Fibrosis quantitative evaluation method and system based on kidney ultrasonic image

Abstract

The invention relates to the technical field of medical image processing, in particular to a kidney ultrasonic image-based fibrosis quantitative evaluation method and system, and the method comprises the steps: obtaining an ultrasonic image, carrying out the standardization processing, and estimating a system response correction function and an optical flow field; selecting a key frame based on an optical flow field, performing registration and constructing pose map optimization, fusing a envelope to establish mapping between a cortical conformal coordinate system and a key frame pixel, and generating a cortical region probability field and a boundary uncertainty field; calculating a sound attenuation coefficient and a scattering statistical parameter in the sliding analysis window, correcting the sound attenuation coefficient and the scattering statistical parameter, projecting the sound attenuation coefficient and the scattering statistical parameter to a conformal coordinate system, carrying out Wasserstein gravity center fusion according to probability and uncertainty weighting, detecting the effective thickness of the cortex, and carrying out consensus optimization to obtain a fibrosis fraction field and a consensus residual field; and generating a non-consistency total score, outputting a confidence interval and a confidence level through conformal prediction calibration, and generating a fallback recalculation or re-collection instruction if necessary. The cross-view, cross-time and cross-equipment consistency is improved, and the closed-loop quality control capability is achieved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of medical image processing, and in particular to a fibrosis quantitative evaluation method and system based on kidney ultrasound images. BACKGROUND

[0002] The progression and prognosis management of chronic kidney disease highly depend on the objective quantification of fibrosis degree. The industry demand for "repeatable, comparable, traceable" non-invasive assessment continues to grow: on the one hand, for stratified management and follow-up decision-making (such as risk stratification, efficacy response tracking, and review interval optimization); on the other hand, for multi-center data consistency and long-term cohort study, to promote standardized management and quality control system construction. Existing ultrasound assessment is often affected by equipment differences, operator scanning angle and path differences, unclear tissue boundaries, and noise interference, resulting in insufficient consistency of results between different perspectives, different time periods, and different examination batches; at the same time, there is a lack of quantitative characterization of "assessment reliability" and executable closed-loop correction mechanisms, making it difficult to form stable and usable quantitative outputs and traceable evidence chains in clinical processes. SUMMARY

[0003] The present application provides a fibrosis quantitative evaluation method and system based on kidney ultrasound images, which at least solves the problem of how to achieve stable and consistent quantitative results of kidney cortical fibrosis under conditions of equipment differences, scanning differences, and boundary uncertainty, and provide reliable and executable correction.

[0004] In a first aspect, the present application provides a fibrosis quantitative evaluation method based on kidney ultrasound images, comprising the following steps: Obtain kidney ultrasound image data and perform standardization processing, estimate system response correction function and optical flow field; Select a key frame set based on the optical flow field, construct and optimize a pose graph based on key frame image registration, generate a mapping relationship between the cortical conformal coordinate system and the key frame pixel coordinates by fusing the kidney envelope, generate a cortical region probability field and a boundary uncertainty field; Calculate the acoustic attenuation coefficient and scattering statistical parameters in the sliding analysis window and perform system response correction, project to the cortical conformal coordinate system using the mapping relationship and execute Wasserstein barycenter fusion according to the cortical region probability field and the boundary uncertainty field, detect the effective thickness of the cortex and obtain the fibrosis score field and the consensus residual field through consensus optimization; Generate a non-consistency total score based on the fibrosis score field and output a confidence interval and a confidence level through a conformal prediction calibration method, and generate a rollback recalculation instruction or a resampling instruction.

[0005] In a second aspect, the present application provides a fibrosis quantitative evaluation system based on kidney ultrasound images, for implementing the fibrosis quantitative evaluation method based on kidney ultrasound images, the system comprising: The collection preprocessing module is used for acquiring kidney ultrasound image data and standardizing processing, and estimating a system response correction function and an optical flow field; The mapping module is used for selecting a key frame set based on the optical flow field, constructing and optimizing a pose graph based on key frame image registration, fusing a kidney envelope to generate a mapping relationship between a cortex conformal coordinate system and a key frame pixel coordinate, generating a cortex region probability field and a boundary uncertainty field, and generating a fibrosis score field and a consensus residual field. The fusion inversion module is used for calculating an acoustic attenuation coefficient and a scattering statistical parameter in a sliding analysis window and performing system response correction, projecting to the cortex conformal coordinate system by using the mapping relationship, and performing Wasserstein barycentric fusion according to the cortex region probability field and the boundary uncertainty field. The calibration decision module is used for generating a non-consistency total score based on the fibrosis score field, outputting a confidence interval and a confidence level by using a conformal prediction calibration method, and generating a backtracking recalculation instruction or a resampling instruction.

[0006] Compared with the prior art, the application has the following advantages and beneficial effects: By means of joint modeling of the system response correction function and the optical flow field, the comparability across devices and frames is improved, and the artifact suppression effect is achieved; by means of key frame selection, pose graph optimization and envelope fusion to establish the cortex conformal coordinate system, the multi-angle and multi-period data are aligned and stably projected in the unified coordinate; by means of the weighted Wasserstein barycentric fusion of the cortex region probability field and the boundary uncertainty field, the robust fusion effect of the unclear boundary and local noise is achieved; by means of the cortex effective thickness detection and consensus optimization, the spatial continuity and consistency of the fibrosis score field are enhanced; by means of the non-consistency total score and the conformal prediction calibration, the explainable output effect of the confidence interval and the confidence level is achieved, and the process closed-loop quality control and the operable correction effect are achieved by the backtracking recalculation instruction or the resampling instruction. BRIEF DESCRIPTION OF DRAWINGS

[0007] Figure 1 FIG. 1 is a schematic diagram of the execution process of the method of the application; Figure 2 FIG. 3 is a schematic diagram of the cortex effective thickness change point detection curve in the specific embodiment of the application; Figure 3 FIG. 5 is a schematic diagram of the fusion parameter field distribution in the specific embodiment of the application; Figure 4 FIG. 7 is a schematic diagram of the fibrosis score field distribution in the specific embodiment of the application; Figure 5 FIG. 9 is a structural block diagram of the system of the application. DETAILED DESCRIPTION

[0008] Embodiments of the present disclosure will be described below with reference to the accompanying drawings. However, it should be understood that these descriptions are merely exemplary and are not intended to limit the scope of the present disclosure. In the following detailed description, numerous specific details are set forth in order to provide a thorough understanding of the embodiments of the present disclosure.

[0009] Kidney ultrasound imaging is one of the most commonly used clinical methods for observing the morphology and echogenicity of the kidney. The echo texture and boundary information of the kidney capsule, cortical region, medullary region, and renal sinus region are usually obtained through different sections and different incident angles, and a multi-view sequence such as longitudinal and transverse sections can be formed during the same examination process. The intensity distribution of ultrasound imaging not only reflects the reflection and scattering characteristics of the tissue interface, but also is related to the imaging parameters such as frequency response, gain, and time gain compensation of the system transmission and reception chain. At the same time, due to the influence of respiration, body position change, and slight displacement of the probe, there are deformations and pose changes between consecutive frames. Therefore, when the ultrasound image is used to carry the quantitative information of the tissue state, it is usually necessary to first standardize and correct the system response of the image, and then register and align the data of different frames and different views, and then perform feature calculation and fusion in a stable anatomical reference system to ensure the consistency and interpretability of the subsequent quantitative results. On this basis, the present application further establishes a unified coordinate expression and quality control mechanism for the renal cortex region to realize the robust quantitative evaluation of fibrosis.

[0010] As shown in Figure 1 , a fibrosis quantitative evaluation method based on kidney ultrasound images includes the following steps: Obtain kidney ultrasound image data and perform standardization processing, estimate system response correction function and optical flow field; Collect the B-mode frame sequence and the synchronous echo signal of the kidney scan, and obtain the phantom calibration echo signal. Perform time gain consistency, gray scale normalization, spatial scale unification, and frame time alignment on the B-mode frame sequence to obtain standardized kidney ultrasound image data. Calculate the power spectrum of the phantom calibration echo signal and divide it by the preset template power spectrum to obtain the system response correction function: ; wherein, is the system response correction function, is the phantom calibration echo power spectrum, is the template power spectrum, is the frequency. Perform optical flow estimation on adjacent standardized B-mode frames to obtain an optical flow field, which is used for subsequent key frame selection and pose graph construction.

