An image processing method, device, medium and program product based on ultrasonic radio frequency signals

Abstract

This invention relates to an image processing method, device, medium, and program product based on ultrasound radio frequency signals, belonging to the field of medical ultrasound image data processing. By automatically identifying the pleural interface line and combining it with a sliding window using dual indicators, it adaptively selects the intersection region with significant pathological features as the target observation window. Based on this localization, a multidimensional quantitative parameter vector is extracted and mapped through a regression model to obtain a quantitative evaluation index. This provides a stable, smooth, and comparable continuous quantitative benchmark for longitudinal data tracking, improving the objectivity and accuracy of long-term disease monitoring.

Description

Technical Field

[0001] This application relates to the field of medical ultrasound image data processing, and in particular to an image processing method, device, medium, and program product based on ultrasound radio frequency signals. Background Technology

[0002] In the diagnosis and treatment of lung diseases, timely and accurate assessment of the extent of lung lesions is extremely important. Ultrasound diagnosis has advantages such as being non-invasive, free of ionizing radiation, real-time, and cost-effective, and can serve as a quick and convenient clinical screening and bedside testing tool.

[0003] In related technologies, some existing semi-quantitative assessment methods, such as scoring systems, score different ultrasound image manifestations to semi-quantitatively assess the severity of lung diseases. This technology, after acquiring an ultrasound image, divides the image into a global region of interest according to preset rules, and then extracts macroscopic image indicators such as grayscale distribution, texture features, or the number of specific echo artifacts within this region. Based on these extracted indicators, the system uses a preset algorithm model to calculate the discrete quantitative score corresponding to the current image and outputs it as the quantitative result of this image data processing, aiming to provide objective reference data for subsequent medical analysis.

[0004] However, when conducting long-term longitudinal data tracking of the same target tissue region, a continuous evolution trajectory of the target region's acoustic properties is required. Because related technologies employ global region segmentation and rely on macroscopic grayscale features for scoring, the extracted macroscopic features are prone to random fluctuations when acquiring image sequences at different time points, influenced by differences in probe scanning sections and tissue physiological movements (such as respiratory rhythms). This results in abrupt data jumps in the time series, making it difficult to establish a smooth and consistent comparative benchmark. Consequently, the evaluation results output by this method lack longitudinal comparability, reducing its reference value in long-term data monitoring. Summary of the Invention

[0005] This application provides an image processing method, device, medium, and program product based on ultrasonic radio frequency signals, which is used to improve the data coherence and longitudinal comparability of ultrasonic image sequences in long-term longitudinal monitoring.

[0006] In a first aspect, this application provides an image processing method based on ultrasound radio frequency signals. The method includes: based on the original ultrasound radio frequency signal of a target scanning area, identifying amplitude jump peaks along the depth dimension in a two-dimensional spatial coordinate system containing depth and lateral dimensions, and defining the continuous boundary formed by connecting these amplitude jump peaks in the lateral dimension as the pleural interface line; traversing the two-dimensional spatial coordinate system based on a sliding window of a preset size, determining the pleural roughness index and the subpleural signal energy index within the sliding window, wherein the pleural roughness index is the depth coordinate variance of the pleural interface line within the sliding window, and the subpleural signal energy index is the region located below the pleural interface line within the sliding window. The average envelope energy of the radio frequency signal is obtained; the intersection region where the subpleural signal energy index is greater than a first preset threshold and the pleural roughness index is greater than a second preset threshold is extracted as the target observation window; after taking the original ultrasound radio frequency signal within the target observation window as the target radio frequency signal, feature extraction is performed on the target radio frequency signal to obtain a multidimensional quantitative parameter vector, which includes envelope statistical parameters, spectral parameters and spatial scale parameters; the multidimensional quantitative parameter vector is input into a pre-trained regression model for mapping to obtain a quantitative evaluation index; the historical evaluation index of the target observation window at historical time nodes is obtained, the difference between the quantitative evaluation index and the historical evaluation index is calculated, and the corresponding visualization trend mark is generated and displayed.

[0007] By employing the aforementioned technical solution, and through automatic identification of the pleural interface line combined with a sliding window using dual indicators, the system can adaptively select intersection regions with significant pathological features as target observation windows. Based on this localization, a multidimensional quantitative parameter vector is extracted and mapped using a regression model to obtain a quantitative assessment index. This provides a stable, smooth, and comparable continuous quantitative benchmark for longitudinal data tracking, improving the objectivity and accuracy of long-term disease monitoring.

[0008] In conjunction with some embodiments of the first aspect, in some embodiments, the step of traversing the sliding window based on a preset size in the two-dimensional spatial coordinate system to determine the pleural roughness index and the subpleural signal energy index within the sliding window specifically includes: determining the set of analytical envelope peaks along the pleural interface line within the sliding window; when the local average value of the analytical envelope peak set is higher than a preset noise floor threshold, calculating the depth coordinate variance of the pleural interface line based on the depth coordinates corresponding to each peak in the analytical envelope peak set to obtain the pleural roughness index; extending a preset distance from the pleural interface line within the sliding window to the depth dimension to construct a subpleural region of interest; performing depth attenuation compensation on the original ultrasound radio frequency signal, and using the mean value of the radio frequency signal envelope energy after depth attenuation compensation within the subpleural region of interest as the subpleural signal energy index.

[0009] By adopting the above technical solution, a preset noise threshold is introduced before calculating roughness, ensuring that the analytical envelope peak is processed only under conditions of sufficient signal-to-noise ratio. The variance is calculated based on the depth coordinates of the effective peak, enabling a more realistic quantification of the pleural irregularity. This enhances the robustness and fidelity of the dual-index screening, ensuring the accuracy and reliability of target observation window positioning under different detection depths and noise environments.

[0010] In conjunction with some embodiments of the first aspect, in some embodiments, the step of extracting the intersection region where the subpleural signal energy index is greater than a first preset threshold and the pleural roughness index is greater than a second preset threshold as the target observation window specifically includes: obtaining the set of subpleural signal energy indices and the set of pleural roughness indices corresponding to all traversed sliding windows within the current target scanning area; statistically sorting the set of subpleural signal energy indices and determining a first preset percentile in the set of subpleural signal energy indices as the first preset threshold; statistically sorting the set of pleural roughness indices and determining a second preset percentile in the set of pleural roughness indices as the second preset threshold; and in the two-dimensional spatial coordinate system, selecting sliding windows that simultaneously satisfy the condition that the subpleural signal energy index is greater than the first preset threshold and the pleural roughness index is greater than the second preset threshold, and determining the spatial intersection region corresponding to the selected sliding windows in the two-dimensional spatial coordinate system as the target observation window.

[0011] By employing the above technical solution, and extracting the first and second preset percentiles as dynamic thresholds, the threshold setting is made dependent on the overall distribution pattern of the current data. The selected spatial intersection region is determined as the target observation window, giving the model a strong generalization ability. When faced with different patient physiological structures and different ultrasound equipment baselines, it can adaptively lock the most representative observation area, ensuring the spatial consistency and effectiveness of quantitative assessment results in complex clinical scenarios.

[0012] In conjunction with some embodiments of the first aspect, in some embodiments, before the step of identifying the amplitude jump peak of the original ultrasound radio frequency signal based on the target scanning area in a two-dimensional spatial coordinate system including depth and lateral dimensions, and determining the continuous boundary formed by connecting the amplitude jump peak in the lateral dimension as the pleural interface line, the method further includes: obtaining anatomical prior constraint information corresponding to the target scanning area, the anatomical prior constraint information including a preset pleural depth range; calculating the normalized cross-correlation coefficient between adjacent sound beams of the original ultrasound radio frequency signal in the lateral dimension; if the normalized cross-correlation coefficient is greater than or equal to a preset correlation threshold, determining the signal segment in the original ultrasound radio frequency signal located within the preset pleural depth range as a candidate signal segment; performing envelope extraction on the candidate signal segment to obtain the analytical envelope signal corresponding to the original ultrasound radio frequency signal, and using the analytical envelope signal as the original ultrasound radio frequency signal to perform the step of identifying the amplitude jump peak of the original ultrasound radio frequency signal along the depth dimension.

