Breast cancer early warning method, device and equipment based on wavelet analysis and storage medium

Abstract

The application discloses a breast cancer early warning method and device based on wavelet analysis, equipment and a storage medium, and relates to the technical field of artificial intelligence. The method comprises the following steps: dividing a target population into healthy people, people with benign breast diseases and people with malignant breast diseases; collecting bilateral breast surface temperature time series of the target population through a temperature sensor; performing data cleaning and standardization processing on the temperature time series to generate a standardized temperature sequence; performing discrete wavelet transform on the standardized temperature sequence to extract multi-level wavelet decomposition coefficients; calculating a feature index set based on the multi-level wavelet decomposition coefficients, wherein the feature index set comprises an amplitude mean value, an over-threshold value count mean value and an over-threshold value count difference; training a machine learning model by using the feature index set and a preset label to generate a breast cancer risk prediction model; and implanting the breast cancer risk prediction model into a micro processing unit of a wearable device to output a breast cancer risk early warning result.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology, and in particular to a method, apparatus, device, and storage medium for early warning of breast cancer based on wavelet analysis. Background Technology

[0002] Breast cancer is one of the most common malignant tumors among women worldwide, and early detection and diagnosis are crucial for reducing mortality rates among breast cancer patients. However, due to a lack of expertise and equipment, the effectiveness of self-examination is limited, making it difficult to achieve the goal of early detection of breast cancer.

[0003] Currently, early screening for breast cancer mainly relies on imaging techniques, especially ultrasound technology, with screenings every three years for high-risk groups. However, a large number of breast cancer cases are still diagnosed in clinical practice only after patients experience significant symptoms at home, often by which time the disease has progressed to the middle or late stages.

[0004] Beneficial effects: The embodiments of this application achieve early warning of breast cancer through non-invasive temperature monitoring, wavelet transform feature extraction, machine learning model training and wearable device integration. Compared with the prior art, it has significant advantages such as non-invasiveness, high sensitivity, low cost, ease of implementation and strong universality, and can effectively improve the early detection rate of breast cancer. Summary of the Invention

[0005] The purpose of this application is to provide a method, device, equipment, and storage medium for early warning of breast cancer based on wavelet analysis.

[0006] To achieve the above objectives, this application provides the following solution:

[0007] Firstly, this application provides a breast cancer early warning method based on wavelet analysis, including:

[0008] The target population is divided into healthy individuals, individuals with benign breast diseases, and individuals with malignant breast diseases;

[0009] The time series of bilateral breast surface temperatures of the target population were collected using temperature sensors.

[0010] The temperature time series is cleaned and standardized to generate a standardized temperature series.

[0011] The standardized temperature sequence is subjected to discrete wavelet transform to extract multi-level wavelet decomposition coefficients;

[0012] Based on the multi-level wavelet decomposition coefficients, a set of feature indicators is calculated, which includes the mean amplitude, the mean over-threshold count, and the difference in over-threshold count.

[0013] A machine learning model is trained using the aforementioned set of feature indicators and preset labels to generate a breast cancer risk prediction model.

[0014] The breast cancer risk prediction model is embedded into the microprocessor unit of a wearable device to output breast cancer risk warning results.

[0015] Optionally, the step of collecting the time series of bilateral breast surface temperatures of the target population using a temperature sensor includes:

[0016] Under preset environmental conditions, temperature data of the surface of both breasts are continuously collected at a sampling frequency specified by a preset standard to generate the temperature time series.

[0017] Optionally, the step of cleaning and standardizing the temperature time series includes:

[0018] Filter out abnormal data points in the temperature time series that are below a preset temperature threshold;

[0019] Identify invalid time periods in the temperature time series caused by sensor detachment, and segment and remove these invalid time periods;

[0020] The left and right side data of the temperature time series are length aligned using linear interpolation to generate the standardized temperature series.

[0021] Optionally, the step of performing discrete wavelet transform on the standardized temperature sequence includes:

[0022] Select a preset wavelet basis function, perform multi-level discrete wavelet decomposition on the standardized temperature sequence, and generate the multi-level wavelet decomposition coefficients.

[0023] The Nyquist frequency is calculated based on the preset sampling frequency, and the frequency range of each level of wavelet decomposition is divided based on the Nyquist frequency.

[0024] The absolute values ​​of the multi-level wavelet decomposition coefficients are extracted to generate a multi-level wavelet amplitude sequence.

[0025] Optionally, the step of calculating the feature index set based on the multi-level wavelet decomposition coefficients includes:

[0026] Calculate the mean amplitude of the multi-level wavelet decomposition coefficients, where the mean amplitude is the average of the absolute values ​​of the left and right wavelet coefficients at the same decomposition level.

[0027] The number of coefficients exceeding a preset intensity threshold in the multi-level wavelet decomposition coefficients is counted, and the percentage of coefficients exceeding the threshold is calculated.

[0028] The weighted summation and difference calculations are performed on the proportions of over-threshold counts at the same decomposition level on the left and right sides to generate the mean of over-threshold counts and the difference of over-threshold counts.

[0029] Optionally, the step of training a machine learning model using the set of feature indicators and preset labels to generate a breast cancer risk prediction model includes:

[0030] A preset classification algorithm is selected based on the sample size, and the preset classification algorithm includes a logistic regression model or an ensemble learning model.

[0031] The set of feature indicators is input into the preset classification algorithm for model training, and regression coefficients and intercept terms are calculated.

[0032] The breast cancer risk prediction model was validated based on preset evaluation indicators, including AUC value, KS value, and PSI population stability index.

[0033] Optionally, the step of outputting the breast cancer risk warning result includes:

[0034] The real-time temperature sequence collected by the user wearing the wearable device is input into the breast cancer risk prediction model to generate a risk score;

[0035] If the risk score exceeds a preset risk threshold, a high-risk warning signal for breast cancer is sent to the user.

[0036] If the risk score is lower than the preset risk threshold, a periodic monitoring alert will be generated.

[0037] Secondly, this application provides a breast cancer early warning device based on wavelet analysis, comprising:

[0038] The acquisition module is used to divide the target population into healthy individuals, individuals with benign breast diseases, and individuals with malignant breast diseases.

[0039] The time series of bilateral breast surface temperatures of the target population were collected using temperature sensors.

