Oral drug precise delivery and curative effect evaluation system based on microfluidics

Abstract

The invention relates to the technical field of microfluidic drug delivery, and discloses an oral drug precise delivery and curative effect evaluation system based on microfluidics. The system comprises a signal acquisition and processing module, an instance identification and feature extraction module, and a path generation and curative effect evaluation module. The signal acquisition and processing module continuously acquires multi-dimensional signals of oral drug delivery through the microfluidic sensing array, and the multi-dimensional signals are integrated into a unified data stream after real-time correction and structure standardization. An instance identification module identifies a delivery instance therefrom, extracts a temporal identity, a spatial location descriptor, and a drug classification identifier. The path generation module divides intervals in combination with the time identifiers, aligns the spatial information to the standard oral cavity model, and generates a drug delivery path time sequence record. And the curative effect evaluation module calculates the drug distribution density of the target area according to drug classification aggregation examples, generates a curative effect evaluation data set, and realizes the precision of drug delivery monitoring and curative effect evaluation.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of microfluidic drug delivery, in particular to a microfluidic-based oral drug precise delivery and efficacy evaluation system. BACKGROUND

[0002] In the treatment of oral diseases, the precision of drug delivery and the objectivity of efficacy evaluation have always been the core problems in clinical practice. As a complex physiological environment, the oral cavity has multiple dynamic interference factors such as saliva flow, mucosal tissue heterogeneity, and masticatory movement, which directly lead to the difficulty of traditional oral drug delivery methods to achieve precise enrichment of drugs in the target lesion area. The commonly used oral drug dosage forms in clinical practice, such as gargle, patch, and gel, generally have the problems of uncontrollable drug release and dispersed action range. Some drugs are quickly lost under the flushing of saliva, which not only reduces the drug utilization efficiency, but also may cause local irritation and other adverse reactions due to the residual of drugs in non-target areas.

[0003] In the monitoring of drug delivery process and the evaluation of efficacy, the existing technology also has obvious limitations. Traditional methods mostly rely on the visual observation of doctors, the subjective feeling feedback of patients, or indirect methods such as ex vivo sample detection to obtain efficacy information. These methods are difficult to capture the dynamic changes in the drug delivery process, and the evaluation results are easily affected by subjective factors, lacking quantitative data support. Even some studies introduce sensing technology for oral signal acquisition, but mostly for single-dimensional sensing, which cannot fully reflect the key information such as time, space, and drug characteristics of drug delivery. At the same time, the signals collected are often distorted due to electromagnetic interference, temperature fluctuations and other factors in the oral environment. There is a lack of effective correction and standardization means in the signal processing process, which makes it difficult to integrate different sources of signals into a unified data stream, further restricting the systematic analysis of the drug delivery process.

[0004] Microfluidic technology has shown great application prospects in the field of biomedical detection due to its miniaturization, integration and low consumption. Some studies have attempted to use it to construct drug delivery systems, but there are still many deficiencies in the existing microfluidic technology in the application of oral drugs. Most microfluidic systems only focus on the realization of drug delivery function, without integrating delivery process monitoring and efficacy evaluation, which leads to the inability to establish a direct correspondence between drug delivery effect and efficacy. In addition, the lack of adaptability design for the special environment of the oral cavity makes the stability of the microfluidic sensing unit poor in the oral cavity, and the continuity and reliability of signal acquisition cannot be guaranteed. At the data processing level, there is a lack of special instance recognition and feature extraction algorithms for oral drug delivery scenarios, which cannot accurately extract the key parameters of drug delivery from complex signals, and thus cannot provide effective data input for efficacy evaluation. These problems together lead to the disconnection between drug delivery and efficacy evaluation in oral drug treatment, which cannot meet the demand for precise treatment in clinical practice, and an integrated system that can realize precise drug delivery monitoring and objective efficacy evaluation is needed. SUMMARY

[0005] The purpose of the present application is to provide a microfluidic-based oral drug precise delivery and efficacy evaluation system to solve the problems raised in the background art.

[0006] To achieve the above purpose, the present application provides a microfluidic-based oral drug precise delivery and efficacy evaluation system, which comprises:

[0007] A signal acquisition and processing module for continuously acquiring multi-dimensional signals of the drug delivery process in the oral cavity through a microfluidic sensing array, real-time correction and structural standardization of the signals, elimination of signal distortion, and integration of unified data flow;

[0008] An instance recognition and feature extraction module for recognizing drug delivery instances from the unified data flow, extracting instance time identifiers, spatial position descriptors and drug classification identifiers;

[0009] A path generation module for dividing time intervals using instance time identifier sequences and aligning spatial position descriptors to a standard oral model to generate drug delivery path time sequence records;

[0010] An efficacy evaluation module for aggregating delivery instances by category according to drug delivery path time sequence records and drug classification identifiers, calculating drug distribution density in the target oral area, and generating efficacy evaluation data sets.

[0011] Preferably, the continuous acquisition of multi-dimensional signals of the drug delivery process in the oral cavity through the microfluidic sensing array, the real-time correction and structural standardization of the signals, the elimination of signal distortion, and the integration of unified data flow comprise:

[0012] The microfluidic sensor array generates multi-source signal streams, which are converted into digital format through signal processing, and the signal levels are adjusted using amplitude normalization algorithm to make the signal dimensions consistent.

[0013] The dimension-consistent signal streams are structurally integrated, the format differences of different signal sources are identified, converted into standard data templates, and the noise components are removed using filtering algorithm to generate a purified signal set.

[0014] Based on the time reference point, the purified signal set is time-synchronized, time delay is compensated, multi-source data is fused, and a unified data stream is output.

[0015] Preferably, the amplitude normalization algorithm is used to adjust the signal level to make the signal dimensions consistent, comprising:

[0016] The peak and valley of the multi-source signal stream are detected, the dynamic range is calculated, the signal is mapped to a common amplitude interval using a scaling function, and signal standardization is achieved.

[0017] Baseline correction is applied to the standardized signal to remove DC offset and ensure signal zero alignment.

[0018] Preferably, the dimension-consistent signal streams are structurally integrated, the format differences of different signal sources are identified, converted into standard data templates, and the noise components are removed using filtering algorithm to generate a purified signal set, comprising:

[0019] The data header information of the dimension-consistent signal stream is parsed, the format descriptor is extracted, the field definition of the standard data template is matched, and the format conversion operation is performed.

[0020] A low-pass filter is applied to suppress high-frequency noise, and an adaptive filter is used to eliminate environmental interference to output the purified signal set.

[0021] Preferably, the drug delivery instances are identified from the unified data stream, the instance time identifier, spatial position descriptor and drug classification identifier are extracted, comprising:

[0022] The event feature pattern in the unified data stream is scanned to detect the instantaneous point of drug delivery occurrence, and the time identifier is recorded.

[0023] The position information is retrieved from the data segment associated with the event feature pattern, the three-dimensional coordinates are parsed through spatial decoding algorithm, and the spatial position descriptor is generated.

[0024] The drug feature code is queried, the classification dictionary is referred to, the drug type is determined, and the drug classification identifier is assigned.

