A system and method for lung injury assessment based on multi-frequency narrowband parameter estimation and machine learning

Abstract

The present application belongs to the technical field of ultrasonic imaging diagnosis, and particularly relates to a lung injury evaluation system and method based on multi-frequency narrow-band parameter estimation and machine learning. In view of the problem that the lung injury type is difficult to be accurately identified in the prior art, the present application obtains a lung ultrasonic radio frequency signal through multi-frequency narrow-band pulse, obtains radio frequency data corresponding to B-line based on the radio frequency signal, extracts information entropy, permutation entropy, bandwidth, skewness, total average energy and Nakagami distribution shape factor from the radio frequency data, inputs the average of the same characteristic parameters of the multi-frequency into a decision tree model, and obtains an evaluation value of the lung tissue characteristic state. The present application can efficiently and accurately identify various lung injury types such as pulmonary edema, pulmonary fibrosis and pneumonia, and has a good clinical application prospect.

Description

Technical Field

[0001] This invention belongs to the field of ultrasound imaging diagnostic technology, specifically relating to a lung injury assessment system and method based on multi-frequency narrowband parameter estimation and machine learning. Background Technology

[0002] Currently, research on the assessment of lung injury types can be broadly categorized into two methods: morphological methods and data-driven methods. While morphological methods offer clinical intuitiveness, their over-reliance on empirically derived features makes it difficult to fully represent complex pathological heterogeneity, especially in scenarios with low signal-to-noise ratios in ultrasound images and ambiguous lesion presentations. Data-driven adaptive methods, while improving the objectivity of feature extraction through mathematical modeling, face new challenges such as unclear feature-pathology correlation mechanisms and insufficient model generalization ability. Compared to the significant differences between B-line and non-B-line data, distinguishing different injury types within the B-line is even more difficult. Currently, a rapid and accurate method for identifying lung injury types remains lacking. Summary of the Invention

[0003] The purpose of this invention is to overcome the shortcomings of the prior art and provide a lung injury assessment system and method based on multi-frequency narrowband parameter estimation and machine learning, so as to solve the problem that there is no method in the prior art that can accurately identify the type of lung injury.

[0004] To achieve the above objectives, the present invention employs the following technical solution: A lung injury assessment system based on multi-frequency narrowband parameter estimation and machine learning, comprising: The data acquisition module is used to acquire lung ultrasound radiofrequency signals, which are obtained based on multi-frequency narrowband pulses; The B-line acquisition module is used to obtain radio frequency data corresponding to the B-line based on the lung ultrasound radio frequency signal; The feature parameter module is used to extract the feature parameters of the B-line from the radio frequency data. The feature parameters include information entropy, permutation entropy, bandwidth, skewness, total average energy, and Nakagami distribution shape factor. The same feature parameter of multiple frequencies is averaged to obtain the average feature parameter. The damage assessment module is used to input the average feature parameters into the decision tree model to obtain the evaluation value of each type of lung tissue feature state.

[0005] A further improvement of the present invention is that: Preferably, in the data acquisition module, the frequency range of the multi-frequency narrowband pulse is 4-10MHz, and the pulse period is 4 pulse periods.

[0006] Preferably, the method by which the B-line acquisition module obtains the radio frequency data corresponding to the B-line is any one of the following: spatial mapping method based on B-mode image, threshold detection method based on radio frequency signal energy, coherence identification method based on multi-frequency characteristics, or using dual-parameter SVM method.

[0007] Preferably, the dual-parameter SVM method includes the following steps: performing Hilbert transform on the obtained lung ultrasound radiofrequency signal to obtain an analytical signal, taking a model of the signal to obtain an envelope signal and an ultrasound image, performing confidence analysis on the ultrasound image to obtain a confidence map, thereby obtaining the position of the pleural line, removing the pleural line and the data above the pleural line, thereby obtaining radiofrequency data without the pleural line.