[0011] The standardization processing includes: performing gray scale normalization and spatial scale unification on the kidney ultrasound image data; the estimating system response correction function includes: obtaining a reference echo signal and determining the system response correction function based on spectral characteristics of the reference echo signal; the estimating optical flow field includes: calculating the optical flow field based on gray scale consistency and smoothness constraints of adjacent frames.

[0012] In an embodiment, obtaining the kidney ultrasound image data includes obtaining a sequence of B-mode frames of a kidney scan and a reference echo signal synchronized with the sequence of B-mode frames. The reference echo signal is an echo signal obtained by scanning a preset phantom or a preset reflector under the condition that the parameters of the ultrasound probe are fixed, and is used to represent the amplitude response difference of the ultrasound system at different frequencies. In order to make the kidney ultrasound image data collected by different batches and different operators have consistent dimensions and contrast, first, the gray scale normalization and spatial scale unification are performed on the kidney ultrasound image data. The gray scale normalization includes performing dynamic range truncation, linear mapping and contrast consistency processing on each B-mode image, wherein the dynamic range truncation is used to suppress the influence of a small number of extreme brightness pixels on the overall mapping, and the contrast consistency is used to make the gray scale distribution of different frames fall into a consistent target interval, thereby reducing the interference of automatic gain and compression curve difference on subsequent analysis. The spatial scale unification includes reading the pixel spacing calibration information and performing resampling on the B-mode image, so that the pixels under different acquisition settings correspond to a unified physical scale, and at the same time, the image is corrected in a unified coordinate direction and cropped in a field of view, so as to ensure that the same anatomical region has an alignable spatial range between different frames.

[0013] In estimating the system response correction function, the reference echo signal is framed according to a window function consistent with subsequent analysis and the power spectrum is calculated, and the system response correction function is generated based on the spectral characteristics of the power spectrum. The system response correction function is used to correct the acoustic attenuation coefficient and scattering statistical parameter obtained from the power spectrum of the echo signal in the frequency domain, so as to reduce the systematic deviation caused by the fluctuation of the probe bandwidth and the change of the receiving link gain with frequency. In order to ensure repeatability, the reference echo signal is obtained with the same center frequency, bandwidth, sampling rate and dynamic range setting as the kidney scan, and a stable echo band is extracted in the same imaging depth range as the correction basis.

[0014] In estimating the optical flow field, the gray consistency between adjacent frames is taken as a data constraint, and the spatial smoothness of the displacement field is taken as a regularization constraint, and the optical flow field is obtained by multi-scale iterative solution. Specifically, first, a pyramid is constructed for the adjacent two B-mode images, and the displacement field is estimated from coarse to fine level by level, and at each scale, the next frame is resampled and aligned according to the current displacement field, and then the displacement field is updated until the residual error converges or the preset iteration number is reached. The obtained optical flow field is used to represent the inter-frame motion of the kidney under the influence of respiration and slight probe jitter, and provides the motion consistency basis for subsequent key frame selection and pose graph construction, and reduces the structure fusion error caused by inter-frame misalignment.

[0015] Based on the optical flow field, a set of key frames is selected, a pose graph is constructed and optimized based on the image registration of the key frames, a mapping relationship between the cortical conformal coordinate system and the key frame pixel coordinates is generated by fusing the kidney capsule, and a cortical region probability field and a boundary uncertainty field are generated; Based on the optical flow field, the displacement amplitude, displacement direction consistency and invalid pixel proportion of adjacent frames are counted, the motion mutation frame and the echo missing frame are removed, and a set of key frames is selected in combination with the time interval and the field of view coverage. Two-level image registration is performed between the key frames, coarse registration is performed first to eliminate the translation and rotation difference, and then fine registration is performed to refine the local alignment relationship, to obtain the relative pose between the key frames and the registration residual error; the relative pose is used as a constraint relationship to construct a pose graph, and the registration residual error of all constraint relationships is minimized and optimized as a whole, and the pose of each key frame in the unified kidney coordinate system is output. The kidney capsule curve is segmented for each key frame, and the kidney capsule curve is aligned and fused into a global kidney capsule curve according to the key frame pose; a cortical conformal coordinate system is established with the arc length of the global kidney capsule curve as the horizontal coordinate and the distance along the normal as the vertical coordinate, forming a mapping relationship between the cortical conformal coordinate system and the key frame pixel coordinates. The cortical region probability field is generated based on the segmentation probability of the kidney capsule curve, and the boundary uncertainty field is generated based on the pose graph optimization residual error and the boundary positioning error of the kidney capsule curve, which is used for subsequent observation weighting and fusion.

[0016] Based on the optical flow field, a set of key frames is selected, a pose graph is constructed and optimized based on the image registration of the key frames, a mapping relationship between the cortical conformal coordinate system and the key frame pixel coordinates is generated by fusing the kidney capsule, and a cortical region probability field and a boundary uncertainty field are generated;

[0017] In one embodiment, the process of selecting a keyframe set based on optical flow field includes calculating the inter-frame motion consistency of a continuous kidney ultrasound frame sequence frame by frame, and retaining frames that meet the motion consistency requirements as candidate keyframes. Specifically, for each pair of adjacent frames, the displacement amplitude distribution and displacement direction distribution within the effective pixel region are statistically analyzed. The displacement amplitude reflects the overall motion intensity, and the displacement direction reflects the concentration of motion. When the displacement amplitude is continuously distributed in space and the displacement direction is consistent within the effective pixel region, the inter-frame is determined to be a stable motion segment, and a candidate keyframe set is extracted from the stable motion segment at time intervals. To avoid misselection caused by optical flow field alone, keyframe image registration is further performed on the candidate keyframe set, and the keyframe set is filtered according to the registration residual rules. The registration residual rule is as follows: after registering any candidate keyframe pair, calculate the mean pixel difference and structural similarity index of the overlapping region after registration. When the mean pixel difference meets the first preset condition and the structural similarity index meets the second preset condition, it is determined that the registration residual of the candidate keyframe pair meets the registration residual rule. When the registration residual of the candidate keyframe and its adjacent candidate keyframe does not meet the registration residual rule, the corresponding candidate keyframe is removed and its previous candidate keyframe is retained as the keyframe.

[0018] When constructing and optimizing the pose graph, the first step is to establish adjacent keyframe constraints within the keyframe set. These constraints consist of the relative poses obtained from registering adjacent keyframe images and the registration residuals. The relative poses describe the relative transformations of adjacent keyframes in a unified kidney coordinate system, while the registration residuals characterize the reliability of this transformation and are used for subsequent constraint weighting. Next, loop closure constraints are established to suppress registration drift that accumulates over time. The loop closure constraints are established as follows: keyframe pairs with a time interval satisfying a preset span are selected as candidate loop closure pairs. Keyframe image registration is performed on these candidate loop closure pairs. When the registration residuals satisfy the registration residual rules, the candidate loop closure pair is added to the pose graph as a loop closure constraint, and the registration residuals determine the loop closure constraint weights. After establishing the adjacent keyframe constraints and loop closure constraints, graph optimization is performed on the pose graph. By iteratively minimizing the weighted registration residuals of all constraints, the keyframe pose of each keyframe in the unified kidney coordinate system is obtained. By using the keyframe selection and pose graph optimization described above, the subsequent renal capsule fusion and cortical conformal coordinate system construction have a stable spatial alignment basis, thereby reducing structural misalignment caused by breathing and probe jitter.

[0019] The mapping relationship between the cortical conformal coordinate system generated by fusing the renal capsule and the keyframe pixel coordinates includes: segmenting the renal capsule curve in the keyframe, obtaining the global renal capsule curve based on the keyframe pose alignment and fusion, generating the cortical conformal coordinate system by parameterizing the arc length of the global renal capsule curve, and determining the mapping relationship by the projection position of the keyframe pixel coordinates to the global renal capsule curve and the inner normal distance; generating the cortical region probability field includes: determining the cortical region probability field based on the segmentation probability of the renal capsule curve; generating the boundary uncertainty field includes: determining the boundary uncertainty field based on the boundary positioning uncertainty of the renal capsule curve and the optimized residual of the pose graph.