[0040] The temperature time series is cleaned and standardized to generate a standardized temperature series.

[0041] The standardized temperature sequence is subjected to discrete wavelet transform to extract multi-level wavelet decomposition coefficients;

[0042] Based on the multi-level wavelet decomposition coefficients, a set of feature indicators is calculated, which includes the mean amplitude, the mean over-threshold count, and the difference in over-threshold count.

[0043] The algorithm module is used to train a machine learning model using the set of feature indicators and preset labels to generate a breast cancer risk prediction model;

[0044] The prediction module is used to embed the breast cancer risk prediction model into the microprocessor unit of the wearable device and output breast cancer risk warning results.

[0045] Optionally, the acquisition module is further configured to:

[0046] Under preset environmental conditions, temperature data of the surface of both breasts are continuously collected at a sampling frequency specified by a preset standard to generate the temperature time series.

[0047] Optionally, the acquisition module is further configured to:

[0048] Filter out abnormal data points in the temperature time series that are below a preset temperature threshold;

[0049] Identify invalid time periods in the temperature time series caused by sensor detachment, and segment and remove these invalid time periods;

[0050] The left and right side data of the temperature time series are length aligned using linear interpolation to generate the standardized temperature series.

[0051] Optionally, the acquisition module is further configured to:

[0052] Select a preset wavelet basis function, perform multi-level discrete wavelet decomposition on the standardized temperature sequence, and generate the multi-level wavelet decomposition coefficients.

[0053] The Nyquist frequency is calculated based on the preset sampling frequency, and the frequency range of each level of wavelet decomposition is divided based on the Nyquist frequency.

[0054] The absolute values ​​of the multi-level wavelet decomposition coefficients are extracted to generate a multi-level wavelet amplitude sequence.

[0055] Optionally, the acquisition module is further configured to:

[0056] Calculate the mean amplitude of the multi-level wavelet decomposition coefficients, where the mean amplitude is the average of the absolute values ​​of the left and right wavelet coefficients at the same decomposition level.

[0057] The number of coefficients exceeding a preset intensity threshold in the multi-level wavelet decomposition coefficients is counted, and the percentage of coefficients exceeding the threshold is calculated.

[0058] The weighted summation and difference calculations are performed on the proportions of over-threshold counts at the same decomposition level on the left and right sides to generate the mean of over-threshold counts and the difference of over-threshold counts.

[0059] Optionally, the algorithm module is further configured to:

[0060] A preset classification algorithm is selected based on the sample size, and the preset classification algorithm includes a logistic regression model or an ensemble learning model.

[0061] The set of feature indicators is input into the preset classification algorithm for model training, and regression coefficients and intercept terms are calculated.

[0062] The breast cancer risk prediction model was validated based on preset evaluation indicators, including AUC value, KS value, and PSI population stability index.

[0063] Optionally, the prediction module is further configured to:

[0064] The real-time temperature sequence collected by the user wearing the wearable device is input into the breast cancer risk prediction model to generate a risk score;

[0065] If the risk score exceeds a preset risk threshold, a high-risk warning signal for breast cancer is sent to the user.

[0066] If the risk score is lower than the preset risk threshold, a periodic monitoring alert will be generated.

[0067] Thirdly, this application provides a computer device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the wavelet analysis-based breast cancer early warning method described in any one of the above.

[0068] Fourthly, this application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the wavelet analysis-based breast cancer early warning method described above.

[0069] Fifthly, this application provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the wavelet analysis-based breast cancer early warning method described above.

[0070] According to the specific embodiments provided in this application, the following technical effects are disclosed:

[0071] This application provides a breast cancer early warning method, device, equipment, and storage medium based on wavelet analysis. Through non-invasive temperature monitoring, wavelet transform feature extraction, machine learning model training, and integration with wearable devices, it achieves early warning of breast cancer. Compared with existing technologies, it has significant advantages such as non-invasiveness, high sensitivity, low cost, ease of implementation, and strong universality, and can effectively improve the early detection rate of breast cancer. Attached Figure Description

[0072] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0073] Figure 1 A flowchart illustrating a breast cancer early warning method based on wavelet analysis, provided as an embodiment of this application;

[0074] Figure 2 A schematic diagram illustrating the principle of a breast cancer early warning method based on wavelet analysis, provided as an embodiment of this application;

[0075] Figure 3 A schematic diagram of the functional modules of a breast cancer early warning device based on wavelet analysis provided in an embodiment of this application;

[0076] Figure 4 This is a schematic diagram of the structure of a computer device provided in an embodiment of this application. Detailed Implementation

[0077] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0078] like Figure 1 As shown, some embodiments of this application provide a breast cancer early warning method based on wavelet analysis. In these embodiments, the method includes:

[0079] Step 101: Divide the target population into healthy individuals, individuals with benign breast diseases, and individuals with malignant breast diseases.

[0080] In this application embodiment, the target population refers to the group of women participating in the breast cancer early warning algorithm research, divided into different categories based on their breast health status: Healthy population: Women with normal breast health and no breast diseases. Benign breast disease population: Women with benign breast lesions (such as breast hyperplasia, breast fibroadenoma, etc.). Malignant breast disease population: Women with breast cancer.

[0081] Reference Figure 2The system first categorizes the target population into three groups based on breast health status: healthy individuals, those with benign breast diseases, and those with malignant breast diseases. This categorization ensures that subsequent data collection and model training cover different breast health states, thereby improving the model's generalization ability and predictive accuracy. The system collaborates with medical institutions to recruit volunteers who meet the research requirements and classifies them according to their breast health status, ensuring a balanced number of participants in each group for subsequent model training and testing.

[0082] Step 102: Collect the time series of bilateral breast surface temperature of the target population using a temperature sensor.

[0083] In this embodiment, the temperature sensor is an electronic device used to measure the surface temperature of an object, capable of converting temperature signals into electrical signals for recording. The bilateral breast surface temperature time series refers to a sequence formed by arranging the surface temperature data of the left and right breasts continuously collected by the temperature sensor in chronological order.

[0084] The system uses high-precision temperature sensors attached to the left and right breasts of the target population, ensuring close contact between the sensors and the skin, and simultaneously collecting temperature data from both breasts under identical environmental conditions. The temperature sensors continuously collected data at a frequency of once every 10 seconds for 12 hours, generating a time series of bilateral breast surface temperatures. The purpose of this step is to obtain dynamic changes in breast surface temperature, providing a foundation for subsequent wavelet analysis and feature extraction.

