A visual data processing system applied to traditional Chinese medicine pediatrics

Abstract

The present application relates to the technical field of data management, and more particularly to a visual data processing system applied to traditional Chinese medicine pediatrics. The system comprises the following modules: a data acquisition module for acquiring pediatric physiological time series data sets; a physiological event anchor point identification module for identifying the physiological event anchor points of the pediatric physiological time series data sets to obtain a set of physiological event anchor points; a time calibration module for performing noise suppression on the pediatric physiological time series data sets according to the set of physiological event anchor points to obtain a set of noise-suppressed physiological time series data; a time calibration module for performing time calibration on the set of noise-suppressed physiological time series data to obtain a set of time-corrected physiological data; a feature extraction module for extracting a set of tongue image sets from the set of time-corrected physiological data; and a tongue feature vector set generation module for generating a set of tongue feature vectors based on the set of tongue image sets. The present application can effectively solve the dynamic correlation analysis problem caused by the multi-dimensional sign data heterogeneity in the processing of traditional Chinese medicine pediatric clinical data.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of data management, and particularly relates to a visual data processing system applied to traditional Chinese pediatric medicine. BACKGROUND

[0002] Early data processing of traditional Chinese pediatric medicine mainly relies on manual recording and simple text description, and it is difficult to systematically manage and deeply analyze complex physiological sign data. With the advancement of medical equipment technology, multi-modal detection technologies such as tongue coating microscopic imaging, pulse waveform detection, and thermal imaging of the body surface have gradually been applied to clinical practice in traditional Chinese pediatric medicine, providing rich data sources for traditional Chinese medicine. However, some problems have been exposed in the development of these technologies, which need to be solved urgently.

[0003] Existing data processing systems for traditional Chinese pediatric medicine often have difficulty in achieving effective dynamic correlation analysis when dealing with multi-dimensional sign data due to the heterogeneity of the data. Traditional visualization systems can usually only display different modalities of data such as tongue coating microscopic images, pulse waveform signals, and body constitution recognition scales in isolation, and lack the ability to construct time-series correlation composite views. For example, in the tracking scenario of children with allergic purpura, it is necessary to observe the distribution of thermal imaging of limb ecchymosis, the microcirculation state of sublingual collateral vessels, and the change trend of pulse slip number. However, due to the time stamp deviation of different detection devices, the system cannot automatically align the collection time nodes of multi-source data, leading to time domain misalignment when comparing the evolution of tongue features and the fluctuation of pulse parameters, which affects the accurate grasp of the development process of the child's condition. SUMMARY

[0004] Therefore, it is necessary to provide a visual data processing system applied to traditional Chinese pediatric medicine to solve at least one of the above technical problems.

[0005] To achieve the above-mentioned purpose, a visual data processing system applied to traditional Chinese pediatric medicine includes the following modules:

[0006] A data acquisition module is configured to acquire a pediatric physiological time series data set, perform physiological event anchor point identification on the pediatric physiological time series data set, and obtain a physiological event anchor point set;

[0007] A time calibration module is configured to perform noise suppression on the pediatric physiological time series data set according to the physiological event anchor point set, obtain a noise-suppressed pediatric physiological time series data set, and perform time calibration on the noise-suppressed pediatric physiological time series data set, to obtain a time-corrected pediatric data set;

[0008] The feature extraction module is used for extracting a tongue image set from the time-corrected physiological data set, generating a tongue feature vector set based on the tongue image set, extracting a pulse signal data set from the time-corrected physiological data set, constructing a pulse characteristic parameter set based on the pulse signal data set, and extracting a thermal imaging image sequence from the time-corrected physiological data set.

[0009] The evolution index construction module is used for constructing a tongue time sequence index sequence based on the tongue feature vector set, generating a pulse time sequence index sequence based on the pulse characteristic parameter set, and generating a body surface temperature rhythm index sequence based on the body region temperature feature table.

[0010] The visualization module is used for constructing a pediatric data visual coding rule based on the tongue time sequence index sequence, the pulse time sequence index sequence and the body surface temperature rhythm index sequence, and visualizing the tongue time sequence index sequence, the pulse time sequence index sequence and the body surface temperature rhythm index sequence according to the pediatric data visual coding rule.

[0011] In the present application, the data acquisition module is used for comprehensively acquiring the pediatric physiological time sequence data set, and the physiological event anchor point recognition provides a key time reference point for subsequent processing. The time calibration module is used for noise suppression and time calibration based on the anchor points, so as to eliminate the timestamp deviation problem between different detection devices and ensure the consistency and accuracy of multi-source data in the time dimension. The feature extraction module is used for extracting the key feature information of tongue, pulse and thermal imaging from the calibrated data and converting it into a feature vector or parameter set for analysis. The evolution index construction module is used for further converting the feature information into a time sequence index sequence, which captures the dynamic change rule of pediatric physiological data over time. The visualization module is used for constructing a pediatric data visual coding rule according to the time sequence index sequence and visually displaying it, so that the evolution process of tongue, pulse and body surface temperature data can be clearly observed. In summary, the present application can effectively solve the dynamic correlation analysis problem caused by the heterogeneity of multi-dimensional sign data in the processing of TCM pediatric clinical data. BRIEF DESCRIPTION OF DRAWINGS

[0012] Other features, objects and advantages of the present application will become more apparent from the following detailed description when read in conjunction with the accompanying drawings:

[0013] Figure 1 FIG. 1 shows a module flow schematic diagram of a TCM pediatric visual data processing system according to an embodiment of the present application. DETAILED DESCRIPTION

[0014] The technical method of the present application will be described clearly and completely below in conjunction with the drawings. Obviously, the described embodiments are part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0015] In addition, the drawings are only schematic illustrations of the present application and are not necessarily drawn to scale. Identical reference signs in the drawings represent identical or similar parts, and thus repeated descriptions thereof will be omitted. Some of the block diagrams shown in the drawings are functional entities, which do not necessarily have to correspond to physically or logically independent entities. The functional entities can be implemented in the form of software, or in one or more hardware modules or integrated circuits, or in different network and / or processor methods and / or microcontroller methods.

[0016] It should be understood that although the terms "first", "second" and the like can be used herein to describe various elements, these elements should not be limited by these terms. These terms are only used to distinguish one element from another. For example, without departing from the scope of the exemplary embodiments, a first element can be referred to as a second element, and similarly a second element can be referred to as a first element. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0017] To achieve the above-mentioned purpose, please refer to Figure 1 The present application provides a visual data processing system applied to pediatric medicine of traditional Chinese medicine, comprising the following modules:

[0018] S1: a data acquisition module for acquiring a pediatric physiological time series data set; performing physiological event anchor point identification on the pediatric physiological time series data set to obtain a physiological event anchor point set;

[0019] S2: a time calibration module for performing noise suppression on the pediatric physiological time series data set according to the physiological event anchor point set to obtain a noise-suppressed physiological time series data set; performing time calibration on the noise-suppressed physiological time series data set to obtain a time-corrected physiological data set;

[0020] S3: a feature extraction module for extracting a tongue image set from the time-corrected physiological data set; generating a tongue feature vector set based on the tongue image set; extracting a pulse signal data set from the time-corrected physiological data set; constructing a pulse feature parameter set based on the pulse signal data set; extracting a thermal imaging image sequence from the time-corrected physiological data set; generating a body region temperature feature table based on the thermal imaging image sequence;

[0021] S4: an evolution index construction module, configured to construct a tongue appearance time sequence index sequence based on the tongue appearance feature vector set, generate a pulse appearance time sequence index sequence based on the pulse appearance characteristic parameter set, and generate a body surface temperature rhythm index sequence based on the body region temperature feature table;

[0022] S5: a visualization module, configured to construct a pediatric data visual coding rule based on the tongue appearance time sequence index sequence, the pulse appearance time sequence index sequence, and the body surface temperature rhythm index sequence, and visualize the tongue appearance time sequence index sequence, the pulse appearance time sequence index sequence, and the body surface temperature rhythm index sequence according to the pediatric data visual coding rule.

[0023] In the present application, the data acquisition module is used to comprehensively acquire the pediatric physiological time sequence data set, and the physiological event anchor point recognition provides a key time reference point for subsequent processing. The time calibration module is used to perform noise suppression and time calibration based on the anchor points, thereby eliminating the timestamp deviation problem between different detection devices and ensuring the consistency and accuracy of the multi-source data in the time dimension. The feature extraction module is used to extract the key feature information of the tongue appearance, pulse appearance, and thermal imaging from the calibrated data and convert the key feature information into feature vectors or parameter sets for analysis. The evolution index construction module is used to further convert the feature information into time sequence index sequences, which capture the dynamic change rule of the pediatric physiological data over time. The visualization module is used to construct a pediatric data visual coding rule according to the time sequence index sequences and visually display the pediatric data visual coding rule, so that the evolution process of the tongue appearance, pulse appearance, and body surface temperature data can be clearly observed. In summary, the present application can effectively solve the dynamic correlation analysis problem caused by the heterogeneity of multi-dimensional sign data in the processing of TCM pediatric clinical data.

[0024] Preferably, the acquisition of the pediatric physiological time sequence data set comprises:

[0025] The multi-source TCM pediatric detection device is used to acquire multi-dimensional physiological information of the child, and a multi-modal TCM pediatric physiological data set is obtained.

[0026] The multi-source TCM pediatric detection device is identified and coded, and a device identification code data table is obtained. A preset network time protocol server is used to synchronize and calibrate the clock of the multi-source TCM pediatric detection device, and a device reference time synchronization record is obtained.

[0027] The multi-modal TCM pediatric physiological data set is analyzed according to the device identification code data table and the device reference time synchronization record, and a device data stream meta-information table is obtained.

[0028] The timestamp field of the device data stream meta-information table is extracted and standardized, and an original acquisition timestamp sequence is obtained. The time deviation values between the detection devices are calculated based on the original acquisition timestamp sequence, and a device time difference parameter table is obtained.

[0029] The computing device time difference parameter table is a table of time difference parameters of a plurality of devices, and the time difference parameters are obtained by measuring the time difference between the time of the device and the reference time of the reference device.

[0030] According to the time difference stability judgment result, the reliability of the device time difference parameter table is quantified, and a time difference reliability evaluation result is obtained.

[0031] Based on the modified device time difference parameter table, a dynamic time warping is performed on a multi-modal pediatric physiological data set to obtain a pediatric physiological time series data set, wherein the pediatric physiological time series data set includes pulse waveform data, tongue coating microscopic image sequences, and body surface thermal imaging data.

