Intracellular electric signal non-invasive reconstruction method and system based on artificial neural network

Abstract

The invention discloses an intracellular electric signal non-invasive reconstruction method and system based on an artificial neural network, and the method comprises the steps: obtaining an extracellular signal, inputting the preprocessed extracellular signal into a trained artificial neural network, and predicting to obtain an intracellular electric signal; the artificial neural network comprises a convolutional neural network and a long-short-term memory network, and the artificial neural network is trained by the following method: acquiring an extracellular sample signal and a paired intracellular sample signal, classifying the extracellular sample signal and the intracellular sample signal, taking the extracellular sample signals and the intracellular sample signals meeting preset classification requirements as a sample set; and training a preset model according to the preprocessed sample set, updating parameters of the preset model according to a preset loss function until preset requirements are met, and obtaining a trained artificial neural network. The embodiment of the invention can stably record the action potential signal in the cell for a long time in a non-invasive manner, and can be widely applied to the technical field of signal detection.

Description

Technical Field

[0001] The present invention relates to the field of signal detection technology, and in particular to a method and system for non-invasive reconstruction of intracellular electrical signals based on artificial neural networks. Background Art

[0002] Recording and analyzing cardiomyocyte action potentials (APs) with single-cell resolution can provide researchers with a more precise understanding of cellular signaling pathways, the signal transduction processes of cellular activity, and the molecular mechanisms of drug-mediated ion channel regulation. Patch clamping is the most commonly used method for detecting intracellular signaling. Patch clamping typically uses a sharp glass pipette with an integrated metal wire as the recording electrode. Negative pressure is applied through the pipette to rupture the cell membrane, allowing for recording of APs from the cytoplasm. In recent years, the combination of microelectrode arrays (MEAs) and cell poration techniques has led to significant progress in high-throughput intracellular recordings from cardiomyocytes. Cell membrane porations generally involve mechanical forces, chemical modification, photoporation, and electroporation. For example, nanowire- or nanotube-based field-effect transistor devices have been used to functionalize fused cell membranes with lipid layers for intracellular recordings. Vertical nanoelectrodes combined with plasmonic photoporation have successfully recorded intracellular APs from mammalian neurons and cardiomyocytes. The combination of MEAs and micro / nano electroporation has been demonstrated to be an effective and stable cell poration technique. Electric pulses <5V applied through the microelectrode generate localized electric fields and induce nanoscale pores in the cell membrane. After electroporation, the microelectrode enters the intracellular environment and is able to record intracellular APs. The main advantage of MEA-electroporation (MEA-EP) is that high-throughput recording of intracellular signals in cell networks is feasible due to the multichannel nature of MEAs.

[0003] For existing patch clamps, cells usually die within a few hours after intracellular recording. Due to the invasiveness of cell membrane perforation, the intracellular recording time of perforation-coupled MEA is quite short (usually <5 minutes). Long-term rupture of the cell membrane may lead to abnormal ion leakage and cytotoxic effects, resulting in changes in intrinsic cellular electrical signals. In addition, after membrane perforation, the cell membrane tends to heal dynamically as part of its self-protection mechanism to prevent toxicity associated with long-term perforation. For example, after electroporation through MEA, intracellular recordings usually only last for a few minutes, and then the signal disappears and reverts to extracellular AP. However, due to the tendency of the perforated cell membrane to heal, these recordings are usually unstable and it is almost impossible to maintain a longer time scale. Summary of the Invention

[0004] In view of this, an object of the embodiments of the present invention is to provide a method and system for non-invasive reconstruction of intracellular electrical signals based on an artificial neural network, which can record intracellular action potential signals in a non-invasive and long-term stable manner.

[0005] In one aspect, an embodiment of the present invention provides a method for non-invasive reconstruction of intracellular electrical signals based on an artificial neural network, comprising:

[0006] Acquire extracellular signals, input the preprocessed extracellular signals into a trained artificial neural network, and predict intracellular electrical signals; the artificial neural network includes a convolutional neural network and a long short-term memory network, and the artificial neural network is trained by the following method:

[0007] Acquiring extracellular sample signals and paired intracellular sample signals, classifying the extracellular sample signals and the intracellular sample signals, and taking the extracellular sample signals and the intracellular sample signals that meet preset classification requirements as a sample set;

[0008] The preset model is trained according to the preprocessed sample set, and the parameters of the preset model are updated according to the preset loss function until the preset requirements are met to obtain a trained artificial neural network.

[0009] Optionally, the extracellular sample signal and the paired intracellular sample signal are obtained by the following method:

[0010] For the extracellular measurement signal, the time point of the trough in each cycle is used as the first reference origin, a first number of data points are extracted before the first reference origin, and a second number of data points are extracted after the first reference origin to obtain the extracellular sample signal;

[0011] For the measurement signal inside the cell, the time point of the peak in each cycle is used as the second reference origin, a first number of data points are extracted before the second reference origin, and a second number of data points are extracted after the second reference origin to obtain the paired intracellular sample signal.

[0012] Optionally, the preset loss function is determined by the following method:

[0013] determining a first loss parameter according to a difference in amplitude between the predicted value of the extracellular sample signal and the paired intracellular sample signal;

[0014] determining a second loss parameter according to a difference in action potential duration between the predicted value of the extracellular sample signal and the paired intracellular sample signal;

[0015] determining a third loss parameter according to a depolarization time difference between the predicted value of the extracellular sample signal and the paired intracellular sample signal;

[0016] determining a fourth loss parameter according to a difference in repolarization time between the predicted value of the extracellular sample signal and the paired intracellular sample signal;

[0017] A preset loss function is determined according to the first loss parameter, the second loss parameter, the third loss parameter, and the fourth loss parameter.

[0018] Optionally, the method further includes:

[0019] Data augmentation is performed on the sample set based on any one or more of amplitude change, signal fusion, or resampling.

[0020] Optionally, performing data expansion on the sample set based on the amplitude change includes:

[0021] The amplitudes of the extracellular sample signals and the intracellular sample signals in the sample set are compressed or stretched according to a preset ratio to perform data expansion on the sample set.