[0085] Step 103: Perform data cleaning and standardization on the temperature time series to generate a standardized temperature series.

[0086] In this embodiment, data cleaning refers to processing the collected raw data to remove noise, outliers, or invalid data, ensuring the accuracy and consistency of the data. Standardization processing refers to converting the data into a unified format or range for subsequent analysis and processing. A standardized temperature series refers to a temperature time series that has undergone data cleaning and standardization, possessing a unified format and range.

[0087] The system cleans the collected bilateral breast surface temperature time series data, removing abnormal data such as low temperature values ​​(≤32℃) and identifying missing test time periods. The original data is then segmented to extract the valid measurement data sequences. Due to potential minor errors in the acquisition time of the left and right sides, the system uses linear interpolation to standardize the data, ensuring consistent data length on both sides. Finally, a standardized temperature sequence is generated, providing clean input data for subsequent wavelet transform.

[0088] Step 104: Perform discrete wavelet transform on the standardized temperature sequence to extract multi-level wavelet decomposition coefficients.

[0089] In this embodiment, Discrete Wavelet Transform (DWT) is a signal processing method that can decompose a signal into sub-bands of different frequencies, suitable for the analysis of non-stationary signals. Multi-level wavelet decomposition coefficients refer to the coefficients obtained after decomposing the signal into multiple frequency sub-bands through Discrete Wavelet Transform, with each sub-band corresponding to a different frequency range.

[0090] The system performs Discrete Wavelet Transform (DWT) on the standardized two-sided temperature sequence, using Symlets wavelet functions as basis functions. Since the sampling frequency is 10 seconds / sample (0.1 Hz), the Nyquist frequency of the wavelet transform is 0.05 Hz. The system performs multi-level wavelet decomposition from level 0 to level 5, plotting the original signal and the DWT coefficients at each level. Discrete Wavelet Transform can capture subtle fluctuations in the temperature sequence, providing multi-level frequency information for subsequent feature extraction.

[0091] Step 105: Based on the multi-level wavelet decomposition coefficients, calculate the feature index set, which includes the mean amplitude, the mean over-threshold count, and the over-threshold count difference.

[0092] In this embodiment, the mean amplitude refers to the average absolute value of the wavelet decomposition coefficients at a certain level, reflecting the average fluctuation intensity of the wavelet coefficients at that level. The mean over-threshold count refers to the ratio of the number of coefficients whose absolute value exceeds a preset threshold to the total number of coefficients at a certain level, reflecting the distribution of the fluctuation intensity of the wavelet coefficients at that level. The over-threshold count difference refers to the difference in the over-threshold count between the left and right breasts at a certain level of wavelet decomposition coefficients, reflecting the symmetry of temperature fluctuations between the two breasts.

[0093] The system extracts a set of feature indices from the multi-level power spectrum after wavelet transform. First, it calculates the mean amplitude of each wavelet level, i.e., averaging the absolute values ​​of the wavelet decomposition coefficients at a given level. Second, it calculates the average count of coefficients exceeding a preset threshold at each wavelet level, i.e., the ratio of the number of coefficients with absolute values ​​exceeding a preset threshold to the total number of coefficients at a given wavelet decomposition level. Finally, it calculates the difference in threshold counts, i.e., the difference between the weighted average of the threshold counts for the left and right breasts. These feature indices effectively reflect the intensity and symmetry of breast temperature fluctuations, providing input features for subsequent model training.

[0094] The mean amplitude of each wavelet level is calculated using the following formula (1):

[0095]

[0096] Where coeff N，i Let k be the i-th coefficient of the Level N wavelet expansion. NThe Level N amplitude mean is the number of coefficients in the N-order expansion. It is actually the average of the absolute values ​​of the N-level DWT expansion coefficients (average fluctuation intensity). The Level N amplitude mean is calculated separately for the left and right sensor data, with each level corresponding to the mean of the left and right sides.

[0097] The calculation method for the average of the coefficient counts above each wavelet threshold is as follows: Formula (2):

[0098]

[0099] Among them, #{|coeff N，i |≥threshold N The} represents the number of coefficients whose absolute values ​​exceed the threshold. This indicator is calculated separately for the left and right sensor data.

[0100] The threshold is set as follows (the same threshold is used for items of the same level):

[0101] threshold0: 0.04

[0102] threshold1: 0.08

[0103] threshold2: 0.16

[0104] The mean amplitude values ​​(feature (1)) of wavelets at Level 0, Level 1, and Level 2 are summed on the left and right sides, and a weighted average is taken for Level 0, Level 1, and Level 2 to obtain the mean amplitude index.

[0105] The average of the above-threshold coefficient counts (feature (2)) of Level 0, Level 1 and Level 2 on the left and right sides is added together, and a weighted average is taken for Level 0, Level 1 and Level 2 to obtain the average above-threshold count index.

[0106] The average of the above-threshold coefficients (feature (2)) of Level 0, Level 1 and Level 2 on the left and right sides is first weighted and then the values ​​obtained from the left and right sides are subtracted and the absolute value is taken to obtain the above-threshold count difference index.

[0107] Step 106: Use the set of feature indicators and preset labels to train a machine learning model to generate a breast cancer risk prediction model.

[0108] In this embodiment, the preset label refers to a known classification label used in model training, such as healthy, benign breast disease, malignant breast disease, etc. A machine learning model refers to a mathematical model that learns patterns from data and makes predictions through algorithms, such as logistic regression models, ensemble learning models, etc. A breast cancer risk prediction model refers to a model generated through training that can predict breast cancer risk based on input features.

[0109] The system trains a machine learning model using a set of feature indicators (mean amplitude, mean overthreshold count, and difference in overthreshold count) and predefined labels (healthy, benign breast disease, malignant breast disease). Based on sample size and data quality, the system selects a logistic regression model for training. During model training, the system obtains regression coefficients by fitting the data and evaluates the model using a test set, calculating evaluation metrics such as AUC and KS values ​​to ensure the model's discriminative power and robustness. The final breast cancer risk prediction model can predict breast cancer risk based on input features.