[0032] In this example, in a child health clinic, a child patient was examined by a pulse detector, a tongue detector and a thermal imaging device to collect pulse waveform data, tongue image sequence and thermal imaging data. The pulse detector was an intelligent pulse detector-PM3000, which was attached to the child's radial artery on the wrist by a pressure sensor and continuously collected pulse waveform signals for 10 minutes at a sampling frequency of 1000 Hz. The signal amplitude range was 0-5V. The tongue detector was an intelligent tongue detector-TS2000, which had a built-in high-definition camera and a standardized light source. When capturing the child's tongue image, it captured a tongue image sequence with a resolution of 2048x1536 pixels at a shooting distance of 30 cm, 2 frames per second, and a total of 5 minutes of image data. The thermal imaging device was a medical-grade infrared thermal imaging device-TH9000, which had an infrared detector resolution of 640x480 pixels and a temperature detection range of 30-45°C. It collected thermal imaging data at a frequency of 1 frame per second at a distance of 1 meter from the child's body surface for 10 minutes. Through these three devices, a multi-modal pediatric physiological data set containing pulse waveform data, tongue image sequence and thermal imaging data was obtained and stored in the clinic's medical data server. A device identification code data table was established using Microsoft Excel to assign unique device identification codes to the pulse detector, tongue detector and thermal imaging device: PM-001, TS-002 and TH-003, respectively. The device information, including device model, purchase date and device serial number, was recorded in the table. At the same time, a dedicated time server configured with an NTP (Network Time Protocol) service was connected to the clinic's local area network. The server synchronized with the time source of the national time center through the Internet, with a time synchronization accuracy of milliseconds. The three detection devices were connected to the local area network through network cables and configured with NTP clients in their device management systems to synchronize and calibrate their clocks with the time server. After calibration, the device reference time synchronization record showed that the pulse detector had a time deviation of 1.2 milliseconds, the tongue detector had a time deviation of 0.8 milliseconds, and the thermal imaging device had a time deviation of 1.5 milliseconds, all within the acceptable synchronization accuracy range. The collected multi-modal pediatric physiological data set was imported into the data processing workstation, which ran the medical data management software Health Data manager. In the software, the data set was analyzed according to the previously established device identification code data table and device reference time synchronization record. In the Health Data manager software, the pulse data file was associated with the device identification code PM-001, the tongue image sequence file was associated with TS-002, and the thermal imaging data file was associated with TH-003. At the same time, the time synchronization calibration parameters of each device, such as the time deviation value, were also input into the software.The software automatically generates a device data stream meta-information table based on the information, which records in detail the device identification code, data type, acquisition start time, acquisition duration, data format, and time synchronization calibration parameter information corresponding to each data file. In the Health Data manager software, the software extracts the timestamp field of each data file in the device data stream meta-information table. For pulse data files, the software reads the timestamp field and finds that the time format is a relative time since the device is turned on (unit: seconds) with millisecond-level time accuracy. Using the built-in timestamp standardization conversion function module of the software, the relative time stamp is converted to an absolute time format conforming to the ISO 8601 standard (for example, 2025-02-20T14:30:25.123+08:00), which contains time zone information (+08:00 indicates East Eight Zone). Similarly, the timestamp extraction and conversion operations are performed on the tongue image sequence file and the body surface thermal imaging data file, and finally the original acquisition timestamp sequence file is obtained, which records the accurate acquisition time of each data point or image frame. Using the Health Data manager software, the time deviation values between the detection devices are calculated based on the original acquisition timestamp sequence. Taking the pulse detector as the reference, the time deviations between the tongue detector and the body surface thermal imager and the pulse detector are calculated. The software extracts the timestamp data of each device from the original acquisition timestamp sequence every 1 minute, and calculates the time deviation sequence between the tongue detector and the pulse detector, for example, in the first 5 measurements, the time deviations are 0.5 milliseconds, 0.6 milliseconds, 0.5 milliseconds, 0.4 milliseconds, and 0.5 milliseconds, respectively; the time deviation sequence between the body surface thermal imager and the pulse detector is 1.0 milliseconds, 1.2 milliseconds, 1.1 milliseconds, 1.3 milliseconds, and 1.0 milliseconds, respectively. The software calculates the time deviation standard deviation of the last 5 measurements of the two time deviation sequences, respectively, and for the tongue detector, the standard deviation is about 0.05 milliseconds; for the body surface thermal imager, the standard deviation is about 0.10 milliseconds. The preset stability threshold is set to 0.2 milliseconds, so the software judges that the time deviations between the tongue detector and the body surface thermal imager and the pulse detector are both in a stable state, and the time deviation stability judgment result shows that the time deviation stability of both is stable, and the result is stored in the device time difference parameter table. Based on the time deviation stability judgment result, the reliability of the device time difference parameter table is quantified in the Health Data manager software. For device time deviation data with stable stability judgment, the software gives a higher weight value (for example, the weight of the tongue detector is 0.8, and the weight of the body surface thermal imager is 0.7), and for data with poor stability (assuming there are other devices), the weight will be reduced accordingly. The software corrects the time deviation data in the original device time difference parameter table by weighted average based on these weight values.For example, for the tongue appearance detector, the time deviation correction value of the tongue appearance detector after weighted average correction is 0.52 milliseconds; for the body surface thermal imager, the corrected time deviation value is 1.06 milliseconds. Using the corrected equipment time difference parameter table, the multi-modal pediatric physiological data set is dynamically time-warped in the Health Data manager software. First, the software takes the time axis of the pulse waveform data collected by the pulse appearance detector as the reference, and adjusts the time axis of the tongue image sequence and the body surface thermal imaging data according to the corrected time difference parameter. For the tongue image sequence, the software adjusts the collection time of each image frame by 0.52 milliseconds forward or backward, so that the time axis of the image sequence is aligned with the time axis of the pulse waveform data. Similarly, the time of each image frame of the body surface thermal imaging data is adjusted according to the deviation of 1.06 milliseconds. After dynamic time warping, a unified time reference pediatric physiological time series data set is obtained, which includes calibrated pulse waveform data, tongue image sequence and body surface thermal imaging data.

[0033] The present application standardizes the data collection process from the source by registering and calibrating the unified equipment identification code of multi-source pediatric detection equipment, and ensures the consistency of the time reference of each device. Through detailed data source analysis and timestamp standardization conversion of multi-modal data set, the time deviation parameters between devices are accurately calculated, and the stability and reliability of the time deviation are further quantified, so that the system can accurately identify and correct the time deviation problem. By dynamically time-warping the original multi-modal data based on the corrected equipment time difference parameter table, a unified pediatric physiological time series data set is generated, effectively integrating pulse waveform, tongue image and body surface thermal imaging physiological information, providing a high-quality, time-consistent data basis for subsequent feature extraction and analysis, thereby significantly improving the efficiency and reliability of the entire pediatric data processing system.

[0034] Preferably, the physiological event anchor point recognition of the pediatric physiological time series data set comprises:

[0035] Real-time peak detection is performed on the pulse waveform data to obtain a pulse peak position sequence;

[0036] The adjacent peak time interval is calculated according to the pulse peak position sequence, and the pulse peak position sequence is screened for physiological effectiveness according to the adjacent peak time interval and the preset physiological cycle threshold range to obtain a standard pulse cycle marker point set;

[0037] The tongue image sequence is compared frame by frame based on the inter-frame difference method to obtain tongue image rapid change frame markers;

[0038] The region temperature gradient of the body surface thermal imaging data is calculated and marked to obtain a temperature sudden change region marker map.

[0039] Based on the temperature sudden change region marking map, the time points of the temperature sudden change region in the continuous frames of the body surface thermal imaging data are counted, and a sequence of thermal imaging event trigger time points is generated;

[0040] The pulse wave peak position sequence, the tongue image rapid change frame marking point and the corresponding event trigger point in the sequence of thermal imaging event trigger time points are recorded as an initial physiological event anchor point set;

[0041] Based on the pulse wave peak position sequence, the tongue image rapid change frame marking point and the sequence of thermal imaging event trigger time points, a cross-modal event time correspondence table is constructed;

[0042] Based on the cross-modal event time correspondence table, each event anchor point in the initial physiological event anchor point set is judged as follows:

[0043] If there are trigger events in at least two modalities within a ±200ms window, the corresponding event anchor point is marked as a high-confidence anchor point, otherwise, the corresponding event anchor point is excluded, and the high-confidence anchor point is recorded as a physiological event anchor point set.

[0044] In this embodiment, Health Signal Analyzer software is used for real-time analysis of pulse waveform data. The built-in peak detection algorithm in the software is based on sliding window technology, with a window length of 0.5 seconds and a step size of 0.1 seconds. When the pulse waveform data is input, the algorithm calculates the first derivative in each sliding window, finds the position where the derivative crosses zero and the waveform amplitude exceeds the baseline by 2 standard deviations as the candidate peak. Further filtering out the effective peaks with a peak interval of 0.5-1.5 seconds, the final pulse peak position sequence is obtained, which records the time stamp and amplitude corresponding to each peak, for example, the peaks are detected at positions 5 seconds, 10.2 seconds, 15.3 seconds, etc., with amplitudes of 2.3 mV, 2.1 mV, 2.4 mV, respectively. In Health Signal Analyzer software, the physiological cycle analysis function module of the software is used to filter the pulse peak position sequence obtained. The software preset physiological cycle threshold range is 0.4-1.6 seconds (corresponding to heart rate range 37.5-150 times / minute, consistent with the normal range of children's heart rate). The software first calculates the time interval of adjacent peaks according to the pulse peak position sequence, for example, the time intervals of the first three peaks are 5.2 seconds, 5.1 seconds. Then compared with the preset threshold, it is found that these intervals all exceed the upper threshold of 1.6 seconds, indicating that there is an abnormality. Further inspection found that the child had limb movement interference during the detection process, causing some peak detection deviation. Adjust the parameters, narrow the sliding window to 0.3 seconds, and re-detect and filter the peaks, finally get the standard pulse period marker set with adjacent peak time interval of 0.6-1.4 seconds, the marker time stamp is 8 seconds, 13.2 seconds, 18.5 seconds, etc., the corresponding heart rate is 43-95 times / minute, which is consistent with the normal heart rate fluctuation range of the child in a calm state. Using Tongue Image Analyzer software, the software analyzes image sequences based on the inter-frame difference method. In the software, the inter-frame difference threshold is set to 0.2 (normalized value), that is, when the gray difference between adjacent two frames of images exceeds this threshold, it is determined that the tongue image has changed. The software compares the tongue image sequence frame by frame, and detects that the inter-frame difference between the 12th and 13th frames is 0.25, which exceeds the set threshold, marking the 13th frame as a tongue image rapid change frame. Further inspection found that when the child slightly opened his mouth or the tongue moved, the tongue image would change, and the software marked 15 tongue image rapid change frames with time stamps distributed at positions 5 seconds, 14 seconds, 23 seconds, etc. in the acquisition sequence. Thermal Scan Processor software is used to analyze the patient's surface thermal imaging data. The software first preprocesses the thermal imaging data, including bad point correction and temperature calibration. In the temperature gradient calculation module, the spatial gradient threshold is set to 0.5°C / cm, and the time gradient threshold is set to 0.3°C / s.The software calculates the temperature difference between each pixel point and its 8 adjacent pixels. If the temperature difference exceeds 0.5℃ / cm, it is marked as a temperature abrupt change area. At the same time, the temperature change rate of the same pixel point in consecutive frames is compared. If it exceeds 0.3℃ / s, it is also marked as a temperature abrupt change area. After calculation, the software generates a temperature abrupt change area marker map, in which different colors represent temperature abrupt change areas. For example, temperature abrupt change appears in the umbilical region of the child's abdomen, indicating changes in gastrointestinal function. Temperature abrupt change is also detected on the forehead and neck, which is related to the fever state. Based on the temperature abrupt change area marker map, the ThermalScan Processor software further counts the time points of the temperature abrupt change area in the consecutive frames of the thermal imaging data. The software sets the minimum time for consecutive temperature abrupt changes to 3 frames (0.3 seconds) to avoid transient noise interference. When a temperature abrupt change area is detected in consecutive frames, the timestamp of the starting frame is recorded. For example, during the 7th-7.3rd second, the temperature abrupt change area appears continuously in the umbilical region of the child's abdomen, and the software marks the 7th second as the thermal imaging event trigger time point. In the 15th-15.4th second, a similar situation occurs in the forehead area, and the trigger time point is also recorded. The final generated thermal imaging event trigger time point sequence contains 8 time points, distributed at the 7th, 15th, 22nd, etc. positions of the acquisition sequence. The pulse wave peak position sequence, tongue image rapid change frame marker points, and thermal imaging event trigger time point sequence are imported into the Health Data Integrator software. This software has cross-modality data fusion capabilities and can read timestamp information of different data types. In the time axis alignment interface of the software, the data time axes of the three modalities are visualized and compared. For example, the pulse wave peaks appear at the 8th, 13.2nd second; the tongue image rapid change frames are at the 5th, 14th second; and the thermal imaging events trigger at the 7th, 15th second. The software automatically associates the event trigger points of the corresponding time points in the three modalities, constructing an initial set of physiological event anchors. The anchor set is presented in a list format, with each anchor containing event type (pulse image, tongue image, thermal imaging), timestamp, and related feature value. For example, anchor 1: pulse image wave peak, timestamp 8 seconds, amplitude 2.3mV; anchor 2: tongue image change, timestamp 5 seconds, frame difference 0.25; anchor 3: thermal imaging event, timestamp 7 seconds, temperature change rate 0.4℃ / s. In the Health Data Integrator software, the cross-modality event analysis module is used to construct an event time correspondence table. The software first extracts the timestamp information from the initial set of physiological event anchors and arranges it in chronological order. For example, the anchor timestamps are 5 seconds (tongue image), 7 seconds (thermal imaging), 8 seconds (pulse image), 13.2 seconds (pulse image), 14 seconds (tongue image), 15 seconds (thermal imaging), etc. The software creates a time correspondence entry for each event, recording the event type, timestamp, and modality source.Meanwhile, the software calculates the time intervals between adjacent events, such as 2 seconds between 5 seconds and 7 seconds, 1 second between 7 seconds and 8 seconds, etc. The generated cross-modal event time correspondence table is displayed in table form, containing columns of event ID, timestamp, event type, modal source, and adjacent event interval. The distribution and association of different modal events on the time axis can be observed intuitively through the table. According to the cross-modal event time correspondence table, the Health Data Integrator software initiates the high-confidence anchor point screening function. The software sets the time window to ±200 milliseconds and judges each event anchor point in the initial physiological event anchor point set. For example, for the pulse wave peak anchor point with a timestamp of 8 seconds, the software checks whether there are events from other modalities within the 200 milliseconds before and after it. It is found that there is a tongue image rapid change frame anchor point at 8.1 seconds, which is outside the time window range (8 seconds + 200 milliseconds = 8.2 seconds), so it does not meet the conditions; but there is a thermal imaging event anchor point at 7.8 seconds, which is within the time window. Therefore, the software marks the 8-second pulse wave peak anchor point as a high-confidence anchor point. After screening all the initial anchor points, the software finally determines 12 high-confidence anchor points, which have at least two events from two modalities within ±200 milliseconds, ensuring the reliability of event association. The high-confidence anchor point set is exported as a.csv file containing event ID, timestamp, and associated modal type information.