[0013] By employing the aforementioned technical solution, anatomical prior constraints are introduced, limiting the search space for signal processing to a preset pleural depth range, effectively shielding against strong reflection interference from superficial skin or deep irrelevant tissues. Simultaneously, the spatial continuity of the signal is quantitatively evaluated by calculating the normalized cross-correlation coefficient between adjacent sound beams. Only when the depth conforms to anatomical norms and the spatial correlation meets a threshold is it confirmed as a candidate signal segment. This interaction between spatial constraints and signal continuity verification can eliminate random noise and isolated artifacts, reducing the false detection rate of subsequent amplitude jump peak identification and improving the overall computational efficiency of the algorithm.

[0014] In conjunction with some embodiments of the first aspect, in some embodiments, after the step of calculating the normalized cross-correlation coefficient between adjacent sound beams of the original ultrasound radio frequency signal in the lateral dimension, the method further includes: if the normalized cross-correlation coefficient is less than the preset correlation threshold, extracting the temporal signal sequence of the original ultrasound radio frequency signal in multiple consecutive image frames; performing multi-frame averaging on the temporal signal sequence to obtain a smoothed radio frequency signal with enhanced signal-to-noise ratio; performing an envelope extraction and filtering algorithm based on the smoothed radio frequency signal to obtain a moving average envelope, and determining the signal segment within the preset pleural depth range in the moving average envelope as the candidate signal segment.

[0015] By employing the above technical solution, and extracting temporal signal sequences from multiple consecutive image frames and performing multi-frame averaging, the information redundancy in the temporal dimension effectively suppresses randomly distributed background noise while coherently enhancing real tissue signals with temporal consistency. The interaction between this cross-frame temporal processing and the subsequent envelope filtering algorithm ensures robust extraction of candidate signal segments even in non-ideal scanning environments such as poor probe contact or limited patient physiological conditions, thus broadening the clinical applicability of this technology.

[0016] In conjunction with some embodiments of the first aspect, in some embodiments, after obtaining the historical evaluation index of the target observation window at a historical time node, calculating the difference between the quantitative evaluation index and the historical evaluation index, generating a corresponding visual trend marker and outputting it for display, the method further includes: obtaining a continuous multi-frame ultrasound image sequence containing the target observation window; performing feature point matching on the continuous multi-frame ultrasound image sequence based on an optical flow tracing algorithm to extract the motion trajectory of the target observation window in the two-dimensional spatial coordinate system; fitting a respiratory cycle according to the motion trajectory to determine the displacement range of the target observation window within the respiratory cycle; generating a static guide marker based on the center coordinates of the displacement range, and superimposing the static guide marker onto the currently output ultrasound image.

[0017] By employing the aforementioned technical solution, this application introduces an optical flow tracing algorithm to address the problem of target tissue deviating from the initial observation window due to physiological movements such as patient respiration. This algorithm can perform feature point matching on multiple consecutive frames of images, capturing the true motion trajectory of the target observation window during dynamic processes. By further fitting the respiratory cycle and determining the displacement range, the center coordinates of this dynamic displacement can be calculated, and a static guide marker can be generated and superimposed on the display screen. This interaction between dynamic tracking and static anchor point generation provides the operator with an intuitive and stable visual reference benchmark. It actively compensates for spatial displacement interference caused by respiratory rhythm, guiding the operator to maintain a highly consistent probe scanning section during long-term longitudinal monitoring, ensuring spatial consistency of data across time points.

[0018] In conjunction with some embodiments of the first aspect, in some embodiments, the step of fitting the respiratory cycle based on the motion trajectory and determining the displacement range of the target observation window within the respiratory cycle specifically includes: performing a frequency domain transformation on the coordinate sequence of the motion trajectory to extract the dominant frequency component corresponding to the respiratory frequency; reconstructing the respiratory motion phase curve based on the dominant frequency component, identifying the end-expiratory phase point and the end-inspiratory phase point in the respiratory motion phase curve; extracting the spatial coordinate extreme values ​​corresponding to the end-expiratory phase point and the end-inspiratory phase point of the target observation window, and determining the spatial envelope region formed by the spatial coordinate extreme values ​​as the displacement range.

[0019] By employing the aforementioned technical solution, irrelevant displacement interference is thoroughly filtered out by extracting the dominant frequency component that matches the respiratory rate in the frequency domain. Reconstructing the phase curve based on this pure dominant frequency component allows for the identification of the extreme phase points at the end of expiration and end of inspiration, two key physiological processes. This interaction between frequency domain isolation and extreme phase point extraction constructs a spatial envelope region that closely resembles the characteristics of real respiratory mechanics, providing robust support for the generation of static guide markers.

[0020] In a second aspect, this application provides an image processing apparatus comprising: one or more processors and a memory; the memory is coupled to the one or more processors and is used to store computer program code, the computer program code including computer instructions, wherein the one or more processors invoke the computer instructions to cause the image processing apparatus to perform the method as described in the first aspect and any possible implementation thereof.

[0021] Thirdly, this application provides a computer program product, including a computer program, characterized in that, when executed by a processor, the computer program causes the image processing device to perform the method described in the first aspect and any possible implementation thereof.

[0022] Fourthly, this application provides a computer-readable storage medium including instructions that, when executed on an image processing device, cause the image processing device to perform the method described in the first aspect and any possible implementation thereof.

[0023] One or more technical solutions provided in the embodiments of this application have at least the following technical effects or advantages:

[0024] 1. By adopting a technique that identifies the pleural interface line based on the original ultrasound radio frequency signal, uses the pleural roughness index and the subpleural signal energy index as a dual-index adaptive screening mechanism to locate the target observation window, and then extracts a multi-dimensional quantitative parameter vector and maps it to a quantitative evaluation index through a regression model, a continuous, stable and highly comparable automated quantitative evaluation of the degree of pleural lesions is achieved, which improves the accuracy and consistency of long-term disease monitoring.

[0025] 2. By employing the technical means of statistically sorting the dual indicators of all sliding windows in the current scanning area, determining the adaptive dynamic threshold with preset percentiles, and selecting the spatial intersection area that simultaneously meets the conditions of the dual indicators as the target observation window, the system can adaptively lock the pathologically representative observation area under different ultrasound equipment baselines, ensuring the spatial consistency and cross-scenario generalization ability of the quantitative assessment results.

[0026] 3. By employing a technique of extracting time-series signal sequences from multiple consecutive images when the normalized cross-correlation coefficient is below a threshold, performing multi-frame averaging to achieve coherent signal-to-noise ratio enhancement, and reconstructing smooth radio frequency signals through envelope extraction and filtering algorithms, this technique effectively solves the technical problem in existing technologies where the quality of a single frame signal is too low under non-ideal conditions such as poor probe contact or limited patient physiological conditions, resulting in the inability to effectively extract candidate signal segments and severely limiting the applicability of the system. This enables robust and reliable extraction of candidate signal segments even in harsh scanning environments, broadening the clinical applicability and fault tolerance of the technique. Attached Figure Description

[0027] Figure 1 This is a schematic flowchart of an image processing method based on ultrasonic radio frequency signals in an embodiment of this application;

[0028] Figure 2 This is another flowchart illustrating an image processing method based on ultrasonic radio frequency signals in an embodiment of this application;

[0029] Figure 3 This is a schematic diagram of an exemplary hardware structure of an image processing device in an embodiment of this application. Detailed Implementation

[0030] The terminology used in the following embodiments of this application is for the purpose of describing particular embodiments only and is not intended to be limiting of this application. As used in the specification and appended claims of this application, the singular expressions “a,” “an,” “the,” “the,” “the,” and “this” are intended to include the plural expressions as well, unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used in this application refers to and includes any or all possible combinations of one or more of the listed items.

[0031] Hereinafter, the terms "first" and "second" are used for descriptive purposes only and should not be construed as implying or suggesting relative importance or implicitly indicating the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature, and in the description of the embodiments of this application, unless otherwise stated, "multiple" means two or more.

[0032] In the clinical diagnosis and treatment of interstitial lung disease, while traditional high-resolution computed tomography (CT) scans can provide detailed images of lung structures, their inherent ionizing radiation risks and the fixed nature of the equipment make them unsuitable for long-term dynamic monitoring of chronic patients. Lung ultrasound technology possesses the inherent advantages of being non-invasive, portable, and bedside operative. Based on this, this application provides an image processing method based on ultrasound radiofrequency signals. This method extracts multidimensional quantitative parameters and outputs a quantitative assessment index through a pre-trained regression model, combining historical data to generate visual trend markers.