[0022] Optionally, data expansion is performed on the sample set based on signal fusion, including:

[0023] The extracellular sample signals and the intracellular sample signals of different source domains in the sample set are fused according to a first preset range and a first preset step size to perform data expansion on the sample set.

[0024] Optionally, performing data expansion on the sample set based on resampling includes:

[0025] The extracellular sample signals and the intracellular sample signals in the sample set are used to generate samples of different time scales according to a second preset range and a second preset step size, so as to perform data expansion on the sample set.

[0026] On the other hand, an embodiment of the present invention provides a non-invasive reconstruction system for intracellular electrical signals based on an artificial neural network, comprising:

[0027] The prediction module is used to obtain extracellular signals, input the preprocessed extracellular signals into a trained artificial neural network, and predict the intracellular electrical signals; the artificial neural network includes a convolutional neural network and a long short-term memory network, and the artificial neural network is trained by the following method:

[0028] Acquiring an extracellular sample signal and a paired intracellular sample signal, classifying the extracellular sample signal and the intracellular sample signal, and taking the extracellular sample signal and the intracellular sample signal that meet preset classification requirements as a sample set;

[0029] The preset model is trained according to the preprocessed sample set, and the parameters of the preset model are updated according to the preset loss function until the preset requirements are met, thereby determining a trained artificial neural network.

[0030] On the other hand, an embodiment of the present invention provides a non-invasive reconstruction system for intracellular electrical signals based on an artificial neural network, comprising:

[0031] at least one processor;

[0032] at least one memory for storing at least one program;

[0033] When the at least one program is executed by the at least one processor, the at least one processor implements the above method.

[0034] On the other hand, an embodiment of the present invention provides a non-invasive reconstruction system of intracellular electrical signals based on an artificial neural network, comprising an extracellular signal acquisition device and a computer device connected to the extracellular signal acquisition device; wherein,

[0035] The extracellular signal acquisition device is used to measure the extracellular signals of single cells;

[0036] The computer device comprises:

[0037] at least one processor;

[0038] at least one memory for storing at least one program;

[0039] When the at least one program is executed by the at least one processor, the at least one processor implements the above method.

[0040] The implementation of the embodiments of the present invention includes the following beneficial effects: in this embodiment, the sample set is quality screened by classification, and the preset model is trained on the preprocessed sample set, which can improve the learning and optimization efficiency of the model; the local key features in the extracellular signal are captured by the convolutional neural network in the artificial neural network, and the long-term time dependence of the extracellular signal is captured by the long short-term memory network in the artificial neural network. Through the gradual learning of the sequence data, the potential relationship between the extracellular signal and the intracellular signal is determined by the artificial neural network, and it has the ability to deduce the intracellular signal based on the extracellular signal, thereby realizing the long-term stable recording of the intracellular action potential signal in a non-invasive manner. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 This is a schematic flow chart of the steps of a method for non-invasive reconstruction of intracellular electrical signals based on an artificial neural network provided by an embodiment of the present invention;

[0042] Figure 2 This is a schematic diagram of a workflow for non-invasive reconstruction of intracellular electrical signals based on an artificial neural network, provided by an embodiment of the present invention;

[0043] Figure 3 This is a calculation flow chart based on an artificial neural network model provided by an embodiment of the present invention;

[0044] Figure 4This is a sample set collected in different batches provided by an embodiment of the present invention;

[0045] Figure 5 A schematic diagram of the results of a non-invasive reconstruction method of intracellular electrical signals based on an artificial neural network provided by an embodiment of the present invention;

[0046] Figure 6 A schematic diagram of the results of another non-invasive reconstruction method of intracellular electrical signals based on an artificial neural network provided by an embodiment of the present invention;

[0047] Figure 7 A schematic diagram of the results of another non-invasive reconstruction method of intracellular electrical signals based on an artificial neural network provided by an embodiment of the present invention;

[0048] Figure 8 A schematic diagram of the results of another non-invasive reconstruction method of intracellular electrical signals based on an artificial neural network provided by an embodiment of the present invention;

[0049] Figure 9 This is a structural block diagram of a non-invasive reconstruction system of intracellular electrical signals based on an artificial neural network provided by an embodiment of the present invention;

[0050] Figure 10 This is another structural block diagram of a non-invasive reconstruction system of intracellular electrical signals based on an artificial neural network provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0051] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. The step numbers in the following embodiments are provided for ease of description only and do not limit the order of the steps. The order of execution of the steps in the embodiments can be adaptively adjusted based on the understanding of those skilled in the art.

[0052] It should be noted that, although the functional modules are divided in the device schematic diagram and the logical order is shown in the flow chart, in some cases, the steps shown or described can be performed in a different order than the module division in the device or the order in the flow chart. The terms "first", "second", etc. in the specification and claims and the above-mentioned drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or precedence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0053] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the art to which this application pertains. The terms used herein are for the purpose of describing the embodiments of this application only and are not intended to limit this application.

[0054] Recording of extracellular and intracellular signaling datasets based on MEA (microelectrode array) electroporation.

[0055] In a specific embodiment, a 32-channel MEA chip is fabricated based on a photolithography process and a magnetron sputtering process. The size of each electrode is 20 microns to ensure coupling between the electrode and the cardiomyocytes. The MEA chip is connected to an electrophysiological signal acquisition system via a printed circuit board and a pin header. The electrophysiological signal acquisition system applies 3V, 20Hz, 200μs square wave electric pulses to the cardiomyocytes through a perforation module for electroporation, so that the electrodes contact the intracellular environment of the cardiomyocytes. The electrophysiological signal acquisition system includes a signal processing module, a customized data collector, and a graphical user interface (computer software) for data visualization. The extracellular and intracellular signals before and after perforation are amplified, filtered, and stored. During the 2-12 days of culture of multiple MEA chips, various treatments (chemical stimulation and electrical stimulation) were performed on the cardiomyocytes and then the extracellular and intracellular signals were recorded once a day based on MEA electroporation.