[0110] For example, using 50 samples from the training set, the mean amplitude (feature (3)), the mean overthreshold count (feature (4)), and the difference in overthreshold count (feature (5)) are selected, and a Logistic regression model is trained in conjunction with the data y-label. The model includes an intercept term. The Logistic regression model formula (3) is:

[0111]

[0112] Where α is the intercept term, and β1 to β3 are the regression coefficients of the mean amplitude, the mean of the over-threshold count, and the difference in the over-threshold count, respectively.

[0113] Based on the model fitting results, the Pseudo R-squared value is 0.523, and the p-value is 6.417 × 10⁻⁶. -6 The model is highly significant, with all feature regression coefficients p-values ​​less than 0.1.

[0114] On the test set, the model has a KS value of 0.70 and an AUC of 0.85, demonstrating good discriminative power. In addition, the model has a PSI of 0.14, indicating that the model is reasonably robust and also shows good robustness in cross-validation.

[0115] In the test set of 300 cases, there were 100 cases each of normal breast tissue, benign breast disease, and breast cancer. That is, 100 cases of breast cancer data and 200 cases of non-breast cancer data. The trained algorithm was used to analyze this data. The algorithm output showed that it accurately predicted 88 breast cancer cases and 190 cases of non-breast cancer cases (including healthy individuals and those with benign breast disease), with 12 false negatives and 10 false positives. Therefore, the overall accuracy of the algorithm is 92.67%, sensitivity is 88%, specificity is 95%, false positive rate is 5%, and false negative rate is 12%. This meets our pre-set qualification criteria. In summary, the model has good discrimination and robustness and meets the standards for practical use.

[0116] Step 107: The breast cancer risk prediction model is embedded into the microprocessor unit of the wearable device, and the breast cancer risk warning result is output.

[0117] In this embodiment, wearable devices refer to electronic devices that can be worn on the body, such as smart bracelets and smart bras, and have data acquisition and processing functions. A microprocessor unit (MCU) refers to the core processor in a wearable device, responsible for executing algorithms and data processing tasks. The breast cancer risk warning result refers to the warning information output after analysis by a breast cancer risk prediction model, indicating to the user whether there is a risk of breast cancer.

[0118] The system embeds a pre-trained breast cancer risk prediction model into the microprocessor unit (MCU) of a wearable device. Users wear the device for 12 hours of home self-examination. The device collects time-series temperature data from both breast surfaces using temperature sensors and analyzes the data using the embedded model. If the breast cancer characteristics of either breast reach or exceed a preset threshold, the device issues a high-risk breast cancer warning, prompting the user to seek medical attention promptly. If the characteristic values ​​of both breasts are below the threshold, the device does not issue a warning but simply reminds the user to continue monthly monitoring.

[0119] This application embodiment achieves early warning of breast cancer through non-invasive temperature monitoring, wavelet transform feature extraction, machine learning model training, and the integration of wearable devices. Compared with existing technologies, it has significant advantages such as being non-invasive, highly sensitive, low-cost, easy to implement, and highly universal, and can effectively improve the early detection rate of breast cancer.

[0120] Optionally, step 102 includes: continuously collecting temperature data of the bilateral breast surfaces at a sampling frequency specified by a preset standard under preset environmental conditions, and generating the temperature time series.

[0121] In this embodiment, preset environmental conditions refer to the constant environmental conditions set by the system during data acquisition, including temperature and humidity, to ensure the consistency and comparability of the data. The preset sampling frequency refers to the data acquisition frequency pre-set by the system according to research needs, i.e., collecting temperature data at fixed intervals. The bilateral breast surfaces refer to the external skin surfaces of the left and right breasts of the human body, to which temperature sensors are attached to collect temperature data. Temperature data refers to the breast surface temperature values ​​measured by temperature sensors, usually in degrees Celsius (°C). A temperature time series refers to a continuous set of temperature data arranged chronologically, reflecting the changes in breast surface temperature over time.

[0122] Under preset environmental conditions, the system continuously collects temperature data from the surfaces of both breasts using temperature sensors at a pre-defined sampling frequency (e.g., once every 10 seconds). The temperature sensors are attached to the breast surfaces to ensure close contact with the skin, and data is collected from both breasts simultaneously under the same environmental conditions. The system generates a time series of breast surface temperatures over 12 hours of continuous data collection. The purpose of this step is to obtain dynamic changes in breast surface temperature, ensuring data continuity and consistency, and providing a foundation for subsequent data processing and analysis.

[0123] Optionally, step 103 includes:

[0124] Step 1031: Filter out abnormal data points in the temperature time series that are below a preset temperature threshold.

[0125] In this embodiment, the temperature time series refers to a continuous set of temperature data arranged in chronological order, reflecting the change in breast surface temperature over time. The preset temperature threshold is a lower limit value pre-set by the system to determine whether the temperature data is abnormal. Temperature data below this threshold is considered abnormal data points. Abnormal data points refer to abnormal temperature values ​​caused by measurement errors, sensor malfunctions, or other external factors, typically manifested as excessively low or high temperature values.

[0126] The system first filters out abnormal data points in the temperature time series. Based on a preset temperature threshold (e.g., 32℃), the system automatically identifies and filters out abnormal data points in the temperature time series that are below this threshold. These abnormal data points may be caused by poor sensor contact with skin, excessively low ambient temperature, or other measurement errors. By filtering these abnormal data points, the system ensures that the temperature data analyzed subsequently has high accuracy and reliability.

[0127] Step 1032: Identify invalid time periods in the temperature time series caused by sensor detachment, and segment and remove the invalid time periods.

[0128] In this embodiment, sensor detachment refers to the temperature sensor separating from the skin during measurement due to improper wearing or other reasons, thus failing to collect temperature data normally. Invalid time periods refer to the time periods during which temperature data is invalid due to sensor detachment or other reasons; temperature data within these periods cannot be used for subsequent analysis. Segmentation and removal refers to separating and deleting invalid time periods from the temperature time series to ensure the continuity and validity of the remaining data.

[0129] The system analyzes temperature time series data to identify invalid time periods caused by sensor detachment. Based on the continuity and variation patterns of temperature data, the system automatically detects periods where the temperature value suddenly drops or remains unchanged for a long time, identifying these as invalid time periods caused by sensor detachment. The system segments and removes these invalid time periods from the temperature time series, ensuring the continuity and validity of the remaining data and providing clean input data for subsequent standardization processing.