[0033] Please see Figure 1 This is a flowchart illustrating an image processing method based on ultrasonic radio frequency signals in an embodiment of this application.

[0034] S101. Based on the original ultrasound radio frequency signal of the target scanning area, in a two-dimensional spatial coordinate system containing depth and lateral dimensions, identify the amplitude jump peak of the original ultrasound radio frequency signal along the depth dimension, and determine the continuous boundary formed by connecting the amplitude jump peak in the lateral dimension as the pleural interface line.

[0035] The target scanning area refers to the lung tissue space covered by the probe's sound beam after the operator places the ultrasound probe at a predetermined intercostal position on the patient's chest wall. This area typically encompasses clinically standardized examination sites such as the anterior chest wall, lateral chest wall, or back. The raw ultrasound radiofrequency signal refers to the complete echo electrical signal received by the ultrasound probe without detection, logarithmic compression, or other conventional B-mode image processing procedures. This signal retains all amplitude and phase information of tissue scattering and reflection. The two-dimensional spatial coordinate system refers to a coordinate reference system established with the ultrasound probe surface as the starting plane, the sound beam propagation direction as the depth dimension, and the probe element arrangement direction as the lateral dimension. Each sound beam corresponds to a coordinate position in the lateral dimension, and each sampling point in the sound beam corresponds to a coordinate position in the depth dimension. The amplitude jump peak refers to the point where the signal amplitude suddenly and significantly increases when traversing the radiofrequency signal of a single sound beam along the depth dimension. This position corresponds to the strong acoustic impedance mismatch reflection generated when the sound beam first encounters the air-filled lung surface after passing through the chest wall soft tissue. The pleural interface line refers to a continuous spatial boundary curve formed by connecting the amplitude jump peaks identified on each sound beam in the lateral dimension. Anatomically, this curve corresponds to the surface projection position of the visceral pleura.

[0036] This step is triggered after the image processing device completes full data acquisition of the target scanning area via the radio frequency signal output interface. Specifically, the image processing device first organizes the acquired raw ultrasound radio frequency signals into a two-dimensional data matrix according to the beam number and sampling depth. One dimension of the matrix corresponds to the beam index in the lateral dimension, and the other dimension corresponds to the time sampling point index in the depth dimension. Subsequently, the device scans the radio frequency signal of each beam point by point along the depth dimension and calculates the amplitude change gradient between adjacent sampling points. When the gradient value exceeds the preset jump detection threshold, the device determines that the position is a candidate point for the amplitude jump peak on the current beam. Due to the significant difference in acoustic impedance between the chest wall soft tissue and the air-containing lung tissue, the reflected echo intensity at the pleural interface is much higher than the echo from the muscle and fat layers above it. Therefore, the first large amplitude jump peak physically corresponds to the depth position of the pleural interface. After the device repeats the above detection process for all sound beams, it obtains a set of peak depth coordinate sequences distributed in the lateral dimension. Then, it fits these discrete coordinate points into a continuous spatial boundary curve in the lateral dimension, namely the pleural interface line, through interpolation or smooth connection algorithms.

[0037] In some embodiments, the automatic identification and determination of the pleural interface line can be achieved in various ways. Optionally, the system can apply bandpass filtering to the original radio frequency signal of each sound beam to suppress high-frequency noise and low-frequency baseline drift, then calculate the Hilbert transform envelope of the filtered signal to obtain the instantaneous amplitude curve, and finally detect the position along the depth dimension where the envelope curve first exceeds a preset multiple of the mean of the entire sound beam envelope as the pleural depth coordinate of that sound beam. Optionally, the system can also calculate a short-time energy sequence along the depth dimension of the radio frequency signal matrix, then apply a first-order difference operation to the short-time energy sequence to highlight the energy change position, finally select the depth coordinate of the first positive maximum in the difference sequence as the amplitude jump peak, and then perform lateral smoothing of the peak coordinates of all sound beams through median filtering to eliminate individual abnormal points before connecting them into a continuous pleural interface line. It is understood that other signal processing methods can also be used to achieve automatic detection of the pleural interface position and construction of continuous boundaries, such as interface detection methods based on autocorrelation functions or peak localization methods based on template matching, which are not limited here.

[0038] S102. Based on the preset size of the sliding window, traverse the two-dimensional spatial coordinate system to determine the pleural roughness index and the subpleural signal energy index within the sliding window.

[0039] The preset-size sliding window refers to a rectangular analysis area defined in a two-dimensional coordinate system with fixed depth and lateral spans. This window moves sequentially in the depth and lateral dimensions with preset steps to cover the area below the pleural interface line. The pleural roughness index is a quantitative representation of the dispersion of the depth coordinates of each point on the pleural interface line relative to the local mean within the lateral range covered by the current sliding window. A higher value indicates a more irregular pleural surface morphology in that area. The subpleural signal energy index is a quantitative representation of the average energy density of the original radio frequency signal below the pleural interface line within the spatial range covered by the current sliding window. A higher value indicates that the subpleural tissue in that area produces stronger scattered or transmitted echoes of ultrasound.

[0040] After the image processing device completes the identification and determination of the pleural interface line in step S101, it uses this pleural interface line as a spatial reference to initiate the sliding window traversal process for this step. It is understandable that the lesions of interstitial lung disease are not uniformly distributed on the lung surface, and the severity of fibrosis varies significantly in different areas. Therefore, it is necessary to perform independent feature evaluation on each local area beneath the pleura through spatial traversal in order to screen out the core area.

[0041] Specifically, the image processing device constructs a rectangular sliding window that extends downwards by a preset distance in the depth dimension and covers a preset width in the lateral dimension, using the pleural interface line as the upper boundary reference. The device drives this sliding window to gradually translate in the lateral dimension with a preset step size, performing two parallel calculations at each new position. The first calculation is for the pleural roughness index: the device extracts the depth coordinate values ​​of all sampling points on the pleural interface line within the lateral range of the current window, calculates the local mean of these coordinate values, and then calculates the sum of squares of the differences between the depth coordinate of each sampling point and the local mean, taking the root mean square or variance. The resulting value is the pleural roughness index at that window position. The surface of a normal, healthy visceral pleura is smooth and flat, with minimal changes in the corresponding depth coordinates, and the roughness index approaches zero; however, when fibrosis causes pleural thickening, traction, or fragmentation, the pleural line shape becomes irregular, the depth coordinate fluctuations increase, and the roughness index rises significantly. The second calculation targets the subpleural signal energy index: the device extracts all original radio frequency signal samples located below the pleural interface line within the current window space and calculates the mean of the squared amplitudes as the average signal energy density of that region. Normal air-filled lung tissue experiences almost total sound reflection due to the extremely high acoustic impedance of the gas interface, resulting in very low radio frequency signal energy below the pleura. However, when interstitial fibrosis causes the gas in the alveoli to be replaced by exudate or fibrous tissue, ultrasound waves can propagate deeper and generate a large number of scattered echoes, significantly increasing the subpleural signal energy. After completing the calculation for all traversed positions within the sliding window, the device obtains two sets of index distribution maps corresponding to the spatial coordinates.

[0042] In some embodiments, the traversal of the sliding window and the calculation of the double exponent can be implemented in various ways. Optionally, based on the probe frequency and clinical experience, the depth span of the sliding window is set to a fixed depth range below the pleural line, and the lateral span is the coverage width of several adjacent sound beams. Then, the window is gradually translated in the lateral dimension with the single sound beam spacing as the step size. At each position, the standard deviation of the pleural line depth coordinates within the window is calculated as the pleural roughness index, and the mean square amplitude of the radio frequency signal within the window is calculated as the subpleural signal energy index. Finally, the double exponent values ​​at all positions are stored as two one-dimensional sequences corresponding to the lateral coordinates. Optionally, depth attenuation compensation is applied to the radio frequency signal below the pleural interface line to restore the true scattering intensity of deep tissues. Then, a two-dimensional sliding window is constructed on the compensated signal and moves simultaneously in both depth and lateral dimensions. Finally, the mean energy of the compensated signal is calculated at each window position as the subpleural signal energy index, and the variance of the pleural line coordinates within the window is calculated simultaneously as the pleural roughness index. It is understood that other size setting strategies or energy calculation methods can also be used to implement the traversal of the sliding window and the determination of the double exponent, which is not limited here.