[0008] Preferably, the formula for calculating the information entropy is: (1) in, It is information entropy. It is the amplitude of a two-dimensional RF signal time series. Signal amplitude The probability density is given by , where n is a fixed sign in the summation, is the upper limit of the summation, p is the time index, and q is the dimension index. The formula for calculating the permutation entropy is: . (4) in, It is a time series signal x The permutation entropy, k It is the total number of reconstructed components. It is the first i a symbol sequence The probability of occurrence.

[0009] Preferably, the formula for calculating the skewness is: (7) in, It is a two-dimensional RF timing signal In the q The envelope of the column, yes skewness, yes The mean, yes variance These are the second-order center distance and the third-order center distance, respectively; The formula for calculating the total average energy is as follows: (8) Where M represents the length of the signal sequence, and y1 represents the amplitude of the i-th sampling point in the signal sequence.

[0010] Preferably, the step of obtaining the bandwidth is as follows: Calculate the power spectral density of the radio frequency signal and find its peak value; Calculate the power point corresponding to a 3dB drop in peak value; Determine the low-frequency and high-frequency intersection points where the power spectral density intersects with the power point; The difference between the high-frequency intersection point and the low-frequency intersection point is used as the bandwidth.

[0011] Preferably, the decision tree model in the damage assessment module is pre-trained in the following manner: Acquire lung ultrasound radiofrequency data under multi-frequency narrowband pulses and construct a dataset; Based on the dataset, obtain the radio frequency data corresponding to line B; Information entropy, permutation entropy, bandwidth, skewness, total average energy, and Nakagami distribution shape factor are extracted from the radio frequency data corresponding to line B as input features, and the corresponding lung tissue feature state labels are labeled. The input features and labels are divided into training and test sets.

[0012] Preferably, during pre-training, the maximum depth parameter is adjusted within the range of 3-20, and the minimum number of split samples parameter is adjusted within the range of 2-20.

[0013] A lung injury assessment method based on multi-frequency narrowband parameter estimation and machine learning includes the following steps: S1, acquire lung ultrasound radiofrequency signals, wherein the lung ultrasound radiofrequency signals are obtained based on multi-frequency narrowband pulses; S2, obtain the radio frequency data corresponding to line B based on the lung ultrasound radio frequency signal; S3, extract the characteristic parameters of the B line from the radio frequency data. The characteristic parameters include information entropy, arrangement entropy, bandwidth, skewness, total average energy, and Nakagami distribution shape factor. Average the same characteristic parameter of multiple frequencies to obtain the average characteristic parameter. S4, the average feature parameters are input into the decision tree model to obtain the evaluation value of each type of lung tissue feature state.

[0014] Compared with the prior art, the present invention has the following beneficial effects: This invention proposes a lung injury assessment system based on multi-frequency narrowband parametric estimation and machine learning. The system transmits multi-frequency narrowband signals to obtain ultrasound radiofrequency echo data at different frequencies. Then, it locates the pleural line using confidence plots, removes the data above the pleural line, and extracts the B-line using a dual-parameter SVM technique for identifying and classifying special lung ultrasound features based on ultrasound echo radiofrequency signals. Based on the extracted B-line, six ultrasound feature parameters are calculated, including information entropy, permutation entropy, bandwidth, skewness, total average energy, and Nakagami shape factor. Finally, the six parameters of data on different lung injury types (such as pulmonary edema, pulmonary fibrosis, pneumonia, etc.) are input into a decision tree model for training to obtain a classification of the lung injury type. This invention's detection method considers multiple features contained in ultrasound radiofrequency data, fully utilizes data features at different frequencies, and combines machine learning methods to obtain good classification results for different lung injury types, showing promising application prospects. This invention obtains radio frequency data corresponding to B-lines by automatically detecting B-lines, extracts the feature parameters, performs independent and identically distributed partitioning, and trains the dataset using a decision tree model. This enables the differentiation of injury types through subtle differences in B-line artifacts, achieving the purpose of assessing various lung injuries such as pulmonary edema, pulmonary fibrosis, and pneumonia.