[0045] The present application ensures the accuracy of pulse period marking by performing real-time peak detection on pulse waveform data and combining physiological cycle threshold for effectiveness screening. By analyzing the tongue fur microscopic image sequence through the inter-frame difference method, the rapid change frame of tongue image is captured, and the temperature gradient calculation of thermal imaging data can accurately locate the temperature sudden change area and its time point. After integrating the pulse wave peak position, tongue image change frame, and thermal imaging event trigger time point into the initial physiological event anchor point set, further screening of high-confidence anchor points is based on the cross-modal event time correspondence table, effectively removing false event anchors caused by noise or device errors. This not only improves the accuracy and reliability of physiological event anchor point identification, but also provides accurate time reference for subsequent time calibration and feature extraction, enabling the system to more effectively capture and analyze key events in pediatric physiological data.

[0046] Preferably, the time calibration on the noise-suppressed physiological time series data set comprises:

[0047] The noise-suppressed physiological time series data set is segmented by a fixed window length of 10 seconds, with an adjacent window overlap rate of 20%, to generate a segmented time series data set;

[0048] The slope and intercept parameters of each segment of data in the segmented time series data set are calculated to generate a segmented linear parameter table;

[0049] Calculate the mean square residual of each segment of data in the segmented timing data set according to the segmented linear parameter table, and perform fitting eligibility determination and screening on the segmented linear parameter table according to the mean square residual and a preset fitting quality screening threshold, to generate an effective segmented linear parameter table;

[0050] Differentially calculate the slopes of adjacent segments in the effective segmented linear parameter table to generate a sequence of adjacent segment slope differences;

[0051] Identify slope mutation points according to the sequence of adjacent segment slope differences to generate a sequence of drift mode switching points;

[0052] Perform mode matching on the effective segmented linear parameter table according to a preset typical drift mode and the sequence of drift mode switching points to generate a drift mode classification table;

[0053] Adaptively smooth the connection of different mode boundary types based on the drift mode classification table to generate a segmented drift characteristic curve, wherein the adaptive smoothing is specifically:

[0054] When the boundary type is a gradually increasing / decreasing mode boundary, a cubic Hermite interpolation is used for smoothing;

[0055] When the boundary type is an oscillation mode boundary, a mean value smoothing filter is used for smoothing;

[0056] Perform cubic spline interpolation on the segmented drift characteristic curve to generate a signal time drift characteristic curve; and construct a time mapping rule according to the signal time drift characteristic curve;

[0057] Perform time coordinate transformation on the noise-reduced physiological timing data set based on the time mapping rule to obtain a time-corrected physiological data set, wherein the time-corrected physiological data set includes calibrated pulse waveform data, calibrated tongue fur microscopic image sequences, and calibrated body surface thermal imaging data.

[0058] In this example, the collected noise-reduced physiological time series dataset is pre-processed for time calibration using the signal processing software Signal Pro X. In the software, a fixed window length of 10 seconds is set, and the overlap rate between adjacent windows is 20% (i.e., there is a 2-second overlap between adjacent windows). The original pulse waveform dataset is segmented into multiple small segments by the data segmentation function module of the software, each segment containing 1000 sampling points (10 seconds x 100 Hz), and sharing 200 sampling points between adjacent segments. The final segmented time series dataset contains 360 segments, which are stored in a dedicated folder, and each segment is saved in the “.seg” format. The segmented time series dataset is analyzed in the Signal Pro X software. The linear fitting parameter extraction function module of the software is used to calculate the slope and intercept parameters of each segment data. This module uses the least squares fitting algorithm to perform linear fitting on the pulse waveform data within each 10-second window. For example, the slope of the fitted line for the first segment is 0.05 mV / s, and the intercept is 1.2 mV; the slope of the second segment is 0.03 mV / s, and the intercept is 1.5 mV. These slope and intercept parameters are automatically organized into a table file, i.e., the segmented linear parameter table, which contains the start time, end time, slope, and intercept values of each segment. After obtaining the segmented linear parameter table, Microsoft Excel is used to determine the fitting eligibility of the data. First, according to the slope and intercept values in the segmented linear parameter table, the mean square residual of each segment data is calculated using the formula function of Excel. The formula for calculating the mean square residual is: where is the actual data point, For fitting the predicted values of the straight line, n is the number of data points. In Excel, the fitting quality filter threshold is set to be the mean square residual less than 0.1 mV². By filtering, the segments with mean square residual exceeding the threshold are excluded, and finally an effective segmented linear parameter table containing 280 segments is obtained. Based on the effective segmented linear parameter table, the data sequence analysis function module of Signal Pro X software is called again. This module can read the slope sequence in the effective segmented linear parameter table and calculate the difference of the slope of adjacent segments. For example, the slope of the first effective segment is 0.05 mV / s, and the slope of the second segment is 0.03 mV / s, then the difference is -0.02 mV / s²; the difference of the slope between the second and third segments is +0.01 mV / s², and so on. The software automatically generates a sequence file of the difference of the slope of adjacent segments, which records the slope change rate between each pair of adjacent segments. In Signal Pro X software, the adjacent segment slope difference sequence is analyzed by using the mutation point detection function module. This module uses a statistical threshold-based method to identify the slope mutation point. According to experience, the absolute value threshold of the slope change rate is set to 0.02 mV / s². The software traverses the difference sequence, and when it detects that the difference value exceeds this threshold, it marks the corresponding position as a drift mode switching point. For example, at the 50th segment pair in the sequence, the difference value reaches -0.03 mV / s², exceeding the threshold, and is identified as a drift mode switching point. After processing, the software generates a drift mode switching point sequence file, which records the positions of all switching points in the original data and the corresponding slope change rates. The drift mode switching point sequence is imported into the pattern recognition function module of Signal Pro X software. This module has several built-in typical drift mode templates, including increasing mode, decreasing mode, and oscillation mode. The software compares the segment data in the effective segmented linear parameter table with the preset typical drift mode by using a pattern matching algorithm. For example, when the slopes of three consecutive segments are 0.05 mV / s, 0.04 mV / s, and 0.03 mV / s, respectively, it matches the decreasing mode; if the segment slopes alternate between positive and negative, such as 0.02 mV / s, -0.01 mV / s, and 0.03 mV / s, it matches the oscillation mode. The software automatically generates a drift mode classification table, which displays the drift mode type of each segment in table form and marks different modes with different colors. Based on the drift mode classification table, the adaptive smoothing function module in Signal Pro X software is called. For the connection between segments with increasing / decreasing mode interface, the software uses the cubic Hermite interpolation method for smoothing processing. For example, at the interface between the increasing mode and the decreasing mode, the software calculates the interpolation points for smooth transition based on the slopes and data points of adjacent segments. For the connection between segments with oscillation mode interface, the average value smoothing filter method is used to reduce the fluctuations caused by oscillation by calculating the average value of adjacent data points.After adaptive smoothing processing, the software generates a segmented drift characteristic curve. Using the advanced interpolation function module of Signal Pro X software, the segmented drift characteristic curve is processed by cubic spline interpolation. The software constructs a smooth and continuous signal time drift characteristic curve according to the key points in the segmented drift characteristic curve, such as the drift mode switching point and the smoothed data points. For example, in the increasing mode segment of the curve, the cubic spline interpolation can accurately depict the slow rising trend of the signal; in the oscillation mode segment, the interpolated curve can better fit the fluctuation characteristics of the original data. Based on the generated signal time drift characteristic curve, the software automatically constructs the time mapping rule. The rule defines the conversion relationship between the original time coordinate and the corrected time coordinate, for example, the signal value at the original time point t will be mapped to the corrected time point t'=t+Δt (Δt is the time offset calculated according to the drift characteristic curve). The time mapping rule is stored in the form of a function or a lookup table in the software's correction parameter file. In the Signal Pro X software, the constructed time mapping rule is applied to the time coordinate transformation of the noise-reduced physiological time series data set. The software reads the original pulse waveform data, tongue coating microscopic image sequence and body surface thermal imaging data, and adjusts the time stamp of each data point according to the conversion relationship in the time mapping rule. For example, a sampling point in the pulse waveform data with an original time of 100 seconds is converted to 100.5 seconds after time mapping rule conversion. After processing, the time-corrected physiological data set is obtained, which includes calibrated pulse waveform data, calibrated tongue coating microscopic image sequence and calibrated body surface thermal imaging data.

[0059] The present application generates a segmented linear parameter table by segmenting the data and calculating the slope and intercept of each segment, providing a basis for subsequent analysis. The reliability of the segmented linear parameters is ensured through the determination of the mean square residual and the fitting quality screening threshold. By differentiating the slopes of adjacent segments and identifying the slope mutation points, the key points of data drift can be accurately located. Based on the pre-set typical drift mode, pattern matching and classification are performed, further improving the recognition ability of the data drift characteristics. Adaptive smoothing techniques are used at the junction of different drift modes, such as cubic Hermite interpolation at the junction of increasing / decreasing modes and mean value smoothing filter at the junction of oscillation modes, effectively eliminating discontinuity and noise interference in the data. The signal time drift characteristic curve generated by cubic spline interpolation constructs an accurate time mapping rule, realizing the time coordinate transformation of the noise-reduced physiological time series data set. This not only improves the time consistency of the data, but also ensures the accurate alignment of the calibrated physiological data set in the time dimension.

[0060] Preferably, generating the tongue image feature vector set based on the tongue image set comprises:

[0061] performing histogram equalization on the tongue image set to obtain an enhanced tongue image set;

[0062] performing semantic segmentation on the enhanced tongue image set to obtain a tongue body region mask image, and performing region extraction on the enhanced tongue image set according to the tongue body region mask image to obtain a tongue body ROI image set;

[0063] performing color space conversion and decomposition on the tongue body ROI image set to obtain a tongue feature channel group;

[0064] performing wavelet transform on the tongue feature channel group to obtain a tongue texture wavelet coefficient table, and performing statistical feature extraction on the tongue texture wavelet coefficient table to obtain a tongue texture feature table;

[0065] calculating a local binary pattern feature based on the tongue body ROI image set to obtain a tongue LBP feature atlas, and performing color distribution feature analysis on the tongue body ROI image set to obtain a tongue color feature vector;

[0066] performing feature fusion on the tongue texture feature table, the tongue LBP feature atlas and the tongue color feature vector to obtain a tongue feature vector set.