[0043] S103. Extract the intersection region where the subpleural signal energy index is greater than the first preset threshold and the pleural roughness index is greater than the second preset threshold, and use it as the target observation window.

[0044] After the image processing device completes the full-space double-exponential traversal calculation in step S102, it enters the region screening stage of this step. It is understandable that fibrotic lesions in interstitial lung disease simultaneously manifest as two accompanying features: pleural surface morphology destruction and increased subpleural tissue density. Regions that meet only a single condition may be due to normal physiological phenomena or probe manipulation artifacts.

[0045] Specifically, to adapt to different device gain conditions, this scheme adopts a percentile adaptive threshold determination strategy: the device statistically analyzes the subpleural signal energy index distribution at all sliding window positions in the current frame, and selects the value located at a preset high percentile as the first preset threshold; similarly, it statistically analyzes the pleural roughness index distribution at all positions, and selects the value located at a preset high percentile as the second preset threshold. Regardless of how the absolute amplitude of the signal changes, the most abnormal region within the current scan frame is selected, thereby eliminating the influence of absolute value differences across devices. Subsequently, the device marks all spatial positions in the two-dimensional spatial coordinate system where the signal energy index is greater than the first preset threshold to form a first candidate set, and marks all spatial positions where the roughness index is greater than the second preset threshold to form a second candidate set. Then, a spatial intersection operation is performed on the two sets to obtain the region that simultaneously satisfies both conditions. Finally, the device determines this intersection region as the target observation window. This dual-threshold intersection strategy ensures that the target observation window is precisely and dynamically aligned with the core area of ​​the current lesion. It eliminates false positive areas where the energy is high only due to slightly higher lung water content under normal conditions but the pleural morphology is normal, as well as false positive areas where the pleural line positioning is shaken only due to probe slippage but the deep tissue is normal.

[0046] In some embodiments, when the number of sliding windows that meet the dual threshold conditions is insufficient to form an effective spatial intersection region, the first preset percentile and / or the second preset percentile are automatically reduced, and the filtering process is re-executed until a valid target observation window is obtained, or if a valid target observation window cannot be obtained, a prompt message is output to the operator to guide the adjustment of the probe scanning position.

[0047] S104. After taking the original ultrasonic radio frequency signal in the target observation window as the target radio frequency signal, perform feature extraction on the target radio frequency signal to obtain a multidimensional quantitative parameter vector.

[0048] The target radio frequency signal refers to the radio frequency signal data segment extracted from the original ultrasound radio frequency signal, whose spatial range is strictly limited to the boundary of the target observation window. This signal segment represents the complete acoustic echo information of the core region of the lesion. The multidimensional quantitative parameter vector refers to a set of objective physical quantities reflecting the microstructural characteristics of the tissue, extracted in parallel or sequentially from the target radio frequency signal. Each component in this vector corresponds to a specific acoustic scattering or attenuation characteristic parameter.

[0049] After the image processing device completes the determination of the target observation window in step S103, the device accurately extracts the signal data of the spatial range corresponding to the target observation window from the original radio frequency signal matrix and enters the multi-dimensional feature extraction stage of this step.

[0050] Specifically, the image processing device performs parallel calculations of three main categories of quantitative parameters on the target radio frequency signal. The first category is the extraction of envelope statistical parameters, which reflect the distribution and structural heterogeneity of tissue scatterers: the device first performs a Hilbert transform on the target radio frequency signal to obtain its analytic signal and then takes the modulus to obtain the envelope signal. Next, the amplitude statistical distribution of the envelope signal is fitted to specific probability density function models, and statistical parameters reflecting the tissue's microstructure are extracted from the fitting results. Specifically, the number of scatterers (μ) and the random scattering ratio (κ) parameter are obtained by fitting a Homodyned-K (HK) distribution model. The number of scatterers (μ) reflects the spatial density of effective acoustic scatterers within the region of interest. As pulmonary fibrosis progresses, normal alveolar structure is destroyed, and the proliferation of numerous fibroblasts and collagen deposition lead to a significant increase in the local concentration of effective scatterers. Therefore, this parameter value increases with the severity of the disease. The random scattering ratio (κ) reflects the degree of disorder in tissue structure. Normal lung tissue interfaces are relatively regular, while the honeycomb-like changes and reticular shadows caused by fibrosis result in highly heterogeneous internal structures, leading to a significant increase in the proportion of random scattering components in ultrasound echoes compared to normal tissue. This causes the parameter value to rise significantly in the lesion region. The shape parameter (m) and scale parameter (Ω) are obtained by fitting a Nakagami distribution model. The shape parameter (m) is closely related to the spatial arrangement and number distribution of effective scatterers within a unit resolution cell. When m is less than 1, it usually indicates that the signal envelope follows a skewed tail distribution, corresponding to a localized aggregation of scatterers in the early stages of fibrosis. When m is greater than or equal to 1, it approximates a Rayleigh or Rice distribution, corresponding to a state where diffuse dense changes in late-stage fibrosis lead to a more uniform and dense distribution of scatterers. The scale parameter (Ω) reflects the total power intensity of the backscattered signal within the region of interest. In fibrotic tissue, collagen deposition leads to an increase in the acoustic impedance difference at the tissue interface, which in turn enhances the backscattered energy. This scale parameter increases significantly with the severity of the lesion.

[0051] The second category involves the extraction of spectral parameters, which reflect the frequency-dependent attenuation and scattering characteristics of tissue to ultrasound. The device applies a window function to the target radio frequency signal and then performs a Fast Fourier Transform or power spectrum estimation to obtain the power spectrum of the radio frequency signal. To eliminate the influence of the ultrasound system itself (including device-related factors such as probe bandwidth response characteristics, system gain settings, and focusing zone diffraction effects), the device uses a reference phantom method to normalize the power spectrum. This involves dividing the power spectrum of the target tissue by the power spectrum of a reference phantom with known acoustic characteristics acquired under the same system settings, thus obtaining a normalized power spectrum that only reflects the acoustic characteristics of the tissue itself. Subsequently, the device performs linear regression fitting on the normalized power spectrum within the effective bandwidth of the probe, extracting three parameters from the fitting results: Mid-band Fit (MBF), Spectral Slope (SS), and Spectral Intercept (SI). The bandgap fit (MBF) is the power value of the fitted straight line at the center frequency. This value comprehensively reflects the scattering intensity and attenuation characteristics of ultrasound waves by the tissue. Collagen deposition in fibrotic tissue significantly increases its scattering cross-section, resulting in a significantly higher MBF value than normal lung tissue with high air content. The spectral slope (SS) is the slope of the fitted straight line, reflecting the rate of change of scattering intensity with ultrasound frequency. In interstitial lung disease, due to the relatively slower attenuation of high-frequency ultrasound components caused by fibrosis or the dominance of enhanced low-frequency scattering, the rate of frequency decrease in the spectrum changes, and the slope often shows a more negative trend. The spectral intercept (SI) is the intercept of the fitted straight line at zero frequency, i.e., the extrapolated value, reflecting the theoretical scattering energy intensity at zero frequency. Its significant increase in the fibrotic lesion area further characterizes the significant increase in scatterer density and scattering efficiency within the lung interstitium. Simultaneously, the device calculates the attenuation coefficient (AC) by comparing the power spectrum at different depths. Specifically, this is achieved by measuring the downward shift of the signal center frequency at different depths or by comparing the signal energy attenuation rate between adjacent depth intervals. Both methods quantify the energy loss of ultrasound waves propagating in tissues. Pulmonary fibrosis-induced tissue compaction alters the penetration and attenuation behavior of ultrasound waves on the lung surface; the AC value can directly quantify this pathological change caused by the disappearance of air-containing structures and the replacement by fibrous tissue.