[0015] Furthermore, this method obtains narrowband pulse ultrasound signals by continuously transmitting four pulse cycles at a time, thereby acquiring original radio frequency signals with high signal-to-noise ratio and stable frequency domain characteristics; it also enhances stability assessment by acquiring ultrasound signals at multiple frequencies for the same lung injury; simultaneously, it enhances the evaluation and recognition capabilities of the decision tree model by extracting six feature parameters: information entropy, permutation entropy, bandwidth, skewness, total average energy, and Nakagami distribution shape factor; ultimately, this invention can efficiently and accurately identify various types of lung injuries such as pulmonary edema, pulmonary fibrosis, and pneumonia, and has broad application prospects. Attached Figure Description

[0016] Figure 1 This is a flowchart of the lung injury assessment method based on multi-frequency narrowband parameter estimation and machine learning of the present invention. Detailed Implementation

[0017] Hereinafter, the terms "first," "second," "third," and "fourth" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature defined as "first," "second," "third," or "fourth" may explicitly or implicitly include one or more of that feature.

[0018] The method provided in this application can be applied to mobile phones, tablets, wearable devices, in-vehicle devices, augmented reality (AR) / virtual reality (VR) devices, laptops, and ultra-mobile personal computers. In this application, the specific type of terminal device is not limited to terminal devices such as mobile personal computers (UMPCs), netbooks, and personal digital assistants (PDAs).

[0019] It should be noted that the terms "first," "second," etc., used in the specification and drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0020] A lung injury assessment method based on multi-frequency narrowband parameter estimation and machine learning includes the following steps: The data acquisition module is used to acquire lung ultrasound radiofrequency signals, which are obtained based on multi-frequency narrowband pulses; The B-line acquisition module is used to obtain radio frequency data corresponding to the B-line based on the lung ultrasound radio frequency signal; The feature parameter module is used to extract the feature parameters of the B-line from the radio frequency data. The feature parameters include information entropy, permutation entropy, bandwidth, skewness, total average energy, and Nakagami distribution shape factor. The same feature parameter of multiple frequencies is averaged to obtain the average feature parameter. The damage assessment module is used to input the average feature parameters into the decision tree model to obtain the evaluation value of each type of lung tissue feature state.

[0021] In some embodiments of the present invention, the data acquisition module uses multi-frequency narrowband pulses to acquire lung ultrasound radio frequency signals for the same lung injury. The method of transmitting multi-frequency narrowband pulses is to continuously transmit ultrasound signals for four pulse cycles each time, and the frequency range of the ultrasound signals is 4-10MHz. For example, the transmission frequency of different pulse cycles can be 4MHz, 6MHz, 8MHz and 10MHz.

[0022] In some embodiments of the present invention, the method by which the B-line acquisition module obtains the radio frequency data corresponding to the B-line includes: spatial mapping method based on B-mode image, threshold detection method based on radio frequency signal energy, coherence identification method based on multi-frequency characteristics, or using dual-parameter SVM method.

[0023] Specifically, the spatial mapping method for the B-mode image is as follows: First, the acquired multi-frequency narrowband lung ultrasound radiofrequency signals are sequentially subjected to Hilbert transform, logarithmic compression, and digital scan transform to reconstruct a B-mode ultrasound image; then, the position of the pleural line is automatically identified in the B-mode image using an edge detection algorithm, and a region of interest with a width of a preset pixel threshold is constructed, starting from the pleural line and extending downwards in a direction perpendicular to the pleural line to the bottom of the image; finally, the image coordinates of the region of interest are inversely mapped back to the original radiofrequency data matrix, and the radiofrequency data at the corresponding position is extracted as the radiofrequency data of the B-line.