[0067] In this embodiment, the tongue image set is imported into the image processing software Image J. In Image J, the histogram equalization function module is selected, and the parameters of histogram equalization are set: the gray level range is set to 0-255, and the default global equalization method is adopted. The software performs histogram equalization processing on the R, G and B channels of each tongue image. After processing, the contrast of the tongue image is obviously improved, and the details of the tongue body, tongue fur and tongue color are clearer. For example, in an original image, the tongue fur color is dark and the boundary with the tongue body color is blurred, and after equalization, the texture and color level of the tongue fur are more distinct. The enhanced tongue image set is imported into the semantic segmentation software SegLab. Based on the deep learning semantic segmentation algorithm, the software pre-trains a model suitable for tongue image analysis. In the software, the tongue body segmentation model is selected, and the confidence threshold is set to 0.8. The software performs semantic segmentation processing on each enhanced tongue image, and outputs a binary mask image of the tongue body region, in which the tongue body region is displayed in white (pixel value 255) and the background region is displayed in black (pixel value 0). For example, in an image, the tongue body region is accurately covered by a white mask, and the edge is clear. Using the region extraction function module of Image J software, the tongue body mask image is pixel-level AND operated with the enhanced tongue image to extract the region of interest (ROI) of the tongue body. The extracted tongue body ROI image set is stored in a separate folder, and each image only contains the tongue body part, removing the background and other interference information. For example, the extracted image clearly shows the shape, color and texture of the tongue body, with a resolution of the size of the original image, but only the details of the tongue body part are retained. Using the color space conversion plug-in of Image J software, the tongue body ROI image set is converted from RGB color space to LAB color space. Select the converted color space as LAB and check the color channel decomposition option. The software performs conversion and decomposition operations on each tongue body ROI image to generate L (luminance), A (green-red channel) and B (blue-yellow channel) single-channel images. For example, the L channel image of a tongue body ROI image after conversion shows the light and dark changes of the tongue body, the A channel highlights the red and green component difference of the tongue body, and the B channel reflects the yellow and blue component difference. The decomposed tongue feature channel group is stored in the form of a folder, with each folder corresponding to a color channel and containing all the gray images of the corresponding channel of the tongue body ROI image. The wavelet transform plug-in Wavelet Tool box is installed in Image J software. The Daubechies wavelet basis function (db4) is selected, and the decomposition layer is set to 3 layers. Each channel image in the tongue feature channel group is subjected to wavelet transform. Taking the A channel image as an example, after wavelet transform, the approximate coefficient and detail coefficient matrices are obtained and stored as a tongue texture wavelet coefficient table. The wavelet coefficient values at different scales are recorded in the table, for example, the approximate coefficient value at the third layer is 0.75, and the detail coefficient at the first layer is 0.32.The statistical analysis function module of Image J was used to extract the statistical features of the wavelet coefficient table. The mean, standard deviation, skewness, and kurtosis of the wavelet coefficients of each channel image were calculated. For example, the mean of the wavelet coefficients of the A channel was 0.52, and the standard deviation was 0.18. These statistical features constitute part of the tongue texture feature table. Finally, the tongue texture feature table integrates the statistical features of the L, A, and B channels. In the Image J software, the Local Binary Patterns plugin was used to calculate the local binary pattern features of the tongue ROI image set. The neighborhood radius of the LBP operator was set to 3 pixels, the number of sampling points was 8, and the default rotation-invariant and gray-scale-invariant mode was used. The software calculates the LBP features for each tongue ROI image and generates a tongue LBP feature atlas. The atlas is displayed in grayscale image form, and each pixel value represents the LBP pattern encoding value at that position. For example, in an LBP feature atlas, the tongue fur area shows a higher texture pattern encoding value, reflecting the microscopic structure characteristics of the tongue fur. At the same time, the color histogram function module of ImageJ was used to analyze the color distribution features of the tongue ROI image set. The software calculates the histogram of each image for the R, G, and B channels, counts the number of pixels at each gray level, and normalizes the results. For example, the tongue red channel histogram shows a higher pixel ratio in the gray level range of 180-255, corresponding to the color characteristics of the tongue. According to the histogram statistics, the color moments (mean, variance, skewness) are calculated as part of the tongue color feature vector. For example, the color mean of the red channel is 0.72, and the variance is 0.08. These values constitute part of the tongue color feature vector. The Feature Fusion Tool software was used. In the software, the tongue texture feature table (containing the mean, standard deviation, skewness, and kurtosis statistical features of the L, A, and B channels, a total of 16 features), the tongue LBP feature atlas (the LBP histogram features obtained by statistics, a total of 256 features), and the tongue color feature vector (containing the color moments of the R, G, and B channels, a total of 9 features) were used as input. The software provides multiple feature fusion methods, and the linear fusion method was selected. The weights of each feature group were set as follows: the tongue texture feature table weight was 0.4, the tongue LBP feature atlas weight was 0.4, and the tongue color feature vector weight was 0.2. The software fused the feature vectors according to the set weights by adding the elements to generate a set of tongue feature vectors. For example, the fused feature vector has a dimension of 16+256+9=281, and each feature vector corresponds to a comprehensive feature description of a tongue ROI image. The fused tongue feature vector set is stored in the output file of the software in matrix form, where each row represents the feature vector of a tongue image, and the columns correspond to different feature dimensions.

[0068] The application enhances the contrast of the tongue image by histogram equalization, so that the image features are more obvious. The application of semantic segmentation can accurately separate the tongue region from the image background, providing accurate area positioning for subsequent feature extraction. The image information is further decomposed into multiple feature channels through color space conversion and decomposition operation, which facilitates feature analysis from different dimensions. The wavelet transform and local binary pattern feature calculation extract key features from texture and color distribution respectively, ensuring that the feature vector can fully reflect the visual information of the tongue image. The generated tongue feature vector set is not only rich in dimensions, but also has stronger ability to capture the detailed features of the tongue image.

[0069] Preferably, constructing the pulse feature parameter set based on the pulse signal data set comprises:

[0070] Performing pulse period detection and segmentation based on the pulse signal data set to obtain a single-cycle pulse segment set;

[0071] Performing continuous wavelet transform on the single-cycle pulse segment set to obtain a pulse time-frequency spectrogram set;

[0072] Extracting frequency band energy features based on the pulse time-frequency spectrogram set to obtain a pulse frequency domain feature set;

[0073] Performing time domain feature parameter calculation on the denoised pulse waveform data to obtain a pulse time domain feature table;

[0074] Constructing the pulse feature parameter set according to the pulse frequency domain feature set and the pulse time domain feature table.

[0075] In this example, Pulse Analyzer Pro software was used to analyze the collected pulse signal dataset. The built-in pulse period detection module in the software uses an adaptive threshold method to identify the start and end points of the pulse wave. The initial threshold was set to 20% of the signal amplitude, the minimum pulse period was set to 0.5 seconds, and the maximum pulse period was set to 2 seconds. The software automatically detected each pulse period and segmented the continuous pulse signal into single-period pulse segments. For example, the first pulse period was detected to start at the 100th sample point and end at the 300th sample point, the second pulse period was detected to start at the 301st sample point and end at the 500th sample point, and so on. A set of 200 single-period pulse segments was obtained, each stored as an independent file. The single-period pulse segment set was imported into Wavelet Toolbox software. This software is a signal processing tool widely used in the field of biomedical signal analysis. Morlet wavelet was chosen as the mother wavelet function, and the scale range was set to 1-50 with a step size of 0.5. The software performed continuous wavelet transform on each single-period pulse segment to generate a pulse time-frequency spectrogram. For example, after transforming the first single-period pulse segment, the time-frequency spectrogram showed the energy distribution of the pulse signal at different time points and frequencies. The time-frequency spectrogram was displayed as a grayscale image, with white areas representing high energy and black areas representing low energy. Each file corresponded to a time-frequency spectrogram of a single-period pulse segment. In Wavelet Toolbox software, the frequency band energy calculation function module was used to analyze the pulse time-frequency spectrogram set. The software divided the frequency range into four bands: 0-2 Hz, 2-5 Hz, 5-10 Hz, and 10-20 Hz. For each pulse time-frequency spectrogram, the software integrated the energy values within each frequency band. For example, in the first pulse time-frequency spectrogram, the energy of the 0-2 Hz band was 12.5, the energy of the 2-5 Hz band was 8.3, the energy of the 5-10 Hz band was 4.2, and the energy of the 10-20 Hz band was 1.8. These energy values were organized into a pulse frequency domain feature set, where each row represented the frequency band energy features of a single-period pulse segment, and the columns corresponded to different frequency bands. The denoised pulse waveform data was imported into Signal Features software. The time domain feature extraction module of the software was used to calculate multiple time domain feature parameters of the pulse waveform. The specific parameters included peak value (maximum amplitude), valley value (minimum amplitude), rise time (time from baseline to peak), fall time (time from peak to baseline), pulse interval (time between adjacent peaks), amplitude difference (difference between peak and valley), root mean square value (RMS), and waveform factor (ratio of root mean square value to average value).For example, in a segment of the pulse waveform data after noise reduction, the peak value is calculated to be 2.3 mV, the valley value is 0.8 mV, the rise time is 0.12 seconds, the fall time is 0.25 seconds, the pulse interval is 0.85 seconds, the amplitude difference is 1.5 mV, the root mean square value is 1.2 mV, and the waveform factor is 1.3. These parameters are arranged into a pulse time domain feature table. Finally, the pulse frequency domain feature set and the pulse time domain feature table are integrated using the Feature Integration Studio software, and the feature splicing method is selected to connect the frequency domain features and the time domain features in sequence into a comprehensive feature vector. For example, the frequency domain features of a single cycle pulse segment are [12.5, 8.3, 4.2, 1.8] (corresponding to the energy of the four frequency bands), and the time domain features are [2.3, 0.8, 0.12, 0.25, 0.85, 1.5, 1.2, 1.3] (corresponding to the above eight time domain parameters). After splicing, the comprehensive feature vector [12.5, 8.3, 4.2, 1.8, 2.3, 0.8, 0.12, 0.25, 0.85, 1.5, 1.2, 1.3] is obtained. The comprehensive feature vectors of all single cycle pulse segments constitute the pulse characteristic parameter set, wherein each feature vector corresponds to the characteristic description of a single cycle pulse segment.

[0076] The present application can accurately extract single cycle pulse segments through pulse period detection and segmentation, providing a basic unit for subsequent feature extraction. The pulse signal is converted from time domain to time-frequency domain through continuous wavelet transform, and the generated pulse time-frequency spectrum set can directly display the time-frequency characteristics of the pulse signal, which helps to capture the subtle changes of the pulse. The energy distribution of the pulse in different frequency bands is further quantified through the extraction of frequency band energy features, enriching the frequency domain feature dimension of the pulse. The time domain feature parameters calculation supplements the core features of the pulse from the time dimension, such as the amplitude and interval of the pulse. The pulse characteristic parameter set constructed by fusing frequency domain features and time domain features not only has rich dimensions and comprehensive information, but also can more accurately reflect the comprehensive characteristics of the pulse, providing high-quality feature data for subsequent time series analysis and visualization, effectively improving the recognition accuracy and analysis ability of the system for pulse features.

[0077] Preferably, generating a body region temperature feature table based on the thermal imaging image sequence comprises:

[0078] Temperature correction and pseudo-color mapping are performed on the thermal imaging image sequence to obtain a standard body surface thermal imaging image set;

[0079] Threshold segmentation is performed based on the standard body surface thermal imaging image set to obtain a body surface region temperature mask image;

[0080] The body surface region temperature mask image is marked for region division to obtain a body part temperature region image;

[0081] The temperature statistical features of each region are calculated based on the body part temperature region map, and a body region temperature feature table is obtained.

[0082] In this embodiment, the Thermal Image Processor software is used to preprocess the collected sequence of body surface thermal imaging images. The software first performs temperature correction on the original thermal imaging images, adjusts the temperature value of each pixel through the built-in temperature calibration curve (provided by the device manufacturer, covering the range of 32-42°C, with an accuracy of ±0.1°C). The corrected temperature range is standardized to 34-38°C. The software applies the pseudo-color mapping function to map the temperature value to the visible light color space, and selects the rainbow color mapping scheme, in which 34°C corresponds to blue, 36°C corresponds to green, and 38°C corresponds to red. The processed standard body surface thermal imaging atlas is saved in the hospital's medical image archive system in the ".png" format. The standard body surface thermal imaging atlas is imported into the image analysis software Image J. The threshold segmentation function is used in the software to perform binary processing on the thermal imaging images. The temperature threshold range is set to 35-37°C, and the pixels in this range are marked as body surface regions (white, pixel value 255), and the remaining background regions are marked as black (pixel value 0). The processed body surface region temperature mask image is stored in a separate folder. The region analysis plug-in of Image J is used to process the body surface region temperature mask image. The software automatically identifies and marks different body part regions. The specific operation is as follows: first, according to the knowledge of human anatomy, manually mark the initial marker points of the main body parts such as head, neck, trunk, and limbs on the first mask image. Based on these marker points, the software automatically divides the complete regions of each body part using the region growing algorithm (with a similarity threshold of 0.85). For example, in one mask image, the head region is marked as label 1, the neck as label 2, the trunk as label 3, and the limbs as label 4 (left upper limb), label 5 (right upper limb), label 6 (left lower limb), and label 7 (right lower limb). These images directly show the division results of different body parts. The body part temperature region map is re-imported into the Thermal Image Processor software, and temperature statistical analysis is performed in combination with the original thermal imaging data. The software automatically identifies each marked region and calculates its temperature statistical features. The specific statistical features include: average temperature, temperature standard deviation, temperature range (maximum temperature - minimum temperature), temperature median, and temperature mode. For example, the analysis results of the trunk region (label 3) show that the average temperature is 36.2°C, the standard deviation is 0.3°C, the range is 0.8°C, the median is 36.1°C, and the mode is 36.0°C. The temperature statistical features of all regions are arranged into a table, where each worksheet corresponds to a body part, and the columns include the timestamp, average temperature, standard deviation, range, median, and mode.