[0020] In one embodiment, after determining the keyframe pose, kidney capsule curve segmentation is performed on each keyframe image. Kidney capsule curve segmentation can be achieved by using an image segmentation model to output a capsule probability map, and then thresholding and extracting the maximum connected component from the capsule probability map to obtain the capsule region. Next, edge tracking is used to obtain a set of kidney capsule curve points arranged in sequence. To improve boundary localization stability, sub-pixel correction can be performed on the capsule curve point set based on gradient extrema during edge tracking, thereby reducing jagged boundaries caused by speckle noise. The kidney capsule curve point sets of each keyframe are transformed to a unified kidney coordinate system according to the keyframe pose, and the curve point sets in overlapping areas are aligned and fused. The fusion method includes equidistant resampling of curves according to arc length and weighted aggregation of curve point sets from multiple frames within the same arc length neighborhood. The weights are jointly determined by the registration residual of the corresponding keyframe and the segmentation probability in the capsule probability map, resulting in a global kidney capsule curve. Arc length parameterization is applied to the global kidney capsule curve. The arc length is used as the horizontal coordinate in the cortical conformal coordinate system, and the inward normal distance of the global kidney capsule curve is used as the vertical coordinate. This establishes a mapping between the cortical conformal coordinate system and the keyframe pixel coordinates. The mapping relationship can be given by the following equation: ; ; in, These are the keyframe pixel coordinates after keyframe pose transformation. for Projection points on the global renal capsule curve Let be the unit normal pointing inwards from the projection point towards the interior of the kidney. Let be the arc length coordinates of the projection point along the global renal capsule curve. The horizontal coordinates are in the cortical conformal coordinate system. The vertical coordinates are defined in the cortical conformal coordinate system. Based on the segmentation probability of the kidney capsule curve and combined with the mapping relationship, the keyframe capsule probability map is mapped to the cortical conformal coordinate system along the projection direction, forming a cortical region probability field. This field characterizes the effective observation confidence of the cortical region in the cortical conformal coordinate system. The boundary uncertainty field is jointly determined by the pose map optimization residual and the boundary positioning uncertainty of the kidney capsule curve. The pose map optimization residual reflects the alignment reliability of the keyframe, while the boundary positioning uncertainty reflects the positioning fluctuation of the capsule edge under insufficient local contrast or speckle interference. After normalization, the two are fused to obtain the boundary uncertainty field, which is used to suppress the observation contribution of low-reliability regions in the subsequent parameter projection and fusion stages. Through the above processing, cortical region observations of different keyframes are made comparable and aligned in a unified cortical conformal coordinate system, providing a stable spatial basis for subsequent weighted fusion and fibrosis fractional field solution.

[0021] The acoustic attenuation coefficient and scattering statistical parameters are calculated within the sliding analysis window and the system response is corrected. The system is then projected onto the cortical conformal coordinate system using the mapping relationship and Wasserstein centroid fusion is performed by weighting the cortical region probability field and the boundary uncertainty field. The effective cortical thickness is detected and the fibrosis fractional field and consensus residual field are obtained through consensus optimization. The echo signals corresponding to each keyframe are used to construct a sliding analysis window along the depth and time directions. Within each sliding analysis window, the acoustic attenuation coefficient is estimated based on the attenuation trend of the echo signal power spectrum with depth, and scattering statistical parameters are obtained by fitting the statistical distribution of the echo envelope amplitude. Subsequently, the power spectrum is corrected using a system response correction function to reduce the impact of differences in system spectral response on the acoustic attenuation coefficient and scattering statistical parameters. The acoustic attenuation coefficient and scattering statistical parameters are projected onto the cortical conformal coordinate system through the mapping relationship between the cortical conformal coordinate system and the keyframe pixel coordinate system. Within the cortical conformal coordinate system, observation weights are generated based on the cortical region probability field and boundary uncertainty field, and after normalization, Wasserstein centroid fusion is performed to obtain the fused parameter field. The observation weights adopt: ; in, For observation weights, The values ​​are taken for the probability field of the cortical region. Values ​​are assigned to the boundary uncertainty field. A depth profile is constructed along the depth direction of the cortical conformal coordinate system for the fusion parameter field, and the location of the change point is determined based on piecewise fitting to obtain the effective cortical thickness. Consensus optimization is performed on the fusion parameter field within the effective cortical thickness range to solve the fibrillation fractional field, and a consensus residual field is generated based on the fitting error to characterize the degree of consistency between local observations and consensus solutions.

[0022] Calculating the acoustic attenuation coefficient within the sliding analysis window includes: estimating the acoustic attenuation coefficient based on the relationship between the echo signal power spectrum and depth at different depth locations; performing system response correction includes: correcting the power spectrum using the system response correction function and then estimating the acoustic attenuation coefficient; calculating scattering statistical parameters within the sliding analysis window includes: determining the scattering statistical parameters based on the statistical distribution fitting of the echo envelope amplitude within the sliding analysis window.

[0023] In one embodiment, after the keyframes are determined, a sliding analysis window is calculated for the echo signal corresponding to the renal cortex region for each keyframe. The sliding analysis window is configured as follows: within the same keyframe, the analysis window moves along the depth direction with a fixed step size, and several adjacent frames are superimposed in the temporal direction to improve stability, thereby reducing random fluctuations caused by speckle noise while ensuring spatial resolution. For each sliding analysis window, the echo signal within the window is extracted and the power spectrum is calculated. Then, the acoustic attenuation coefficient is estimated based on the relationship between the power spectrum and depth at different depth locations. The acoustic attenuation coefficient is estimated using a depth-difference method: the power spectra of two echo signals located at different depth locations are calculated separately and the logarithmic difference is taken to obtain the attenuation trend of the logarithmic power with depth. The acoustic attenuation coefficient is then determined based on this attenuation trend, so that the acoustic attenuation coefficient reflects the combined effect of absorption and scattering of ultrasound energy by the cortical tissue. To reduce the offset caused by differences in system spectral response, system response correction is performed before estimating the acoustic attenuation coefficient. System response correction includes frequency domain correction of the power spectrum using a system response correction function, making the amplitude response of the corrected power spectrum more consistent at different frequencies, thereby improving the comparability of acoustic attenuation coefficients for different subjects and equipment settings. The system response correction function is obtained in the same way as described above. Acquisition of the acoustic attenuation coefficient is performed after power spectrum correction to avoid misinterpreting equipment response errors as tissue attenuation characteristics.

[0024] In calculating the scattering statistical parameters, envelope detection is performed on the echo signal within each sliding analysis window to obtain the echo envelope amplitude sequence. The scattering statistical parameters are then determined based on the statistical distribution fitting of the echo envelope amplitude sequence. The statistical distribution fitting can employ a commonly used speckle statistical model to fit and solve for the mean, variance, and quantile characteristics of the echo envelope amplitude sequence, thereby obtaining scattering statistical parameters that characterize the variations in scatterer density and scattering intensity. To ensure the stability of the fitting results, outlier removal and robust normalization are performed on the echo envelope amplitude sequence before fitting. Outlier removal removes extreme amplitudes caused by strong reflection points or blood vessel walls, while robust normalization reduces the impact of overall gain differences on the fitting results. The resulting acoustic attenuation coefficient and scattering statistical parameters respectively characterize tissue energy attenuation and speckle statistical properties. These two parameters are then used together to construct a more sensitive and robust parameter representation of the degree of fibrosis when projected onto the cortical conformal coordinate system and fused.

[0025] The weighting of the cortical region probability field and the boundary uncertainty field includes: increasing the weight of observations within the cortical region based on the cortical region probability field and decreasing the weight of observations with large boundary uncertainties based on the boundary uncertainty field, and normalizing the observation weights; the Wasserstein barycentric fusion includes: constructing an empirical distribution of the acoustic attenuation coefficient and scattering statistical parameters projected onto the cortical conformal coordinate system and obtaining the Wasserstein barycentric distribution through iterative solution, and determining the fusion parameter field based on the Wasserstein barycentric distribution.