[0056] See Figure 1 The embodiment of the present invention provides a method for non-invasive reconstruction of intracellular electrical signals based on an artificial neural network, comprising:

[0057] S100 , obtaining extracellular sample signals and paired intracellular sample signals, classifying the extracellular sample signals and the intracellular sample signals, and taking the extracellular sample signals and the intracellular sample signals that meet preset classification requirements as a sample set.

[0058] It should be noted that the standards, methods, and number of categories for classifying extracellular and intracellular sample signals are determined according to actual applications and are not specifically limited in this embodiment. The standards for pre-set classification requirements are determined according to actual applications and are not specifically limited in this embodiment.

[0059] In one specific embodiment, the quality of all extracellular and intracellular sample signals was classified into four categories: A, B, C, and D, based on SNR (Signal-to-Noise Ratio) and waveform distortion. To further analyze and classify signal quality, a convolutional neural network (CNN)-based model was proposed to automatically extract multi-scale temporal features from the signals, enabling efficient and accurate signal quality classification. Finally, a fully connected layer with four output nodes generated a classification distribution, and the CNN classification model classified the signal quality into four categories: A, B, C, and D. To improve the effectiveness of model training, extracellular signals with "A" and "B" quality were used and paired with intracellular signals with "A" quality to form a sample dataset. This strategy effectively reduced the noise introduced by signal quality variations, thereby improving the model's generalization ability and prediction accuracy, as well as the signal-to-noise ratio. While this screening process may reduce the yield of intracellular signals from the multi-channel MEA system, it significantly improved the overall quality of the final sample dataset. Before training, all extracellular and intracellular sample signals were normalized in amplitude to ensure consistency and comparability of signal characteristics between different samples.

[0060] S200. Train the preset model according to the preprocessed sample set, and update the parameters of the preset model according to the preset loss function until the preset requirements are met to obtain a trained artificial neural network; the artificial neural network includes a convolutional neural network and a long short-term memory network.

[0061] It should be noted that the preset loss function and preset requirements are determined according to actual applications and are not specifically limited in this embodiment.

[0062] The artificial neural network combines a CNN (convolutional neural network) and an LSTM (long short-term memory network). The model captures key local features and temporal dependencies in extracellular electrical signals, thereby improving prediction accuracy. To improve model fit during training, enhance generalization, and reduce training time, the data was downsampled from 10,000 to 6,000 points. Subsequently, the cellular electrical signal data was normalized to eliminate amplitude differences that could affect model training, allowing the model to learn and optimize more effectively.

[0063] Use CNN network to extract local features in the signal. CNN structure consists of multiple convolutional layers and pooling layers. The output of the convolutional layer is represented by the following equation:

[0064]

[0065] Among them, y i,j is the output of the convolutional layer, f is the activation function (such as ReLU), xi,j is the input signal, w m,n is the weight of the convolution kernel, and b is the bias. By applying multiple convolution kernels, the model learns the key waveform features of the signal. After processing through the pooling layer, the model further reduces the feature size, usually using max pooling, and the output is as follows:

[0066]

[0067] In this model, after two convolution and pooling operations, the dimensionality of the feature map is gradually reduced from the original signal dimension, resulting in a more compact and information-rich feature representation. After extracting features through the CNN layer, the processed signal data is passed to the LSTM layer. The LSTM layer uses three main gating mechanisms: forget gate, input gate, and output gate. The calculation of the forget gate is expressed as:

[0068] f t =σ(W f ·[h t-1 ,x t ]+b f )

[0069] The input gate is calculated as follows:

[0070] i t =σ(W i ·[h t-1 ,x t ]+b i )

[0071] The output gate is calculated as follows:

[0072] o t =σ(W o ·[h t-1 ,x t ]+b o )

[0073] Among them, W f 、W i and W o is the weight matrix corresponding to each gate, b f 、b i and b o is the bias, σ is the sigmoid activation function. The update of the cell state in LSTM is given by:

[0074]

[0075] The output is calculated as follows:

[0076] h t =o t tanh(C t )

[0077] In one specific embodiment, the number of hidden units in the LSTM layer is set to 256, with three stacked layers, giving the network a powerful ability to capture complex temporal patterns. The input signal at each time step is processed using information from both the current time step and historical data, enhancing the ability to identify underlying patterns in the signal. This is crucial for analyzing the relationship between extracellular and intracellular signals.

[0078] S300 , obtaining extracellular signals, inputting the pre-processed extracellular signals into a trained artificial neural network, and predicting intracellular electrical signals.

[0079] It should be noted that the preprocessing methods of extracellular signals and extracellular sample signals are the same.

[0080] Initialize the artificial neural network model, allowing it to load the trained weights file. Extracellular signals from a control group (normal culture) over multiple days (data extraction method identical to that used for the pretraining and experimental groups) are normalized. The processed extracellular signals are fed into the model to predict intracellular signals. The predicted depolarization time, repolarization time, and APD10-90 (action potential duration) are calculated.

[0081] Optionally, the extracellular sample signal and the paired intracellular sample signal are obtained by the following method:

[0082] S010. For the extracellular measurement signal, take the time point of the trough in each cycle as the first reference origin, extract a first number of data points before the first reference origin, and extract a second number of data points after the first reference origin to obtain an extracellular sample signal;

[0083] S020. For the measurement signal inside the cell, the time point of the peak in each cycle is used as the second reference origin, a first number of data points are extracted before the second reference origin, and a second number of data points are extracted after the second reference origin to obtain paired intracellular sample signals.

[0084] It should be noted that the first quantity and the second quantity are determined according to actual applications and are not specifically limited in this embodiment.