[0083] The application ensures the temperature data of the thermal imaging image to be accurate and the visual effect to be better through temperature correction and pseudo-color mapping. The temperature distribution of the body surface region is clearly divided through the application of threshold segmentation, and the generated body surface region temperature mask diagram can accurately identify different temperature regions. Further regional division markers subdivide the body surface into specific body part temperature regions, so that the temperature analysis is more targeted and clinically meaningful. By calculating the temperature statistical characteristics of each region, the generated body region temperature feature table not only provides comprehensive temperature data, but also reflects the distribution rule and change trend of the body surface temperature, providing high-quality feature data for subsequent body surface temperature rhythm analysis.

[0084] Preferably, the tongue appearance time sequence index sequence is constructed based on the tongue appearance feature vector set, comprising:

[0085] The tongue appearance feature vector set is color corrected to obtain a tongue appearance color corrected feature vector set;

[0086] Tongue body and tongue fur boundary feature extraction is performed based on the tongue appearance color corrected feature vector set to obtain a tongue body-tongue fur boundary contour sequence;

[0087] Tongue appearance contour evolution simulation is performed based on the tongue body-tongue fur boundary contour sequence to obtain tongue appearance contour evolution trajectory data;

[0088] Tongue body stretching and contracting dynamic parameters are calculated based on the tongue appearance contour evolution trajectory data to obtain a tongue body motion index set;

[0089] Wavelet packet decomposition is performed on the tongue appearance color corrected feature vector set to obtain a tongue appearance texture wavelet energy time sequence spectrum;

[0090] Tongue fur roughness and direction correlation index table is obtained by performing tongue fur roughness anisotropy analysis on the tongue appearance texture wavelet energy time sequence spectrum;

[0091] Regional humidity index calculation is performed on the tongue body-tongue fur boundary contour sequence to obtain a tongue body humidity index time sequence;

[0092] The tongue appearance time sequence index sequence is generated based on the tongue body motion index set, the tongue fur roughness and direction correlation index table, and the tongue body humidity index time sequence.

[0093] In this embodiment, the tongue feature vector set is color corrected using Adobe Photoshop software. First, import the tongue_roi_001.jpg image corresponding to the tongue feature vector set into Photoshop. Observe the histogram of the image and find that the color distribution is yellowish, which is caused by the lighting conditions during shooting. Select Image > Adjust > Color Curves and set the input color curves' shadow, midtones, and highlights to 0, 1.0, and 255 respectively to enhance the image contrast. Select Image > Adjust > Color Balance and adjust the red, green, and blue sliders in the midtones option to +15, -5, and +10 respectively to correct the overall yellowish tone problem. Finally, select Image > Adjust > Saturation and increase the saturation to +10 to make the colors more vivid. Save the processed image as tongue_corrected_001.jpg and the corresponding feature vector with updated color values as tongue_vector_corrected_001.csv. Through the above operations, the tongue color correction feature vector set is obtained. Import the color-corrected tongue image into Image J software. In the software, select the edge detection plugin and use the Sobel operator for edge enhancement processing. Set the threshold to 0.2 (normalized gradient amplitude) and mark the edge points above the threshold as white (pixel value 255) and below the threshold as black (pixel value 0). Through edge detection, the preliminary boundary profile of tongue coating and tongue coating is obtained. Use the analyze particles function, set the size to 50-500 pixels, and the circularity to 0.5-1.0 to perform morphological analysis on the edge to remove small noise points and false contours, and finally obtain a clear tongue-tongue coating boundary profile sequence. In the tongue profile evolution simulation step, use MATLAB software. Import the tongue-tongue coating boundary profile sequence into MATLAB and use the active contour function in the Image Processing Toolbox to simulate. Set the evolution iteration number to 100 times and save the intermediate result every 10 iterations. For example, at the 10th iteration, the tongue profile starts to change slightly and the contour line gradually smoothens; at the 50th iteration, the contour changes tend to be stable, showing the evolution trend of the tongue profile. By observing the contour changes at different iteration numbers, generate the tongue profile evolution trajectory data, which contains the contour coordinate information at each iteration. This data records the dynamic changes of the tongue profile during the simulation. Use the previously generated tongue profile evolution trajectory data to write a script in MATLAB to calculate the tongue stretching dynamic parameters. The script first extracts the contour coordinates from the trajectory data and calculates the area and perimeter of the tongue body at each time point. For example, at time point t1, the tongue area is A1 and the perimeter is P1; at time point t2, the tongue area is A2 and the perimeter is P2.The stretching dynamic parameters of the tongue body are obtained by calculating the area change rate (such as (A2-A1) / A1) and the perimeter change rate (such as (P2-P1) / P1) of adjacent time points. These parameters are arranged into a table, which contains columns such as timestamp, area, perimeter, area change rate, and perimeter change rate. The tongue image color correction feature vector set is decomposed using the wavelet toolbox of MATLAB. The db4 wavelet basis function is selected, and the decomposition level is set to 3 layers. The RGB values of each pixel point in the tongue image feature vector are decomposed to obtain the wavelet packet coefficients. For example, for a pixel point with RGB values (R=120, G=150, B=80), the wavelet packet coefficients at different frequencies and scales are obtained after decomposition. These coefficients reflect the contribution of the pixel point to different texture details. The wavelet packet coefficients of all pixel points are arranged into a table, which contains pixel position, RGB value, and corresponding wavelet packet coefficients. The tongue texture wavelet energy time series spectrum is imported into Image J software. The anisotropy analysis plugin of the software is used to analyze the spectrum in different directions. The analysis directions are set to 0°, 45°, 90°, and 135°, and the energy distribution in each direction is calculated. For example, in the tongue texture wavelet energy time series spectrum, the total energy in the 0° direction is 1200, in the 45° direction is 800, in the 90° direction is 1000, and in the 135° direction is 900. According to these energy values, the energy proportion of each direction is calculated, and the tongue fur roughness-direction correlation index table is obtained, which is saved in the form of an Excel file and contains information such as analysis direction, total energy, and energy proportion. The tongue body-tongue fur boundary contour sequence is imported into Image J software. The regional analysis function of the software is used to calculate the moisture index of the tongue body region. According to the correlation model between tongue color and moisture (which is based on a large amount of clinical data statistics and stored in the software database), the software automatically analyzes the RGB values of the tongue color and converts them into a moisture index. For example, the average RGB values of the tongue body region of a tongue image are R=180, G=160, and B=130, and the moisture index calculated according to the model is 0.65 (the moisture index range is 0-1, and the higher the value, the greater the humidity). The moisture index calculation results are saved as an Excel file, which contains the timestamp of each tongue image and the corresponding moisture index value. The tongue body motion index set, tongue fur roughness-direction correlation index table, and tongue body moisture index time series sequence are integrated using SPSS Statistics software. First, align the three data sets according to the timestamp to ensure consistency of the data at each time point. Select the data>merge file function to combine the three data sets into a comprehensive data set. In the merged data set, each time point contains tongue body motion parameters (area change rate, perimeter change rate), tongue fur roughness indicators (energy proportion in each direction), and tongue body moisture index. Use the analysis>descriptive statistics function of SPSS to calculate the mean, standard deviation, and other statistical quantities of each indicator.The finally generated tongue image time sequence index sequence contains comprehensive index data of all time points.

[0094] The tongue color information accuracy is preserved by color correction, and a reliable foundation is provided for subsequent feature analysis. Through tongue body and tongue fur boundary feature extraction and tongue image contour evolution simulation, subtle changes in tongue image morphology can be accurately captured, and the generated tongue image contour evolution trajectory data reflects the dynamic evolution process of tongue body morphology. The motion characteristics of the tongue body are further quantified through the calculation of the tongue body stretching dynamic parameters, providing important indicators for analyzing the physiological state of the tongue body. The texture features of the tongue image are deeply mined through wavelet packet decomposition and anisotropy analysis of tongue fur roughness, providing detailed information about the surface structure of the tongue fur. The change in the moisture level of the tongue image is reflected from another dimension through the calculation of the regional humidity index. The finally generated tongue image time sequence index sequence comprehensively reflects the dynamic characteristics of tongue body motion, tongue fur texture and tongue body humidity, and fully reflects the evolution law of the tongue image over time, providing more detailed, quantitative and dynamic feature data for tongue diagnosis analysis in pediatric TCM.

[0095] Preferably, generating the pulse image time sequence index sequence based on the pulse image characteristic parameter set comprises:

[0096] Zero-point drift correction is performed on the pulse image characteristic parameter set to obtain a corrected pulse image parameter time sequence;

[0097] Based on the corrected pulse image parameter time sequence, pulse wave form feature recognition is performed to generate a pulse wave form time sequence feature sequence;

[0098] Time-varying form clustering is performed on the pulse wave form time sequence feature sequence to generate a pulse wave time-varying form clustering label set;

[0099] Based on the pulse wave time-varying form clustering label set, a pulse image time-frequency joint feature tensor data is constructed;

[0100] Pulse image dynamic index calculation is performed on the pulse image time-frequency joint feature tensor data to generate a pulse image dynamic index time sequence table;

[0101] Multi-lead signal synchronization is performed according to the pulse image dynamic index time sequence table to generate a multi-lead pulse image phase difference feature atlas;

[0102] Based on the multi-lead pulse image phase difference feature atlas, a blood vessel elasticity parameter is calculated to generate a blood vessel elasticity index sequence;

[0103] Time-space evolution modeling is performed on the blood vessel elasticity index sequence to generate a pulse image time sequence index sequence.