[0026] After projecting the acoustic attenuation coefficient and scattering statistics onto the cortical conformal coordinate system via a mapping relationship, multiple sets of observations from different keyframes are aggregated at each spatial location in the cortical conformal coordinate system. To suppress the interference of mismatches and segmentation instabilities near the capsule boundary on the observations and to increase the contribution of observations within the cortical region to the fusion results, observation weights are first constructed and normalized based on the cortical region probability field and boundary uncertainty field. The calculation of the observation weights uses: ; ; in, For the first The observation weight of each observation. For the first The probability field values ​​of the cortical region corresponding to each observation. For the first The boundary uncertainty field values ​​corresponding to each observation are as follows: These are the normalized observation weights. By constructing these weights, locations with higher probability field values ​​in the cortical region receive higher weights, while locations with higher boundary uncertainty field values ​​receive lower weights. This reduces the impact of boundary positioning fluctuations and pose alignment residuals on the fusion process.

[0027] When performing Wasserstein centroid fusion, empirical distributions are constructed for the sound attenuation coefficient and scattering statistical parameters, and the centroid distribution is obtained. The fusion parameter field is then determined from the centroid distribution. The empirical distribution is constructed as follows: at each spatial location in the cortical conformal coordinate system, the observations from each keyframe are weighted or statistically analyzed according to normalized observation weights to form a discrete histogram representation. The horizontal axis of the histogram represents the binning of the observations, and the vertical axis represents the cumulative weight value, normalized to probability. To facilitate implementation by those skilled in the art, the empirical distribution can adopt a histogram with fixed bins, and a globally uniform range and step size can be used for the bin boundaries to ensure comparability of fusion results between different spatial locations.

[0028] Wasserstein centroid fusion is used to obtain more stable fusion results in the presence of outliers or multimodal distributions. Its computational objective can be expressed as: in, It is a center-of-gravity distribution. For the first The empirical distribution corresponding to each observation For distribution With distribution The Wasserstein distance between them. In actual implementation, the centroid distribution is obtained iteratively. The iterative process includes initializing the centroid distribution as a weighted average histogram, and then repeatedly performing distribution alignment and centroid update until convergence. Distribution alignment is used to calculate the mass transfer relationship from the centroid distribution to each empirical distribution on the histogram bins. Centroid update is used to synthesize the aligned shapes of each empirical distribution according to the normalized observation weights to obtain a new centroid distribution. The iteration termination condition can be that the difference between two adjacent iterations of the centroid distribution is less than a preset threshold or the preset number of iterations is reached.

[0029] After obtaining the centroid distributions of the acoustic attenuation coefficient and the scattering statistics, the weighted mean or median of the centroid distributions is taken as the fusion output at the corresponding location to form a fusion parameter field. The fusion parameter field includes at least the fusion acoustic attenuation coefficient field and the fusion scattering statistics field, and is used as input in subsequent cortical effective thickness detection and consensus optimization. This ensures that different keyframes and different local observations obtain consistent and interference-resistant parameter representations under a unified cortical conformal coordinate system, thereby improving the stability and repeatability of the fibrillation fractional field.

[0030] Detecting the effective cortical thickness includes: constructing a depth profile of the fusion parameter field along the depth direction of the cortical conformal coordinate system, and determining the location of the change point of the depth profile based on piecewise model fitting, and determining the location of the change point as the effective cortical thickness; obtaining the fibrillation fraction field and consensus residual field through consensus optimization includes: constructing a monotonic mapping relationship between the fibrillation fraction field and the fusion parameter field, solving the fibrillation fraction field through iterative optimization, and determining the consensus residual field based on the fitting error of the monotonic mapping relationship.

[0031] In one embodiment, effective cortical thickness detection is performed based on a fusion parameter field within the cortical conformal coordinate system. Specifically, along the depth direction of the cortical conformal coordinate system, depth sequences of the fusion acoustic attenuation coefficient field and the fusion scattering statistical parameter field are extracted for each arc length position. The depth sequences are then de-isolated and smoothed in one dimension to suppress local spikes caused by speckle noise. The smoothed depth sequence is used as a depth profile, and a piecewise model with a single change point is used for fitting. The location of the change point is determined by traversing candidate change points and comparing the fitting residuals. The change point is used to characterize the transition from the cortical region to the medullary region or the low effective echo region, thereby determining the location of the change point as the effective cortical thickness. The piecewise model can be solved using the following residual minimum criterion: ; in, For the location of the change point, For depth profile at the first The value at each depth sampling point The piecewise fitted values ​​before the point of change. These are the piecewise fitted values ​​after the point of change. For the candidate change point locations, This is the depth sampling sequence number. Using this effective cortical thickness to limit the subsequent solution range can reduce the interference of medullary structures and areas with excessive echo attenuation on fibrosis assessment.

[0032] After obtaining the effective cortical thickness, the fibrosis fractional field and consensus residual field are solved only within the depth range covered by the effective cortical thickness. First, the fused acoustic attenuation coefficient field and the fused scattering statistical parameter field are robustly normalized within the cortical range, and a fused characterization quantity is formed according to a preset rule, ensuring that the fused characterization quantity maintains a monotonic response to fibrosis-related tissue changes. Then, a monotonic mapping relationship between the fibrosis fractional field and the fused characterization quantity is constructed. This monotonic mapping relationship is implemented using a piecewise linear monotonic function, and the fibrosis fractional field is solved through iterative optimization. Iterative optimization includes alternating execution of mapping updates and fractional field updates: in the mapping update stage, while keeping the fibrosis fractional field unchanged, a monotonic regression is performed on the monotonic mapping relationship to satisfy the monotonic constraint; in the fractional field update stage, while keeping the monotonic mapping relationship unchanged, the monotonic mapping relationship is applied to the fused characterization quantity to obtain initial fractional values, and the initial fractional values ​​are smoothly updated with spatial consistency constraints to obtain a fibrosis fractional field that is continuous within the cortical conformal coordinate system and does not excessively diffuse at the boundaries. The consensus residual field is determined by the fitting error between the fibrosis fractional field and the monotonic mapping relationship, and can be calculated using the following formula: ; in, For the consensus residual field values, For the fractional field of fiberization, The output of the mapping relationship between the monotonic mapping relation and the fused representation quantity is shown. The consensus residual field serves as a fusion characterization measure, indicating the degree of consistency between local observations and the consensus solution, thus providing a basis for subsequent confidence assessment and backtracking recalculation.

[0033] The non-consistent total score is generated based on the fiber fractional field, and the confidence interval and confidence level are output through the conformal prediction calibration method. Then, a rollback recalculation instruction or a re-sampling instruction is generated.

[0034] Along the depth direction of the cortical conformal coordinate system, depth profiles are extracted for each arc length position of the fusion parameter field. Median filtering and one-dimensional smoothing are applied to the depth profiles to suppress isolated noise. For each candidate depth position, the depth profile is divided into a capsule-side segment and a medullary-side segment. The sum of the fitting residuals for the two segments is calculated using least-squares linear fitting. The candidate depth position with the smallest sum of residuals is selected as the change point position, and this change point position is determined as the effective cortical thickness to limit the solution range of the subsequent fibrosis fraction field. Within the effective cortical thickness, robust normalization is performed on the fusion parameter field, and a fusion representation quantity is generated. A piecewise linear monotonic mapping relationship from the fusion representation quantity to the fibrosis fraction field is constructed. Iterative mapping updates and score updates are performed. The mapping update uses monotonic regression to ensure that the mapping relationship satisfies the monotonic constraint, and the score update uses neighborhood consistency constraints to smooth the scores and maintain consistency with the mapping of the fusion representation quantity, until the score change between two adjacent iterations is less than a preset threshold. The consensus residual field is determined by the following formula: .

