they are expressed as:

[0139] Y=[SCD,ICD]

[0140] The mapping relationship between X and Y can be expressed as:

[0141] FX=Y

[0142] Specific structure such as Figure 4 As shown:

[0143] The network consists of an input layer, a hidden layer, and an output layer. The input layer has M input neurons, and the output layer has N output neurons.

[0144] The input layer contains parameters x1,x2,...,x m Corresponding to the M input parameters, the parameters of the output layer are y1,y2,...,y n Corresponding to N output parameters.

[0145] χ l,k,ρ represents the weight value between the pth input parameter in the input layer and the kth neuron in the lth layer. The output of the kth neuron in the lth layer can be expressed as:

[0146]

[0147] in,

[0148] is the output of the kth neuron in the lth layer;

[0149] is the activation function;

[0150] The input parameter p is the input to the k-th neuron in the l-th layer;

[0151] S5: training neural network to predict cavitation intensity;

[0152] This embodiment constructs an initial data set based on experimental data, and enriches the model training samples through data enhancement (such as time transformation, amplitude scaling, etc.). The network weights are optimized, and more samples are generated through data enhancement technology to enrich the training set.

[0153] The first seventy groups of data construct an Excel table to construct the training set, and the last thirty groups of data construct an Excel table to construct the validation set. The training set is used to train the ANN to obtain the prediction model, and the validation set is used to verify the accuracy of the prediction model.

[0154] First, normalize the data, the formula is as follows:

[0155]

[0156] Among them, x,y and χ,y represent the data before and after normalization respectively, x min ,y min ,x max and max are the minimum and maximum values ​​in the sample data respectively;

[0157] After multiplying the weight ratio and adding the offset, an activation function is used. The activation function uses the Sigmoid function. The formula is as follows:

[0158]

[0159] The back propagation algorithm is used for training, and the experimental results in the table are used to train the network. During the training process, the connection weights between neurons are updated by minimizing the error back propagation. The formula is as follows:

[0160]

[0161] in,

[0162] s is the number of training times;

[0163] η is the training efficiency;

[0164] δ l,ρ is the learning error, which is determined by the following formula:

[0165]

[0166] is the output of the pth neuron in the lth layer;

[0167] K is the number of nodes in the hidden layer;

[0168] S6: Compare the predicted value with the true value to verify the accuracy of the model;

[0169] Use the validation set to evaluate the prediction performance of the model, adjust the model structure according to the validation results, and adjust the number of neural network layers, the number of neurons in each layer, the learning rate, etc. according to the validation results to ensure the generalization ability, efficiency and accuracy of the model;

[0170] The prediction method provided in this embodiment obtains a signal through an experiment, and then processes the signal to obtain two types of values, SCD and ICD. At the same time, the obtained two types of values, SCD and ICD, are put into a neural network to train the neural network model.

[0171] Specifically, the SCD and ICD after passive cavitation detection signal processing and analysis are compared with the SCD and ICD after the neural network model prediction. The loss function introduces relative error and root mean square error as evaluation indicators to verify the accuracy of the method. The relative error calculation formula is as follows:

[0172]

[0173] The root mean square error calculation formula is as follows:

[0174]

[0175] in, Represents the calculation result of the g-th sample;

[0176] Represents the experimental results of the g-th sample.

[0177] The above-described embodiments are only preferred embodiments for fully illustrating the present invention, and the protection scope of the present invention is not limited thereto. Equivalent substitutions or changes made by those skilled in the art based on the present invention are within the protection scope of the present invention. The protection scope of the present invention shall be subject to the claims.

Claims

1. A method and system for predicting the effect of integrated diagnosis and treatment based on artificial neural network, characterized by: The following steps are involved: S1: Determine the auxiliary variables in the neural network model; S2: By means of passive cavitation detection method, the system is adjusted to adjust the ultrasonic excitation of microbubble cavitation and the cavitation signal is received by the transducer; S3: performing signal processing on the cavitation signal; S4: Preliminary construction of neural network model based on auxiliary variables and experimental results; S5: training neural network to predict cavitation intensity; S6: Compare the predicted value with the true value to verify the accuracy of the model.

2. The method for predicting the treatment effect of integrated diagnosis and treatment based on artificial neural network according to claim 1, characterized in that: The auxiliary variables include at least one of ultrasound pulse intensity, ultrasound pulse driving sound pressure, driving frequency, pulse duration, pulse repetition frequency, microbubble concentration and ultrasound contrast agent type.

3. The method for predicting the treatment effect of integrated diagnosis and treatment based on artificial neural network according to claim 1, characterized in that: The passive cavitation detection method in step S2 is performed in the following manner: S21: Acquiring a signal to generate a pulsed sinusoidal signal with adjustable voltage amplitude, pulse duration and pulse repetition frequency; S22: After being amplified by the power amplifier, the focused transducer is driven to emit sound waves, and the ultrasonic waves are transmitted to the microbubble solution. The microbubbles undergo steady-state cavitation or transient cavitation under the stimulation of the sound waves, and the generated sound signals are transmitted to the surroundings; S23: the acoustic signal is received by another receiving transducer, the acoustic signal is converted into an electrical signal through the inverse piezoelectric effect, and the cavitation signal is displayed on an oscilloscope; S24: Save the data and perform cavitation signal processing.

4. The method for predicting the treatment effect of integrated diagnosis and treatment based on artificial neural network according to claim 1, characterized in that: The signal processing of the cavitation signal in step S3 is performed in the following manner: The collected cavitation signals are processed, the spectrum features are extracted using fast Fourier transform, and the time-frequency domain features are extracted using filters to determine the key information related to the microbubble cavitation behavior and the therapeutic effect; the extracted key information includes subharmonic and superharmonic components; based on the processed data, the cavitation behavior characteristics related to the therapeutic effect are identified.