[0052] The third category involves the extraction of scale-related parameters, which reflect the physical size and spatial distribution of the scatterer. Based on the aforementioned normalized power spectrum, the device further combines theoretical scattering models with nonlinear curve fitting to invert the physical size of the microstructure within the tissue. The theoretical scattering models used include, but are not limited to, the Gaussian morphology factor model and the fluid sphere model. The former assumes that the spatial autocorrelation function of the scatterer follows a Gaussian distribution and is suitable for scatterers with relatively regular shapes in soft tissues, while the latter assumes that the scatterer is spherical and that its acoustic properties differ from the surrounding medium, making it suitable for lesions containing cystic structures. The device performs curve matching between the normalized power spectrum and the theoretical spectrum of the selected scattering model, and inverts the backscattering coefficient, effective scatterer size, and effective acoustic concentration from the matching results. Backscattering coefficient (BSC) directly quantifies the backscattering cross section generated per unit solid angle and per unit volume of tissue. In normal lung tissue, due to the significant difference in acoustic impedance between the gas and tissue interfaces, incident ultrasound is almost totally reflected, preventing sound waves from effectively penetrating the lung parenchyma. Therefore, the BSC in normal air-containing lung tissue is extremely low. However, in fibrotic tissue, the reduced air content allows sound waves to penetrate the lung surface and generate strong scattering within the dense fibrous structure, resulting in a significantly higher BSC value compared to normal tissue. Effective scattering volume (ESD) reflects the average equivalent physical radius of the tissue structure that dominates ultrasound scattering. During the fibrotic process, bronchiolar dilation, the formation of microcellular cysts, and thickened interlobular septa collectively constitute new large-scale scattering sources, leading to a significantly higher measured ESD value compared to normal tissue. Effective acoustic concentration (EAC) is defined as the product of the number density of scatterers and the square of the acoustic impedance fluctuation. This parameter comprehensively reflects the increase in the number of scatterers in fibrotic lesions and the intensification of the acoustic impedance difference at the interface between collagen tissue and residual gas-containing structures. When fibrosis progresses, the EAC value increases significantly due to the dual driving forces of increased scatterer density and enhanced impedance contrast.

[0053] The device arranges and combines the above three categories of parameters in a preset order to form a complete multidimensional quantitative parameter vector. Each component in the vector corresponds to a specific acoustic scattering or attenuation characteristic parameter, which together constitute a quantitative characterization of the microacoustic properties of lung tissue within the target observation window from three orthogonal dimensions: envelope statistics, spectral response, and physical scale.

[0054] In some embodiments, multidimensional feature extraction of the target RF signal can be achieved in various ways. Optionally, the target RF signal is subjected to Hilbert transform to obtain the envelope signal, and the Nakagami distribution and Homodyned-K distribution are fitted using the maximum likelihood estimation method to obtain various statistical parameters. Then, a Hanning window is applied to the target RF signal, a fast Fourier transform is performed, and the logarithmic power spectrum is fitted with least squares linearly within the negative six dB bandwidth of the probe to obtain the spectral slope, spectral intercept, and frequency band fitting values. Finally, the normalized power spectrum is fitted with the theoretical backscattering model of the Gaussian scatterer using nonlinear least squares to obtain the effective scatterer size and backscattering coefficient. Optionally, the target observation window is divided into multiple equal-depth sub-regions along the depth dimension, and the power spectrum of each sub-region is calculated independently and the mean is taken to reduce the variance of the spectrum estimation. Then, the system transfer function is normalized to the power spectrum using the reference phantom method to eliminate equipment-related bias. Finally, linear regression and scattering model fitting are performed simultaneously on the normalized power spectrum to obtain spectral parameters and scale parameters.

[0055] It is understandable that other envelope distribution models or spectral analysis methods can also be used to extract multidimensional quantitative parameters, and no limitation is made here.

[0056] S105. Input the multidimensional quantitative parameter vector into the pre-trained regression model for mapping to obtain the quantitative evaluation index.

[0057] The pre-trained regression model refers to a machine learning regression model that has undergone supervised learning training based on clinical sample data before the deployment of this method. This model uses a multidimensional quantitative parameter vector as input features and a pulmonary fibrosis severity label annotated by a high-resolution computed tomography semi-quantitative scoring system as the training target, learning a nonlinear mapping relationship between ultrasound radiofrequency features and fibrosis degree. The quantitative assessment index is a continuous real number output by the model, ranging from zero to thirty, defined as the pulmonary fibrosis index. A higher value indicates a more severe degree of fibrosis in the assessed area.

[0058] After the image processing device completes the extraction of the multidimensional quantitative parameter vector in step S104, the device will send the vector into the deployed regression model for inference calculation.

[0059] Specifically, the image processing device first performs normalization processing on the multidimensional quantitative parameter vector obtained in step S104 to eliminate the influence of differences in the dimensions and orders of magnitude of different parameters on model inference. The statistics used for normalization are derived from the mean and standard deviation of each parameter calculated and stored on the training sample set during the model training phase. The normalized parameter vector is input into the pre-trained regression model, which has been constructed during the training phase through the following process: a batch of patients with interstitial lung disease confirmed by high-resolution computed tomography and healthy controls are collected as training samples; the high-resolution computed tomography images of each subject completed within a specified time window before and after ultrasound examination are manually labeled with the degree of fibrosis using a semi-quantitative visual scoring system to form a gold standard label; then, with the corresponding ultrasound radiofrequency multidimensional parameter vector as input and the gold standard label as the supervision signal, the hyperparameters are tuned using cross-validation method, with the goal of minimizing the mean square error between the predicted value and the label to complete the learning of model parameters. During model inference, the device performs nonlinear mapping operations on the normalized parameter vector through the model's decision function or ensemble tree structure, and finally outputs a continuous pulmonary fibrosis index as a quantitative assessment result of the current examination area.

[0060] In some embodiments, the pre-trained regression model specifically employs either a support vector regression model or a random forest regression model. When using a support vector regression model, the model maps the normalized multidimensional parameter vector to a high-dimensional feature space using a kernel function. Within this space, it searches for a regression hyperplane that minimizes the error between the predicted output and the true label while maintaining acceptable model complexity. When using a random forest regression model, the model constructs multiple randomly sampled decision trees. Each tree performs independent regression predictions based on different feature subsets and sample subsets of the parameter vector. The final ensemble output is the mean of all tree predictions. The gold standard label used during model training is derived from the Warrick scoring system. This system uses high-resolution computed tomography images as the basis for evaluation. Experienced radiologists perform semi-quantitative visual scoring of the degree of pulmonary fibrosis, ranging from 0 to 30 points, where 0 indicates no visible fibrosis and 30 indicates diffuse severe pulmonary fibrosis. This score comprehensively considers the proportion of lung lobes affected by fibrosis and the severity of the lesions. During model training, K-fold cross-validation was used to systematically tune the model's hyperparameters. The objective function was to minimize the mean squared error between the predicted values ​​on the training set and the gold standard label of the Warrick score, ensuring that the model's generalization performance was optimal under different parameter configurations. It should be noted that although the original labels of the Warrick score are integers, the regression decision surface constructed by the regression model (such as support vector regression) in the high-dimensional feature space through kernel function mapping is itself continuous. Therefore, the model's predicted output for any point in the input feature space is a continuous real value.

[0061] In some embodiments, the mapping from a multidimensional parameter vector to a quantitative evaluation index can be achieved in several ways. Optionally, the multidimensional parameter vector is normalized using Z-score normalization to ensure each parameter component has zero mean and unit variance. The normalized vector is then input into a pre-trained support vector regression model, mapped to a high-dimensional space using a radial basis function kernel, and regression decisions are performed. Finally, the model output value is truncated to an effective range of 0 to 30 as the quantitative evaluation index. Alternatively, the multidimensional parameter vector is scaled to a range of 0 to 1 using min-max normalization. The normalized vector is then input into a pre-trained random forest regression model, where multiple decision trees provide predictions, and the average is taken as the ensemble output. Finally, the output value is constrained to an effective range and used as the quantitative evaluation index.

[0062] It is understandable that other types of regression models or normalization strategies can also be used to achieve the mapping from multidimensional parameter vectors to continuous evaluation indices, which is not limited here.

[0063] S106. Obtain the historical evaluation index of the target observation window at historical time nodes, calculate the difference between the quantitative evaluation index and the historical evaluation index, generate the corresponding visualization trend mark and output it.

[0064] Here, "historical time node" refers to the time record corresponding to when the same patient underwent this method during previous visits; the system stores the assessment results of the patient at each historical time node. "Historical assessment index" refers to the pulmonary fibrosis index value obtained by the same patient at each historical time node using the same methodological process. "Visual trend marker" refers to an intuitive graphical symbol generated based on the direction and magnitude of the difference between the current assessment index and the historical assessment index.