[0104] In this embodiment, MATLAB software is used to correct the zero-point drift of the pulse characteristic parameter set. Load the pulse characteristic parameter set file, which contains multiple characteristic parameters of the pulse signal, such as pulse period, amplitude, rise time. In MATLAB, call the detrend function in the Signal Processing Toolbox and select the constant option to remove the DC offset in the signal. For example, for a pulse signal segment, the original amplitude range is 0.5V to 2.5V, and after the detrend function processing, the zero-point drift of the signal is corrected, and the new amplitude range becomes -0.5V to 1.5V, and the baseline of the signal is stable around 0V. The corrected pulse parameter time sequence is saved as a new.mat file. Import the corrected pulse parameter time sequence into the Python environment and use the Biosppy library to identify the main wave form features of the pulse. Load the new.mat file and use the pulse_processing function in Biosppy. In the function, set the waveform type to radial (radial artery pulse) and use the default peak detection algorithm. For example, for a corrected pulse signal, the program identifies the peak position of the main wave at the 100th sampling point and the trough position at the 150th sampling point. The feature extraction results include the rise slope, fall slope, peak height and trough depth of the main wave, which are arranged into a pulse main wave form time sequence feature sequence and saved as a CSV file main_wave_features.csv. Use the scikit-learn library of Python to perform time-varying morphological clustering on the pulse main wave form time sequence feature sequence. Load the main_wave_features.csv file, select the KMeans clustering algorithm, set the number of clusters to 3, and the initialization method to k-means++. For example, for the extracted rise slope, fall slope and peak height features, the program assigns each time window's main wave form feature vector to the nearest cluster center. After multiple iterations, the program generates a pulse wave time-varying morphological clustering label set, each label corresponds to a main wave form type (such as label 0 for sharp peak, label 1 for flat, and label 2 for deep valley). The clustering label sequence is saved as a CSV file pulse_cluster_labels.csv. In MATLAB, use the Signal Processing Toolbox to build a pulse time-frequency joint feature tensor data. Load the raw pulse signal file raw_pulse_signal.mat and the clustering label file pulse_cluster_labels.csv. Use the spectrogram function to perform time-frequency analysis on the pulse signal, set the window length to 1 second (corresponding to 1000 sampling points at a sampling frequency of 1000 Hz), and the overlap length to 0.5 seconds.The calculated time-frequency map data is combined with the cluster labels to form a three-dimensional tensor data with dimensions of time, frequency, and cluster features. For example, at time point t=5 seconds and frequency f=5 Hz, the tensor data contains the energy value at that frequency and the corresponding cluster label. The tensor data is saved as a.mat file named tensor_data.mat. Load the tensor_data.mat file and use MATLAB's Tensor Toolbox to calculate the pulse dynamic indicators. Choose the Higher-Order Singular Value Decomposition (HOSVD) method to decompose the tensor data. Set the rank of the decomposed tensor to [10,10,3], meaning that the first 10, first 10, and first 3 singular values are retained for the time, frequency, and feature dimensions, respectively. The core tensor and factor matrices obtained from the decomposition are used to calculate the pulse dynamic indicators. For example, the sum of the singular values in the time dimension represents the stability of the pulse signal over time, and the distribution of the singular values in the frequency dimension reflects the energy distribution characteristics of the pulse. The results of the dynamic indicator calculation are saved as an Excel file named pulse_dynamic_indicators.xlsx. Import the pulse dynamic indicators time series table into the Python environment and use NumPy and Matplotlib libraries for multi-lead signal synchronization processing. Load the pulse_dynamic_indicators.xlsx file and assume that the data contains pulse signals from 3 leads. Use the Dynamic Time Warping (DTW) algorithm to calculate the optimal time alignment path between different lead signals. For example, for lead 1 and lead 2, the calculated alignment path shows that lead 2 is delayed by 5 sampling points relative to lead 1. Based on the alignment results, calculate the phase difference and generate a multi-lead pulse phase difference feature map. The map displays the phase difference distribution between different leads in the form of a heat map and is saved as a PNG image file named phase_difference_map.png. In MATLAB, use the Image Processing Toolbox to analyze phase_difference_map.png. Load the image file and convert it to a grayscale image. Use the graythresh function to calculate the global threshold of the image and binarize it to highlight the main feature area. Perform morphological processing (such as dilation and erosion operations) on the binarized image to extract morphological features related to blood vessel elasticity. Use a pre-established empirical formula (based on clinical experimental data) that relates phase difference to blood vessel elasticity parameters to calculate a sequence of blood vessel elasticity indicators. For example, within a certain time window, the blood vessel elasticity modulus corresponding to the phase difference is 15 kPa. Save the calculation results as a text file named artery_elasticity.txt. Use the PyTorch library in Python to model the spatio-temporal evolution of the blood vessel elasticity indicator sequence.Load the artery_elasticity.txt file and convert the data into tensor format. Construct a one-dimensional convolutional neural network (CNN) model with a kernel size of 3, a stride of 1, and an output channel number of 16. The model input is the time series of vascular elasticity indicators, and the output is the feature map extracted by convolution. Through training the model, the features reflecting the spatiotemporal changes of vascular elasticity are extracted. Combine these features with the previous pulse dynamic indicators to generate a pulse time series indicator sequence, containing timestamps and comprehensive pulse indicator values.

[0105] The present application effectively eliminates the baseline drift problem in pulse parameters through zero point drift correction, improving the stability of the data. Through pulse main waveform feature recognition and subsequent time-varying morphology clustering analysis, not only the key morphological features of pulse waveform are captured, but also the variation law over time is revealed. The constructed pulse time-frequency joint feature tensor data further integrates the time domain and frequency domain information, providing a more comprehensive feature basis for subsequent analysis. Through pulse dynamic indicator calculation and multi-lead signal synchronization processing, the dynamics of pulse and its phase difference can be quantified, providing a reliable basis for the calculation of vascular elasticity parameters. The finally generated pulse time series indicator sequence integrates the pulse dynamics and vascular elasticity information, comprehensively reflecting the dynamic change process of pulse, providing more detailed, quantitative and dynamic feature data for pulse diagnosis analysis in traditional Chinese pediatric medicine.

[0106] Preferably, generating the body surface temperature rhythm indicator sequence based on the body region temperature feature table comprises:

[0107] Perform dynamic baseline correction on the body region temperature feature table to obtain a corrected body surface temperature time series table;

[0108] Perform spatiotemporal interpolation on the corrected body surface temperature time series table to generate a body surface temperature spatiotemporal continuous distribution field;

[0109] Extract body surface thermal features from the body surface temperature spatiotemporal continuous distribution field to generate a body surface heat conduction parameter sequence;

[0110] Construct a temperature feature attractor graph according to the body surface heat conduction parameter sequence;

[0111] Perform body surface temperature complexity analysis based on the temperature feature attractor graph to generate a body surface temperature complexity indicator table;

[0112] Perform oscillation mode decomposition on the body surface temperature complexity indicator table to obtain a temperature oscillation feature set;

[0113] Perform wavelet coherence analysis on the temperature oscillation feature set to generate a body surface-internal organ temperature intensity time series graph; generate a body surface temperature rhythm indicator sequence based on the body surface-internal organ temperature intensity time series graph.

[0114] In this embodiment, the Thermal Analysis Pro software is used for dynamic baseline correction of the body region temperature feature table. This feature table contains the body surface temperature data of the patient during the detection process, covering multiple regions such as the head, neck, torso, and limbs. The software reads the original temperature data file, which records the temperature changes of each region during the detection period. In the software, select the dynamic baseline correction function module, set the time window length to 5 minutes, and the moving step to 1 minute. The software calculates the temperature median in each time window as the baseline reference value, and the difference between the original temperature data and the baseline reference value as the corrected temperature value. For example, in the torso region, the temperature median of the first 5-minute window is 36.2°C, and the original temperature at a certain time point is 36.5°C, then the corrected temperature is 0.3°C. After processing, the corrected body surface temperature time series table is obtained. The corrected body surface temperature time series table is imported into the Spatial Interpolation Toolbox software. This software is a spatio-temporal interpolation tool widely used in the biomedical field. In the software, select the Kriging interpolation method as the interpolation algorithm, and set the interpolation grid spacing to 2 cm x 2 cm. The software generates a spatio-temporal continuous distribution field of the body surface temperature based on the corrected temperature data and the geometric position information of each region of the body. For example, in the torso region, the software calculates the temperature value of each grid point based on the data collected by adjacent temperature sensors, forming a smooth temperature distribution surface. In the Spatial Interpolation Toolbox software, the heat force feature extraction function module is used to analyze the spatio-temporal continuous distribution field of the body surface temperature. The software automatically calculates the thermal conductivity, thermal diffusivity, and heat capacity, etc. For example, for the temperature distribution field of the torso region, the software calculates the thermal conductivity as 0.2 W / (m·K), the thermal diffusivity as 0.1 cm² / s, and the heat capacity as 3.5 J / (g·K). These parameters reflect the thermal conduction characteristics of different regions of the body surface. The sequence of body surface thermal characteristic parameters is saved in CSV file format, which contains the time stamp, region identifier, and corresponding thermal characteristic value. The body surface thermal characteristic parameter sequence is imported into the Chaos Analysis Suite software. In the software, select the phase space reconstruction function module, set the delay time to 3, and the embedding dimension to 4. The software constructs a temperature feature attractor graph based on the thermal characteristic parameter sequence. For example, in the thermal conductivity time series of the torso region, the software generates a three-dimensional attractor graph through phase space reconstruction, which shows the trajectory of thermal conductivity in the phase space, and the graph uses different colors to represent the density of the trajectory, directly showing the internal law of the body surface temperature change. In the Chaos Analysis Suite software, the complexity analysis function module is used to further analyze the temperature feature attractor graph. The software calculates the fractal dimension and Lyapunov exponent of the attractor, and evaluates the complexity of the body surface temperature change.For example, the fractal dimension of the torso region temperature feature attractor is 2.3, and the Lyapunov exponent is 0.15, indicating that the temperature change in this region has moderate complexity and chaotic characteristics. The complexity analysis results are saved as an Excel file containing the fractal dimension and Lyapunov exponent indicators of each region. The body surface temperature complexity indicator table is imported into the Vibration Analysis Toolbox software. In the software, the Empirical Mode Decomposition (EMD) method is selected to decompose the body surface temperature time series data into multiple Intrinsic Mode Functions (IMF). For example, the temperature time series data of the neck region is decomposed into 5 IMF components and a residual component. Each IMF component represents a temperature oscillation pattern in a different frequency band. The software extracts the oscillation frequency, amplitude, and energy features of each IMF component to form a temperature oscillation feature set. The temperature oscillation feature set and the internal temperature reference data (assumed to be obtained from the esophageal temperature sensor) are imported into the Wavelet Coherence Analyzer software. In the software, the Morlet wavelet is selected as the analysis mother wavelet, and the scale range is set to 1-30 days with a step size of 0.5 days. The software performs wavelet coherence analysis on the body surface temperature and internal temperature data to generate a body surface-internal temperature intensity time series graph. For example, the software shows that during the detection period of 5-10 days, there is significant coherence between the body surface temperature and the internal temperature in the 2-5 day cycle range, with a coherence coefficient above 0.7. According to the time series graph, the software extracts the coherence intensity and phase difference features to generate a body surface temperature rhythm indicator sequence, which contains time stamps, coherence intensity, and phase difference indicators.

[0115] The present application effectively eliminates the reference drift in the body surface temperature data through dynamic baseline correction, improving the stability and accuracy of the data. Through spatiotemporal interpolation processing, a continuous distribution field of body surface temperature is further generated, providing a more complete and smooth data basis for subsequent thermal feature extraction. Through body surface thermal feature extraction and construction of temperature feature attractor graph, not only the conduction characteristics of body surface temperature are captured, but also the internal rules and complexity of temperature change are revealed. The body surface temperature complexity indicator table generated through complexity analysis based on the attractor graph further quantifies the complexity characteristics of the body surface temperature change. Through oscillation mode decomposition and wavelet coherence analysis, the periodicity characteristics of the body surface temperature and its correlation with the internal temperature are further explored, and the body surface-internal temperature intensity time series graph generated intuitively reflects the dynamic relationship between the body surface and internal temperature. The final body surface temperature rhythm indicator sequence integrates the complexity and oscillation characteristics of the body surface temperature, comprehensively reflecting the rhythmic changes of the body surface temperature over time.

[0116] Especially important is that the pediatric data visual coding rule is constructed based on the tongue image time evolution indicator sequence, the pulse image evolution time indicator sequence, and the body surface temperature rhythm indicator sequence, which includes:

[0117] A pediatric data visual coding rule is constructed based on the tongue appearance time sequence evolution index sequence, the pulse appearance evolution time sequence index sequence, and the body surface temperature rhythm index sequence;

[0118] A physiological index evolution trajectory graph is constructed based on the tongue appearance time sequence evolution index sequence, the pulse appearance evolution time sequence index sequence, and the body surface temperature rhythm index sequence;

[0119] A time sequence fan-shaped navigation basic framework is constructed based on the physiological index evolution trajectory graph;

[0120] A time sequence unified index architecture is obtained by performing multi-modal data interface design on the time sequence fan-shaped navigation basic framework;

[0121] An interactive trigger event flow is designed based on the time sequence unified index architecture, and a user intention recognition framework is obtained;

[0122] A multi-view interactive feedback mechanism is obtained by constructing a view linkage controller according to the user intention recognition framework;

[0123] Focus-context visual enhancement schemes are obtained by performing focus data enhancement and context data reduction on the multi-view interactive feedback mechanism;

[0124] A key time node highlighting scheme is obtained by performing time key node dynamic detection based on the focus-context visual enhancement scheme;

[0125] A pediatric data visual coding rule is obtained by performing visual coding design on the key time node highlighting scheme.