[0130] Step 1033: Use linear interpolation to perform length alignment processing on the left and right side data of the temperature time series to generate the standardized temperature series.

[0131] In this embodiment, linear interpolation is a mathematical method used to estimate the value of unknown data points based on the linear relationship between known data points. Left and right side data refer to the temperature time series of the left and right breast surfaces collected separately by temperature sensors. Length alignment processing refers to using interpolation or other methods to make the lengths of the left and right side data consistent for subsequent synchronous analysis. A standardized temperature series refers to a temperature time series that has undergone data cleaning, invalid time period removal, and length alignment processing, possessing a uniform format and length.

[0132] The system performs length alignment processing on the left and right temperature time series. Due to potential minor errors in the data acquisition times of the left and right sides, the data lengths may differ. The system employs linear interpolation to interpolate the shorter data set, aligning its length with the longer set. Linear interpolation estimates the values ​​of missing data points based on the linear relationship between known data points, ensuring consistent data lengths on both sides. The resulting standardized temperature series has a uniform format and length, providing standardized input data for subsequent wavelet transform and feature extraction.

[0133] The embodiments of this application can perform abnormal data filtering, invalid time period removal, and length alignment processing on temperature time series to generate standardized temperature series.

[0134] Optionally, step 104 includes:

[0135] Step 1041: Select a preset wavelet basis function, perform multi-level discrete wavelet decomposition on the standardized temperature sequence, and generate the multi-level wavelet decomposition coefficients.

[0136] In this embodiment, the preset wavelet basis function refers to the wavelet transform basis function pre-selected by the system for signal decomposition. Common wavelet basis functions include Symlets, Daubechies, etc. The standardized temperature sequence refers to a temperature time series after data cleaning, invalid time period removal, and length alignment, possessing a uniform format and length. Multilevel discrete wavelet decomposition refers to the process of decomposing a signal into multiple frequency sub-bands through discrete wavelet transform (DWT), each sub-band corresponding to a different frequency range. Multilevel wavelet decomposition coefficients refer to the coefficients obtained after decomposing the signal into multiple frequency sub-bands through discrete wavelet transform, each sub-band corresponding to a different frequency range.

[0137] The system first selects a preset wavelet basis function (such as the Symlets wavelet function) to perform multi-level discrete wavelet decomposition on the standardized temperature sequence. The system decomposes the temperature sequence into multiple frequency sub-bands, each corresponding to a different wavelet decomposition level (e.g., level 0 to level 5). Through discrete wavelet transform, the system generates multi-level wavelet decomposition coefficients, which reflect the fluctuation characteristics of the temperature sequence within different frequency ranges. Multi-level wavelet decomposition can capture subtle changes in the temperature sequence, providing multi-layered frequency information for subsequent feature extraction.

[0138] Step 1042: Calculate the Nyquist frequency based on the preset sampling frequency, and divide the frequency range of each level of wavelet decomposition based on the Nyquist frequency.

[0139] In this embodiment, the preset sampling frequency refers to the sampling frequency set by the system during data acquisition, i.e., collecting temperature data once at fixed intervals (e.g., once every 10 seconds). The Nyquist frequency is the highest effective frequency in signal processing, equal to half the sampling frequency. According to the Nyquist sampling theorem, the sampling frequency of a signal must be at least twice the highest frequency of the signal to accurately reconstruct the signal. The frequency range refers to the distribution range of the signal in the frequency domain, usually expressed in Hertz (Hz).

[0140] The system calculates the Nyquist frequency based on a preset sampling frequency (e.g., 0.1Hz, meaning data is collected every 10 seconds). The Nyquist frequency is half the sampling frequency, i.e., 0.05Hz. The system then divides the frequency range of each wavelet decomposition level based on the Nyquist frequency. Since the frequency range of each wavelet decomposition level is half that of the previous level, the system performs multi-level decomposition from level 0 to level 5, with the frequency range of each level decreasing by half. For example, the frequency range of level 0 is 0-0.05Hz, the frequency range of level 1 is 0-0.025Hz, and so on. By dividing the frequency range, the system can more accurately analyze the fluctuation characteristics of temperature sequences in different frequency bands.

[0141] Step 1043: Extract the absolute values ​​of the multi-level wavelet decomposition coefficients to generate a multi-level wavelet amplitude sequence.

[0142] In this embodiment, the system extracts the absolute values ​​of the multi-level wavelet decomposition coefficients to generate a multi-level wavelet amplitude sequence. First, the system takes the absolute value of each level of wavelet decomposition coefficient to obtain the fluctuation intensity of that level's wavelet coefficient. Then, the system arranges the absolute values ​​of each level of wavelet coefficients in chronological order to generate a multi-level wavelet amplitude sequence. These amplitude sequences reflect the fluctuation intensity of the temperature sequence within different frequency ranges, providing important input data for subsequent feature extraction and model training.

[0143] The embodiments of this application can perform multi-level discrete wavelet decomposition on a standardized temperature sequence, generate multi-level wavelet decomposition coefficients, and divide the frequency range of each wavelet decomposition level according to the Nyquist frequency. The system generates a multi-level wavelet amplitude sequence by extracting the absolute values ​​of the multi-level wavelet decomposition coefficients.

[0144] Optionally, step 105 includes:

[0145] Step 1051: Calculate the mean amplitude of the multi-level wavelet decomposition coefficients, where the mean amplitude is the average of the absolute values ​​of the left and right wavelet coefficients at the same decomposition level.

[0146] In this embodiment, multi-level wavelet decomposition coefficients refer to the coefficients obtained after decomposing the signal into multiple frequency sub-bands through discrete wavelet transform, with each sub-band corresponding to a different frequency range. The amplitude mean refers to the average absolute value of a certain level of wavelet decomposition coefficients, reflecting the average fluctuation intensity of that level of wavelet coefficients. The left and right side wavelet coefficients refer to the coefficients obtained after wavelet decomposition of the temperature time series of the left and right breast surfaces collected by temperature sensors.