[0065] After the image processing device completes the calculation of the current quantitative assessment index in step S105, the device enters the longitudinal comparison and trend visualization stage of this step. It is understandable that, as a chronic progressive disease, interstitial lung disease, while the absolute values ​​of a single examination have reference value, clinical decisions also depend on the disease's evolution trend over time. Therefore, it is necessary to place the current results in the patient's individual longitudinal timeline for comparative analysis.

[0066] Specifically, the image processing device queries the system database based on the patient's unique identifier to obtain the patient's historical assessment index stored at the most recent or multiple historical time points. Then, the device calculates the numerical difference between the current quantitative assessment index and the historical assessment index. Based on the positive or negative direction and absolute magnitude of the difference, the device generates corresponding visual trend markers according to preset rules: when the difference is positive and exceeds a preset change threshold, an upward arrow is generated to indicate fibrosis progression; when the difference is negative and exceeds the preset change threshold, a downward arrow is generated to indicate fibrosis improvement; when the absolute value of the difference does not exceed the preset change threshold, a horizontal arrow is generated to indicate relatively stable condition. Finally, the device outputs the current pulmonary fibrosis index value, the corresponding fibrosis grade, the visual trend markers, and the contribution information of optional key parameters to the display interface for clinicians to comprehensively interpret and make decisions. Simultaneously, the device automatically stores the current assessment index and its corresponding timestamp in the database for later retrieval during subsequent examinations.

[0067] In some embodiments, the output of the assessment results includes a structured presentation of the following three layers: The first layer is the joint display of the quantitative assessment index value and the corresponding fibrosis grade. The image processing device presents the pulmonary fibrosis index value of the current examination area in digital form on the display interface, and simultaneously displays the fibrosis severity level corresponding to the value according to the preset grading threshold. The grading includes at least four levels: normal, mild fibrosis, moderate fibrosis, and severe fibrosis. The boundary thresholds between each level can be preset by the device at the factory or customized by the clinical user according to specific disease subtypes and patient groups. The second layer is a visualization of the longitudinal trend. The device automatically generates corresponding visual trend markers based on the direction and magnitude of the difference between the current assessment index and the historical assessment index. The marker forms include, but are not limited to, directional arrows, numerical line trend graphs, or color gradient bar graphs. The directional arrows indicate fibrosis progression when upward, improvement when downward, and stability when horizontal. The line graph displays the evolution curve of all historical examination records of the patient with time as the horizontal axis and the assessment index as the vertical axis. The color gradient bar graph intuitively maps the level of the assessment index with color depth. The third layer is the output of auxiliary interpretability information. The device simultaneously presents the measured values ​​of the top-ranking key ultrasound quantitative parameters that contribute to the current pulmonary fibrosis index on the display interface. The contribution can be determined based on the importance of the regression model features or based on the degree to which each parameter deviates from the normal reference range. The purpose of this auxiliary information is to enable experienced clinicians to verify and understand the rationality of the model output results from the underlying physical parameter level, thereby enhancing the clinical credibility and transparency of the entire assessment system.

[0068] In this embodiment, the dual-threshold intersection strategy with the pleural interface line as the anatomical anchor point is used to automatically locate the target observation window and extract multidimensional quantitative parameter vectors within it. These vectors are then mapped to quantitative evaluation indices through a regression model. Combined with longitudinal historical comparisons, trend markers are generated, thereby achieving the technical effect of radiation-free, bedside, and repeatable dynamic quantitative monitoring of pulmonary fibrosis.

[0069] In the above embodiments, the image processing device can achieve quantitative assessment and longitudinal trend tracking of pulmonary fibrosis through dual-threshold intersection localization and multidimensional radiofrequency parameter regression mapping. In practical applications, when performing the aforementioned image processing method based on ultrasound radiofrequency signals, due to the patient's respiratory movements, the lung tissue region corresponding to the target observation window will shift back and forth in the depth dimension with the respiratory cycle. This makes it difficult for the operator to stably align the probe with the same anatomical position during continuous acquisition, affecting the spatial consistency between multiple measurements and the repeatability of quantitative results. This image processing method based on ultrasound radiofrequency signals can solve this technical problem by introducing a respiratory motion cycle extraction and static guide marker generation mechanism based on optical flow tracing, improving the spatial positioning stability of the target observation window under respiratory motion interference and the operator's real-time guidance capability.

[0070] Please see Figure 2 This is another flowchart illustrating an image processing method based on ultrasonic radio frequency signals in an embodiment of this application.

[0071] S201. Based on the original ultrasound radio frequency signal of the target scanning area, in a two-dimensional spatial coordinate system containing depth and lateral dimensions, identify the amplitude jump peak of the original ultrasound radio frequency signal along the depth dimension, and determine the continuous boundary formed by connecting the amplitude jump peak in the lateral dimension as the pleural interface line.

[0072] S202. Based on the preset size of the sliding window, traverse the two-dimensional spatial coordinate system to determine the pleural roughness index and the subpleural signal energy index within the sliding window.

[0073] S203. Extract the intersection region where the subpleural signal energy index is greater than the first preset threshold and the pleural roughness index is greater than the second preset threshold, and use it as the target observation window.

[0074] S204. After taking the original ultrasonic radio frequency signal in the target observation window as the target radio frequency signal, feature extraction is performed on the target radio frequency signal to obtain a multidimensional quantitative parameter vector.

[0075] S205. Input the multidimensional quantitative parameter vector into the pre-trained regression model for mapping to obtain the quantitative evaluation index.

[0076] S206. Obtain the historical evaluation index of the target observation window at historical time nodes, calculate the difference between the quantitative evaluation index and the historical evaluation index, generate the corresponding visualization trend mark and output it.

[0077] Steps S201 to S206 and Figure 1 In the illustrated embodiment, steps S101 to S106 are similar, and can be found in the description of steps S101 to S106, which will not be repeated here.

[0078] S207. Obtain a sequence of multiple consecutive ultrasound images containing the target observation window.

[0079] Among them, a continuous multi-frame ultrasound image sequence refers to a set of ultrasound B-mode image frames that are arranged in an orderly manner in the time dimension and are continuously recorded by the image processing equipment at the same time or within a similar time window for the same scanning area. Each frame in the sequence contains the anatomical space region corresponding to the target observation window.

[0080] After the image processing device completes the evaluation result output in step S206, in order to further improve the accuracy and repeatability of subsequent examinations, the device initiates this step to obtain dynamic image data reflecting the spatial displacement of the target observation window region due to respiratory motion. It is understandable that lung tissue undergoes periodic reciprocating displacement in the depth direction with respiratory motion, causing the spatial position of the target observation window to continuously change in the real-time image.

[0081] Specifically, after completing the quantitative analysis of the radio frequency signal, the image processing equipment retrieves a sequence of consecutive multi-frame ultrasound B-mode images containing the spatial location of the target observation window from the equipment's image buffer or real-time acquisition channel. The frame rate of this sequence should be sufficient to capture the temporal resolution requirements of tissue motion at a normal respiratory rate, typically covering at least one complete respiratory cycle. Once the equipment confirms that each frame in the sequence is valid and that the spatial coverage fully encompasses the target observation window, it proceeds to the next step for motion analysis.

[0082] S208. Based on the optical flow tracing algorithm, feature point matching is performed on a continuous multi-frame ultrasound image sequence to extract the motion trajectory of the target observation window in a two-dimensional spatial coordinate system.

[0083] Optical flow tracing is a computer vision motion estimation method based on the assumption of image brightness conservation. It estimates the displacement vector of feature points in an image by analyzing local changes in pixel grayscale distribution between adjacent frames. Feature point matching refers to the process of establishing cross-frame correspondences between pixels with significant grayscale gradient features within or around the target observation window across consecutive frames. Motion trajectory refers to the sequence of coordinate positions traversed by the spatial representative point of the target observation window over time in a series of consecutive image frames. This trajectory reflects the periodic displacement of tissues caused by respiratory motion.