[0024] Specifically, the threshold detection method based on radio frequency signal energy involves directly performing envelope detection on the lung ultrasound radio frequency signal to obtain the amplitude envelope curve of the radio frequency signal; for each scan line, calculating its energy distribution in the depth direction and setting an energy threshold; when the energy value of a certain scan line exceeds the threshold in a continuous depth range, and the ratio of its energy peak value to the average energy value of the surrounding scan lines is greater than a preset ratio, then the entire radio frequency data of that scan line is marked as the radio frequency data corresponding to line B.

[0025] Specifically, the process of the coherence identification method based on multi-frequency features is to extract the radio frequency signal under each frequency component and calculate the autocorrelation function of the signal at the same spatial location under each frequency component. If a certain spatial location shows an autocorrelation coefficient higher than that of the background region under different frequency components and its coherence is continuously distributed in the vertical direction, then it is determined that there is a B-line artifact at that location, and the corresponding radio frequency data is extracted as the radio frequency data of the B-line.

[0026] The specific method mentioned is the method in the authorized patent "A method for identifying and classifying special signs of lung ultrasound based on ultrasound echo radio frequency signals" with announcement number CN114492519B.

[0027] The specific process is as follows: The lung ultrasound radiofrequency signal obtains the radiofrequency data corresponding to the B-line; the obtained lung ultrasound radiofrequency signal undergoes a Hilbert transform to obtain an analytical signal; the signal is then modulated to obtain an envelope signal, thereby obtaining an ultrasound image. Confidence analysis is performed on the obtained ultrasound image to obtain a confidence map, which in turn determines the position of the pleural line. Data above and below the pleural line is removed, resulting in radiofrequency data excluding the pleural line, further improving the accuracy of obtaining the radiofrequency data corresponding to the B-line.

[0028] Specifically, the calculation method for each feature parameter in the feature parameter module is as follows: Information entropy (InEn) is a parameter that reflects the uncertainty assessment of a signal: a highly chaotic signal tends to have a larger information entropy, calculated as follows: (1) in, It is information entropy. It is the amplitude of a two-dimensional RF signal time series. Signal amplitude The probability density is given by , where n is a fixed sign in the summation, is the upper limit of the summation, p is the time index, and q is the dimension index.

[0029] Permutation entropy (PerEn) is a method used to detect abrupt changes and randomness in time series data. It captures subtle details in time series and provides a quantitative measure of random noise in the sequence. The steps to calculate permutation entropy are as follows: (2) in, It is the embedded dimension. t It is the delay time, matrix Y Each line is a reconstructed component, totaling [number missing]. k One reconstructed component, k = N ( 1)·t N is the original time series length, and q is the index dimension.

[0030] Each reconstructed component is arranged in ascending order to obtain the column index of each element position in the vector, thus forming a set of symbol sequences: !. (3) in, Indicates the first i List, It is a sequence of symbols.

[0031] M The total number of phase space mappings is There are 10 different symbol sequences. The number of occurrences of each symbol sequence is divided by the total number of occurrences. The probability of the occurrence of the symbol sequence is obtained by !. After obtaining the probability, the permutation entropy of element x is calculated using the following formula: . (4) in, It is a time series signal x The permutation entropy,k It is the probability of the total number of reconstructed components occurring. It is the first i a symbol sequence The probability of occurrence.

[0032] Bandwidth is a measure of the spectral distribution range of a radio frequency (RF) signal, reflecting the degree of energy concentration and frequency spread characteristics of the signal in the frequency domain. A narrow bandwidth indicates that the signal's energy is concentrated primarily on a few frequency components, suggesting a compact spectral distribution. Conversely, a wide bandwidth indicates that the signal's energy is dispersed over a wider frequency range, indicating a more dispersed spectral distribution. Bandwidth is obtained by calculating the difference between the frequencies corresponding to a 3dB drop from the maximum energy value in the signal's power spectrum. The calculation process is as follows: First, calculate the power spectral density S( of the radio frequency signal). Find its peak value. , Calculate the -3dB power point: (5) Then find S( )= The two intersections and Finally, the bandwidth is calculated: (6) in, and These are the lower half-power frequency and the upper half-power frequency, respectively.