[0085] First, separate the single extracellular and intracellular signals. The extracellular signals recorded before MEA-electroporation and the stable intracellular signals (recorded for >10 seconds after electroporation) were extracted. The duration of one cycle of the extracellular sample signal is defined as Pe, and for the intracellular signal, it is defined as Pi. In most experiments, due to the safety of the electroporation process, the cycle remains essentially unchanged after MEA-electroporation, so Pe≈Pi is usually the case. For the extracellular sample signal, the time point of the trough in each signal cycle is used as the first reference origin. Starting from this first reference origin, Pe / 4 of the signal is taken before the trough, and 3Pe / 4 is taken after the trough. Taking into account the variation in peak width between different batches, in order to ensure the consistency of data length and capture the true time characteristics, 2500 data points before the trough and 7500 data points after the trough are further extracted, and the data is used as an extracellular signal. Similarly, for the intracellular sample signal, the peak value in each signal cycle is used as the second reference point. From this second reference point, Pi / 4 of the signal precedes the peak value and 3Pi / 4 follows it. An additional 2500 data points are extracted before the peak value and 7500 data points after the peak value to obtain the complete intracellular sample signal. The extracellular and intracellular sample signals are paired and output as sample data.

[0086] Optionally, the preset loss function is determined by the following method:

[0087] S030, determining a first loss parameter according to the amplitude difference between the predicted value of the extracellular sample signal and the paired intracellular sample signal;

[0088] S040, determining a second loss parameter according to a difference in action potential duration between a predicted value of the extracellular sample signal and a paired intracellular sample signal;

[0089] S050, determining a third loss parameter according to a depolarization time difference between the predicted value of the extracellular sample signal and the paired intracellular sample signal;

[0090] S060, determining a fourth loss parameter according to a repolarization time difference between the predicted value of the extracellular sample signal and the paired intracellular sample signal;

[0091] S070. Determine a preset loss function based on the first loss parameter, the second loss parameter, the third loss parameter, and the fourth loss parameter.

[0092] It should be noted that the first loss parameter, the second loss parameter, the third loss parameter and the fourth loss parameter are determined according to actual applications and are not specifically limited in this embodiment.

[0093] As LSTM gradually learns from time series data, it identifies the underlying patterns in the signal, providing rich temporal features. To improve the interpretability and applicability of the model, additional loss function metrics are designed. In this model, a comprehensive loss function is designed to optimize the performance of the CNN-LSTM network, better capture the characteristics of cellular electrical signals and promote model learning. The loss function is designed to take into account the error between the predicted value and the true value, as well as biologically relevant parameters. First, the standard mean square error (bias) loss is used to calculate the difference between the predicted signal and the true signal, as follows:

[0094]

[0095] Among them, y i is the true signal value, is the signal value predicted by the model, and n is the number of samples. Bias is effective in capturing the overall prediction accuracy of the signal. In addition to bias, electrophysiological-related loss terms are also introduced, particularly for calculating APD50 and pacemaker points. These metrics are expressed by the following formulas:

[0096] APD50=t 50 -t start

[0097] Among them, t 50 is the time point when the signal reaches 50% of its peak value, t start is the starting point of the signal. This metric reflects the duration of the cell's electrical signal, which is biologically meaningful.

[0098] The loss function of APD50 is defined as:

[0099]

[0100] Among them, t APD50 The time point representing the actual APD50, represents the time point of APD50 predicted by the model.

[0101] For the depolarization and repolarization times, the loss function is defined as:

[0102] L depolar =|t depolar -t avg_depolar |

[0103] L repolar =|t repolar -t avg_repolar |

[0104] Among them, t depolar is the actual depolarization time, t avg_depolar is the depolarization time predicted by the model, trepolar is the actual repolarization time, t avg_repolar is the repolarization time predicted by the model. The comprehensive loss function is expressed as:

[0105] L=λ1·MSE+λ2·L APD50 +λ3·L repolar +λ4·L depolar

[0106] Among them, λ1, λ2, λ3, and λ4 are coefficients that balance the importance of each loss term. This design allows the model to not only optimize prediction accuracy but also capture biologically important parameters, thereby improving the biological interpretability of the model.

[0107] By jointly optimizing these loss functions, the model comprehensively learns the temporal characteristics and biological significance of the signal, thereby enhancing the understanding of the relationship between extracellular and intracellular sample signals. The combination of CNN and LSTM in this model improves the accuracy of signal prediction and lays a solid foundation for subsequent biological analysis.

[0108] Optionally, the non-invasive reconstruction method of intracellular electrical signals based on artificial neural networks further includes:

[0109] Perform data augmentation on the sample set based on any one or more of amplitude change, signal fusion, or resampling.

[0110] To address the discrepancies between different chip data, a calibration method based on transfer learning was proposed. Initially, data augmentation techniques were employed to generate a more diverse set of target domain samples, alleviating the problem of insufficient target domain data. Specifically, these augmentation methods include, but are not limited to, three methods: amplitude variation, signal fusion, and resampling.

[0111] Optionally, data augmentation is performed on the sample set based on amplitude changes, including:

[0112] The amplitudes of the extracellular sample signals and the intracellular sample signals in the sample set are compressed or stretched according to a preset ratio to perform data expansion on the sample set.

[0113] It should be noted that the preset magnification is determined based on actual application and is not specifically limited in this embodiment. In a specific embodiment, the amplitude variation method adjusts the overall amplitude of the signal (e.g., from 0.8 times to 1.5 times the original amplitude) to simulate changes under different experimental conditions, generating 50 samples to enhance the model's adaptability to amplitude fluctuations.

[0114] Optionally, based on signal fusion, data expansion is performed on the sample set, including:

[0115] The extracellular sample signals and intracellular sample signals of different source domains in the sample set are fused according to a first preset range and a first preset step size to perform data expansion on the sample set.

[0116] It should be noted that the first preset range and the first preset step size are determined according to actual applications and are not specifically limited in this embodiment.

[0117] In a specific embodiment, the signal fusion method combines samples from two source domains in different proportions (ranging from 0.1 to 0.9, with a step size of 0.01, generating 81 samples) to create new samples with diversified features. These new samples combine the stability of the source domain and the personalized features of the target domain, thereby improving the generalization ability of the model.

[0118] Optionally, data augmentation is performed on the sample set based on resampling, including:

[0119] The extracellular sample signals and the intracellular sample signals in the sample set are used to generate samples of different time scales according to a second preset range and a second preset step size, so as to perform data expansion on the sample set.