5. The method for predicting the treatment effect of integrated diagnosis and treatment based on artificial neural network according to claim 1, characterized in that: The signal processing of the cavitation signal in step S3 is performed in the following manner: Collect the cavitation signal f(t) (t=0, 1, ···, N-1) collected within a pulse output time, where N represents the number of sampling points within a sampling period T; Perform fast Fourier transform (FFT) on the cavitation signal to obtain F(f): Wherein, F(f) represents the Fourier transform frequency domain characteristic signal of the cavitation signal; The amplitudes of superharmonics and higher harmonics are obtained according to the obtained frequency domain characteristic signals, and the superharmonics and higher harmonics are extracted through a bandpass comb peak filter; used to characterize the steady-state cavitation dose SCD; Subharmonics, fundamental waves, superharmonics and higher harmonics are filtered out by a bandpass notch filter to obtain x'(t), and then according to the formula: The broadband noise RMS value is calculated to characterize the inertial cavitation dose ICD; where Where, t represents the tth pulse; U RMS represents the RMS value of broadband noise; x'(i) represents the signal after comb filtering to remove subharmonics, fundamental waves, superharmonics and higher harmonics; i represents the i-th data point; F(f) represents the result after FFT of the original signal x(t); x(t) represents the original signal.

6. The method for predicting the treatment effect of integrated diagnosis and treatment based on artificial neural network according to claim 1, characterized in that: The neural network model in step S4 is constructed in the following manner; Determine an input layer, the input layer includes a variety of parameters related to the ultrasonic cavitation process, and combine the parameters in the following manner as a characteristic input item X of the model: X = [Ps, f, PD, PRF, N, UCA]; wherein, the ultrasonic pulse driving sound pressure Ps, the driving frequency f, the pulse duration PD, the pulse repetition frequency PRF, the microbubble concentration N, and the microbubble type UCA; The output variable Y is determined by the following combination: Y = [SCD, ICD]; where SCD and ICD are important indicators of the intensity of reaction cavitation; Determine the mapping relationship between X and Y as: FX = Y; where F is the neural network prediction model; Construct a neural network model consisting of an input layer, a hidden layer, and an output layer as follows: The input layer has M input neurons, and the output layer has N output neurons; The input layer contains parameters x1,x2,...,x m Corresponding to the M input parameters, the parameters of the output layer are y1,y2,...,y n Corresponding to N output parameters; l,k,o represents the weight value between the pth input parameter in the input layer and the kth neuron in the lth layer; the output of the kth neuron in the lth layer is expressed as: in, is the output of the kth neuron in the lth layer; is the activation function; The pth input parameter is the input to the kth neuron in the lth layer.

7. The method for predicting the treatment effect of integrated diagnosis and treatment based on artificial neural network according to claim 1, characterized in that: The neural network is trained and the cavitation intensity is predicted in the following manner in step S5: Construct an initial data set and obtain model training samples through data enhancement; Constructing a training set and a validation set, wherein the training set is used to train the ANN to obtain a prediction model, and the validation set is used to verify the accuracy of the prediction model; The data were normalized according to the following formula: Among them, x,y and χ,y represent the data before and after normalization respectively, x min ,y min ,x max and max are the minimum and maximum values ​​in the sample data respectively; After multiplying the weight ratio and adding the offset, an activation function is passed. The activation function uses the Sigmoid function. The formula is as follows: The back propagation algorithm is used for training, and the data in the training set is used to train the network. During the training process, the connection weights between neurons are updated by minimizing the error back propagation. The formula is as follows: in, s is the number of training times; η is the training efficiency; δ l,p is the learning error, which is determined by the following formula: is the output of the pth neuron in the lth layer; K is the number of nodes in the hidden layer.

8. The method for predicting the treatment effect of integrated diagnosis and treatment based on artificial neural network according to claim 1, characterized in that: The verification model in step S6 is performed in the following manner: Use the validation set to evaluate the prediction performance of the model, adjust the model structure according to the validation results, and adjust the number of neural network layers, the number of neurons in each layer, the learning rate, etc. according to the validation results to ensure the generalization ability, efficiency and accuracy of the model; The SCD and ICD after passive cavitation detection signal processing and analysis were compared with the SCD and ICD predicted by the neural network model. The loss function introduced relative error and root mean square error as evaluation indicators to verify the accuracy of the method. The relative error calculation formula is as follows: The root mean square error calculation formula is as follows: in, Represents the calculation result of the g-th sample; Represents the experimental results of the g-th sample.

9. The diagnosis and treatment integrated treatment effect prediction system based on artificial neural network is characterized by: Including signal generator, power amplifier, focusing transducer, receiving transducer, oscilloscope, signal processing system; The signal generator is used to generate a pulsed sinusoidal signal; The power amplifier, the pulsed sinusoidal signal is input into the transmitting transducer after passing through the power amplifier; The focusing transducer is used to generate and transmit sound waves to the microbubble solution; The receiving transducer is used to receive the cavitation signal of the sound wave in the microbubble cavitation area; The oscilloscope is used to display the cavitation signal; The signal processing system is used to receive the cavitation signal and perform analysis and processing; the signal processing system is performed in the following manner: The collected cavitation signals are processed, the spectral features are extracted using fast Fourier transform, and the time-frequency domain features are extracted using filters to determine the key information related to the microbubble cavitation behavior and the therapeutic effect; the extracted key information includes subharmonics, superharmonics, and broadband noise; based on the processed data, the cavitation behavior characteristics related to the therapeutic effect are identified.

10. The diagnosis and treatment integrated treatment effect prediction system based on artificial neural network according to claim 9, characterized in that: The training samples of the neural network model are obtained by data enhancement.

Images