[0033] Skewness is a measure of the direction and degree of skewness in the distribution of statistical data, reflecting the asymmetry of the amplitude of the radio frequency (RF) signal envelope. If the skewness value is less than 0, there are relatively fewer data points with smaller values ​​in the RF signal envelope, indicating that the distribution is skewed towards higher values. Conversely, if the skewness value is greater than 0, there are relatively fewer data points with larger values ​​in the RF signal envelope, indicating that the distribution is skewed towards lower values.

[0034] The skewness of the radio frequency data envelope is calculated using the following formula: (7) in, It is a two-dimensional RF timing signal In the r The envelope of the column, yes skewness, yes The mean, yes variance It refers to the second-order and third-order center distances.

[0035] Total Average Energy (TAE) is a measure of the overall strength level of a radio frequency (RF) signal, reflecting the degree of energy concentration in the time or frequency domain. A lower TAE indicates a generally smaller amplitude in the RF signal, suggesting a weaker signal. Conversely, a higher TAE indicates a generally larger amplitude in the RF signal, suggesting a stronger signal. It is calculated by dividing the sum of the squares of the amplitudes at all points on the envelope of the RF scan line signal by the length of the signal sequence. (8) Where M represents the length of the signal sequence, and y1 represents the amplitude of the i-th sampling point in the signal sequence.

[0036] The Nakagami distribution shape factor is a general form of the Rayleigh distribution, and its probability density is: (9) in, It is a random parameter. It is the gamma function. m These are shape parameters. These are scale factors, which can be calculated using the following formula: (10) (11) In this formula, X is a random variable in probability statistics, representing the amplitude of the echo signal, and (u, v) represents the spatial coordinates. This represents the local estimation window, indicating that the formula is used in... Calculations are performed within the specified range.

[0037] In the formula, the shape factor *m* controls the tail decay of the probability density function. When *m* = 1, the Nakagami distribution is the same as the Rayleigh distribution. When *m* is greater than 1, the tail decay of the probability density function is slower, and vice versa. *m* reflects the shape of the probability distribution curve of the echo signal amplitude.

[0038] Furthermore, the feature parameter module extracts six feature parameters from radio frequency signals at multiple frequencies, and averages the six feature parameters corresponding to multiple frequencies to obtain the six average feature parameters of line B corresponding to a single lung injury.

[0039] In the evaluation process, this invention considers six parameters. Instead of relying on vague morphological descriptions, this invention designs a multidimensional feature space with clear physical meaning, and each parameter corresponds to a clear pathological-acoustic mechanism.

[0040] Information entropy and permutation entropy capture the degree of disorder in the distribution of scatterers from the perspectives of global statistics and local temporal sequence, respectively. For pulmonary edema tissue, the alveolar cavities are filled with a homogeneous fluid medium, forming a uniform acoustic interface. This uniform structure reduces the randomness of the backscattered signal, resulting in a signal sequence with strong regularity and predictability, thus placing the information entropy and permutation entropy values ​​in the lower range. For pulmonary fibrosis tissue, a complex network of cross-linked fibers exists within the tissue, increasing the non-uniformity and spatial variability of the acoustic scattering interface. This complex fiber network significantly enhances the disorder of the backscattered signal, placing the entropy value in the higher range. For pneumonia tissue, strong random backscattering sources exist (such as scattered inflammatory cells or exudate), but the spatial distribution of scatterers is relatively uniform. This "uniformly random distribution" acoustic characteristic places the signal disorder between regular and completely random signals, thus placing the entropy value in the middle range.

[0041] Bandwidth is used to detect the scale diversity of scatterers. For pulmonary edema and pneumonia, the signal is dominated by uniform scattering events, with frequency components concentrated in the mid-to-low frequency range and no obvious high-frequency absorption or reflection, resulting in a narrow bandwidth. In contrast, pulmonary fibrosis: heterogeneous structures induce multi-scale scattering, with strong reflection and weak scattering coexisting. The fibrotic interface generates high-frequency oscillating components, significantly broadening the bandwidth. This allows it to be effectively distinguished from pulmonary edema and pneumonia, which are dominated by uniform scattering and have concentrated energy.