[0035] The calculation of the non-consistency total score based on the fibrosis fractional field includes: generating predicted acoustic attenuation coefficients and predicted scattering statistical parameters based on the fibrosis fractional field and calculating predicted residual statistics with the fused parameter field; calculating the viewpoint consistency index based on the fibrosis fractional fields corresponding to longitudinal and transverse renal ultrasound image data; calculating the time stability index based on the fibrosis fractional field obtained by dividing the keyframe set into a front keyframe set and a back keyframe set according to time sequence; calculating the alignment consistency index based on the closure constraint relationship residuals and boundary uncertainty field of the pose graph; and fusing the predicted residual statistics, viewpoint consistency index, time stability index, and alignment consistency index to obtain the non-consistency total score. The output of confidence intervals and confidence levels through conformal prediction calibration includes: determining calibration rules based on the non-consistency total score distribution of calibration samples and outputting confidence intervals and confidence levels accordingly.

[0036] In one embodiment, after obtaining the fiber fractional field and the fusion parameter field, a predicted acoustic attenuation coefficient and a predicted scattering statistical parameter are constructed based on the fiber fractional field, and a consistency check is performed with the fusion parameter field to generate predicted residual statistics. Specifically, using the monotonic mapping relationship determined in the consensus optimization phase, the fiber fractional field is reverse-mapped to obtain the predicted acoustic attenuation coefficient field and the predicted scattering statistical parameter field; the acoustic attenuation residual field is obtained by point-by-point subtraction of the predicted acoustic attenuation coefficient field and the fusion acoustic attenuation coefficient field, and the scattering residual field is obtained by point-by-point subtraction of the predicted scattering statistical parameter field and the fusion scattering statistical parameter field. The median absolute deviation and upper quantile of the residual field are calculated as predicted residual statistics to characterize the degree of explanation of the fusion parameter field by the fiber fractional field.

[0037] To characterize the inconsistencies introduced by differences in scanning perspective, the aforementioned process was performed on longitudinal and transverse renal ultrasound images to obtain longitudinal and transverse fibrosis fractional fields, respectively. These fields were then aligned to the same arc length coordinate range using a cortical conformal coordinate system. Within the overlapping range, the absolute difference statistics and correlation coefficients of the two fields were calculated, and these statistics were used together to determine the perspective consistency index. To characterize temporal drift, the keyframe set was divided into a pre-set and a post-set keyframes according to chronological order, yielding pre-set and post-set fibrosis fractional fields. The difference statistics between these two fields were calculated within the same cortical conformal coordinate system as a temporal stability index. To characterize alignment error, the residuals of closure constraints in the pose diagram were summarized within the arc length range with higher probability fields in the cortical region, and combined with the statistics of the boundary uncertainty field to determine the alignment consistency index. This ensures that the alignment consistency index simultaneously reflects the geometric closure of the closure constraints and the instability of the capsule boundary positioning.

[0038] The prediction residual statistics, perspective consistency index, time stability index, and alignment consistency index are normalized and then fused to obtain the total inconsistency score: ; in, For inconsistent total scores, To predict residual statistics, As a consistency indicator of perspective, For time stability indicators, To align consistency metrics, To predict the residual statistical fusion coefficient, The perspective consistency fusion coefficient, The time-stability fusion coefficient, This represents the alignment consistency fusion coefficient.

[0039] When outputting confidence intervals and confidence levels, a conformal prediction calibration method is used to establish a mapping rule from inconsistency total scores to confidence outputs. During the calibration phase, calibration samples are acquired, and the inconsistency total score is calculated for each sample. Simultaneously, the difference in the total fiber score from repeated scans of the same subject is calculated. A monotonic correspondence is established between the inconsistency total score and the difference in the total fiber score, and the quantile threshold set is determined from the distribution of the inconsistency total score in the calibration samples. In the application phase, after calculating the inconsistency total score for the subject to be evaluated, the confidence level is determined based on the quantile threshold set, and the half-width of the confidence interval is determined from the monotonic correspondence. The confidence interval for the total fiber score is then output. ; in, Let be the confidence interval. For total fibrosis, This represents the half-width of the confidence interval determined by the inconsistency total score. Through the above inconsistency total score and conformal prediction calibration, the confidence output is adaptively adjusted as data consistency changes, thereby providing an interpretable basis for rollback recalculation and re-sampling decisions.

[0040] The rollback and recalculation instructions include: identifying loop constraint relationships in the pose graph whose residuals satisfy the residual rules as invalid loop constraint relationships and reducing or eliminating invalid loop constraint relationships; re-optimizing the pose graph to update the mapping relationship between the cortical conformal coordinate system and the keyframe pixel coordinates; and re-executing Wasserstein centroid fusion, detecting effective cortical thickness, and consensus optimization. The re-acquisition instructions include: determining the probe incident angle adjustment scheme and the scanning path adjustment scheme based on the non-consistent total score; and re-acquiring the renal ultrasound image data.

[0041] In one embodiment, after outputting the inconsistency total score, confidence interval, and confidence level, the system enters a closed-loop processing phase to correct unreliable results caused by alignment anomalies or insufficient scanning conditions, or to guide re-sampling. The closed-loop processing includes two paths: generating rollback recalculation instructions and generating re-sampling instructions. Both are triggered by the inconsistency total score and its decomposition items, and can be executed separately depending on the triggering reason.

[0042] When generating the rollback and recalculation instruction, the residuals of each loop constraint in the pose graph are first calculated. These residuals characterize whether geometric closure conflicts occur after loop constraints are formed between different keyframes at the same spatial location. The residual rules are statistically obtained during the offline calibration phase. Specifically, the distribution of loop constraint residuals is statistically analyzed in multiple stable scan samples, and residual thresholds and abnormality judgment intervals are determined accordingly. If a loop constraint residual satisfies the residual rules, it is identified as an invalid loop constraint and subjected to weight reduction or elimination. Weight reduction involves lowering the constraint weight of invalid loop constraints to a preset proportion of the constraint weights of constraints in adjacent keyframes, retaining loop candidates but reducing their impact on the overall pose graph optimization. Elimination involves removing invalid loop constraints from the pose graph so they do not participate in subsequent optimization. After weight reduction or elimination, the pose graph is re-optimized to update the keyframe poses. Based on the updated keyframe pose, the alignment and fusion of the kidney capsule curves are re-performed to obtain an updated global kidney capsule curve. The mapping relationship between the cortical conformal coordinate system and the keyframe pixel coordinates is then updated using the global kidney capsule curve. Subsequently, the acoustic attenuation coefficient and scattering statistics are re-projected onto the cortical conformal coordinate system using the updated mapping relationship, and Wasserstein centroid fusion is re-performed to obtain an updated fusion parameter field. Based on the updated fusion parameter field, the effective cortical thickness is re-detected, and consensus optimization is re-performed within the effective cortical thickness range to obtain an updated fibrosis fraction field and consensus residual field. Through this backtracking and recalculation process, abnormal closure conflicts in closure constraints can be removed or weakened from the pose graph, reducing the systematic shift in the mapping relationship. This results in a more stable spatial continuity of the fusion parameter field and the fibrosis fraction field within the cortical conformal coordinate system, thereby reducing the overall inconsistency score.

[0043] When generating a re-acquisition command, the main causes of insufficient scan quality are determined based on the source of the total inconsistency score and its decomposition items, and an executable probe incident angle adjustment scheme and scan path adjustment scheme are formed. The probe incident angle adjustment scheme is determined by locating the arc-length interval where the predicted residual statistical anomalies are concentrated in the cortical conformal coordinate system, and combining this with the continuity evaluation of the capsule boundary in the original image, determining the direction in which the incident angle needs to be changed, so that the renal capsule curve region obtains clearer boundary echoes and a more stable capsule segmentation probability. The scan path adjustment scheme is determined by locating the arc-length interval where the perspective consistency index or time stability index is abnormal in the cortical conformal coordinate system, specifying the scan segments to be repeatedly covered and the scan segments to be enhanced, and specifying the scan speed and coverage order, so that the keyframe set forms more sufficient overlapping observations within this interval and reduces inter-frame motion inconsistency. After the re-acquisition command is output, the operator re-acquires renal ultrasound image data according to the probe incident angle adjustment scheme and the scan path adjustment scheme, and executes the above process again to obtain a fibrosis fractional field, confidence interval, and confidence level that meet the consistency requirements. By introducing backtracking recalculation and resampling instructions into closed-loop processing, unreliable outputs caused by alignment anomalies and scanning differences can be self-corrected or guided to resampling without adding extra hardware, thereby improving the repeatability and interpretability of quantitative assessment results of fibrosis.