[0084] After acquiring a sequence of multiple consecutive ultrasound images, the image processing device performs optical flow motion analysis on the sequence to quantify the spatiotemporal displacement patterns of the target observation window region. Specifically, in the first frame of the sequence, the image processing device automatically selects a set of feature points with good tracking characteristics within or near the target observation window region. These feature points are typically located at positions with significant gray-level gradient changes to ensure stable tracking in subsequent frames. Then, the device sequentially executes an optical flow tracing algorithm on adjacent frame pairs in the sequence: for each pair of adjacent frames, the algorithm calculates the corresponding positions of each feature point in the previous frame in the next frame based on the gray-level distribution consistency constraint of the local image patch, obtaining the displacement vector of each feature point within the frame interval. The device accumulates the displacement vectors of all frames frame by frame to obtain the cumulative displacement trajectory of each feature point from the starting frame of the sequence to each subsequent frame. Finally, the device spatially aggregates the trajectories of all feature points within the target observation window, using the mean or median of their displacements to represent the overall motion trajectory of the target observation window, forming a two-dimensional coordinate sequence that varies over time.

[0085] In some embodiments, motion trajectory extraction based on optical flow tracing can be achieved in several ways. Optionally, a corner detection algorithm is used to automatically detect several corner features within the target observation window region of the first frame as an initial set of tracking points. Then, for each pair of adjacent frames in the sequence, a pyramid-style Lucas-Kanade optical flow algorithm is used to calculate the inter-frame displacement vector of each tracking point and accumulate them to obtain the temporal trajectory of each point. Finally, the spatial median of the trajectories of all tracking points is taken to eliminate the influence of individual point tracking failures, thereby obtaining the overall motion trajectory of the target observation window. Optionally, five tracking seed points are manually or automatically set at the geometric center and four corner positions of the target observation window. Then, a dense optical flow algorithm is used to calculate the optical flow field of the entire frame image and extract the displacement vectors at the seed point positions, accumulating them frame by frame to form their respective trajectories. Finally, a weighted average of the five trajectories is performed to represent the overall motion of the window.

[0086] It is understandable that other motion estimation methods or feature tracking strategies can also be used to extract the motion trajectory of the target observation window, and this is not limited here.

[0087] S209. Fit the respiratory motion cycle based on the motion trajectory to determine the displacement range of the target observation window within the respiratory motion cycle.

[0088] Among them, the respiratory cycle refers to the time cycle parameter that reflects the patient's respiratory rhythm, which is obtained by fitting the motion trajectory data. A complete respiratory cycle includes the tissue repositioning process corresponding to one inspiratory phase and one expiratory phase.

[0089] After the image processing device acquires the motion trajectory of the target observation window, it extracts the time series of the depth dimension component from the trajectory. Since respiratory motion is mainly manifested as the periodic rise and fall of lung tissue along the depth direction, this component sequence exhibits an approximately sinusoidal periodic fluctuation characteristic. The device performs frequency domain analysis on this time series, determining its dominant frequency component through Fourier transform or autocorrelation analysis. The reciprocal of this dominant frequency is the duration of the respiratory motion cycle. Subsequently, based on the fitted respiratory cycle, the device segments the motion trajectory to determine the maximum and minimum depth coordinate positions of the target observation window within a complete cycle. The difference between these two values ​​represents the maximum displacement range caused by respiratory motion. Determining this displacement range provides the necessary spatial parameters for calculating the center position of the static guide marker in the next step.

[0090] In some embodiments, the fitting of the respiratory movement cycle and the determination of the displacement range can be achieved in various ways. Optionally, a low-pass filter is applied to the depth component time series of the motion trajectory to remove high-frequency interference components such as heartbeat, retaining the signal within the respiratory frequency range. Then, a fast Fourier transform is performed on the filtered sequence, and the frequency corresponding to the maximum peak value in the power spectrum is extracted as the respiratory frequency to determine the respiratory movement cycle. Finally, the difference between the global maximum and global minimum values ​​of the sequence over the entire recording time is calculated as the displacement range. Optionally, peak detection is performed on the depth component time series to locate all local maximum and local minimum time points. Then, the median of the time interval between adjacent extreme points of the same type is calculated as the estimated value of the respiratory movement cycle. Finally, the difference between the mean of all maximum positions and the mean of all minimum positions is calculated as the stable displacement range within the respiratory movement cycle. It is understood that other time-frequency analysis methods or cycle estimation algorithms can also be used to fit the respiratory movement cycle and quantify the displacement range; this is not limited here.

[0091] S210. Generate static guide markers based on the center coordinates of the displacement range, and superimpose the static guide markers onto the currently displayed ultrasound image.

[0092] The center coordinate of the displacement range refers to the geometric center of the extreme displacement range of the target observation window within the respiratory cycle. This coordinate represents the equivalent static position of the target observation window after eliminating respiratory motion fluctuations.

[0093] After the image processing device completes the determination of the respiratory motion cycle and displacement range in step S209, the device proceeds to this step to convert the motion analysis results into visual guidance information that can be directly used by the operator. Understandably, in subsequent re-examinations, the operator needs to precisely align the probe with the same anatomical position as in the previous examination to ensure spatial consistency in longitudinal contrast, and static guide markers provide the operator with a stable positioning target undisturbed by breathing.

[0094] Specifically, the image processing device calculates the center coordinates of the displacement range, taking the arithmetic mean of the highest and lowest coordinates of the target observation window in the depth direction during the respiratory cycle as the center depth coordinates, and the median coordinate in the lateral direction as the center lateral coordinates. Based on these center coordinates, the device generates a static guide marker of a predefined shape, whose geometric center coincides with the calculated center coordinates, and whose size can be set to match the spatial dimensions of the target observation window. Subsequently, the device renders this static guide marker as a graphic overlay onto the currently displayed real-time ultrasound image. Because the position of this marker is calculated based on the center position of the respiratory motion rather than following the tissue movement in the real-time image, it appears as a stable reference frame with a fixed position on the display interface. During subsequent examinations, the operator can adjust the probe to align the current target area with this static guide marker, thereby ensuring good spatial consistency of the acquisition area for each examination.

[0095] In some embodiments, the generation and overlay display of static guide markers can be achieved in various ways. Optionally, the center depth coordinates are determined by taking the arithmetic mean of the upper and lower extreme values ​​of the displacement range and combining it with the lateral median coordinates of the target observation window to determine the complete center position. Then, a rectangular or cross-shaped semi-transparent graphic marker adapted to the size of the target observation window is generated with this center position as the center. Finally, this marker is overlaid as a fixed layer on the real-time ultrasound display screen and rendered with a striking contrast color for operator identification. Optionally, the center coordinates of the displacement range are calculated and stored together with the size information of the target observation window as the patient's spatial positioning file. Then, during subsequent follow-up examinations, the center coordinates and size information are automatically retrieved from the file to generate guide markers and overlay them onto the real-time image. Finally, text prompts are displayed synchronously next to the markers to guide the operator to adjust the probe position until the target area is aligned with the markers. It is understood that other marker forms or overlay methods can also be used to generate and display static guide markers, which are not limited here.

[0096] Understandably, steps S207 to S210 generate static guide markers by introducing optical flow tracing and fitting the respiratory motion cycle, thereby achieving stable guidance for the spatial positioning of the target observation window under respiratory motion interference. In some embodiments, if the inspection only requires a single acquisition and there is no need for subsequent re-examination and alignment, steps S207 to S210 may not be executed. In this case, the output of the entire evaluation process can be completed after step S206, which is not limited here.

[0097] In this embodiment, by employing a technique based on optical flow tracing to extract the motion trajectory of the target observation window and fitting the respiratory cycle to determine the center coordinates of the displacement range, thereby generating a static guide marker, the system can transform the target area location information that originally changed with respiratory motion into a fixed spatial reference marker. In subsequent re-examinations, the operator can quickly align the probe with the same anatomical position as in previous examinations based on this marker. This effectively solves the technical problems in the prior art where spatial alignment of the target area is difficult between multiple examinations due to respiratory motion and the repeatability of quantitative results is limited. Thus, it achieves the technical effects of ensuring spatial consistency across examination time nodes and improving the reliability of longitudinal quantitative assessment.

[0098] The exemplary image processing device 300 provided in the embodiments of this application is described below. Figure 3 This is an exemplary hardware structure diagram of the image processing device 300 provided in the embodiments of this application.