[0042] Skewness measures the asymmetry of echo amplitude distribution, enabling the sensitive capture of strong reflection signals generated by a small number of collagen fibers in fibrotic lesions. In a homogeneous scattering medium, the histogram of the RF backscattered echo envelope amplitude approximates a Rayleigh distribution. Skewness increases with the presence of medium heterogeneity. For pulmonary edema, fluid accumulation enhances tissue homogenization, resulting in an echo amplitude distribution approaching a Rayleigh distribution with low skewness. For pulmonary fibrosis, collagen fiber proliferation leads to increased tissue heterogeneity, resulting in a significantly positively skewed echo amplitude (skewness > 0). For pneumonia, the tissue structure is relatively homogeneous but exhibits minor variations, with skewness falling between that of pulmonary edema and pulmonary fibrosis.

[0043] For pulmonary edema, high fluid content leads to high scatterer density, but the scattering efficiency of the liquid medium is low, the intensity of individual scattering is weak, and the overall energy value is generally low. For pneumonia, the scatterer density and scattering intensity are moderate, and the energy value is at a medium level. For pulmonary fibrosis, the heterogeneous structure induces strong and frequent reverberant reflections, and the local energy is significantly increased. Therefore, the total average energy is selected as one of the judgment factors.

[0044] The lower the scatterer concentration or the more non-uniform the spatial distribution (such as mixed reflections and other signal types), the smaller the m value. For pulmonary edema and pneumonia, the scatterer concentration is high and the spatial distribution is uniform, so the m value is large. For pulmonary fibrosis, a large amount of collagen fibers are deposited, the normal structure is destroyed, and highly non-uniform and discontinuous fibrotic plaques or nodules are formed. Random backscattering deviates from the Rayleigh distribution, and the m value is significantly reduced. Therefore, the Nakagami distribution shape factor is selected.

[0045] The multidimensional cross-validation network formed by the coordinated operation of these six parameters greatly enhances the robustness and discriminative power of the identification model. For example, the combination of high energy and low entropy strongly points to "chaotic but strongly scattering" pulmonary fibrosis, rather than "orderly but weakly scattering" pulmonary edema. The simultaneous occurrence of significantly increased skewness and significantly decreased m-factor, a characteristic combination of "high skewness and low-density scattering," provides strong statistical evidence for the structure of fibrotic plaques. Furthermore, the synergy of high bandwidth and high permutation entropy simultaneously confirms the high complexity of the tissue in both the frequency and time domains, a combination particularly typical in pulmonary fibrosis, further eliminating potential misjudgments from a single dimension. Therefore, when all six parameters are considered simultaneously, the design, which involves cross-validation from multiple physical dimensions, enables the model to capture and amplify minute acoustic differences within B-line artifacts that are difficult to detect with the naked eye, ultimately achieving effective assessment and classification of pulmonary edema, pulmonary fibrosis, and pneumonia.

[0046] The damage assessment module inputs the average feature parameters into the decision tree model to obtain the evaluation value of each type of lung tissue feature state, where the lung tissue feature state represents the type of lung injury, and each evaluation value represents the probability value of the type of lung injury.

[0047] In a specific example, a training method for a lung injury assessment system based on multi-frequency narrowband parameter estimation and machine learning includes the following steps: S1. For the same lung injury, multi-frequency narrow-band pulses are used to acquire lung ultrasound radiofrequency signals. The method of transmitting multi-frequency narrow-band pulses is to continuously transmit ultrasound signals for four pulse cycles each time, and the frequency of the ultrasound signals is 4-10MHz. The above method is repeated to acquire lung tissue features, including radiofrequency data of various lung injuries such as pulmonary edema, pulmonary fibrosis and pneumonia, to construct a dataset.