[0147] The system first processes the multi-level wavelet decomposition coefficients, calculating the mean amplitude of each level. The system then takes the absolute value of the wavelet decomposition coefficients for both the left and right breasts to obtain the fluctuation intensity of the left and right wavelet coefficients. Next, the system averages the absolute values ​​of the left and right wavelet coefficients at the same decomposition level to obtain the mean amplitude of the wavelet decomposition coefficients at that level. The mean amplitude reflects the average fluctuation intensity of the wavelet coefficients at that level, providing basic data for subsequent feature extraction.

[0148] Step 1052: Count the number of coefficients in the multi-level wavelet decomposition coefficients that exceed the preset intensity threshold, and calculate the percentage of coefficients exceeding the threshold.

[0149] In this embodiment, the preset intensity threshold refers to a lower limit value for intensity set in advance by the system, used to determine whether the wavelet decomposition coefficients have significant fluctuation intensity. Wavelet coefficients exceeding this threshold are considered to have significant fluctuation. The percentage of coefficients exceeding the threshold is the ratio of the number of coefficients exceeding the preset intensity threshold in a certain level of wavelet decomposition coefficients to the total number of wavelet coefficients in that level, reflecting the distribution of fluctuation intensity of the wavelet coefficients in that level.

[0150] The system performs threshold-exceeding statistics on multi-level wavelet decomposition coefficients. First, based on preset intensity thresholds (e.g., 0.04, 0.08, 0.16), the system counts the number of coefficients exceeding a certain threshold at each level. Then, the system calculates the threshold-exceeding percentage, which is the ratio of the number of coefficients exceeding the threshold to the total number of wavelet coefficients at that level. This threshold-exceeding percentage reflects the intensity distribution of fluctuations in the wavelet coefficients at that level, providing an important indicator for subsequent feature extraction.

[0151] Step 1053: Perform weighted summation and difference calculation on the proportion of over-threshold counts at the same decomposition level on the left and right sides to generate the mean of over-threshold counts and the difference of over-threshold counts.

[0152] The system performs weighted summation and difference calculation on the proportion of overthreshold counts at the same decomposition level on both sides. First, the system performs a weighted summation of the proportions of overthreshold counts at the same decomposition level on both sides to obtain the mean overthreshold count. The weighting method can be set according to specific needs, for example, assigning different weights to level 0, level 1, and level 2. Then, the system calculates the difference between the proportions of overthreshold counts at the same decomposition level on both sides, taking the absolute value to obtain the overthreshold count difference. The mean overthreshold count reflects the overall fluctuation intensity of both breasts, while the overthreshold count difference reflects the symmetry of fluctuation intensity between the two breasts. These indicators provide important feature inputs for subsequent model training.

[0153] This application's embodiments can calculate the mean amplitude and the proportion of out-of-threshold counts of multi-level wavelet decomposition coefficients, and generate the mean and difference of out-of-threshold counts. These characteristic indicators can effectively reflect the intensity, distribution, and symmetry of breast temperature fluctuations, providing multi-level and multi-dimensional input data for subsequent machine learning model training. Through these features, the system can more accurately identify early characteristics of breast cancer, thereby improving the accuracy and reliability of breast cancer early warning.

[0154] Optionally, step 106 includes:

[0155] Step 1061: Select a preset classification algorithm based on the sample size. The preset classification algorithm includes a logistic regression model or an ensemble learning model.

[0156] In this embodiment, sample size refers to the number of data samples used for model training. The size of the sample size directly affects the selection and performance of the model. The pre-defined classification algorithm refers to the machine learning algorithm pre-set by the system for the classification task. Common classification algorithms include logistic regression models and ensemble learning models. A logistic regression model is a linear model used for binary or multi-class classification problems. It obtains regression coefficients by fitting data to predict the class probability of a sample. An ensemble learning model is a machine learning method that improves model performance by combining multiple base learners (such as decision trees). Common ensemble learning models include XGBoost and Random Forest.

[0157] The system selects a pre-defined classification algorithm based on the sample size. If the sample size is small, the system chooses a logistic regression model for training because it exhibits good stability and interpretability with small sample sizes. If the sample size is large, the system chooses an ensemble learning model (such as XGBoost) for training because ensemble learning models can handle large amounts of data and have high prediction accuracy under complex data distributions. The system automatically selects an appropriate classification algorithm based on sample size and data quality to ensure the efficiency and effectiveness of model training.

[0158] Step 1062: Input the set of feature indicators into the preset classification algorithm for model training, and calculate the regression coefficients and intercept terms.

[0159] In this embodiment, the system inputs a set of feature indicators (such as mean amplitude, mean overthreshold count, and difference in overthreshold count) into a preset classification algorithm for model training. If a logistic regression model is selected, the system calculates the regression coefficient and intercept term for each feature indicator by fitting the data. The regression coefficient reflects the contribution of each feature indicator to breast cancer risk, while the intercept term is used to adjust the model's prediction results. The system optimizes the regression coefficient and intercept term by minimizing the loss function, making the model's prediction results as consistent as possible with the actual labels. The final generated model can predict breast cancer risk based on the input feature indicators.

[0160] Step 1063: Validate the breast cancer risk prediction model based on preset evaluation indicators, including AUC value, KS value, and PSI population stability index.

[0161] In this embodiment, the AUC value refers to the area under the ROC curve, used to evaluate the classification performance of the model. The closer the AUC value is to 1, the better the classification performance of the model. The KS value refers to the Kolmogorov-Smirnov statistic, used to evaluate the maximum discrimination of the model. The larger the KS value, the stronger the discrimination ability of the model. The PSI population stability index is used to evaluate the stability of the model in different populations. The smaller the PSI value, the better the stability of the model.

[0162] The system uses pre-defined evaluation metrics to validate the breast cancer risk prediction model. First, the system calculates the model's AUC value to assess its classification performance. A value closer to 1 indicates a stronger ability to distinguish between breast cancer risk and non-breast cancer risk. Second, the system calculates the KS value to assess the model's maximum discrimination. A higher KS value indicates a stronger ability to distinguish between different categories (e.g., healthy, benign breast disease, malignant breast disease). Finally, the system calculates the PSI (Population Stability Index) to assess the model's stability across different populations. A lower PSI value indicates more stable prediction results across different populations. The system comprehensively evaluates the model's performance using these metrics to ensure its accuracy, discrimination, and stability.