[0044] In one specific embodiment, the following set of exemplary embodiments is provided to fully demonstrate the execution chain, data calculation, and output format. For ease of reproduction, the example data is constructed using a reproducible simulation / de-identification method, intended to illustrate the implementation and effect presentation, and does not constitute a clinical conclusion.

[0045] The overall flow of this embodiment is shown in the appendix. Figure 1 A sequence of 120 grayscale ultrasound images from the same kidney scan was acquired at a frame rate of 25 frames per second, with each frame measuring 512×512 pixels. The corresponding echo signal amplitude spectrum or radio frequency data frequency domain representation was simultaneously saved for acoustic parameter estimation. In the standardization process, grayscale normalization and spatial scale unification were first performed on each frame: after truncating speckle spikes by pixel grayscale quantiles, the grayscale was mapped to 0 to 1; the horizontal and vertical scales were unified to 0.12 mm per pixel using the device-calibrated pixel spacing. Taking one frame as an example, the mean was 84 and the standard deviation was 22 before normalization, while the mean was 0.48 and the standard deviation was 0.13 after normalization, thus reducing cross-frame offset caused by different gains and dynamic ranges.

[0046] The system response correction function is obtained through a reference echo signal: Under the same equipment settings, a reference echo signal is acquired from a uniform phantom or a reference area at a fixed depth. The amplitude spectrum of the reference echo signal is calculated and normalized to form the system response correction function. Subsequently, frequency domain correction is performed on the power spectrum of the target echo signal to ensure uniformity of the system gain across different frequency channels. To ensure operability, this embodiment uses 64 equally spaced frequency points to form a discrete representation of the system response correction function, and linear interpolation is used between these frequency points.

[0047] The optical flow field is obtained by considering the grayscale consistency and smoothing constraints of adjacent frames: grayscale residuals are established between two adjacent frames, and a smoothing constraint term is added to the spatial gradient of the optical flow field. A multi-scale pyramid iterative solution is then used to obtain the pixel-by-pixel displacement. The optical flow field is used to characterize probe micro-motion and tissue deformation. In this embodiment, the average inter-frame optical flow amplitude is 1.6 pixels, and the median is 1.2 pixels.

[0048] Subsequently, a set of keyframes was selected based on the optical flow field: the consistency of optical flow amplitude and direction was used to form an inter-frame motion consistency index, and keyframes were screened in conjunction with registration residual rules. In this embodiment, 16 keyframes were selected from 120 frames, with keyframe indices of 0, 7, 14, 22, 29, 36, 44, 52, 60, 68, 76, 84, 92, 100, 108, and 116. Image registration was performed between keyframes, and a pose graph was constructed: adjacent keyframes established adjacency constraints, and loop closure constraints were established when there was visual overlap. Graph optimization was used to iteratively update the pose of keyframes. In this embodiment, the pose graph contains 15 adjacency constraints and 2 loop closure constraints. The average registration residual before pose graph optimization was 1.9 pixels, which was reduced to 0.7 pixels after optimization.

[0049] The renal capsule curve is segmented from keyframes and fused into a global renal capsule curve based on keyframe pose alignment. Arc length parameterization is performed on the global renal capsule curve to generate a cortical conformal coordinate system: the arc length normalized coordinates range from 0 to 1, and the inward normal depth ranges from 0 to 15 mm. The mapping relationship is determined by the projection position from the keyframe pixel coordinates to the global renal capsule curve and the inward normal distance. The cortical region probability field is obtained by mapping and aggregating the pixel-level probabilities of the capsule segmentation; the boundary uncertainty field is jointly determined by the pose map optimization residual and the capsule boundary positioning uncertainty, used for subsequent weighted suppression of unreliable observations.

[0050] Acoustic parameter estimation and fusion are performed within a cortical conformal coordinate system. This embodiment uses a sliding analysis window with dimensions of 5 mm horizontally and 8 mm vertically, with a step size of 2 mm. The acoustic attenuation coefficient is estimated based on the relationship between the echo signal power spectrum and depth, and the power spectrum is first corrected using a system response correction function. Taking a representative analysis window as an example, the integral values ​​of the corrected power spectrum at depths of 10, 20, 30, and 40 mm are 8.9, 5.2, 3.1, and 1.8, respectively. A logarithmic domain linear fitting is used to estimate the attenuation slope. The core calculation expression is as follows: ; ; in, For depth The power spectrum integral value at that point; Depth, in millimeters; The intercept is the fitting angle. The slope of the logarithmic domain decay; This is an estimate of the sound attenuation coefficient, expressed in decibels per centimeter. This window estimate yields... .

[0051] The scattering statistics are determined based on a fitting of the statistical distribution of the echo envelope amplitude. In this embodiment, the Nakagami shape parameter is used as one of the scattering statistics, calculated using the mean and variance of the squared envelope amplitude. ; in, This represents the echo envelope amplitude. These are scattering statistical parameters. The statistics within this window are obtained... , ,therefore .

[0052] After projecting the sound attenuation coefficient and scattering statistics onto the cortical conformal coordinate system through a mapping relationship, the observation weights within the cortical region are increased based on the cortical region probability field, and the observation weights at locations with higher uncertainty are decreased based on the boundary uncertainty field. The observation weights are then normalized to obtain the final observation weights. Subsequently, an empirical distribution is constructed for each arc-length coordinate position, and Wasserstein barycentric fusion is performed to obtain the fused parameter field. To reduce the terminology burden, this embodiment implements Wasserstein barycentric fusion as a weighted minimization of the optimal transmission distance of the empirical distribution: ; in, For the first The empirical distribution of the observations; To normalize the observation weights; The second-order Wasserstein distance; This shows the centroid distribution obtained through fusion. See the appendix for the fusion results. Figure 3 The image shows the two-dimensional distribution of the fusion parameter field within a conformal cortical coordinate system. The horizontal axis represents the arc-length normalized coordinate, ranging from 0 to 1, indicating the position along the arc-length direction of the renal capsule; the vertical axis represents the depth (mm), with a scale range of approximately 0 to 12, indicating the depth of entry into the cortex from the capsule inwards. The image uses color coding to encode the magnitude of the fusion parameters, with the color bar on the right indicating the numerical range (approximately 0.16–0.26), and colors ranging from cool to warm indicating that the fusion parameters are from low to high. A distinct high-value "hotspot" elliptical region is visible in the image, around an arc-length normalized coordinate of approximately 0.6 and a depth of approximately 4–6 mm, representing a stronger acoustic / scattering composite characterization after fusion at this location. The values ​​gradually decrease in the surrounding area, exhibiting a smooth gradient distribution transitioning from the hotspot outwards. This distribution illustrates the local enhancement and spatial continuity of the fusion result in the arc-length-depth two-dimensional space and provides an input field for thickness detection and fibrosis fraction mapping.

[0053] The effective cortical thickness is determined by the depth profile change points of the fused parameter field: a profile is taken along the inner normal depth direction, and the change point location is selected as the effective cortical thickness using the principle of minimizing the residual of the two-segment piecewise linear model. In this embodiment, the depth corresponding to the change point at the arc-length normalized coordinate of 0.64 is 6.21 mm. (See Appendix) Figure 2 This diagram displays a depth profile of the fusion parameters along the depth direction to determine the location of change points corresponding to the effective cortical thickness. The horizontal axis represents depth (millimeters), with a scale range of approximately 0–12; the vertical axis represents the depth profile of the fusion parameters (dimensionless). The diagram uses broken lines and dots to show the fusion parameter values ​​at each depth sampling location, showing an overall increasing trend with depth: the increase is slower in the shallow layers (approximately 0–6 mm), and accelerates significantly in the middle and deeper layers. A vertical reference line is drawn at approximately 6 mm to mark the location of change points; this location corresponds to the estimated depth of the "effective cortical thickness," i.e., the boundary point where the fusion parameters transition from a gradual to a steep change with depth. The shape of this curve supports the logic for determining the change point with the minimum piecewise linear residual and provides a visual basis for subsequent thickness output.