[0048] S2 performs a Hilbert transform on the radio frequency signal obtained from S1 to obtain an analytical signal, and then takes the modulus of the analytical signal to obtain an envelope signal, thereby obtaining an ultrasound image.

[0049] S3 adopts step S2 from the authorized patent with publication number CN114492519B, performs confidence analysis on the ultrasound image obtained in S2, obtains a confidence map, and then obtains the position of the pleural line. The pleural line and the data above the pleural line are removed, and radiofrequency data without the pleural line are obtained.

[0050] The specific process of removing the part above the pleural line in S3 is as follows: confidence analysis of the ultrasound image is performed through a random walk frame to obtain a confidence map, thereby obtaining the position of the pleural line, and then the pleural line and the area above it are removed. S4. Feature parameters of the B-line are extracted from radio frequency data. Six feature parameters are extracted from multiple frequencies: information entropy, permutation entropy, bandwidth, skewness, total average energy, and Nakagami distribution shape factor. The average of these six feature parameters across multiple frequencies is then used to obtain the six feature parameters corresponding to a single lung injury in the B-line. Furthermore, the feature parameters of all scan lines within the entire B-line region are averaged to obtain the feature parameters corresponding to a single B-line. The six extracted feature parameters are information entropy, permutation entropy, bandwidth, skewness, total average energy, and Nakagami distribution shape factor.

[0051] In step S5, the feature parameters obtained in step S4 are independently and identically distributed into a training set and a test set of 5:2 size. The feature parameter data corresponding to line B are independent of each other. These are then input into the decision tree for parameter tuning and optimization. The parameters with the best performance are selected for model training to obtain an optimized decision tree model. The parameter tuning method involves adjusting the maximum depth within the range of 3-20 and the minimum number of split samples within the range of 2-20. Finally, the trained decision tree model is tested and validated using the test set.

[0052] Furthermore, during the testing process, if the accuracy meets the requirements for clinical use, the model weights and structure are solidified and integrated into the lung injury assessment system to achieve automatic injury assessment and judgment of unknown lung ultrasound data.

[0053] This invention discloses a lung injury assessment method based on multi-frequency narrowband parameter estimation and machine learning, the method comprising the following steps: S1, acquire lung ultrasound radiofrequency signals, wherein the lung ultrasound radiofrequency signals are obtained based on multi-frequency narrowband pulses; S2, obtain the radio frequency data corresponding to line B based on the lung ultrasound radio frequency signal; S3, extract the characteristic parameters of the B line from the radio frequency data. The characteristic parameters include information entropy, arrangement entropy, bandwidth, skewness, total average energy, and Nakagami distribution shape factor. Average the same characteristic parameter of multiple frequencies to obtain the average characteristic parameter. S4, the average feature parameters are input into the decision tree model to obtain the evaluation value of each type of lung tissue feature state.

[0054] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0055] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0056] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0057] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A lung injury assessment system based on multi-frequency narrowband parameter estimation and machine learning, characterized in that, include: The data acquisition module is used to acquire lung ultrasound radiofrequency signals, which are obtained based on multi-frequency narrowband pulses; The B-line acquisition module is used to obtain radio frequency data corresponding to the B-line based on the lung ultrasound radio frequency signal; The feature parameter module is used to extract the feature parameters of the B-line from the radio frequency data. The feature parameters include information entropy, permutation entropy, bandwidth, skewness, total average energy, and Nakagami distribution shape factor. The same feature parameter of multiple frequencies is averaged to obtain the average feature parameter. The damage assessment module is used to input the average feature parameters into the decision tree model to obtain the evaluation value of each type of lung tissue feature state.

2. The lung injury assessment system based on multi-frequency narrowband parameter estimation and machine learning according to claim 1, characterized in that, In the data acquisition module, the frequency range of the multi-frequency narrowband pulse is 4-10MHz, and the pulse period is 4 pulse periods.