[0163] The system in this application embodiment can select a suitable classification algorithm based on the sample size, input the feature index set into the model for training, and verify the breast cancer risk prediction model based on preset evaluation indicators.

[0164] Optionally, step 107 includes:

[0165] Step 1071: Input the real-time temperature sequence collected by the user wearing the wearable device into the breast cancer risk prediction model to generate a risk score.

[0166] In this embodiment, the system collects real-time temperature sequences through a wearable device worn by the user and inputs these sequences into a breast cancer risk prediction model. The model analyzes the real-time temperature sequences, extracts feature indicators (such as mean amplitude, mean number of exceedances, and difference in number of exceedances), and calculates a risk score based on these feature indicators. The risk score is a numerical result output by the model, reflecting a quantitative assessment of the user's current breast cancer risk. The system uses the risk score to determine whether the user has a risk of breast cancer.

[0167] Step 1072: If the risk score exceeds a preset risk threshold, a high-risk warning signal for breast cancer is sent to the user.

[0168] In this embodiment, the system determines whether a risk score generated by a breast cancer risk prediction model exceeds a preset risk threshold. If the risk score exceeds the preset threshold, the system sends a high-risk breast cancer warning signal to the user. The warning signal can be sent to the user via the wearable device's display screen, vibration alerts, or mobile application notifications. The warning signal typically includes a notification that the user has a high risk of breast cancer and a recommendation that the user visit a hospital for further examination within one month.

[0169] Step 1073: If the risk score is lower than the preset risk threshold, a periodic monitoring prompt message is generated.

[0170] In this embodiment, if the risk score is lower than a preset risk threshold, the system generates a periodic monitoring reminder and sends it to the user via the wearable device's display or a mobile application notification. The reminder typically includes a recommendation that the user wear the wearable device monthly for breast health monitoring to ensure early detection of potential risks. The purpose of the periodic monitoring reminder is to help users maintain awareness of their breast health and promptly detect any abnormal changes through continuous monitoring.

[0171] The embodiments of this application can generate a breast cancer risk score based on the real-time temperature sequence collected by the user wearing a wearable device, and send a high-risk breast cancer warning signal or regular monitoring reminder information to the user based on the score results.

[0172] Based on the same inventive concept, this application also provides a wavelet analysis-based breast cancer early warning device for implementing the aforementioned wavelet analysis-based breast cancer early warning method. The solution provided by this device is similar to the implementation described in the above method; therefore, the specific limitations of one or more wavelet analysis-based breast cancer early warning device embodiments provided below can be found in the limitations of the wavelet analysis-based breast cancer early warning method described above, and will not be repeated here.

[0173] In one exemplary embodiment, such as Figure 3 As shown, a breast cancer early warning device 20 based on wavelet analysis is provided, comprising:

[0174] The acquisition module 201 is used to divide the target population into healthy people, people with benign breast diseases, and people with malignant breast diseases;

[0175] The temperature time series of the bilateral breast surface of the target population was collected using a temperature sensor;

[0176] The temperature time series is cleaned and standardized to generate a standardized temperature series.

[0177] The standardized temperature sequence is subjected to discrete wavelet transform to extract multi-level wavelet decomposition coefficients;

[0178] Based on the multi-level wavelet decomposition coefficients, a set of feature indicators is calculated, which includes the mean amplitude, the mean over-threshold count, and the difference in over-threshold count.

[0179] Algorithm module 202 is used to train a machine learning model using the set of feature indicators and preset labels to generate a breast cancer risk prediction model;

[0180] The prediction module 203 is used to embed the breast cancer risk prediction model into the microprocessor unit of the wearable device and output the breast cancer risk warning result.

[0181] Optionally, the acquisition module 201 is further configured to:

[0182] Under preset environmental conditions, temperature data of the surface of both breasts are continuously collected at a sampling frequency specified by a preset standard to generate the temperature time series.

[0183] Optionally, the acquisition module 201 is further configured to:

[0184] Filter out abnormal data points in the temperature time series that are below a preset temperature threshold;

[0185] Identify invalid time periods in the temperature time series caused by sensor detachment, and segment and remove these invalid time periods;

[0186] The left and right side data of the temperature time series are length aligned using linear interpolation to generate the standardized temperature series.

[0187] Optionally, the acquisition module 201 is further configured to:

[0188] Select a preset wavelet basis function, perform multi-level discrete wavelet decomposition on the standardized temperature sequence, and generate the multi-level wavelet decomposition coefficients.

[0189] The Nyquist frequency is calculated based on the preset sampling frequency, and the frequency range of each level of wavelet decomposition is divided based on the Nyquist frequency.

[0190] Extract the absolute values ​​of the multi-level wavelet decomposition coefficients to generate a multi-level wavelet amplitude sequence.

[0191] Optionally, the acquisition module 201 is further configured to:

[0192] Calculate the mean amplitude of the multi-level wavelet decomposition coefficients, where the mean amplitude is the average of the absolute values ​​of the left and right wavelet coefficients at the same decomposition level.

[0193] The number of coefficients exceeding a preset intensity threshold in the multi-level wavelet decomposition coefficients is counted, and the percentage of coefficients exceeding the threshold is calculated.

[0194] The weighted summation and difference calculations are performed on the proportions of over-threshold counts at the same decomposition level on the left and right sides to generate the mean of over-threshold counts and the difference of over-threshold counts.

[0195] Optionally, the algorithm module 202 is further configured to:

[0196] A preset classification algorithm is selected based on the sample size, and the preset classification algorithm includes a logistic regression model or an ensemble learning model.

[0197] The set of feature indicators is input into the preset classification algorithm for model training, and regression coefficients and intercept terms are calculated.

[0198] The breast cancer risk prediction model was validated based on preset evaluation indicators, including AUC value, KS value, and PSI population stability index.

[0199] Optionally, the prediction module 203 is further configured to:

[0200] The real-time temperature sequence collected by the user wearing the wearable device is input into the breast cancer risk prediction model to generate a risk score;

[0201] If the risk score exceeds a preset risk threshold, a high-risk warning signal for breast cancer is sent to the user.

[0202] If the risk score is lower than the preset risk threshold, a periodic monitoring alert will be generated.