[0120] It should be noted that the second preset range and the second preset step size are determined according to actual applications and are not specifically limited in this embodiment.

[0121] In a specific embodiment, the resampling method adjusts the signal length (ranging from 8000 to 24000, with a step size of 100, generating 101 samples) to produce samples of different time scales, helping the model capture the temporal characteristics of the target domain signal and improving its adaptability to rhythm changes or inconsistent sampling rates.

[0122] These augmented samples constitute a new transfer learning training set, providing more diverse samples for subsequent transfer learning. Given that the signal characteristics of the new chip differ from those of the old chip, directly applying the source-domain model could result in significant bias. Therefore, transfer learning is employed to effectively adapt the model from the source domain to the target domain. Specifically, the model is fine-tuned using augmented target-domain data to ensure it adapts to the data distribution of the new chip. The CNN component of this model primarily extracts spatial features, while the LSTM component models the temporal relationships of the signal. During transfer learning, the first convolutional layer of the CNN is frozen to preserve its essential feature extraction capabilities learned from the source domain, while the weights of subsequent convolutional layers and LSTM components are unfrozen to adapt the model to target-domain samples. This strategy, by locking in low-level feature extraction capabilities, reduces the risk of overfitting due to insufficient target-domain samples while enhancing the model's sensitivity to target-domain signal characteristics. By combining data augmentation and transfer learning, prediction error on new chip data is successfully reduced, cross-domain signal prediction performance is improved, and the effectiveness and applicability of this calibration strategy are demonstrated.

[0123] The implementation of the embodiments of the present invention includes the following beneficial effects: in this embodiment, the sample set is quality screened by classification, and the preset model is trained on the preprocessed sample set, which can improve the learning and optimization efficiency of the model; the local key features in the extracellular signal are captured by the convolutional neural network in the artificial neural network, and the long-term time dependence of the extracellular signal is captured by the long short-term memory network in the artificial neural network. Through the gradual learning of the sequence data, the potential relationship between the extracellular signal and the intracellular signal is determined by the artificial neural network, and it has the ability to deduce the intracellular signal based on the extracellular signal, thereby realizing the long-term stable recording of the intracellular action potential signal in a non-invasive manner.

[0124] The following describes a specific example of the process of measuring action potential signals in myocardial cells.

[0125] See Figure 2 and Figure 3 Based on the fusion of artificial intelligence (AI) and MEA electroporation system (AI-MEA-EP), ultra-long-term and high-throughput monitoring of AP in cardiomyocytes is achieved. First, based on the micromotor array (MEA) and electroporation intracellular recording system, a rich variety of waveforms of extracellular sample signals and intracellular sample signal data pairs are recorded as a data set. Figure 4 , Figure 4 Middle (A) is a schematic diagram showing multiple rounds of cell culture on different MEA chips to obtain extracellular / intracellular AP signal pairs on different days of different experimental batches; cells on different MEA chips are treated with different culture conditions or stimuli, resulting in cells showing different AP distributions. Figure 4(B) shows a typical dataset of pre-processed extracellular and intracellular signals collected from different batches of cell recording experiments. Then, the paired cell electrophysiological datasets are quality screened, normalized and segmented as training sets to improve the learning and optimization efficiency of the model. Next, the training set is used for convolutional neural network-long short-term memory network (CNN-LSTM) model training, so that the model has the ability to capture local key features and long-term temporal dependencies in cell electrophysiological signals. In this process, the CNN network extracts local features in the signal and learns various features in the training set by applying multiple 1-dimensional convolution kernels and activation functions. The subsequent pooling layer and dropout layer reduce the feature size and the number of parameters, thereby improving computational efficiency and preventing overfitting. After the CNN layer extracts features, the processed signal data is passed to the LSTM layer. The LSTM network captures It identifies the long-term correlation in APs and effectively models their temporal information. The memory unit of LSTM enables it to retain previously input information, which improves the overall predictive performance of the model when processing the temporal characteristics of electrophysiological signals. By gradually learning from sequence data, the LSTM network determines the potential relationship between extracellular and intracellular signals and has the ability to deduce intracellular signals based on extracellular signals. In addition, to further improve the prediction accuracy of intracellular signals across different batches, we incorporate new intracellular signals recorded by electroporation of several batches of MEAs into the pre-trained model for data augmentation and transfer learning, leveraging existing knowledge to accelerate the convergence of the CNN-LSTM model and improve the accuracy of intracellular signals. In the application, the waveforms of extracellular APs recorded by MEAs are quality screened, normalized, and segmented before being input into the CNN-LSTM model for deriving intracellular AP waveforms.

[0126] Drug treatment of cardiomyocytes was as follows:

[0127] As a test group, cardiomyocytes were pre-cultured on MEAs for 8 days and then treated with isoproterenol hydrochloride every 24 hours. Isoproterenol hydrochloride was dissolved in DMSO solution. The drug was then diluted to the desired concentration using complete culture medium (Dulbecco's modified Eagle's medium containing 10% fetal bovine serum (ThermoFisher Scientific, USA), 2 mm L-glutamine (Gibco, USA) and 1% penicillin (Gibco, USA)). Before recording, the drug solution was incubated with the cells for 10 minutes. For each drug tested at three different concentrations, electrophysiological signals were recorded using MEAs. After recording, the medium containing the drug was aspirated, the cells were washed with PBS, and fresh culture medium was added. In drug testing, the intracellular signals of all 32 channels were derived by the AI model, and 8 / 32 channels with the best signal quality (scored by the CNN classification model) were used for analysis. During statistical analysis, cells on MEAs treated with DMSO were used as a control group, and all drug-induced changes were normalized to those of the DMSO-treated control group. All changes studied were calculated as fold changes relative to the control baseline.