[0025] Preferably, the event feature pattern in the unified data stream is scanned to detect the instantaneous point of drug delivery occurrence, and the time identifier is recorded, comprising:

[0026] Adopting pattern matching algorithm to identify mutation points in the signal, verifying the consistency of mutation points and drug delivery logic, and marking timestamps; Further sorting and deduplicating the timestamps.

[0027] Preferably, the position information is retrieved from the data segment associated with the event feature pattern, the three-dimensional coordinates are analyzed by a spatial decoding algorithm, and the spatial position descriptor is generated, including:

[0028] Extracting the original reading of the position sensor, and converting the reading into a three-dimensional space point by applying a coordinate solving algorithm;

[0029] Cluster analysis is performed on the three-dimensional space point to eliminate outliers and generate a robust spatial position descriptor.

[0030] Preferably, the time interval is divided using the instance time identifier sequence, and the spatial position descriptor is aligned to the standard oral model to generate the drug delivery path time record, including:

[0031] According to the distribution of the instance time identifier, the data is divided into equal length time blocks to form continuous time intervals;

[0032] For each time interval, the spatial position descriptor is transformed from the local coordinate system to the global coordinate system of the standard oral model, and the point-to-point mapping is realized using a registration algorithm;

[0033] The transformed position data is organized in chronological order, the drug delivery attribute is added, and the drug delivery path time record is constructed.

[0034] Preferably, the spatial position descriptor is transformed from the local coordinate system to the global coordinate system of the standard oral model, and the point-to-point mapping is realized using a registration algorithm, including:

[0035] Loading the coordinate system definition of the standard oral model, calculating the transformation parameters between the local coordinate system and the global coordinate system;

[0036] Applying the transformation parameters to re-project each spatial position descriptor to the global coordinate system, verifying the projection accuracy and adjusting the error.

[0037] Preferably, according to the drug delivery path time record and the drug classification identifier, the delivery instances are aggregated by category, the drug distribution density in the target oral area is calculated, and the therapeutic effect evaluation data set is generated, including:

[0038] According to the drug classification identifier, the drug delivery path time record is grouped into multiple category subsets;

[0039] For each category subset, the delivery frequency and drug amount in the target oral area are calculated, and the density value is calculated in combination with the area;

[0040] The change of the polymeric density value over time generates a therapeutic efficacy evaluation data set.

[0041] Compared with the prior art, the present application has the following advantages:

[0042] The signal acquisition and processing module adopts a microfluidic sensing array to construct a multi-dimensional acquisition system. Compared with the traditional single sensing method, it can simultaneously capture multiple signals such as drug concentration, temperature, pH value and drug spatial position in the oral cavity, and realize comprehensive perception of the drug delivery process. The real-time correction mechanism built-in the module can dynamically correct the signal distortion caused by interference factors such as saliva flow and temperature fluctuations in the oral environment, combined with standardized processing of the structure, to ensure the consistency of various signals in format and accuracy. The unified data stream formed finally provides a high-quality data basis for subsequent analysis, avoiding the problem of difficult integration of traditional multi-source signals.

[0043] The instance recognition and feature extraction module develops a special recognition algorithm for the unified data stream, which can accurately locate the start and end nodes of the drug delivery instance, and extract effective information from complex data. The time identifier, spatial position descriptor and drug classification identifier provide core parameters for quantitative analysis of the drug delivery process. The time identifier sequence clearly records the action time of the drug in the oral cavity, the spatial position descriptor accurately reflects the distribution area of the drug, and the drug classification identifier provides a basis for subsequent efficacy analysis according to drug characteristics. The extraction of these characteristic parameters makes the drug delivery process change from fuzzy experience judgment to quantifiable objective data record, solving the problem of missing key information in traditional methods.

[0044] The path generation module combines time identifiers with spatial position information, realizes time interval division to time sequence decomposition of the drug delivery process, and aligns and calibrates the spatial position descriptor with the standard oral cavity model. The generated time sequence record of the drug delivery path intuitively presents the dynamic migration process of the drug in the oral cavity. This visual path record can clearly show whether the drug reaches the target lesion area, the residence time in the target area and other key information, providing a direct basis for judging the accuracy of drug delivery. Compared with the defect that the traditional method cannot track the dynamic distribution of the drug, the time sequence record formed by this module makes the drug delivery process traceable and analyzable, providing a specific direction for optimizing the delivery scheme.

[0045] The efficacy evaluation module aggregates the delivery instances by drug categories based on the path timing records and drug classification identifiers, and realizes the quantitative evaluation of the efficacy by measuring the drug distribution density in the target area. This module discards the traditional evaluation method which relies on subjective feeling and visual observation, and takes the actual distribution of the drug in the target area as the core index of the efficacy judgment. The efficacy evaluation data set formed contains quantitative information such as drug delivery accuracy and target area drug enrichment degree. This evaluation method can objectively reflect the drug treatment effect and avoid the evaluation deviation caused by subjective factors. At the same time, the design of this module makes the efficacy data of different drugs and different patients comparable, providing an objective reference for clinical drug selection. BRIEF DESCRIPTION OF DRAWINGS

[0046] Figure 1 The timing diagram of the microfluidic-based oral drug precise delivery and efficacy evaluation system described in the application;

[0047] Figure 2 The flow chart of the signal acquisition and processing module;

[0048] Figure 3 The flow chart of signal structure integration and noise filtering;

[0049] Figure 4 The adaptive filtering and original signal energy attenuation characteristic comparison analysis diagram;

[0050] Figure 5 The oral drug total dose distribution column chart. DETAILED DESCRIPTION

[0051] The technical solutions in the embodiments of the application will be described clearly and completely below with reference to the drawings in the embodiments of the application. Obviously, the described embodiments are only part of the embodiments of the application, rather than all the embodiments. Based on the embodiments in the application, all other embodiments obtained by those skilled in the art without creative labor fall within the protection scope of the application.

[0052] Please refer to Figure 1The application provides a microfluidic-based oral drug precise delivery and efficacy evaluation system, which comprises a signal acquisition and processing module, an instance recognition and feature extraction module, a path generation module and an efficacy evaluation module. The signal acquisition and processing module continuously acquires multi-dimensional signals of the drug delivery process in the oral cavity through a microfluidic sensing array, performs real-time correction and structural standardization on the signals, eliminates signal distortion, and integrates unified data flow. The instance recognition and feature extraction module identifies drug delivery instances from the unified data flow, extracts instance time identifiers, spatial position descriptors and drug classification identifiers. The path generation module divides time intervals using the instance time identifier sequence, aligns the spatial position descriptors to a standard oral model, and generates a drug delivery path time record. The efficacy evaluation module aggregates delivery instances by category according to the drug delivery path time record and the drug classification identifier, calculates the drug distribution density in the target oral area, and generates an efficacy evaluation dataset.