[0054] The fiber fraction field is obtained through consensus optimization: a monotonic mapping relationship is established between the fiber fraction and the fusion parameter field, ensuring that the fiber fraction does not decrease with increasing fusion parameters. The fitting error is used to form a consensus residual field to characterize local inconsistencies. See the appendix for the fiber fraction field. Figure 4 This displays the two-dimensional distribution of the fractional fibrillation field within the same cortical conformal coordinate system. The horizontal axis represents arc length normalized coordinates (0–1), and the vertical axis represents depth (mm) (approximately 0–12). The meanings of the coordinates are the same as those in the appendix. Figure 3Consistent. The image uses color coding to encode the fiber fraction size, with the color bar on the right indicating the fraction range (approximately 0.40–0.60, warm colors indicating higher fractions and cool colors indicating lower fractions). A high-resolution "hotspot" region is also observed in the image around 0.6 arc-length normalized coordinates and at a depth of approximately 3–6 mm, its spatial location being consistent with the surrounding area. Figure 3 The high-value regions of the fusion parameters correspond to each other, reflecting that the consensus optimization gave a higher fibrillation score in that local area; the score gradually decreases as you move away from this area, and the overall distribution remains continuous and smooth. This attached figure is used to visually present the joint distribution characteristics of the fibrillation score in the arc length and depth directions, supporting the distinction between "local high-risk areas" and "background stable areas," and facilitating spatial comparison and interpretation with thickness detection results. In this embodiment, the statistical mean of the fibrillation score field is 0.26, with a local peak value reaching 0.74, and the mean of the consensus residual field is 0.12, indicating that the main differences are concentrated around the arc length normalized coordinates of 0.55 to 0.70.

[0055] The overall inconsistency score is obtained by fusing four types of consistency indices: the prediction residual statistics take the mean of the consensus residual field (0.12); the viewpoint consistency index is obtained from the difference in fiber fraction fields between the longitudinal and transverse scanning directions (0.18); the temporal stability index is obtained from the difference in fiber fraction fields between the preceding and following keyframes (0.14); and the alignment consistency index is obtained from the combination of the loop closure constraint residual and the boundary uncertainty field (0.21). A weighted summation is used to obtain the overall inconsistency score. ; in, The total score is inconsistent. For the first Category index weights; For the first Class indicator value. This embodiment takes... ,get Subsequently, a conformal prediction calibration method is used to map the inconsistency total score to a confidence interval and a confidence level. In this embodiment, the output confidence level is high, and the confidence interval is 0.20 to 0.32. Since the inconsistency total score did not trigger the resampling threshold, the system does not generate a resampling instruction; at the same time, the residual of the pose graph closure constraint relationship does not exceed the invalid threshold, so the system does not generate a backtracking recalculation instruction. Thus, this embodiment achieves stable output within a single scan, and the output reliability is quantitatively interpreted through the inconsistency total score and calibration results, reducing the result fluctuations caused by differences in operator experience.

[0056] The core code snippet that can be directly used to reproduce the computational chain and output format of this embodiment is as follows: import numpy as np def estimate_attenuation(depth_mm, power): y = np.log(np.maximum(power, 1e-12)) A = np.vstack([np.ones_like(depth_mm), depth_mm]).T beta, *_ = np.linalg.lstsq(A, y, rcond=None) k_hat = -beta[1] return float(k_hat * 10 / np.log(10)) # dB / cm (example conversion) def change_point_depth(depth_mm, y): x = np.arange(len(y)) def sse(cp): b1 = np.polyfit(x[:cp+1], y[:cp+1], 1) b2 = np.polyfit(x[cp:], y[cp:], 1) return float(np.sum((y[:cp+1]-np.polyval(b1, x[:cp+1]))**2) + np.sum((y[cp:]-np.polyval(b2, x[cp:]))**2)) cp_hat = min(range(4, len(y)-4), key=sse) return float(depth_mm[cp_hat]) like Figure 5 As shown, a quantitative assessment system for fibrosis based on renal ultrasound imaging is used to implement the quantitative assessment method for fibrosis based on renal ultrasound imaging. The system includes: The acquisition and preprocessing module is used to acquire and standardize renal ultrasound image data, and estimate the system response correction function and optical flow field. The module includes an ultrasound probe, a transmit / receive front-end circuit, a beamforming and image reconstruction circuit, a frame acquisition interface, and a computational processing unit. The ultrasound probe and transmit / receive front-end circuit acquire renal echo signals. The transmit / receive front-end circuit includes transmit pulse drive, receive protection switching, low-noise amplification, variable gain amplification, and analog filtering to form an analog signal that satisfies the imaging link. The beamforming and image reconstruction circuit performs delay summation, envelope detection, and grayscale imaging on the multi-channel echo signals, outputting renal ultrasound image data. The frame acquisition interface and frame buffer storage perform timestamp marking and caching on the image frame sequence. The computational processing unit performs standardization processing on the renal ultrasound image data, including grayscale normalization and spatial scale unification, and generates a system response correction function based on the reference echo signal and corresponding frequency domain representation. Simultaneously, it calculates the optical flow field based on adjacent frame image data, providing input for subsequent keyframe selection and mapping.

[0057] The mapping module is used to select a set of keyframes based on the optical flow field, construct and optimize the pose map based on the keyframe image registration, and fuse the mapping relationship between the kidney capsule cortical conformal coordinate system and the keyframe pixel coordinates to generate the cortical region probability field and boundary uncertainty field. The mapping module includes a keyframe management unit, an image registration acceleration unit, a pose map solving unit, a kidney capsule segmentation unit, and a mapping generation unit. The keyframe management unit receives the optical flow field and performs statistical analysis on the motion consistency between frames. It then filters the keyframe set based on the image registration residual rules and completes keyframe caching and indexing. The image registration acceleration unit performs registration calculations on the keyframe images and outputs the registration constraints between keyframes. The pose graph solving unit establishes the constraint relationships and closure constraints between adjacent keyframes and performs graph optimization to obtain the keyframe poses. It also outputs the pose graph optimization residuals for uncertainty measurement. The kidney capsule segmentation unit outputs the kidney capsule curve and its segmentation probability in the keyframes. The mapping generation unit obtains the global kidney capsule curve based on the keyframe pose alignment and fusion. It performs arc length parameterization on the global kidney capsule curve to generate a cortical conformal coordinate system. Based on the projection position and inner normal distance from the keyframe pixel coordinates to the global kidney capsule curve, it determines the mapping relationship between the cortical conformal coordinate system and the keyframe pixel coordinates. It also generates the cortical region probability field and boundary uncertainty field and outputs them to shared storage.

[0058] The fusion inversion module is used to calculate the acoustic attenuation coefficient and scattering statistics within the sliding analysis window and perform system response correction. It projects the system onto the cortical conformal coordinate system using the mapping relationship and performs Wasserstein centroid fusion by weighting the cortical region probability field and boundary uncertainty field. It detects the effective cortical thickness and obtains the fibrous fractional field and consensus residual field through consensus optimization. The fusion inversion module includes a sliding analysis window calculation unit, a system response correction execution unit, a coordinate projection unit, a weighted fusion unit, and an inversion solution unit. The sliding analysis window calculation unit extracts echo statistics from keyframe image data within the sliding analysis window and calculates the acoustic attenuation coefficient and scattering statistical parameters. The system response correction execution unit calls the system response correction function to correct relevant frequency domain features or statistical features to reduce the impact of imaging link differences on parameter estimation. The coordinate projection unit projects the acoustic attenuation coefficient and scattering statistical parameters to the cortical conformal coordinate system based on the mapping relationship between the cortical conformal coordinate system and the keyframe pixel coordinate system. The weighted fusion unit generates observation weights based on the cortical region probability field and boundary uncertainty field and performs normalization, then performs Wasserstein centroid fusion on the projected parameter distribution to form a fused parameter field. The inversion solution unit detects the effective cortical thickness on the fused parameter field and outputs a fiber fraction field and a consensus residual field through consensus optimization, where the consensus residual field characterizes the spatial distribution of the consistency between the inversion results and observations.