[0128] The process of AI-MEA-EP inferring intracellular AP waveforms from extracellular FP signals is as follows:

[0129] In order to test the overall performance of the developed AI model, 80% of the dataset was randomly selected as the training set and the remaining 20% as the test set. Figure 5 , Figure 5 (A) shows representative results of the test group, MEA-recorded extracellular (cyan) and extracellular signals and the corresponding AI-MEA-EP-derived intracellular signals (blue). For comparison, the normalized intracellular signal (brown) obtained by MEA-electroporation was used as a reference signal. The waveform of the intracellular signal derived by AI-MEA-EP accurately reproduced the key features of the intracellular waveform obtained by MEA-electroporation. Next, the waveform accuracy of the intracellular APs derived by AI-MEA-EP was analyzed. Specifically, the waveform indicators such as deviation and R2 value were calculated. Figure 5 As shown in (B), the deviation is 0.0027±0.0031, indicating high waveform accuracy, while the R2 value is 0.9442±0.0702, confirming the good correlation of the derived intracellular APs. This example also compares intracellular AP parameters, including depolarization time, repolarization time, and APD10-90. The AP depolarization times obtained by MEA-EP and AI-MEA-EP were 0.03±0.01s and 0.03±0.03s, respectively. Figure 5(C), while the repolarization times were 0.15±0.03s and 0.16±0.04s, respectively. Figure 5 (D). Figure 5 As shown in (E), the maximum absolute difference in APD10-90 is only approximately 0.3 ms, further highlighting the high fidelity of the derived waveform parameters. These findings demonstrate that intracellular signals obtained through AI-MEA-EP not only have the unique advantage of being non-invasive but also exhibit high accuracy in waveform and feature reproduction.

[0130] The procedure for continuous and ultra-long-term intracellular AP recordings via AI-MEA-EP is as follows:

[0131] Each of the continuously recorded extracellular AP signals was independently fed into the AI model and converted into a corresponding intracellular AP signal. Next, the individual intracellular waveforms were concatenated based on the time series of the extracellular signals, and the beat frequency of the intracellular signals was aligned with the beat frequency of the extracellular signals. Figure 6 (A) shows a continuous extracellular waveform of approximately 7 s recorded from a representative microelectrode, the intracellular AP derived by AI-MEA-EP, and the intracellular AP recorded by MEA-electroporation. Intracellular APs were recorded transiently for minutes after MEA electroporation, after which the intracellular signal gradually decayed and eventually disappeared over time, reverting to the extracellular signal. This is mainly attributed to the short-term healing of the cell membrane after electroporation, which limits the duration of intracellular entry. The results show that the intracellular AP waveforms highly overlap with the experimentally recorded waveforms. Notably, the intracellular signals obtained by AI-MEA-EP remained stable in amplitude during the continuous recording period, without the gradual decay or disappearance observed in the MEA-electroporation experiments.

[0132] Figure 6 The results in (B) further demonstrate the ability to continuously record APs over approximately 2 seconds on multiple channels of the MEA chip. Representative extracellular APs were successfully recorded on 8 channels, demonstrating the system's ability to simultaneously capture AP signals at multiple sites ( Figure 6 Middle (B)-(i)). After MEA-electroporation, intracellular APs were temporarily recorded for 60 seconds ( Figure 6 (B)-(ii)). The intracellular AP signal derived from AI-MEA-EP showed a waveform similar to that of the intracellular AP recorded by MEA-electroporation ( Figure 6 To evaluate the single precision in multi-channel, the R value was analyzed to be 0.832 ± 0.154 and the deviation was 0.012 ± 0.009, indicating a high fit and low deviation in the overall waveform profile. Figure 6(C). This example also analyzes waveform parameters, including beat frequency, depolarization time, repolarization time, and APD10-90. The intracellular APs derived from AI-MEA-EP have a beat frequency of 2.85 bpm. The depolarization time is 0.034 ± 0.009 s, which is consistent with the MEA electroporation recording value of 0.025 ± 0.010 s, see Figure 6 Middle (D). The repolarization time obtained by AI-MEA-EP was 0.165 ± 0.014 s, which is very consistent with the MEA electroporation result of 0.157 ± 0.016 s. Figure 6 Middle (E). The maximum absolute difference in APD10-90 was only about 2 ms, further confirming the high similarity between the intracellular signaling of AI-MEA-EP and that recorded via MEA-electroporation, see Figure 6 Middle (F).

[0133] This example further investigates the feasibility of using the AI-MEA-EP system to continuously record intracellular APs for more than 6 days in cell culture. Figure 7 Middle (A), Representative extracellular APs recorded by MEA chip and the corresponding intracellular APs derived by AIMEA-EP. To minimize potential differences caused by changes in MEA chip or cell culture conditions, one or two sets of extra / intracellular AP pairs obtained by MEA electroporation were used at the beginning of the experiment to calibrate the AI model by transfer learning. The calibration results showed high consistency with the waveforms recorded by MEA-electroporation. Specifically, the deviation of the intracellular signal obtained by the uncalibrated AIMEA-EP was 0.0107±0.0064, and the R value was 0.7366±0.2363. After transfer learning calibration, the deviation was further reduced to 0.0101±0.0063, while the R value increased to 0.7527±0.2403, see Figure 7 (B) and (C). The results of the 6-day recording period were analyzed based on AP duration, depolarization time, repolarization time, and beat frequency. Figure 7 Specifically, the mean absolute errors of the calibrated AI-MEA-EP were 0.0123 ± 0.0010 ms, 0.0160 ± 0.0010 ms, and 0.0222 ± 0.0016 ms for APD30, APD50, and APD70, respectively, while these errors were 0.0099 ± 0.0006 ms, 0.0157 ± 0.0010 ms, and 0.0216 ± 0.0014 ms for the calibrated and uncalibrated AI-MEA-EP, respectively. The results of depolarization time were slightly higher than those of MEA-electroporation by about 0.01 s, see Table 2. Figure 7 (D), while the results of repolarization time were more consistent, with a mean deviation of <0.1ms, see Figure 7Middle (E). Based on the experimental results, as the culture days increase, the duration of APD (e.g., APD30, APD50, and APD70) gradually increases, while the beating frequency gradually decreases.