[0053] Embodiment 1: refer to Figure 2 The microfluidic sensing array generates a multi-source signal stream, which is derived from micro-sensor nodes arranged at different anatomical positions inside the oral cavity. These nodes are distributed in an array form on the occlusal surface of the teeth, the gingival margin groove and the mucosal surface of the tongue back. The microfluidic sensing array uses a flexible circuit substrate to achieve conformal fitting with the oral cavity. The multi-source signal stream includes electrochemical impedance signals, optical reflection signals and micro-pressure resistance signals. The electrochemical impedance signals reflect changes in drug ion concentration, the optical reflection signals represent changes in mucosal color and moisture, and the micro-pressure resistance signals monitor mechanical forces during drug release. The signal conversion processing link uses a multi-channel analog-to-digital converter to convert analog signals to digital format. The analog-to-digital converter has a 16-bit resolution and a sampling rate of 1000 times per second. The digital format signals are temporarily stored in a first-in-first-out buffer in the form of fixed-length data packets. The amplitude normalization algorithm adjusts the amplitude of the multi-source signal stream. The algorithm detects the peak and valley values of each signal channel in real time. The peak detection uses a sliding window local maximum value identification method, and the valley detection uses a local minimum value scanning algorithm. The dynamic range calculation is based on the difference between the maximum and minimum values in the window. The scaling function uses a linear mapping relationship to transform the original signal amplitude to the standardized interval of zero to one. The slope and intercept of the linear mapping are calculated in real time based on the dynamic range, achieving consistent signal dimension. The standardized signals enter the baseline correction stage, which is achieved through a digital high-pass filter. The cutoff frequency of the digital high-pass filter is set to 0.5 Hz, effectively removing the direct current offset component in the signal and aligning the signal waveform to zero.

[0054] The dimensionally consistent signal flow enters the structure integration stage, and the structure integration module parses the header information of each data packet, which includes the sensor node number, sampling timestamp and signal quality identifier. The format descriptor extracts the signal source characteristics from the header information, which includes the byte sequence marker, data precision identifier and checksum field. The standard data template uses a unified field definition, including absolute timestamp, signal channel number, three-dimensional signal value and status flag bit. The format conversion operation performs field mapping and data reorganization, field mapping matches the field order of the standard data template according to the format descriptor, and data reorganization processes signal interpolation alignment of different sampling rates. The filter algorithm performs noise reduction processing on the structure integrated signal, and the low-pass filter uses a fourth-order Butterworth filter structure, with the cutoff frequency dynamically configured according to the signal type, the electrochemical signal cutoff frequency set to 10 Hz, the optical signal cutoff frequency set to 50 Hz, and the pressure signal cutoff frequency set to 100 Hz. The adaptive filter uses a multi-input single-output structure, the main input signal comes from the microfluidic sensing array, and the reference input signal comes from the environmental temperature and humidity sensor. The coefficient update of the adaptive filter uses the normalized least mean square algorithm, and the step size parameter is set to 0.01. The purified signal set is subjected to time synchronization processing, which is based on a global clock source that provides a time reference using a temperature-compensated crystal oscillator. The time delay compensation calculates the transmission delay of each signal channel, which is determined by measuring the propagation time of the signal from the sensor to the processing unit, and the interpolation method uses cubic spline interpolation to achieve time alignment of multi-channel signals. The multi-source data fusion uses a weighted average algorithm, and the weight coefficient is dynamically allocated according to the signal signal-to-noise ratio. The output unified data stream is stored in a structured binary format.

[0055] The specific implementation of the amplitude normalization algorithm includes a signal preprocessing step, which divides the continuous signal into fixed-length analysis frames, with an analysis frame length of 256 sampling points and a frame overlap rate of 50%. The window function applies a Hanning window for windowing processing, and the Hanning window coefficient is multiplied by the signal samples in each analysis frame. Dynamic range calculation is performed on the windowed signal frame, peak detection uses amplitude threshold comparison method, and amplitude threshold is set to three times the root mean square value in the frame. Valley detection uses negative threshold comparison, and the threshold is set symmetrically with peak detection. The linear transformation formula of the scaling function uses floating point operation, and the transformation coefficient is updated in real time according to the dynamic range in the frame. The direct current offset estimation of baseline correction uses the moving average method, and the moving average window length is set to 100 sampling points. The direct current offset is subtracted from the original signal. After zero alignment, the signal is subjected to amplitude verification, which checks whether the signal exceeds the standardized interval. The signal exceeding the interval is subjected to soft limiting processing.

[0056] The data header information analysis in the structure integration stage adopts a hierarchical analysis strategy. The hierarchical analysis first identifies the data packet start flag, then analyzes the metadata area, and finally extracts the payload. The format descriptor matching operation uses a preloaded configuration mapping table, which stores the data format specifications of different types of sensors. The field definition of the standard data template uses byte alignment, and the fixed length field occupies continuous byte space, while the variable length field uses length prefix coding. The format conversion operation handles byte order conversion, which selects big-endian or little-endian mode according to the characteristics of the target platform. The low-pass filter in the filtering algorithm is implemented using a direct II-type digital filter structure, and the filter coefficients are calculated offline and stored in a lookup table. The environmental interference elimination of the adaptive filter is realized by multi-order filtering, and the main signal path and the reference signal path go through the same preprocessing process. The time synchronization of the purified signal set uses hardware timestamp marking, which is generated by a dedicated timing circuit. The timestamp alignment algorithm uses the nearest neighbor matching strategy, and the matching tolerance is set to 1 millisecond. The data association of multi-source data fusion is based on the spatial position relationship of sensors, and the data of adjacent sensors is preferentially fused. The output interface of the unified data stream uses a high-speed serial bus, and the data transmission rate reaches 100 megabytes per second.

[0057] The physical implementation of the microfluidic sensor array adopts a multilayer flexible circuit manufacturing process, and the circuit layer includes signal acquisition electrodes, optical waveguides, and micro piezoresistive films. The electrochemical impedance signal measurement adopts a four-electrode method, which eliminates the influence of electrode polarization, and the frequency scanning range of the excitation signal is from 100 Hz to 10 kHz. The optical reflection signal measurement uses a combination of dual-wavelength light-emitting diodes and photodetectors, with wavelengths of 660 nm and 880 nm selected for oxyhemoglobin and deoxyhemoglobin absorption peaks, respectively. The micro piezoresistive signal measurement uses a Wheatstone bridge structure, and the temperature compensation circuit of the Wheatstone bridge is integrated inside the sensor node. The analog-to-digital converter in the signal conversion processing has a programmable gain amplifier with a gain range from 1 to 1000 automatically adjusted. The dynamic range calculation of the amplitude normalization algorithm introduces a hysteresis mechanism to avoid frequent fluctuations of peak and valley values. The linear mapping of the scaling function supports a non-linear segmented mapping mode, which is used to process saturated signals. The digital high-pass filter for baseline correction uses an infinite impulse response filter structure, and the filter coefficients are designed using the bilinear transformation method.

[0058] The hardware implementation of the structure integration module uses a field programmable gate array, which processes multiple signal streams in parallel. The format descriptor matching operation uses a content addressable memory, which stores format feature templates. The storage of standard data templates uses a dual-port random access memory, which allows simultaneous read and write access. The low-pass filter coefficient calculation of the filtering algorithm uses a frequency sampling method, which is designed based on ideal filter responses. The reference signal input of the adaptive filter contains multiple environmental parameters, including temperature, humidity, and motion acceleration. The timestamp synchronization of the purified signal set uses the precision time protocol, which achieves microsecond-level precision through network clock synchronization. The weighted coefficient calculation of multi-source data fusion is based on signal quality indicators, including signal-to-noise ratio and signal stability. The storage format of the unified data stream uses a time series database structure, which supports fast time range queries.