[0126] In this embodiment, pediatric data visual encoding rules are constructed using Medical Design Studio software. First, load the tongue image time evolution index sequence file, pulse image evolution time sequence index file and body surface temperature rhythm index sequence file. According to the established visualization requirements, set the tongue image index to red, the pulse image index to blue, and the temperature index to yellow. In the software, set the data range: tongue image index range 0-100, pulse image index range 0-200, temperature index range 34-38℃. According to these ranges, the software automatically generates color mapping rules to convert data values into corresponding color shades. For example, a tongue image value of 50 corresponds to a medium red color, a pulse image value of 100 corresponds to a medium blue color, and a temperature of 36℃ corresponds to a medium yellow color. Finally, the completed pediatric data visual encoding rules are saved in a file that defines the visualization representation of different types of data. The previously constructed visual encoding rule file is imported into the Trajectorymapper software. This software is a trajectory atlas generation tool. In the software, select the three-dimensional visualization mode with time as the horizontal axis and different physiological indicators as the vertical axis. Set the time axis range to the total duration from the start to the end of the detection. The software draws the physiological indicator evolution trajectory atlas according to the visual encoding rules and physiological indicator data. For example, the tongue image index trajectory line is displayed in red, and the trajectory line color changes accordingly as the tongue image value changes; the pulse image index is displayed in a blue trajectory line, and the color depth is adjusted according to the numerical value change. The atlas is displayed in the form of an interactive three-dimensional model, supporting rotation, scaling and other operations. The completed trajectory atlas is saved as a file and stored in the hospital's visualization resource library. In the Trajectorymapper software, use the fan-shaped navigation plug-in to construct the time sequence fan-shaped navigation basic framework. Set the fan-shaped center point to correspond to the start time of the detection, and the fan-shaped radius represents the time progress. From the center to the outside, there are key time nodes such as the 1st day, the 3rd day, the 7th day, the 14th day, etc. Each time node is represented by a small fan-shaped area, and the angle size is allocated according to the time interval. The total angle of the entire fan-shaped navigation is set to 270 degrees. The software projects the physiological indicator evolution trajectory atlas onto the fan-shaped navigation, represented by different colored dots for each physiological indicator state. For example, at the 3rd day position, red dots represent tongue image state and blue dots represent pulse image state. The preliminary constructed navigation framework is saved as a fan_framework.fnav file. In the Fan Framework Designer software, design the multi-modal data interface for the time sequence fan-shaped navigation basic framework. Choose to add a hovering prompt function on the fan-shaped area. When the mouse hovers over a small fan-shaped area at a certain time point, a pop-up window displays the detailed physiological indicator data at that time point. Set the data loading delay to 300 milliseconds to avoid interface lag caused by frequent operations.Meanwhile, the database connection settings are integrated to connect to the PediatricData2025 database of the hospital, and data at corresponding time points are loaded from the Tongue, Pulse, and Temperature tables. After the interface design is completed, the software automatically generates a time sequence unified index architecture, using timestamps as unified index keys to associate the data of each modality. The final saved architecture file is UnifiedIndex.ui. The interaction trigger event flow is designed in the Interaction Flow Architect software. On the time sequence sector navigation, the single-click event is set to view detailed data, and the double-click event is set to compare data at adjacent time points. The keyboard shortcut Ctrl+D is added for quick mode switching. The software records the frequency of user operations, and when the single-click operation frequency exceeds 5 times per minute, the intelligent recommendation function is triggered to display the associated data of interest. The user intent recognition framework includes detailed definitions of event types, trigger conditions, and response actions. For example, when the user frequently switches the position to view on the 7th day and the 14th day, the system automatically recommends to display trend analysis data in this time range, improving user operation efficiency. In the Multi View Linker software, the view linkage controller is constructed. The main view is set to sector navigation, and the auxiliary views are detailed data tables and trend line charts. When the user selects a time point on the sector navigation, the detailed data table automatically updates to display the specific values of the tongue, pulse, and temperature at that time point; the trend line chart highlights the selected time point and displays the trend lines of adjacent time points. A smooth transition animation is added, and a 0.5-second fade effect is used when switching views to enhance visual coherence. The constructed multi-view interaction feedback mechanism is saved as a ViewLink.vlc file. For example, after the user clicks on the 10th day position, the detailed data table smoothly transitions to display the data of the 10th day within 0.5 seconds, and the highlighted part of the trend line chart is also updated synchronously, providing a more intuitive display of data association for the user. In the FocusContext Enhancer software, the focus data enhancement and context data reduction are performed. On the sector navigation, the currently selected time point is set as the focus area, which is displayed in a larger size, while the non-focus area is displayed with a transparency of 50%, guiding the user to focus on the focus data. In the detailed data table, the focus data row is highlighted with bold font and increased row height. For the context data, only the simplified view of the key indicators is displayed, such as only the integer part of the temperature, and the main trend labels of the tongue and pulse. After the enhancement scheme is saved as a FocusScheme.fcs file, when displayed in the interface, the focus area is automatically enlarged and highlighted, and the context data is presented in a simplified form, improving the information hierarchy of the interface and helping the user quickly locate the key information. In the Key Pointmarker software, dynamic detection of time key nodes is performed based on the focus-context visual enhancement scheme.A key node detection algorithm is set, and when the physiological indicators change by more than a preset threshold (such as tongue appearance changes greater than 20%, pulse appearance changes greater than 30%, temperature changes greater than 1°C), it is automatically marked as a key time node. In the fan-shaped navigation, the key nodes are marked with a prominent star, and the marker color corresponds to the visual coding rules according to the indicator type. For example, tongue appearance key nodes are marked with a red star, pulse appearance is marked with a blue star, and temperature is marked with a yellow star. After the scheme is saved as a KeyPoints.kps file, the key time nodes are automatically highlighted with a star-shaped marker on the fan-shaped navigation, guiding the user to quickly identify important change moments in the data and improving data interpretation efficiency. In the Medical Design Studio software, the key time node highlighting scheme is finally designed for visual coding. The size of the star-shaped marker is dynamically adjusted according to the key degree, and the change amplitude is more than 150% of the threshold, the key node is displayed as a large star, the change amplitude is between 100%-150%, the key node is displayed as a medium star, and the change amplitude is between the preset threshold and 100%, the key node is displayed as a small star. At the same time, add a hover prompt function, when the mouse hovers over the star-shaped marker, display the detailed information of the key node, including the type of the changing indicator, the change amplitude and the clinical significance. The final pediatric data visual coding rule integrates all visual elements such as color coding, node marking, and interactive prompts.

[0127] Especially important is that visualizing the tongue appearance time sequence evolution indicator sequence, the pulse appearance evolution time sequence indicator sequence, and the body surface temperature rhythm indicator sequence according to the pediatric data visual coding rule includes:

[0128] Visualizing the tongue appearance time sequence evolution indicator sequence, the pulse appearance evolution time sequence indicator sequence, and the body surface temperature rhythm indicator sequence according to the pediatric data visual coding rule;

[0129] A hierarchical detail display mechanism is constructed according to the pediatric data visual coding rule to obtain a pediatric data detail display framework;

[0130] Based on the pediatric data detail display framework, a modal navigation channel is established to obtain a seamless modal navigation unit;

[0131] Based on the physiological indicator evolution trajectory atlas and the seamless modal navigation unit, a dynamic symptom correlation force guide graph is drawn to obtain a dynamic symptom correlation force guide graph;

[0132] A semantic enhanced symptom interaction relationship model is obtained by mapping the dynamic symptom correlation force guide graph to a human three-dimensional model;

[0133] Based on the semantic enhanced symptom interaction relationship model, a multi-angle comparison view framework is constructed to obtain a pediatric data space-time comparison analysis framework;

[0134] According to the pediatric data spatiotemporal comparison analysis framework, a symptom evolution visualization component is formulated to obtain a symptom intensity spatiotemporal distribution map;

[0135] The symptom intensity spatiotemporal distribution map is subjected to syndrome evolution visualization modeling to obtain a dynamic visualization view of syndrome evolution; and the symptom intensity spatiotemporal distribution map and the dynamic visualization view of syndrome evolution are visualized through a preset data display interface.

[0136] In this embodiment, Tableau software is used to visualize the tongue, pulse, and body temperature data. The tongue time series evolution index sequence file, pulse evolution time series index sequence file, and body temperature rhythm index sequence file are imported into Tableau. According to the previously designed pediatric data visual encoding rules, the tongue index is represented by red, the pulse index is represented by blue, and the temperature index is represented by yellow. In Tableau, three line graphs are created, corresponding to the tongue, pulse, and temperature indexes respectively. For example, the tongue index line graph shows the change trend of the tongue value in the range of 0-100, the pulse index line graph shows the change trend of the pulse value in the range of 0-200, and the temperature index line graph shows the change trend of the temperature in the range of 34-38℃. The final generated visualization chart is saved as a Tableau workbook file. In the Tableau workbook, a pediatric data detail display framework is constructed. Using the hierarchical view function of Tableau, the data is displayed in layers according to the time hierarchy (hour, day, week). The default view is set to display week-level data, and a filter is added to the view to allow users to view day-level and hour-level data by drilling down. For example, users can see the overall trend of tongue, pulse, and temperature in the week view, and by clicking on the data of a certain week, the detailed data of each day in that week is automatically displayed. The constructed hierarchical detail display framework is saved in the workbook file, ensuring that users can flexibly switch between different levels of details. In the Tableau workbook, a modal navigation channel is established using the action function. Set the tongue details, pulse details, and temperature details as the navigation targets. When the user selects a data point (such as a red data point with a tongue value exceeding 80) at a certain time point in the main view, the filter action of the tongue details worksheet is triggered, automatically jumping to and displaying the detailed information of the tongue at that time point. At the same time, a button is set to return to the main view. The navigation channel settings are saved in the workbook file, ensuring that users can smoothly navigate between different modal data and fully understand the multi-dimensional information of pediatric data. In Gephi software, load the physiological index evolution trajectory graph file and the interactive log file of the seamless inter-modal navigation unit. Set the nodes to represent the physiological index state at different time points, and the edges to represent the association strength between indexes. The association strength is calculated according to the correlation coefficient between indexes, and the index pair with a correlation coefficient greater than 0.6 is considered to have strong association. For example, between the time points with high tongue value (deep red) and high temperature value (deep yellow), a thick edge is drawn to represent strong association. The software automatically lays out to generate a symptom association force directed graph, showing the association network between different physiological indexes. The directed graph is saved in Gephi native format. In Blender software, import the symptom association force directed graph data. Use the human body model plugin of Blender to load a standard child human body model.The nodes in the guide map are mapped to the corresponding parts of the human model, for example, the tongue image nodes are mapped to the oral cavity position, the pulse image nodes are mapped to the wrist position, and the temperature nodes are mapped to the torso surface. The size and color of the nodes are dynamically changed according to the symptom intensity, and the associated edges are displayed in semi-transparent lines. The finally generated semantic enhanced symptom interaction model is saved in the Blender file format, supporting three-dimensional rotation and zoom viewing. In the 3dsmax software, the semantic enhanced symptom interaction model file is loaded. Using the multi-view function of 3dsmax, four view windows are created, corresponding to the front view, side view, top view and perspective view respectively. The camera parameters of each view are set, the front view displays the front symptom distribution of the human body, the side view displays the side symptom distribution, the top view displays the top symptom distribution, and the perspective view provides a three-dimensional perspective. For example, in the front view, the tongue image nodes are displayed at the oral cavity position, and the pulse image nodes are displayed at the wrist position; in the perspective view, the connection mode of the associated edges between the nodes in the three-dimensional space can be clearly seen. The completed multi-angle comparison view framework is saved as a 3dsmax scene file. In the Unity engine, a symptom intensity spatiotemporal distribution graph is created. Using the Timeline function of Unity, the time axis range is set to the total duration of the detection period. The symptom intensity data is mapped to the column chart in the three-dimensional space, the tongue intensity is represented by a red column chart, the pulse intensity is represented by a blue column chart, and the temperature intensity is represented by a yellow column chart. For example, at the 3rd day position of the time axis, the column chart with high tongue intensity value is displayed as a higher red column. Through the animation system of Unity, the dynamic effect of the evolution of symptom intensity over time is made, and the finally generated symptom intensity spatiotemporal distribution graph is saved as a Unity scene file. In the Unity engine, the symptom intensity spatiotemporal distribution graph file is loaded. Using the animation curve function of Unity, a syndrome evolution model is constructed according to the TCM syndrome classification standard. The combination features of red tongue, fast pulse and high temperature are set for wind-heat syndrome, and the combination features of pale tongue, slow pulse and low temperature are set for wind-cold syndrome. Through the keyframe animation technology, the evolution process of the syndrome in the time sequence is displayed. For example, the wind-cold syndrome features are displayed in the first 3 days of detection, and gradually transition to the wind-heat syndrome features from the 4th day. The finally generated dynamic visualization view of syndrome evolution is saved as a Unity Prefab file, supporting reuse in different projects. In the Unity engine, a comprehensive data display interface is created. Using the UI system of Unity, an interface layout containing two main panels is designed. The left panel is used to display the symptom intensity spatiotemporal distribution graph, and the right panel is used to display the dynamic visualization view of syndrome evolution. A time slider controller is added to allow users to drag the slider on the bottom time axis to view the data at different time points. Interactive buttons such as play, pause and reset are set to facilitate user operation. The final visualization interface is packaged as a Unity executable file and deployed on the workstation of the hospital.

[0137] Therefore, the embodiments should be regarded, at any point, as being exemplary and not limiting, the scope of the application being defined by the appended claims and not by the above description, and all changes which come within the meaning and range of equivalency of the claims are therefore intended to be embraced therein.