3. The lung injury assessment system based on multi-frequency narrowband parameter estimation and machine learning according to claim 1, characterized in that, The method by which the B-line acquisition module obtains the radio frequency data corresponding to the B-line is any one of the following: spatial mapping method based on B-mode image, threshold detection method based on radio frequency signal energy, coherence identification method based on multi-frequency characteristics, or using dual-parameter SVM method.

4. The lung injury assessment system based on multi-frequency narrowband parameter estimation and machine learning according to claim 1, characterized in that, The dual-parameter SVM method includes the following steps: performing Hilbert transform on the obtained lung ultrasound radiofrequency signal to obtain an analytical signal, taking a modulus on the signal to obtain an envelope signal and an ultrasound image, performing confidence analysis on the ultrasound image to obtain a confidence map, thereby obtaining the position of the pleural line, removing the pleural line and the data above the pleural line, thereby obtaining radiofrequency data without the pleural line.

5. A lung injury assessment system based on multi-frequency narrowband parameter estimation and machine learning according to claim 1, characterized in that, The formula for calculating the information entropy is: (1) in, It is information entropy. It is the amplitude of a two-dimensional RF signal time series. Signal amplitude The probability density is given by n, where n is a fixed sign in the summation, q is the upper limit of the summation, p is the time index, and q is the dimension index. The formula for calculating the permutation entropy is: . (4) in, It is a time series signal x The permutation entropy, k It is the total number of reconstructed components. It is the first i a symbol sequence The probability of occurrence.

6. The lung injury assessment system based on multi-frequency narrowband parameter estimation and machine learning according to claim 1, characterized in that, The formula for calculating the skewness is: (7) in, It is a two-dimensional RF timing signal In the q The envelope of the column, yes skewness, yes The mean, yes variance These are the second-order center distance and the third-order center distance, respectively; The formula for calculating the total average energy is as follows: (8) Where M represents the length of the signal sequence, and y1 represents the amplitude of the i-th sampling point in the signal sequence.

7. The lung injury assessment system based on multi-frequency narrowband parameter estimation and machine learning according to claim 1, characterized in that, The steps for obtaining the bandwidth are as follows: Calculate the power spectral density of the radio frequency signal and find its peak value; Calculate the power point corresponding to a 3dB drop in peak value; Determine the low-frequency and high-frequency intersection points where the power spectral density intersects with the power point; The difference between the high-frequency intersection point and the low-frequency intersection point is used as the bandwidth.

8. A lung injury assessment system based on multi-frequency narrowband parameter estimation and machine learning according to claim 1, characterized in that, The decision tree model in the damage assessment module is pre-trained using the following method: Acquire lung ultrasound radiofrequency data under multi-frequency narrowband pulses and construct a dataset; Based on the dataset, obtain the radio frequency data corresponding to line B; Information entropy, permutation entropy, bandwidth, skewness, total average energy, and Nakagami distribution shape factor are extracted from the radio frequency data corresponding to line B as input features, and the corresponding lung tissue feature state labels are labeled. The input features and labels are divided into training and test sets.

9. A lung injury assessment system based on multi-frequency narrowband parameter estimation and machine learning according to claim 8, characterized in that, During pre-training, the maximum depth parameter is adjusted within the range of 3-20, and the minimum number of split samples parameter is adjusted within the range of 2-20.

10. A lung injury assessment method based on multi-frequency narrowband parameter estimation and machine learning, characterized in that, Includes the following steps: S1, acquire lung ultrasound radiofrequency signals, wherein the lung ultrasound radiofrequency signals are obtained based on multi-frequency narrowband pulses; S2, obtain the radio frequency data corresponding to line B based on the lung ultrasound radio frequency signal; S3, extract the characteristic parameters of the B line from the radio frequency data. The characteristic parameters include information entropy, arrangement entropy, bandwidth, skewness, total average energy, and Nakagami distribution shape factor. Average the same characteristic parameter of multiple frequencies to obtain the average characteristic parameter. S4, the average feature parameters are input into the decision tree model to obtain the evaluation value of each type of lung tissue feature state.

Images