[0059] The power supply system of the microfluidic sensor array uses a micro button battery, and the battery life is extended through dynamic power management. The analog-to-digital converter of the signal conversion processing has an automatic calibration function, and the automatic calibration period is set to once every hour. The peak detection of the amplitude normalization algorithm introduces digital filter preprocessing, which eliminates pulse noise interference. The mapping interval of the scaling function is configurable, and the mapping interval is adjusted according to the type of drug. The moving average window length of the baseline correction can be adaptively changed based on signal frequency component analysis. The data header information parsing of the structure integration supports multiple protocol formats, including custom binary protocols and standard medical device protocols. The data verification of the format conversion operation uses cyclic redundancy check, which verifies the data integrity. The calculation optimization of the filtering algorithm uses a pipeline architecture, which improves data processing throughput. The coefficient initialization of the adaptive filter uses a preset initial value, which is set based on typical environmental conditions. The clock source of the time synchronization has a backup battery that maintains clock operation when the main power fails. The algorithm of multi-source data fusion uses a distributed computing framework that executes in parallel on multi-core processors. The output of the unified data stream uses compression encoding, which reduces storage space occupancy.

[0060] Example 2: see Figure 3, the structure integration operation of the dimensionally consistent signal stream starts the data header information parsing process, the data header information is located at the starting position of each signal data packet, and the data header information includes a signal source device identifier, a data acquisition timestamp, a signal sampling frequency parameter, and a data format version number. The format descriptor extraction module scans specific fields in the data header information, and the format descriptor explicitly identifies the structural characteristics of the signal stream, including the data encoding method, the byte arrangement order, and the checksum algorithm type. The standard data template adopts a predefined field definition specification, and the field definition specification includes fixed-length data fields and optional extended field areas. The format conversion operation performs data mapping and recombination processes, the data mapping establishes a mapping table according to the correspondence relationship between the format descriptor and the standard data template, and the mapping table stores field name, data type, and byte offset information. The signal stream after structure integration enters the filtering processing stage, and the low-pass filter is designed in the form of a digital filter, and the cutoff frequency parameter of the digital filter is dynamically configured according to the signal frequency band characteristics. The adaptive filter adopts a multi-channel filtering architecture, and the main signal channel of the multi-channel filtering architecture receives the original signal of the microfluidic sensing array, and the reference signal channel is connected to the environmental monitoring sensor network. The purified signal set is subjected to time marking processing, the time marking is synchronized with the global clock source, and the output signal is stored in a standardized data format. The process of identifying a drug delivery instance from the unified data stream starts the event feature pattern scanning mechanism, the event feature pattern is constructed based on multi-dimensional signal features, and the multi-dimensional signal features include signal amplitude mutation patterns, frequency spectrum characteristic changes, and waveform morphology distortion characteristics. The pattern matching algorithm adopts a dynamic time warping calculation method, which can compensate for the stretching and deformation on the time axis of the signal. The mutation point detection module monitors the first-order difference sequence of the signal, and the threshold value judgment of the first-order difference sequence uses an adaptive threshold value algorithm, which adjusts the threshold value size according to the historical statistical characteristics of the signal. The consistency verification link of the mutation point and the drug delivery logic introduces multi-condition judgment logic, which checks the signal amplitude change rate, event duration, and spatial position correlation. The time stamp recording function uses a high-precision clock chip, and the time synchronization signal of the high-precision clock chip is derived from a satellite navigation system. The timestamp sorting operation is realized using a quicksort algorithm, and the timestamp deduplication function is based on the accurate comparison of timestamp values, and the merging of repeated timestamps uses a time window matching method.

[0061] The spatial position descriptor generation module retrieves position information data from the data segment associated with the event feature pattern, the position information data is derived from a positioning sensor unit embedded in the microfluidic sensing array, the positioning sensor unit adopts a combination scheme of ultrasonic ranging module and inertial measurement unit. The spatial decoding algorithm performs a three-dimensional coordinate solving process, the three-dimensional coordinate solving process is based on a multi-sensor data fusion algorithm, the multi-sensor data fusion algorithm adopts an extended Kalman filtering method. The three-dimensional spatial point clustering analysis uses a density clustering algorithm, the density clustering algorithm can identify dense regions in the spatial point cloud. The outlier rejection module applies a statistical outlier detection technique, the statistical outlier detection technique calculates the distance distribution characteristics of each data point and the adjacent points. The robust spatial position descriptor output is a spatial coordinate sequence with a confidence index, the confidence index is derived from the residual analysis result of the coordinate solving. The drug classification identifier determination process starts the drug feature code query mechanism, the drug feature code is derived from the signal spectrum analysis feature vector, the signal spectrum analysis feature vector is obtained by fast Fourier transform calculation. The classification dictionary stores the mapping relationship between drug types and signal features, the update and maintenance of the classification dictionary is realized through a machine learning model, the machine learning model adopts a support vector machine classification algorithm. The drug classification identifier allocation process uses a unique identifier generation algorithm, the unique identifier generation algorithm ensures that each drug delivery instance has an independent classification label. The instance record storage module packages the time identifier, spatial position descriptor and drug classification identifier into a complete data structure, the data structure is stored in binary format to optimize the storage space.

[0062] The detailed implementation of the event feature mode scanning includes a signal cache management mechanism, the signal cache management mechanism adopts a ring buffer structure, and the size of the ring buffer is dynamically adjusted according to the signal sampling rate. In the dynamic time warping implementation of the pattern matching algorithm, the warping path constraint condition limits the maximum proportion of time scaling, and the maximum proportion of time scaling is set to plus or minus 20%. The Sobel operator convolution calculation of the edge detection uses a separate convolution optimization method, and the separate convolution optimization method reduces the computational complexity. The multi-condition check of the consistency verification is integrated in the rule engine, the rule engine loads a configurable verification rule set, and the verification rule set includes time constraint conditions, space constraint conditions and signal feature constraint conditions. The coordinate calculation link in the space position descriptor generation adopts sensor calibration parameters, the sensor calibration parameters are obtained through a precise calibration process, and the precise calibration process is performed on a reference point with a known coordinate. The density clustering algorithm parameters of the clustering analysis are set based on the features of oral anatomical structure, and the minimum neighborhood radius parameter is determined according to the tooth spacing. The statistical method of the outlier rejection adopts multi-dimensional outlier detection, and the multi-dimensional outlier detection considers the space coordinate value and the signal intensity value at the same time. The matching algorithm implementation of the drug feature code query realizes similarity calculation, and the similarity calculation uses the cosine similarity measurement method. The mapping rule update mechanism of the classification dictionary introduces an incremental learning function, and the incremental learning function allows the system to gradually optimize the classification accuracy during the running process.