[0134] The AI-MEA-EP test process for evaluating drug-induced AP changes in myocardial cells is as follows:

[0135] This example uses the AI-MEA-EP system to further explore the changes in APs in cardiomyocytes under the influence of ion channel drugs. The electrophysiological signals of cardiomyocytes treated with different concentrations of isoproterenol hydrochloride showed obvious rhythm and waveform changes. Figure 8 For isoproterenol hydrochloride treatment, the AI-MEA-EP derived waveform had a bias of 0.007 ± 0.003 and an R value of 0.839 ± 0.100, indicating high waveform fitting accuracy and low error, see Figure 8 Middle (E). With increasing concentrations of isoproterenol (100 nM, 500 nM, 1 μM), intracellular APs exhibited accelerated depolarization and repolarization, and the APD decreased. Specifically, the repolarization time recorded by MEA-electroporation decreased by 34.2%, while the repolarization time derived by AI-MEA-EP decreased by 9.1%, see Figure 8 Middle (F). In terms of depolarization time, MEA electroporation recorded a 20.5% reduction, while AI-MEA-EP achieved a 21.2% reduction, see Figure 8 In addition, the APD30, APD50, and APD70 recorded by MEA electroporation decreased by 22.4%, 22.2%, and 22.0%, respectively, while those recorded by AI-MEA-EP decreased by 8.57%, 14.04%, and 14.29%, respectively. Figure 8 Middle (H). These results indicate that cardiomyocyte excitability is significantly enhanced with increasing isoproterenol concentrations, and AI-MEA-EP exhibits high accuracy in inferring intracellular AP changes.

[0136] The embodiments of the present invention have the following effects: The AI-MEA-EP system exhibits extraordinary accuracy, and the intracellular AP derived by AI-MEA-EP is highly consistent with the intracellular AP physically obtained by MEA-EP, with a waveform profile deviation of <0.3% and an APD10-90 error of <0.3ms. The AI-MEA-EP system exhibits extraordinary stability and can achieve ultra-long-term recording (>10 consecutive days). In addition, AI-MEA-EP can reliably capture drug-induced AP changes, including isoproterenol-triggered APD shortening (deviation <0.007).

[0137] An embodiment of the present invention provides a non-invasive reconstruction system for intracellular electrical signals based on an artificial neural network, comprising:

[0138] The prediction module is used to obtain extracellular signals, input the preprocessed extracellular signals into the trained artificial neural network, and predict the intracellular electrical signals; the artificial neural network includes a convolutional neural network and a long short-term memory network, and the artificial neural network is trained by the following method:

[0139] Acquiring extracellular sample signals and paired intracellular sample signals, classifying the extracellular sample signals and the intracellular sample signals, and taking the extracellular sample signals and the intracellular sample signals that meet preset classification requirements as a sample set;

[0140] The preset model is trained according to the preprocessed sample set, and the parameters of the preset model are updated according to the preset loss function until the preset requirements are met, and the trained artificial neural network is determined.

[0141] It can be seen that the contents of the above method embodiments are all applicable to the present system embodiments. The functions specifically implemented by the present system embodiments are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0142] See Figure 9 The embodiment of the present invention provides a non-invasive reconstruction system of intracellular electrical signals based on an artificial neural network, comprising:

[0143] at least one processor;

[0144] at least one memory for storing at least one program;

[0145] When at least one program is executed by at least one processor, the at least one processor implements the above method.

[0146] Among them, the memory is a non-transient computer-readable storage medium that can be used to store non-transient software programs and non-transient computer executable programs. The memory may include a high-speed random access memory and may also include a non-transient memory, such as at least one disk storage device, a flash memory device, or other non-transient solid-state storage device. In some embodiments, the memory optionally includes a remote memory remotely arranged relative to the processor, and these remote memories can be connected to the processor via a network. Examples of the above-mentioned network include but are not limited to the Internet, an intranet, a local area network, a mobile communication network and a combination thereof.

[0147] It can be seen that the contents of the above method embodiments are all applicable to the present system embodiments. The functions specifically implemented by the present system embodiments are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0148] In addition, the embodiments of the present application further disclose a computer program product or computer program, which is stored in a computer-readable storage medium. The processor of a computer device can read the computer program from the computer-readable storage medium, and the processor executes the computer program, so that the computer device performs the above-mentioned method. Similarly, the contents of the above-mentioned method embodiment are all applicable to the present storage medium embodiment, and the functions specifically implemented by the present storage medium embodiment are the same as those of the above-mentioned method embodiment, and the beneficial effects achieved are also the same as those achieved by the above-mentioned method embodiment.

[0149] An embodiment of the present invention further provides a computer-readable storage medium, which stores a program executable by a processor. The program executable by the processor is used to implement the above method when executed by the processor.

[0150] It is understood that all or some steps, systems in the disclosed method above can be implemented as software, firmware, hardware and appropriate combinations thereof. Some physical components or all physical components can be implemented as software by a processor, such as a central processing unit, a digital signal processor or a microprocessor, or implemented as hardware, or implemented as an integrated circuit, such as an application specific integrated circuit. Such software can be distributed on a computer-readable medium, and the computer-readable medium can include computer storage media (or non-transitory media) and communication media (or temporary media). As known to those of ordinary skill in the art, the term computer storage medium is included in any method or technology for storing information (such as computer-readable instructions, data structures, program modules or other data) and is volatile and non-volatile, removable and non-removable media. Computer storage media includes but is not limited to RAM, ROM, EEPROM, flash memory or other memory technology, CD-ROM, digital versatile disk (DVD) or other optical disk storage, magnetic cassette, magnetic tape, disk storage or other magnetic storage device, or can be used to store desired information and any other medium that can be accessed by a computer. Furthermore, as is well known to those skilled in the art, communication media typically embodies computer-readable instructions, data structures, program modules, or other data in a modulated data signal such as a carrier wave or other transport mechanism, and may include any information delivery media.

[0151] See Figure 10 The embodiment of the present invention provides a non-invasive reconstruction system of intracellular electrical signals based on artificial neural networks, including a cell signal acquisition device and a computer device connected to the cell signal acquisition device; wherein,

[0152] Cell signal acquisition equipment, used to measure extracellular and intracellular signals of single cells;

[0153] Computer equipment includes:

[0154] at least one processor;

[0155] at least one memory for storing at least one program;

[0156] When at least one program is executed by at least one processor, the at least one processor implements the above method.