[0099] In some embodiments, the image processing device 300 is a computer device or includes a computer device. The computer device includes a processor, memory, and a network interface connected via a system bus. The processor of the computer device provides computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores an operating system, computer programs, and a database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The database of the computer device stores data. The network interface of the computer device is used to communicate with other external terminals or servers via a network connection. In some embodiments, the network interface can be a wired network interface; in some embodiments, the network interface can also be a wireless network interface. When the computer program is executed by the processor, it implements the methods in the embodiments of this application.

[0100] Those skilled in the art will understand that Figure 3The 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.

[0101] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit it. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of this application.

[0102] As used in the above embodiments, depending on the context, the term "when..." can be interpreted as meaning "if...", "after...", "in response to determining...", or "in response to detecting...". Similarly, depending on the context, the phrase "when determining..." or "if (the stated condition or event) is interpreted as meaning "if determining...", "in response to determining...", "when (the stated condition or event) is detected", or "in response to detecting (the stated condition or event)".

[0103] In the above embodiments, implementation can be achieved entirely or partially through software, hardware, firmware, or any combination thereof. When implemented using software, it can be implemented entirely or partially in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, fiber optic, digital subscriber line) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that integrates one or more available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium (e.g., solid-state drive), etc.

[0104] 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. This program can be stored in a computer-readable storage medium, and when executed, it can include the processes described in the above method embodiments. The aforementioned storage medium includes various media capable of storing program code, such as ROM or random access memory (RAM), magnetic disks, or optical disks.

Claims

1. An image processing method based on ultrasonic radio frequency signals, characterized in that, The method includes: Based on the original ultrasound radio frequency signal of the target scanning area, in a two-dimensional spatial coordinate system including depth and lateral dimensions, the amplitude jump peak of the original ultrasound radio frequency signal is identified along the depth dimension, and the continuous boundary formed by connecting the amplitude jump peak in the lateral dimension is determined as the pleural interface line. A sliding window of a preset size is traversed in the two-dimensional spatial coordinate system to determine the pleural roughness index and the subpleural signal energy index within the sliding window. The pleural roughness index is the depth coordinate variance of the pleural interface line within the sliding window, and the subpleural signal energy index is the mean value of the radio frequency signal envelope energy in the region below the pleural interface line within the sliding window. The intersection region where the subpleural signal energy index is greater than a first preset threshold and the pleural roughness index is greater than a second preset threshold is extracted as the target observation window; After taking the original ultrasonic radio frequency signal within the target observation window as the target radio frequency signal, feature extraction is performed on the target radio frequency signal to obtain a multidimensional quantitative parameter vector, which includes envelope statistical parameters, spectral parameters and spatial scale parameters. The multidimensional quantitative parameter vector is input into a pre-trained regression model for mapping to obtain a quantitative evaluation index; Obtain the historical evaluation index of the target observation window at historical time points, calculate the difference between the quantitative evaluation index and the historical evaluation index, generate corresponding visual trend markers, and output and display them.

2. The method according to claim 1, characterized in that, The step of traversing the sliding window of a preset size in the two-dimensional coordinate system to determine the pleural roughness index and the subpleural signal energy index within the sliding window specifically includes: Determine the set of analytical envelope peaks within the sliding window that run along the pleural interface line; When the local average value of the analytical envelope peak set is higher than the preset noise floor threshold, the depth coordinate variance of the pleural interface line is calculated based on the depth coordinates corresponding to each peak in the analytical envelope peak set to obtain the pleural roughness index. Starting from the pleural interface line within the sliding window, a preset distance is extended to the depth dimension to construct the subpleural region of interest. The original ultrasound radio frequency signal is subjected to depth attenuation compensation, and the average value of the radio frequency signal envelope energy after depth attenuation compensation in the subpleural region of interest is used as the subpleural signal energy index.

3. The method according to claim 1, characterized in that, The step of extracting the intersection region where the subpleural signal energy index is greater than a first preset threshold and the pleural roughness index is greater than a second preset threshold, and using it as the target observation window, specifically includes: Obtain the set of subpleural signal energy indices and the set of pleural roughness indices corresponding to all traversed sliding windows within the current target scanning area; The set of subpleural signal energy indices is statistically sorted, and the first preset percentile in the set of subpleural signal energy indices is determined as the first preset threshold. The set of pleural roughness indices is statistically sorted, and the second preset percentile in the set of pleural roughness indices is determined as the second preset threshold. In the two-dimensional spatial coordinate system, sliding windows that simultaneously satisfy the conditions of a subpleural signal energy index greater than the first preset threshold and a pleural roughness index greater than the second preset threshold are selected, and the spatial intersection region corresponding to the selected sliding windows in the two-dimensional spatial coordinate system is determined as the target observation window.

4. The method according to claim 1, characterized in that, Before the step of identifying amplitude jump peaks of the original ultrasound radio frequency signal based on the target scanning area in a two-dimensional spatial coordinate system including depth and lateral dimensions along the depth dimension, and determining the continuous boundary formed by connecting the amplitude jump peaks in the lateral dimension as the pleural interface line, the method further includes: Obtain the anatomical prior constraint information corresponding to the target scanning area, wherein the anatomical prior constraint information includes a preset pleural depth range; Calculate the normalized cross-correlation coefficient between adjacent sound beams of the original ultrasonic radio frequency signal in the lateral dimension; If the normalized cross-correlation coefficient is greater than or equal to a preset correlation threshold, the signal segment in the original ultrasound radio frequency signal that is located within the preset pleural depth range is determined as a candidate signal segment. Envelope extraction is performed on the candidate signal segments to obtain the analytical envelope signal corresponding to the original ultrasonic radio frequency signal, and the analytical envelope signal is used as the original ultrasonic radio frequency signal to perform the step of identifying the amplitude jump peak of the original ultrasonic radio frequency signal along the depth dimension.

5. The method according to claim 4, characterized in that, After the step of calculating the normalized cross-correlation coefficient between adjacent sound beams of the original ultrasonic radio frequency signal in the lateral dimension, the method further includes: When the normalized cross-correlation coefficient is less than the preset correlation threshold, the temporal signal sequence of the original ultrasound radio frequency signal in multiple consecutive image frames is extracted; The time-series signal sequence is subjected to multi-frame averaging to obtain a smoothed radio frequency signal with enhanced signal-to-noise ratio; Based on the smoothed radio frequency signal, an envelope extraction and filtering algorithm is performed to obtain a moving average envelope, and the signal segments in the moving average envelope that are located within the preset pleural depth range are determined as the candidate signal segments.

6. The method according to claim 1, characterized in that, After the steps of obtaining the historical evaluation index of the target observation window at historical time points, calculating the difference between the quantitative evaluation index and the historical evaluation index, generating corresponding visual trend markers, and outputting and displaying them, the method further includes: Acquire a sequence of consecutive multi-frame ultrasound images containing the target observation window; Based on the optical flow tracing algorithm, feature point matching is performed on the continuous multi-frame ultrasound image sequence to extract the motion trajectory of the target observation window in the two-dimensional spatial coordinate system; Based on the motion trajectory, fit the respiratory motion cycle and determine the displacement range of the target observation window within the respiratory motion cycle; A static guide mark is generated based on the center coordinates of the displacement range, and the static guide mark is superimposed on the currently displayed ultrasound image.

7. The method according to claim 6, characterized in that, The step of fitting the respiratory motion cycle based on the motion trajectory and determining the displacement range of the target observation window within the respiratory motion cycle specifically includes: The coordinate sequence of the motion trajectory is transformed in the frequency domain to extract the dominant frequency component corresponding to the breathing frequency; Reconstruct the respiratory motion phase curve based on the dominant frequency component, and identify the end-expiratory phase point and end-inspiratory phase point in the respiratory motion phase curve. Extract the spatial coordinate extreme values ​​of the target observation window corresponding to the end-expiratory phase point and the end-inspiratory phase point, and determine the spatial envelope region formed by the spatial coordinate extreme values ​​as the displacement range.

8. An image processing device, characterized in that, The image processing device includes: one or more processors and a memory; the memory is coupled to the one or more processors, the memory is used to store computer program code, the computer program code including computer instructions, and the one or more processors call the computer instructions to cause the image processing device to perform the method as described in any one of claims 1-7.

9. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the method of any one of claims 1-7.

10. A computer-readable storage medium comprising instructions, characterized in that, When the instructions are executed on an image processing device, the image processing device performs the method as described in any one of claims 1-7.

Images