[0063] The positioning sensor unit hardware implementation of the microfluidic sensor array adopts miniaturized design, which integrates the ultrasonic transducer and the inertial measurement unit in a millimeter-level package. The extended Kalman filter model of the spatial decoding algorithm includes a state prediction equation and a measurement update equation. The state prediction equation describes the sensor motion model, and the measurement update equation fuses multi-source observation data. The three-dimensional spatial point smoothing filter adopts a Kalman filter framework, and the state variables of the Kalman filter framework include position coordinates and motion velocity. The algorithm selection of cluster analysis is based on the point cloud density characteristics. The high-density area adopts a grid clustering method, and the low-density area adopts a hierarchical clustering method. The algorithm implementation of outlier rejection includes a noise point marking function, which marks suspicious data points as a to-be-verified state. The logic of drug classification identifier assignment includes a conflict resolution mechanism, which handles feature code matching uncertainty. The database module of instance record storage adopts a time series database engine, which optimizes time range query performance. The implementation of the ring buffer of the signal cache management mechanism includes an overflow protection function, which triggers a data transfer operation when the buffer is full. The calculation optimization of the pattern matching algorithm adopts multi-thread parallel processing, which allocates signal segments to different processor cores. The threshold adaptive algorithm of mutation point detection uses a sliding window statistical method to calculate the mean and variance of the signal in the window. The rule engine of consistency verification supports dynamic rule loading, which allows remote updating of verification logic. The sensor data preprocessing of spatial coordinate solving includes temperature compensation, which corrects the temperature influence on the ultrasonic propagation speed. The result output of cluster analysis includes cluster center coordinates and cluster radius parameters, which are used to define the drug delivery area range. The signal processing flow of drug feature code extraction includes wavelet transform analysis, which provides signal time-frequency localization features. The storage of the classification dictionary uses a distributed database architecture, which ensures high availability of data access. The data structure design of instance records includes a data compression function, which reduces storage space occupation and improves transmission efficiency.

[0064] Referring to Figure 4 In the adaptive filtering and original signal energy attenuation characteristic comparison analysis, the energy attenuation curve is drawn based on multi-channel signal processing in a logarithmic coordinate system. In specific operations, the energy curve of the original signal in the low frequency band (10 0The low-pass filter curve exhibits a normalized energy value close to 1.0 (-10¹Hz), showing a fluctuating attenuation trend with increasing frequency, reflecting the fundamental spectral characteristics of the signal. An energy valley of 0.4 appears near 50Hz, indicating the design limitation of its fixed cutoff frequency. The adaptive filter curve shows a steeper attenuation slope in the high-frequency range (>30Hz) and multiple energy oscillation nodes in the 20-40Hz interval, characterizing the dynamic suppression mechanism of the adaptive algorithm against non-stationary noise, achieving noise suppression through real-time adjustment of filter parameters. During parameter configuration, the horizontal axis frequency uses a logarithmic scale (10¹⁰). 0 The normalized energy range on the vertical axis is set to 0 to 1.0, and the grid spacing is 0.2 units to enhance readability. This figure, calculated using the area under the curve integral, verifies that adaptive filtering improves high-frequency noise suppression efficiency by 15%-20% compared to traditional low-pass filtering while preserving effective signal bandwidth.

[0065] Example 3: Scanning event feature patterns in the unified data stream is implemented using a pattern matching algorithm. This algorithm is based on dynamic time warping, which can handle nonlinear deformations of signals along the time axis. The event feature patterns are defined by typical signal characteristics generated during drug delivery, including step changes in electrochemical signals, abrupt changes in reflectivity of optical signals, and pulse characteristics of pressure signals. The implementation of the pattern matching algorithm includes a signal preprocessing step, which divides the unified data stream into fixed-length analysis windows, with the window length set to 256 sampling points. Dynamic time warping calculates the minimum cumulative distance between the current window signal and the reference template signal, using a dynamic programming algorithm. A mutation point detection module monitors local feature changes in the signal, which are enhanced by calculating the first and second differences of the signal. Consistency verification between mutation points and drug delivery logic employs a multi-level verification strategy, including time correlation checks, spatial consistency verification, and signal feature matching degree evaluation.

[0066] The dynamic time-warped distance calculation in pattern matching algorithms uses the following mathematical expression:

[0067] in: Indicates the current cumulative distance. This represents the Euclidean distance between the current pair of points. and These represent the time indices of the test signal and the reference template signal, respectively. This recursive formula uses dynamic programming to find the optimal normalized path, and the cumulative distance value of the optimal normalized path is used for similarity judgment. When the cumulative distance is lower than a set threshold, an event feature pattern is considered to have been detected.

[0068] The mutation point detection employs a multi-scale analysis method, which uses a Gaussian difference pyramid to detect local extreme points of the signal. The first-order difference calculation employs a central difference formula, which provides an accurate estimate of the derivative. The second-order difference calculation highlights the curvature changes of the signal, and the curvature change regions often correspond to drug delivery events. An adaptive threshold algorithm adjusts the detection sensitivity according to the background noise level of the signal, which is estimated by the historical statistics of the signal. The time correlation check of the consistency verification ensures that the mutation points appear within the expected drug delivery time window, and the spatial consistency verification confirms that the detection results of multiple sensors have geometric reasonableness. The process of retrieving the position information from the data segments associated with the event feature pattern initiates the spatial decoding process, which processes the multi-source positioning data collected by the microfluidic sensing array. The original readings of the position sensor include ultrasonic time-of-flight data, acceleration and angular velocity measurements of the inertial measurement unit, and direction data of the magnetometer. The spatial decoding algorithm employs a sensor fusion framework, which integrates multi-source observation data through an extended Kalman filter. The three-dimensional coordinate calculation is based on the geometric positioning principle, which uses the intersection positioning method of ultrasonic ranging. The cluster analysis applies a density clustering algorithm, which can discover natural groupings in spatial data. The outlier rejection uses a statistical distribution-based outlier detection method, which calculates the Mahalanobis distance of each data point from the cluster center.

[0069] The specific implementation of the spatial decoding algorithm includes a coordinate transformation step that converts the measurements in the sensor local coordinate system to the global coordinate system. The state vector of the extended Kalman filter contains position, velocity, and attitude parameters, and the prediction of the state vector is based on the numerical integration of inertial measurement data. The observation update stage fuses ultrasonic ranging data and magnetometer direction data, and the observation update corrects the state estimate by minimizing the difference between the prediction and the observation. The geometric positioning of the three-dimensional coordinate solution uses least squares optimization, which reduces the influence of measurement errors. The density clustering algorithm of the cluster analysis uses a grid-based fast implementation that divides the space into a uniform grid of voxels. The Mahalanobis distance calculation of the outlier rejection considers the covariance structure of the data, which reflects the correlation between dimensions. The preprocessing of the position sensor data includes a signal quality assessment step that checks the integrity and reasonableness of the data. The ultrasonic time-of-flight data is corrected by temperature compensation, which adjusts the sound speed parameter according to the ambient temperature. The integration error of the inertial measurement unit data is corrected by zero velocity update, which resets the velocity error in the stationary phase. The calibration of the magnetometer data eliminates hard and soft iron interference, and the hard iron interference compensation is achieved by offset correction. The real-time implementation of the spatial decoding algorithm uses incremental calculation, which reduces the computational complexity and improves the processing speed. The result optimization of the cluster analysis uses hierarchical clustering refinement, which merges small clusters that are over-segmented. The threshold selection of the outlier rejection is based on statistical significance testing, which uses the chi-square distribution to determine the critical value.