[0157] Specifically, the cell signal acquisition device includes but is not limited to a fusion device based on artificial intelligence (AI) and MEA electroporation system (AI-MEA-EP); and as for the computer device, it can be different types of electronic devices, including but not limited to desktop computers, laptops and other terminals.

[0158] It can be seen that the contents of the above method embodiments are all applicable to the present system embodiments. The functions specifically implemented by the present system embodiments are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0159] It should be understood that in this application, "at least one (item)" means one or more, and "plurality" means two or more. "And / or" is used to describe the association relationship of associated objects, indicating that three relationships may exist. For example, "A and / or B" can mean: only A exists, only B exists, and A and B exist at the same time, where A and B can be singular or plural. The character " / " generally indicates that the previous and next associated objects are in an "or" relationship. "At least one of the following items" or similar expressions refers to any combination of these items, including any combination of single items or plural items. For example, at least one of a, b or c can mean: a, b, c, "a and b", "a and c", "b and c", or "a and b and c", where a, b, c can be single or multiple.

[0160] In the several embodiments provided in this application, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are merely schematic. For example, the division of the units is merely a logical function division. In actual implementation, there may be other division methods, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be an indirect coupling or communication connection through some interfaces, devices or units, which can be electrical, mechanical or other forms.

[0161] The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of these units may be selected to achieve the purpose of this embodiment according to actual needs.

[0162] The above is a specific description of the preferred implementation of the present invention, but the invention is not limited to the embodiments. Those skilled in the art can make various equivalent modifications or substitutions without violating the spirit of the present invention. These equivalent modifications or substitutions are all included in the scope defined by the claims of this application.

Claims

1. A non-invasive reconstruction method of intracellular electrical signals based on artificial neural networks, characterized in that: include: Acquire extracellular signals, input the preprocessed extracellular signals into a trained artificial neural network, and predict intracellular electrical signals; the artificial neural network includes a convolutional neural network and a long short-term memory network, and the artificial neural network is trained by the following method: Acquiring extracellular sample signals and paired intracellular sample signals, classifying the extracellular sample signals and the intracellular sample signals, and taking the extracellular sample signals and the intracellular sample signals that meet preset classification requirements as a sample set; The preset model is trained according to the preprocessed sample set, and the parameters of the preset model are updated according to the preset loss function until the preset requirements are met to obtain a trained artificial neural network.

2. The method according to claim 1, characterized in that The extracellular sample signal and the paired intracellular sample signal are obtained by the following method: For the extracellular measurement signal, the time point of the trough in each cycle is used as the first reference origin, a first number of data points are extracted before the first reference origin, and a second number of data points are extracted after the first reference origin to obtain the extracellular sample signal; For the measurement signal inside the cell, the time point of the peak in each cycle is used as the second reference origin, a first number of data points are extracted before the second reference origin, and a second number of data points are extracted after the second reference origin to obtain the paired intracellular sample signal.

3. The method according to claim 1, characterized in that The preset loss function is determined by the following method: determining a first loss parameter according to a difference in amplitude between the predicted value of the extracellular sample signal and the paired intracellular sample signal; determining a second loss parameter according to a difference in action potential duration between the predicted value of the extracellular sample signal and the paired intracellular sample signal; determining a third loss parameter according to a depolarization time difference between the predicted value of the extracellular sample signal and the paired intracellular sample signal; determining a fourth loss parameter according to a difference in repolarization time between the predicted value of the extracellular sample signal and the paired intracellular sample signal; A preset loss function is determined according to the first loss parameter, the second loss parameter, the third loss parameter, and the fourth loss parameter.

4. The method according to claim 1, wherein The method further comprises: Data augmentation is performed on the sample set based on any one or more of amplitude change, signal fusion, or resampling.

5. The method according to claim 4, characterized in that Based on the amplitude change, data expansion is performed on the sample set, including: The amplitudes of the extracellular sample signals and the intracellular sample signals in the sample set are compressed or stretched according to a preset ratio to perform data expansion on the sample set.

6. The method according to claim 4, characterized in that Based on signal fusion, data expansion is performed on the sample set, including: The extracellular sample signals and the intracellular sample signals of different source domains in the sample set are fused according to a first preset range and a first preset step size to perform data expansion on the sample set.

7. The method according to claim 4, characterized in that Based on resampling, data expansion is performed on the sample set, including: The extracellular sample signals and the intracellular sample signals in the sample set are used to generate samples of different time scales according to a second preset range and a second preset step size, so as to perform data expansion on the sample set.

8. A non-invasive reconstruction system of intracellular electrical signals based on artificial neural networks, characterized in that: include: The prediction module is used to obtain extracellular signals, input the preprocessed extracellular signals into a trained artificial neural network, and predict the intracellular electrical signals; the artificial neural network includes a convolutional neural network and a long short-term memory network, and the artificial neural network is trained by the following method: Acquiring an extracellular sample signal and a paired intracellular sample signal, classifying the extracellular sample signal and the intracellular sample signal, and taking the extracellular sample signal and the intracellular sample signal that meet preset classification requirements as a sample set; The preset model is trained according to the preprocessed sample set, and the parameters of the preset model are updated according to the preset loss function until the preset requirements are met, thereby determining a trained artificial neural network.

9. A non-invasive reconstruction system of intracellular electrical signals based on artificial neural networks, characterized in that: include: at least one processor; at least one memory for storing at least one program; When the at least one program is executed by the at least one processor, the at least one processor implements the method according to any one of claims 1 to 7.

10. A non-invasive reconstruction system of intracellular electrical signals based on artificial neural networks, characterized in that: It includes an extracellular signal acquisition device and a computer device connected to the extracellular signal acquisition device; wherein, The extracellular signal acquisition device is used to measure the extracellular signals of single cells; The computer device comprises: at least one processor; at least one memory for storing at least one program; When the at least one program is executed by the at least one processor, the at least one processor implements the method according to any one of claims 1 to 7.

Images