[0138] The foregoing is considered as illustrative only of the principles of the application. Numerous modifications and changes will readily occur to those skilled in the art, and the generic principles defined herein can be applied to other embodiments without departing from the spirit or scope of the application. Therefore, the application is not intended to be limited to the embodiments shown herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A visual data processing system applied to traditional Chinese medicine pediatrics, characterized in that, The method comprises the following modules: a data acquisition module for acquiring pediatric physiological time series data sets; physiological event anchor point identification is performed on the pediatric physiological time series data sets to obtain a set of physiological event anchor points; wherein the pediatric physiological time series data sets are acquired by: acquiring multi-dimensional physiological information of a child patient through multi-source pediatric TCM detection equipment to obtain a multi-modal pediatric TCM physiological data set; registering and encoding the equipment identification codes of the multi-source pediatric TCM detection equipment to obtain an equipment identification code data table; and synchronizing and calibrating the clocks of the multi-source pediatric TCM detection equipment by using a pre-set network time protocol server to obtain a device reference time synchronization record; performing data source analysis on the multi-modal pediatric TCM physiological data set according to the equipment identification code data table and the device reference time synchronization record to obtain a device data stream meta-information table; extracting the time stamp field from the device data stream meta-information table and performing standardized conversion to obtain an original acquisition time stamp sequence; and calculating the time deviation values between the detection equipment based on the original acquisition time stamp sequence to obtain a device time difference parameter table; calculating the standard deviation of the time deviation values of five consecutive measurements in the device time difference parameter table, and performing stability judgment according to the standard deviation and a pre-set stability threshold to obtain a time deviation stability judgment result; quantifying the reliability of the device time difference parameter table according to the time deviation stability judgment result to obtain a time difference reliability evaluation result; and performing weighted average correction on the device time difference parameter table according to the time difference reliability evaluation result to obtain a corrected device time difference parameter table; performing dynamic time warping on the multi-modal pediatric TCM physiological data set based on the corrected device time difference parameter table to obtain the pediatric physiological time series data set, wherein the pediatric physiological time series data set comprises pulse waveform data, tongue coating microscopic image sequences, and body surface thermal imaging data; wherein the physiological event anchor point identification on the pediatric physiological time series data set comprises: performing real-time peak detection on the pulse waveform data to obtain a pulse wave peak position sequence; calculating the time interval between adjacent wave peaks according to the pulse wave peak position sequence, and performing physiological validity screening on the pulse wave peak position sequence according to the time interval between adjacent wave peaks and a pre-set physiological cycle threshold range to obtain a set of standard pulse cycle marker points; performing frame-by-frame comparison on the tongue coating microscopic image sequences based on the inter-frame difference method to obtain tongue image rapid change frame markers; performing regional temperature gradient calculation and marking on the body surface thermal imaging data to obtain a temperature sudden change region marker map; statistically calculating the time points of the temperature sudden change region in the continuous frames of the body surface thermal imaging data based on the temperature sudden change region marker map to generate a thermal imaging event trigger time point sequence; the corresponding event trigger points in the pulse wave peak position sequence, the tongue image rapid change frame markers, and the thermal imaging event trigger time point sequence are recorded as an initial set of physiological event anchor points; a cross-modal event time correspondence table is constructed based on the pulse wave peak position sequence, the tongue image rapid change frame markers, and the thermal imaging event trigger time point sequence; each event anchor point in the initial set of physiological event anchor points is judged based on the cross-modal event time correspondence table as follows: If at least two modalities have a trigger event within a ±200 ms window, the corresponding event anchor point is marked as a high-confidence anchor point, otherwise, the corresponding event anchor point is discarded, and the high-confidence anchor points are recorded as a set of physiological event anchor points; a time calibration module, configured to perform noise suppression on the pediatric physiological time series data set according to the set of physiological event anchor points to obtain a noise-suppressed physiological time series data set, and perform time calibration on the noise-suppressed physiological time series data set to obtain a time-corrected physiological data set; a feature extraction module, configured to extract a tongue appearance image set from the time-corrected physiological data set, generate a tongue appearance feature vector set based on the tongue appearance image set, extract a pulse appearance signal data set from the time-corrected physiological data set, construct a pulse appearance characteristic parameter set based on the pulse appearance signal data set, and extract a thermal imaging image sequence from the time-corrected physiological data set, and generate a body region temperature feature table based on the thermal imaging image sequence; an evolution index construction module, configured to construct a tongue appearance time series index sequence based on the tongue appearance feature vector set, generate a pulse appearance time series index sequence based on the pulse appearance characteristic parameter set, and generate a body surface temperature rhythm index sequence based on the body region temperature feature table; a visualization module, configured to construct a pediatric data visual coding rule based on the tongue appearance time series index sequence, the pulse appearance time series index sequence, and the body surface temperature rhythm index sequence, and visualize the tongue appearance time series index sequence, the pulse appearance time series index sequence, and the body surface temperature rhythm index sequence according to the pediatric data visual coding rule.

2. The visual data processing system for Chinese medicine pediatrics of claim 1, wherein, The time calibration on the noise-suppressed physiological time series data set includes: segmenting the noise-suppressed physiological time series data set into segmented time series data set with a fixed window length of 10 seconds and an overlapping rate of 20% between adjacent windows; calculating the slope and intercept parameters of each segment of the segmented time series data set to generate a segmented linear parameter table; calculating the mean square residual of each segment of the segmented time series data set according to the segmented linear parameter table, and performing fitting eligibility determination and screening on the segmented linear parameter table according to the mean square residual and a preset fitting quality screening threshold to generate an effective segmented linear parameter table; differentially calculating the slopes of adjacent segments in the effective segmented linear parameter table to generate an adjacent segment slope difference sequence; identifying slope mutation points according to the adjacent segment slope difference sequence to generate a drift mode switching point sequence; performing mode matching on the effective segmented linear parameter table according to a preset typical drift mode and the drift mode switching point sequence to generate a drift mode classification table; performing adaptive smoothing on the connection of different mode boundary types based on the drift mode classification table to generate a segmented drift characteristic curve, wherein the adaptive smoothing is specifically: when the boundary type is a gradually increasing / decreasing mode boundary, performing smoothing by using a cubic Hermite interpolation; when the boundary type is an oscillation mode boundary, performing smoothing by using a mean value smoothing filter; performing cubic spline interpolation on the segmented drift characteristic curve to generate a signal time drift characteristic curve; and constructing a time mapping rule according to the signal time drift characteristic curve; performing time coordinate transformation on the noise-suppressed physiological time series data set based on the time mapping rule to obtain the time-corrected physiological data set, wherein the time-corrected physiological data set includes calibrated pulse waveform data, calibrated tongue fur microscopic image sequence, and calibrated body surface thermal imaging data.

3. The visual data processing system for Chinese medicine pediatrics of claim 1, wherein, The generating of the tongue feature vector set based on the tongue image set comprises: The tongue image set is subjected to histogram equalization to obtain an enhanced tongue image set; The enhanced tongue image set is subjected to semantic segmentation to obtain a tongue body region mask image; the enhanced tongue image set is subjected to region extraction according to the tongue body region mask image to obtain a tongue body ROI image set; The tongue body ROI image set is subjected to color space conversion and decomposition to obtain a tongue feature channel group; The tongue feature channel group is subjected to wavelet transform to obtain a tongue texture wavelet coefficient table; the tongue texture wavelet coefficient table is subjected to statistical feature extraction to obtain a tongue texture feature table; The tongue body ROI image set is subjected to local binary pattern feature calculation to obtain a tongue LBP feature atlas; the tongue body ROI image set is subjected to color distribution feature analysis to obtain a tongue color feature vector; The tongue texture feature table, the tongue LBP feature atlas and the tongue color feature vector are subjected to feature fusion to obtain the tongue feature vector set.

4. The visual data processing system for Chinese medicine pediatrics of claim 1, wherein, The constructing of the pulse feature parameter set based on the pulse signal data set comprises: The pulse signal data set is subjected to pulse period detection and segmentation to obtain a single-period pulse segment set; The single-period pulse segment set is subjected to continuous wavelet transform to obtain a pulse time-frequency spectrum set; The pulse time-frequency spectrum set is subjected to frequency band energy feature extraction to obtain a pulse frequency domain feature set; The denoised pulse waveform data is subjected to time domain feature parameter calculation to obtain a pulse time domain feature table; The pulse feature parameter set is constructed according to the pulse frequency domain feature set and the pulse time domain feature table.

5. The visualized data processing system for Chinese medicine pediatrics of claim 1, wherein, The generating of a body region temperature feature table based on a thermal imaging image sequence comprises: The thermal imaging image sequence is subjected to temperature correction and pseudo-color mapping to obtain a standard body surface thermal imaging image set; The standard body surface thermal imaging image set is subjected to threshold segmentation to obtain a body surface region temperature mask image; The body surface region temperature mask image is subjected to region division marking to obtain a body part temperature region image; The body part temperature region image is subjected to calculation of region temperature statistical features to obtain a body region temperature feature table.

6. The visual data processing system for Chinese medicine pediatrics of claim 1, wherein, The constructing of a tongue time sequence index sequence based on a tongue feature vector set comprises: The tongue feature vector set is subjected to color correction to obtain a tongue color corrected feature vector set; The tongue color corrected feature vector set is subjected to tongue body and tongue fur demarcation feature extraction to obtain a tongue body-fur demarcation contour sequence; The tongue body-fur demarcation contour sequence is subjected to tongue contour evolution simulation to obtain tongue contour evolution trajectory data; The tongue contour evolution trajectory data is subjected to tongue body stretching and shrinking dynamic parameter calculation to obtain a tongue body motion index set; The tongue color corrected feature vector set is subjected to wavelet packet decomposition to obtain a tongue texture wavelet energy time sequence spectrum; The tongue texture wavelet energy time sequence spectrum is subjected to tongue fur roughness anisotropy analysis to obtain a tongue fur roughness-direction correlation index table; The tongue body-fur demarcation contour sequence is subjected to region humidity index calculation to obtain a tongue body humidity index time sequence; The tongue time sequence index sequence is generated based on the tongue body motion index set, the tongue fur roughness-direction correlation index table and the tongue body humidity index time sequence.

7. The visualized data processing system for Chinese medicine pediatrics of claim 1, wherein, The generating of a pulse time sequence index sequence based on a pulse feature parameter set comprises: The pulse feature parameter set is subjected to zero point drift correction to obtain a corrected pulse parameter time sequence; The pulse wave main wave form feature recognition is performed based on the corrected pulse condition parameter time sequence, and a pulse wave main wave form time sequence feature sequence is generated. The pulse wave main wave form time sequence feature sequence is subjected to time-varying form clustering, and a pulse wave time-varying form clustering label set is generated. The pulse wave time-varying form clustering label set is used to construct a pulse condition time-frequency joint feature tensor data. The pulse condition time-frequency joint feature tensor data is subjected to pulse condition dynamic index calculation, and a pulse condition dynamic index time sequence table is generated. The multi-lead signal synchronization is performed according to the pulse condition dynamic index time sequence table, and a multi-lead pulse condition phase difference feature atlas is generated. The blood vessel elasticity parameter is calculated based on the multi-lead pulse condition phase difference feature atlas, and a blood vessel elasticity index sequence is generated. The blood vessel elasticity index sequence is subjected to time-space evolution modeling, and a pulse condition time sequence index sequence is generated.

8. The visualized data processing system for Chinese medicine pediatrics of claim 1, wherein, The body surface temperature rhythm index sequence is generated based on the body region temperature feature table, including: The dynamic baseline correction is performed on the body region temperature feature table, and a corrected body surface temperature time sequence table is obtained. The time-space interpolation is performed on the corrected body surface temperature time sequence table, and a body surface temperature time-space continuous distribution field is generated. The body surface thermal feature extraction is performed on the body surface temperature time-space continuous distribution field, and a body surface heat flow conduction parameter sequence is generated. The temperature feature attractor graph is constructed according to the body surface heat flow conduction parameter sequence. The body surface temperature complexity analysis is performed based on the temperature feature attractor graph, and a body surface temperature complexity index table is generated. The oscillation mode decomposition is performed on the body surface temperature complexity index table, and a temperature oscillation feature set is obtained. The wavelet coherence analysis is performed on the temperature oscillation feature set, and a body surface-internal organ temperature intensity time sequence graph is generated. The body surface temperature rhythm index sequence is generated based on the body surface-internal organ temperature intensity time sequence graph.

Images