[0070] The extraction of the drug signature code uses a time-frequency analysis method that simultaneously captures the time and frequency domain features of the signal. The fast Fourier transform calculates the power spectral density of the signal, which shows the frequency distribution characteristics of the signal. The wavelet transform analysis provides a multi-resolution signal representation that is suitable for analyzing non-stationary signals. The feature vector construction combines time domain statistics and frequency domain descriptors, including mean, variance, and skewness, and frequency domain descriptors include spectral centroid and bandwidth parameters. The construction of the classification dictionary uses a supervised learning method that uses labeled training data to establish the mapping relationship between features and categories. The support vector machine classifier finds the optimal classification hyperplane that maximizes the interval between classes. The signal processing flow of the drug signature code extraction includes dimension reduction operations that use principal component analysis to reduce the feature dimension. The feature selection algorithm evaluates the contribution of each feature to classification, and the contribution evaluation is based on the information gain criterion. The update mechanism of the classification dictionary supports online learning, which allows the system to gradually improve the classification performance. The kernel function selection of the support vector machine classifier uses the radial basis function, which handles non-linear separable problems. The confidence calculation of the classification result is based on the distance from the point to the hyperplane, which reflects the degree of certainty of the classification.

[0071] The event detection sensitivity of the microfluidic sensor array is configurable, achieved by adjusting the detection threshold. The reference template library of the pattern matching algorithm supports dynamic updating, which optimizes the templates based on new observation data. The multi-scale analysis of the mutation point detection employs a pyramid structure, which provides signal representations at different resolutions. The rule set of the consistency verification is customizable, adapting to different drug delivery scenarios. The coordinate reference system definition of the spatial decoding algorithm is based on a standard oral model, which provides anatomical structure benchmarks. The parameter adaptive adjustment of the clustering analysis optimizes automatically based on point cloud density. The algorithm robustness of the outlier rejection is enhanced, achieved by employing robust statistical methods. The standardization of the drug feature code eliminates dimensional influence, making different features comparable. The version management of the classification dictionary ensures consistency, recording the modification history of the dictionary. The computational optimization of the event feature mode scanning uses parallel processing techniques, which distribute signal segments to multiple processing units. The acceleration of the pattern matching algorithm employs an early termination strategy, which terminates the search early when the cumulative distance exceeds the threshold. The real-time performance of the mutation point detection is guaranteed by a sliding window mechanism, which avoids repeated calculations of processed data. The parallel checking of the consistency verification reduces processing delays, verifying multiple conditions simultaneously. The accuracy of the spatial decoding algorithm is improved by sensor calibration, which compensates for individual differences and installation errors. The computational efficiency of the clustering analysis is improved by spatial indexing, which accelerates the search for neighboring points. The batch processing of the outlier rejection optimizes memory usage, processing multiple data points at once. The algorithm selection of the drug feature code extraction is based on signal characteristics, considering computational complexity and feature discriminability. The compressed storage of the classification dictionary reduces memory usage, using dictionary encoding techniques.

[0072] Embodiment 4: Refer to Table 1, the work of dividing time intervals with example time identifier sequence starts from the output data of the receiving instance identification and feature extraction module, the example time identifier sequence is a set of time stamps arranged in chronological order, each time stamp marks the exact time when a drug delivery instance is detected. The time interval division adopts the equal division strategy with fixed length, the equal division strategy divides the entire observation time period into continuous and non-overlapping time windows, the duration length of each time window is pre-set according to the drug delivery kinetics characteristics. The boundary of the time window is aligned to the integral second scale of the absolute time coordinate, this alignment ensures the regularity of the time interval and avoids the boundary ambiguity. The process of aligning the spatial position descriptor to the standard oral model involves coordinate system transformation, the standard oral model is a digitized model containing the three-dimensional geometric information of teeth, gums, tongue and oral mucosa surface. The local coordinate system is a follow-up coordinate system attached to the microfluidic sensing array, its origin is usually defined at the geometric center of the array. The global coordinate system is the world coordinate system fixed on the standard oral model, its origin is defined at the anatomical center point of the oral cavity. The transformation parameter calculation needs to solve the rotation matrix and translation vector from the local coordinate system to the global coordinate system.

[0073] Table 1: Corresponding relationship of feature points for coordinate system transformation parameter calculation

[0074]

[0075] The point cloud registration algorithm realizes the accurate alignment between the local coordinate system and the global coordinate system by using the iterative closest point method, which finds the optimal rigid transformation to minimize the average distance between the two point sets through iterative calculation. For the spatial position descriptor in each time interval, the calculated transformation parameters are used to re-project its coordinate values from the local coordinate system to the global coordinate system of the standard oral model. The projection accuracy verification is evaluated by calculating the re-projection error, which is the Euclidean distance between the projected point and the nearest point on the standard oral model. The drug delivery path time sequence record organizes these spatial position points after coordinate system transformation in chronological order, each record point is attached with the corresponding drug delivery instance attribute information. The preprocessing of the example time identifier sequence includes the normalization of time stamps, which uniformly converts different sources of time information into milliseconds calculated from the start of the experiment. The equal-length time block segmentation algorithm traverses the entire normalized time stamp sequence to divide continuous time intervals with fixed interval length. The time interval boundary processing adopts the upward rounding strategy to ensure that each time stamp is uniquely attributed to a specific time interval. The data structure of the spatial position descriptor contains three-dimensional coordinate values and direction vectors, which describe the local posture information during drug delivery. The definition of the global coordinate system of the standard oral model follows the medical imaging standards, and the coordinate system axis is consistent with the human anatomy orientation.

[0076] The detailed steps of the transformation parameter calculation starts from the selection of corresponding point pairs, which are anatomical landmark points that can be clearly identified in both the local coordinate system and the global coordinate system of the standard dental model. Table 1 lists the correspondence of four typical feature points used to calculate the transformation parameters. Based on these corresponding point pairs, the least squares method is used to solve the optimal rotation matrix R and translation vector T, so that the objective function is minimized. The iterative process of the iterative closest point method includes four steps: finding corresponding points, calculating transformations, applying transformations, and evaluating errors. The loop is executed until the average registration error is less than the set threshold or the maximum number of iterations is reached. The reprojection error is evaluated by calculating the distance between each projection point and the model surface, and the distance calculation uses a spatial search algorithm to speed up. The construction of the time series record of the drug delivery path uses a time series database model, which efficiently stores and retrieves path point data indexed by time. Each path point data record contains a timestamp field, X coordinate, Y coordinate, Z coordinate field in the global coordinate system, drug classification identifier field, delivery dose estimate field and confidence field. The storage format of the time series record uses columnar storage to optimize query performance, which facilitates time range queries and spatial area queries. The flexibility of time interval division is reflected in the configurable time window length, which allows adjustment according to different drug delivery rates. For fast-release drugs, a shorter time window is used to capture details, while for slow-release drugs, a longer time window is used to observe overall trends. The alignment accuracy of the spatial position descriptor is affected by the installation stability of the microfluidic sensing array, which is ensured by the flexible fitting design of the array substrate and the fixing device. The personalized adaptation of the standard dental model is achieved by acquiring the actual geometry of the user's mouth through three-dimensional scanning technology, and the standard model is deformed non-rigidly to match individual characteristics. The robustness of the point cloud registration algorithm is enhanced by introducing the random sample consensus algorithm, which effectively eliminates the influence of false matching point pairs.