[0059] The calibration decision module is used to generate an inconsistent total score based on the fiber fraction field and output the confidence interval and confidence level through the conformal prediction calibration method, and generate a rollback recalculation instruction or a re-sampling instruction. The calibration decision module includes a consistency assessment unit, a conformal prediction calibration unit, an instruction generation and human-computer interaction unit, and a result storage unit. The consistency assessment unit reads information related to the fiberized fraction field, consensus residual field, and pose graph optimization residual, generates the input for calculating the non-consistent total score, and completes the total score generation. The conformal prediction calibration unit executes the conformal prediction calibration method based on the non-consistent total score distribution of the calibration samples, outputs confidence intervals and confidence levels, and provides a traceable reliability measure for the output results. The instruction generation and human-computer interaction unit generates rollback recalculation instructions or re-acquisition instructions based on the confidence level and the non-consistent total score. The rollback recalculation instructions trigger the update of the pose graph and mapping relationship and re-execute the fusion inversion process. The re-acquisition instructions output the probe incident angle adjustment scheme and the scanning path adjustment scheme and drive the display terminal to prompt the operator. The result storage unit saves the non-consistent total score, confidence interval, confidence level, and instruction trigger records in a versioned manner to support subsequent review and quality control.

[0060] The above are merely embodiments of the present invention and are not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principle of the present invention should be included within the scope of the claims of the present invention.

Claims

1. A method for quantitative assessment of fibrosis based on renal ultrasound imaging, characterized in that, Includes the following steps: Renal ultrasound imaging data were acquired and standardized to estimate the system response correction function and optical flow field. The keyframe set is selected based on the optical flow field, the pose map is constructed and optimized based on the keyframe image registration, and the mapping relationship between the kidney capsule cortical conformal coordinate system and the keyframe pixel coordinates is fused to generate the cortical region probability field and boundary uncertainty field. The acoustic attenuation coefficient and scattering statistical parameters are calculated within the sliding analysis window and the system response is corrected. The system is then projected onto the cortical conformal coordinate system using the mapping relationship and Wasserstein centroid fusion is performed by weighting the cortical region probability field and the boundary uncertainty field. The effective cortical thickness is detected and the fibrosis fractional field and consensus residual field are obtained through consensus optimization. The non-consistent total score is generated based on the fiber fractional field, and the confidence interval and confidence level are output through the conformal prediction calibration method. Then, a rollback recalculation instruction or a re-sampling instruction is generated.

2. The method according to claim 1, characterized in that, The standardization process includes: performing grayscale normalization and spatial scale unification on the renal ultrasound image data; estimating the system response correction function includes: acquiring the reference echo signal and determining the system response correction function based on the spectral characteristics of the reference echo signal; estimating the optical flow field includes: calculating the optical flow field based on the grayscale consistency and smoothing constraints of adjacent frames.

3. The method according to claim 1, characterized in that, Selecting a set of keyframes based on optical flow fields includes: determining the motion consistency between frames based on the amplitude and direction consistency of the optical flow field, and filtering the set of keyframes based on the registration residual rules of keyframe image registration; constructing and optimizing a pose graph based on keyframe image registration includes: establishing constraint relationships and closure constraint relationships between adjacent keyframes, and obtaining the pose of keyframes through graph optimization.

4. The method according to claim 1, characterized in that, The mapping relationship between the cortical conformal coordinate system generated by fusing the renal capsule and the keyframe pixel coordinates includes: segmenting the renal capsule curve in the keyframe, obtaining the global renal capsule curve based on the keyframe pose alignment and fusion, generating the cortical conformal coordinate system by parameterizing the arc length of the global renal capsule curve, and determining the mapping relationship by the projection position of the keyframe pixel coordinates to the global renal capsule curve and the inner normal distance; generating the cortical region probability field includes: determining the cortical region probability field based on the segmentation probability of the renal capsule curve; generating the boundary uncertainty field includes: determining the boundary uncertainty field based on the boundary positioning uncertainty of the renal capsule curve and the optimized residual of the pose graph.

5. The method according to claim 1, characterized in that, Calculating the acoustic attenuation coefficient within the sliding analysis window includes: estimating the acoustic attenuation coefficient based on the relationship between the echo signal power spectrum and depth at different depth locations; performing system response correction includes: correcting the power spectrum using the system response correction function and then estimating the acoustic attenuation coefficient; calculating scattering statistical parameters within the sliding analysis window includes: determining the scattering statistical parameters based on the statistical distribution fitting of the echo envelope amplitude within the sliding analysis window.

6. The method according to claim 1, characterized in that, The weighting of the cortical region probability field and the boundary uncertainty field includes: increasing the weight of observations within the cortical region based on the cortical region probability field and decreasing the weight of observations with large boundary uncertainties based on the boundary uncertainty field, and normalizing the observation weights; the Wasserstein barycentric fusion includes: constructing an empirical distribution of the acoustic attenuation coefficient and scattering statistical parameters projected onto the cortical conformal coordinate system and obtaining the Wasserstein barycentric distribution through iterative solution, and determining the fusion parameter field based on the Wasserstein barycentric distribution.

7. The method according to claim 1, characterized in that, Detecting the effective cortical thickness includes: constructing a depth profile of the fusion parameter field along the depth direction of the cortical conformal coordinate system, and determining the location of the change point of the depth profile based on piecewise model fitting, and determining the location of the change point as the effective cortical thickness; obtaining the fibrillation fraction field and consensus residual field through consensus optimization includes: constructing a monotonic mapping relationship between the fibrillation fraction field and the fusion parameter field, solving the fibrillation fraction field through iterative optimization, and determining the consensus residual field based on the fitting error of the monotonic mapping relationship.

8. The method according to claim 1, characterized in that, The calculation of the non-consistency total score based on the fibrosis fractional field includes: generating predicted acoustic attenuation coefficients and predicted scattering statistical parameters based on the fibrosis fractional field and calculating predicted residual statistics with the fused parameter field; calculating the viewpoint consistency index based on the fibrosis fractional fields corresponding to longitudinal and transverse renal ultrasound image data; calculating the time stability index based on the fibrosis fractional field obtained by dividing the keyframe set into a front keyframe set and a back keyframe set according to time sequence; calculating the alignment consistency index based on the closure constraint relationship residuals and boundary uncertainty field of the pose graph; and fusing the predicted residual statistics, viewpoint consistency index, time stability index, and alignment consistency index to obtain the non-consistency total score. The output of confidence intervals and confidence levels through conformal prediction calibration includes: determining calibration rules based on the non-consistency total score distribution of calibration samples and outputting confidence intervals and confidence levels accordingly.

9. The method according to claim 1, characterized in that, The rollback and recalculation instructions include: identifying loop constraint relationships in the pose graph whose residuals satisfy the residual rules as invalid loop constraint relationships and reducing or eliminating invalid loop constraint relationships; re-optimizing the pose graph to update the mapping relationship between the cortical conformal coordinate system and the keyframe pixel coordinates; and re-executing Wasserstein centroid fusion, detecting effective cortical thickness, and consensus optimization. The re-acquisition instructions include: determining the probe incident angle adjustment scheme and the scanning path adjustment scheme based on the non-consistent total score; and re-acquiring the renal ultrasound image data.

10. A quantitative assessment system for fibrosis based on renal ultrasound imaging, used to implement the quantitative assessment method for fibrosis based on renal ultrasound imaging as described in any one of claims 1-9, characterized in that, The system includes: The acquisition and preprocessing module is used to acquire and standardize renal ultrasound image data, and estimate the system response correction function and optical flow field. The mapping module is used to select a set of keyframes based on the optical flow field, construct and optimize the pose map based on the keyframe image registration, fuse the mapping relationship between the kidney capsule to generate the cortical conformal coordinate system and the keyframe pixel coordinates, and generate the cortical region probability field and boundary uncertainty field. The fusion inversion module is used to calculate the acoustic attenuation coefficient and scattering statistics within the sliding analysis window and perform system response correction. It is projected onto the cortical conformal coordinate system using the mapping relationship and performs Wasserstein centroid fusion by weighting the cortical region probability field and boundary uncertainty field. It detects the effective cortical thickness and obtains the fibrosis fraction field and consensus residual field through consensus optimization. The calibration decision module is used to generate an inconsistent total score based on the fiber fraction field and output the confidence interval and confidence level through the conformal prediction calibration method, and generate a rollback recalculation instruction or a re-sampling instruction.

Images