[0203] This application embodiment achieves early warning of breast cancer through non-invasive temperature monitoring, wavelet transform feature extraction, machine learning model training, and the integration of wearable devices. Compared with existing technologies, it has significant advantages such as being non-invasive, highly sensitive, low-cost, easy to implement, and highly universal, and can effectively improve the early detection rate of breast cancer.

[0204] In one exemplary embodiment, a computer device is provided, which may be a server or a terminal, and its internal structure diagram may be as follows. Figure 4As shown, this computer device includes a processor, memory, input / output (I / O) interfaces, and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the I / O interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides the environment for the operating system and computer programs stored in the non-volatile storage media. The database stores breast cancer early warning data based on wavelet analysis. The I / O interfaces are used for information exchange between the processor and external devices. The communication interface is used for communication with external terminals via a network connection. When the computer program is executed by the processor, it implements a breast cancer early warning method based on wavelet analysis.

[0205] Those skilled in the art will understand that Figure 4 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0206] In one exemplary embodiment, a computer device is also provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps in the above-described method embodiments.

[0207] In one exemplary embodiment, a computer-readable storage medium is provided storing a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0208] In one exemplary embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0209] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data must comply with relevant regulations.

[0210] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM).

[0211] The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.

[0212] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0213] This document uses specific examples to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. Furthermore, those skilled in the art will recognize that, based on the ideas of this application, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A breast cancer early warning method based on wavelet analysis, characterized in that, The breast cancer early warning method based on wavelet analysis includes: The target population is divided into healthy individuals, individuals with benign breast diseases, and individuals with malignant breast diseases; The time series of bilateral breast surface temperatures of the target population were collected using temperature sensors. The temperature time series is cleaned and standardized to generate a standardized temperature series. The standardized temperature sequence is subjected to discrete wavelet transform to extract multi-level wavelet decomposition coefficients; Based on the multi-level wavelet decomposition coefficients, a set of feature indicators is calculated, which includes the mean amplitude, the mean over-threshold count, and the difference in over-threshold count. A machine learning model is trained using the aforementioned set of feature indicators and preset labels to generate a breast cancer risk prediction model. The breast cancer risk prediction model is embedded into the microprocessor unit of a wearable device to output breast cancer risk warning results.

2. The breast cancer early warning method based on wavelet analysis according to claim 1, characterized in that, The step of collecting the time series of bilateral breast surface temperature of the target population using a temperature sensor includes: Under preset environmental conditions, temperature data of the surface of both breasts are continuously collected at a sampling frequency specified by a preset standard to generate the temperature time series.

3. The breast cancer early warning method based on wavelet analysis according to claim 1, characterized in that, The steps of cleaning and standardizing the temperature time series include: Filter out abnormal data points in the temperature time series that are below a preset temperature threshold; Identify invalid time periods in the temperature time series caused by sensor detachment, and segment and remove these invalid time periods; The left and right side data of the temperature time series are length aligned using linear interpolation to generate the standardized temperature series.

4. The breast cancer early warning method based on wavelet analysis according to claim 1, characterized in that, The step of performing discrete wavelet transform on the standardized temperature sequence includes: Select a preset wavelet basis function, perform multi-level discrete wavelet decomposition on the standardized temperature sequence, and generate the multi-level wavelet decomposition coefficients. The Nyquist frequency is calculated based on the preset sampling frequency, and the frequency range of each level of wavelet decomposition is divided based on the Nyquist frequency. The absolute values ​​of the multi-level wavelet decomposition coefficients are extracted to generate a multi-level wavelet amplitude sequence.

5. The breast cancer early warning method based on wavelet analysis according to claim 1, characterized in that, The step of calculating the feature index set based on the multi-level wavelet decomposition coefficients includes: Calculate the mean amplitude of the multi-level wavelet decomposition coefficients, where the mean amplitude is the average of the absolute values ​​of the left and right wavelet coefficients at the same decomposition level. The number of coefficients exceeding a preset intensity threshold in the multi-level wavelet decomposition coefficients is counted, and the percentage of coefficients exceeding the threshold is calculated. The weighted summation and difference calculations are performed on the proportions of over-threshold counts at the same decomposition level on the left and right sides to generate the mean of over-threshold counts and the difference of over-threshold counts.

6. The breast cancer early warning method based on wavelet analysis according to claim 1, characterized in that, The step of training a machine learning model using the set of feature indicators and preset labels to generate a breast cancer risk prediction model includes: A preset classification algorithm is selected based on the sample size, and the preset classification algorithm includes a logistic regression model or an ensemble learning model. The set of feature indicators is input into the preset classification algorithm for model training, and regression coefficients and intercept terms are calculated. The breast cancer risk prediction model was validated based on preset evaluation indicators, including AUC value, KS value, and PSI population stability index.

7. The breast cancer early warning method based on wavelet analysis according to claim 1, characterized in that, The steps for outputting breast cancer risk warning results include: The real-time temperature sequence collected by the user wearing the wearable device is input into the breast cancer risk prediction model to generate a risk score; If the risk score exceeds a preset risk threshold, a high-risk warning signal for breast cancer is sent to the user. If the risk score is lower than the preset risk threshold, a periodic monitoring alert will be generated.

8. A breast cancer early warning device based on wavelet analysis, characterized in that, The breast cancer early warning device based on wavelet analysis includes: The acquisition module is used to divide the target population into healthy individuals, individuals with benign breast diseases, and individuals with malignant breast diseases. The time series of bilateral breast surface temperatures of the target population were collected using temperature sensors. The temperature time series is cleaned and standardized to generate a standardized temperature series. The standardized temperature sequence is subjected to discrete wavelet transform to extract multi-level wavelet decomposition coefficients; Based on the multi-level wavelet decomposition coefficients, a set of feature indicators is calculated, which includes the mean amplitude, the mean over-threshold count, and the difference in over-threshold count. The algorithm module is used to train a machine learning model using the set of feature indicators and preset labels to generate a breast cancer risk prediction model; The prediction module is used to embed the breast cancer risk prediction model into the microprocessor unit of the wearable device and output breast cancer risk warning results.

9. A computer device, comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that the processor executes the computer program to implement the steps of the wavelet analysis-based breast cancer early warning method according to any one of claims 1-7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the computer program implements the steps of the wavelet analysis-based breast cancer early warning method according to any one of claims 1-7.

Images