[0077] The projection accuracy verification step initiates the adjustment process when the error exceeds the tolerance, and the adjustment process improves the registration result by locally optimizing the transformation parameters or introducing more corresponding point pairs. The data structure of the drug delivery path time record supports dynamic updating, which allows the historical path to be corrected after obtaining more accurate positioning data. The visualization of the time record superimposes the path points on the three-dimensional oral model and encodes different colors to represent different drug types or time information. The local coordinate system of the microfluidic sensor array is calibrated before each use, and the coordinate system calibration process requires the user to perform a series of standard actions to collect feature point data with known relative positions. The global coordinate system of the standard oral model maintains a transformation parameter lookup table that stores transformation parameters corresponding to different user individuals to enable fast switching. The computational efficiency of the iterative closest point method is optimized by the spatial partition tree data structure, which accelerates the nearest point search process. The compressed storage of the drug delivery path time record uses a lossy compression algorithm that reduces storage space occupancy while maintaining path shape features. The output of the entire path generation module is a structured drug delivery path time record file, which uses an open standard format to store data for easy exchange with other medical information systems. The analysis functions of the path data include path length calculation, speed analysis, and regional coverage statistics.

[0078] Example 5: Operation of category aggregation according to drug delivery path time record and drug classification identifier. The drug delivery path time record contains a large amount of spatiotemporal information of individual delivery instances, each with a unique drug classification identifier. The system reads the path time record database and groups instances with the same identifier into the same set according to the numerical value of the drug classification identifier. For example, identifier "A001" represents sodium fluoride anticaries gel, and identifier "B002" represents chlorhexidine antibacterial mouthwash. The system will create two separate group containers to store all delivery instances of these two types of drugs. After grouping, multiple category subsets are formed, each containing all delivery locations and dosage data of a specific drug within a certain time period. The definition of the target oral region is based on the anatomical partition of the standard oral model, which divides the oral cavity into multiple quantifiable regions, such as the upper anterior teeth region, the lower left molar region, the tongue dorsum region, etc. Each region has clear geometric boundaries in the model, which are defined by a set of three-dimensional mesh vertices. To calculate the drug distribution density, the surface area of the target region needs to be determined. The system calculates the sum of the areas of all triangles that make up the region mesh to obtain the exact surface area value. For example, the surface area calculation of the lower left molar region will accumulate the areas of all small triangles covering the region, resulting in an area value in square millimeters.

[0079] For each category subset, the system calculates the delivery frequency within the target oral region, which refers to the number of drug delivery events per unit time. The system iterates through each instance in the category subset and checks whether its spatial coordinates fall within the geometric boundary of the target region. If the coordinates of an instance are determined to be inside the target region by ray-casting, the instance is counted as an effective delivery. For example, if the sodium fluoride anticaries gel is detected 15 effective deliveries in the dorsal tongue region within a ten-minute observation period, the delivery frequency is 1.5 per minute. The estimation of drug amount is based on the dose information recorded for each delivery instance, which comes from the quantitative measurement of drug concentration by the microfluidic sensing array. The system aggregates the dose values of all effective delivery instances within the target region to obtain the total drug amount. For example, the dose values of the 15 delivery instances are 0.2 microliter, 0.25 microliter,..., and the sum is 3.75 microliter. The density value calculation combines the delivery frequency or drug amount with the area of the region, and the distribution density can be expressed as either frequency density or mass density. The system performs a division operation to divide the total delivery frequency or total drug amount by the surface area of the target region. For example, if the surface area of the dorsal tongue region is 500 square millimeter and the total drug amount of the sodium fluoride anticaries gel is 3.75 microliter, the mass density is 0.0075 microliter per square millimeter. The change of density value over time is captured by a sliding time window, and the system divides the entire observation time axis into consecutive, possibly overlapping time segments, and repeats the above statistical and calculation processes independently for each time segment. For example, with a window length of one minute and a sliding step of 30 seconds, the density value is calculated for each one-minute window, resulting in a sequence of density values over time.

[0080] Generating the efficacy evaluation dataset requires aggregating density values of all categories, all regions, and all time points. The dataset is stored in a multi-dimensional table structure. The rows of the table represent different time points, and the columns represent different combinations of drug categories and oral regions. Each cell is filled with the corresponding density value. For example, at time point T1, the frequency density of sodium fluoride anticaries gel in the maxillary anterior tooth region is 0.002 times per minute per square millimeter, and the mass density of chlorhexidine antibacterial mouthwash in the tongue dorsum region is 0.005 microliters per square millimeter. The efficacy evaluation dataset also contains metadata that describes the generation parameters, time range, region definition, and drug classification criteria of the dataset. The hash grouping algorithm for category aggregation uses linear probing to solve hash collisions, which ensures high grouping performance even with a large number of instances. The geometric boundary detection of the target oral region uses the point-in-polyhedron test algorithm, which accurately determines the relationship between the point and the polygonal face. The surface area calculation uses the Gauss divergence theorem to convert the surface area integral into a linear integral, which improves the calculation efficiency, especially for complex curved surfaces. The dose aggregation operation uses high-precision floating-point accumulation, which uses the Kahan summation algorithm to reduce rounding errors. The density calculation introduces normalization to prevent numerical overflow, which scales the density value to a reasonable numerical range. The time series aggregation uses the aggregation query function of the database, which optimizes the query speed using indexes.

[0081] The query interface of the drug delivery path time sequence record supports filtering by time and spatial range, allowing the efficacy evaluation module to quickly retrieve relevant instances. The partition information of the standard oral model is stored in a configuration file, which supports dynamic loading and modification to adapt to different clinical areas of interest. The visualization module of the distribution density maps the density value to color coding, which is rendered on the three-dimensional oral model to form a heat map. The export format of the efficacy evaluation dataset supports CSV and JSON standards, which facilitates further processing by other data analysis tools. The system records detailed operation logs during processing, which are used to track each step of data conversion and possible error sources. The category subset is stored in memory using a linked list structure, which facilitates dynamic insertion and deletion of instance data. The boundary box of the target region is precomputed to accelerate spatial queries, which establishes a spatial index to achieve fast preliminary screening. The smoothing of the density time series uses a moving average filter to eliminate random fluctuations and highlight trend changes. The version management of the efficacy evaluation dataset records the parameters and results of each generation, supporting the reproducibility and comparative analysis of the results. The final dataset is transmitted to the central storage server through an encrypted channel, which ensures the security and privacy of patient data.

[0082] Referring to Figure 5The cumulative dose data output by the efficacy evaluation module is visualized by a stacked column chart. In the specific operation, the horizontal axis of the chart is marked with six standard oral anatomical regions, including maxillary anterior teeth region, maxillary posterior teeth region, mandibular anterior teeth region, mandibular posterior teeth region, tongue dorsum region and buccal mucosa region, and the vertical axis is scaled with total dose (microliters). The column of each region is composed of the dose values of the four types of drugs in order: the bottom layer is compound chlorhexidine gargle, followed by sodium fluoride anticaries gel, chlorhexidine antibacterial mouthwash and metronidazole oral ulcer gel on the top layer. The color blocks clearly represent the distribution differences corresponding to the drug classification identifiers. After the time sequence record of the drug delivery path is aggregated by categories, the total dose quantifies the cumulative amount of drugs delivered in each region. The data shows that the total dose of the buccal mucosa region and the maxillary anterior teeth region is more than 100 microliters, indicating that these regions are high-frequency targets for drug delivery, while the total dose of the tongue dorsum region is the lowest (50 microliters), revealing spatial heterogeneity of distribution.

[0083] While embodiments of the application have been shown and described, it is to be understood that the embodiments described are merely exemplary and that changes in form and detail can be made without departing from the spirit and scope of the application. The scope of the application is defined by the appended claims and their equivalents.

Claims

1. A microfluidic-based oral drug precision delivery and efficacy evaluation system, characterized in that, The system comprises: a signal acquisition and processing module for continuously acquiring multi-dimensional signals of the drug delivery process in the oral cavity through a microfluidic sensing array, correcting and structurally normalizing the signals in real time, eliminating signal distortion, and integrating unified data flow; an instance recognition and feature extraction module for recognizing drug delivery instances from the unified data flow, extracting instance time identifiers, spatial position descriptors, and drug classification identifiers; a path generation module for dividing time intervals using instance time identifier sequences, aligning spatial position descriptors to a standard oral cavity model, and generating drug delivery path time records; an efficacy evaluation module for aggregating delivery instances by category according to the drug delivery path time records and drug classification identifiers, calculating drug distribution density in the target oral cavity area, and generating efficacy evaluation data sets.

2. The microfluidic-based oral drug precise delivery and efficacy evaluation system according to claim 1, wherein, The continuously acquiring multi-dimensional signals of the drug delivery process in the oral cavity through a microfluidic sensing array, correcting and structurally normalizing the signals in real time, eliminating signal distortion, and integrating unified data flow comprises: The microfluidic sensing array generates multi-source signal flow, which is converted into digital format through signal conversion processing, the signal level is adjusted using amplitude normalization algorithm to make the signal dimension consistent; The dimension-consistent signal flow is structurally integrated, the format differences of different signal sources are identified, converted into standard data templates, and the noise components are removed using filtering algorithm to generate a purified signal set; Based on the time reference point, the purified signal set is time-synchronized, time delay is compensated, multi-source data is fused, and unified data flow is output.

3. The microfluidic-based oral drug precision delivery and efficacy evaluation system according to claim 2, wherein, The application of amplitude normalization algorithm to adjust the signal level to make the signal dimension consistent comprises: Detecting the peak and valley values of the multi-source signal flow, calculating the dynamic range, mapping the signal to a common amplitude interval using a scaling function, and realizing signal standardization; Baseline correction is applied to the standardized signal to remove DC offset and ensure signal zero alignment.

4. The microfluidic-based oral drug precision delivery and efficacy evaluation system according to claim 2, wherein, The structural integration of the dimension-consistent signal flow, the identification of the format differences of different signal sources, the conversion into standard data templates, and the removal of noise components using filtering algorithm to generate a purified signal set comprises: Parsing the data header information of the dimension-consistent signal flow, extracting the format descriptor, matching the field definition of the standard data template, and performing format conversion operation; A low-pass filter is applied to suppress high-frequency noise, and an adaptive filter is used to eliminate environmental interference to output the purified signal set.

5. The microfluidic-based oral drug precision delivery and efficacy evaluation system according to claim 1, wherein, The identification of drug delivery instances from the unified data flow, the extraction of instance time identifiers, spatial position descriptors, and drug classification identifiers comprises: Scanning the event feature pattern in the unified data flow, detecting the instantaneous point of drug delivery occurrence, and recording the time identifier; Retrieving position information from the data segment associated with the event feature pattern, analyzing three-dimensional coordinates through spatial decoding algorithm, and generating spatial position descriptors; Querying the drug feature code, determining the drug type by referring to the classification dictionary, and assigning the drug classification identifier.

6. The microfluidic-based oral drug precision delivery and efficacy evaluation system according to claim 5, wherein, The scanning of the event feature pattern in the unified data flow, the detection of the instantaneous point of drug delivery occurrence, and the recording of the time identifier comprises: Using a pattern matching algorithm to identify the mutation point in the signal, verifying the consistency of the mutation point with the drug delivery logic, and labeling the timestamp; further sorting and removing the timestamp.

7. The microfluidic-based oral drug precision delivery and efficacy evaluation system according to claim 5, wherein, The position information is retrieved from the data segment associated with the event feature pattern, the three-dimensional coordinates are analyzed by a spatial decoding algorithm, and a spatial position descriptor is generated, including: Extracting the original readings of the position sensor, converting the readings into three-dimensional space points by using a coordinate calculation algorithm; Performing cluster analysis on the three-dimensional space points, eliminating abnormal points, and generating robust spatial position descriptors.

8. The microfluidic-based oral drug precision delivery and efficacy evaluation system according to claim 1, wherein, The time interval is divided by using the instance time identifier sequence, and the spatial position descriptor is aligned to the standard oral model to generate the drug delivery path time record, including: According to the distribution of the instance time identifier, the data is divided into equal length time blocks to form continuous time intervals; For each time interval, the spatial position descriptor is transformed from the local coordinate system to the global coordinate system of the standard oral model, and the point-to-point mapping is realized by using a registration algorithm; Organize the transformed position data in chronological order, add drug delivery attributes, and construct the drug delivery path time record.

9. The microfluidic-based oral drug precision delivery and efficacy evaluation system according to claim 8, wherein, The spatial position descriptor is transformed from the local coordinate system to the global coordinate system of the standard oral model, and the point-to-point mapping is realized by using a registration algorithm, including: Load the coordinate system definition of the standard oral model, calculate the transformation parameters between the local coordinate system and the global coordinate system; Apply the transformation parameters to re-project each spatial position descriptor into the global coordinate system, verify the projection accuracy, and adjust the error.

10. The microfluidic-based oral drug precision delivery and efficacy evaluation system according to claim 1, wherein, According to the drug delivery path time record and the drug classification identifier, the delivery instances are aggregated by category, the drug distribution density in the target oral area is calculated, and the therapeutic effect evaluation data set is generated, including: According to the drug classification identifier, the drug delivery path time record is grouped into multiple category subsets; For each category subset, the delivery frequency and the amount of drug in the target oral area are calculated, and the density value is calculated in combination with the area; Aggregate the density value over time to generate the therapeutic effect evaluation data set.

Images