Ultrasound image quality assessment and automatic correction system for nerve block

Abstract

This invention relates to the field of medical image processing and intelligent control technology of ultrasound equipment, specifically, to an ultrasound image quality assessment and automatic correction system for nerve blocks. It includes an ultrasound image acquisition unit for controlling the ultrasound imaging equipment to acquire ultrasound images of the nerve block region in real time in B-mode; a similarity risk detection unit for generating a visual confusion risk score characterizing nerves and blood vessels, thereby identifying high-risk confusion areas in the ultrasound image; a multimodal verification triggering unit for performing blood flow signal verification on high-risk confusion areas, thereby distinguishing between nerve tissue and vascular tissue; and an adaptive correction unit for generating parameter adjustment instructions based on the results of the blood flow signal verification. This system can automatically identify and quantify the visual confusion risk between nerves and blood vessels, seamlessly triggering Doppler verification when the risk exceeds a threshold, greatly reducing the risk of misjudgment due to human oversight or lack of experience, and improving operational safety and efficiency.

Description

Technical Field

[0001] This invention relates to the field of medical image processing and intelligent control technology of ultrasound equipment, and more specifically, to an ultrasound image quality assessment and automatic correction system for nerve block. Background Technology

[0002] Ultrasound-guided nerve block has become a core technique for regional anesthesia due to its ability to achieve precise localization, improve success rates, and reduce complications. This technique primarily relies on ultrasound imaging equipment to provide real-time anatomical views, with B-mode ultrasound being the fundamental imaging modality for observing nerves and surrounding tissue structures.

[0003] Under B-mode ultrasound, peripheral nerves (usually appearing as honeycomb or bundle-like hypoechoic structures) and blood vessels (anechoic or hypoechoic lumens) are easily confused in terms of morphology and grayscale, especially small nerves and small veins, and nerves and blood vessels in cross-sections. Currently, clinical practice relies on two main methods to resolve this confusion: one is for the operator to subjectively differentiate them based on their anatomical knowledge and image interpretation experience; the other is for the operator to manually operate the ultrasound equipment, temporarily switching to color Doppler or power Doppler mode, and identifying blood vessels by detecting blood flow signals when indeterminacy is not possible.

[0004] However, existing technologies have significant shortcomings: 1. The procedure is highly dependent on subjective experience, with significant differences in skill levels among different doctors and varying judgment standards. Furthermore, in a high-pressure, fast-paced surgical environment, visual fatigue and distraction can easily lead to omissions or misjudgments. 2. Manually switching to Doppler mode requires interrupting the normal scanning and puncture process, which is cumbersome, inefficient, and requires professional knowledge to interpret Doppler signals. 3. Existing ultrasound imaging equipment is inherently "passive." The optimization of its imaging parameters (such as gain, dynamic range, and focus) relies heavily on repeated manual adjustments by the user. The equipment cannot intelligently perceive specific confusion risks in the image, nor does it have the ability to automatically trigger verification processes based on image content and drive adaptive correction of parameters accordingly.

[0005] Therefore, an ultrasound image quality assessment and automatic correction system for nerve block was designed. Summary of the Invention

[0006] The purpose of this invention is to provide an ultrasound image quality assessment and automatic correction system for nerve blocks, in order to solve the problems mentioned in the background art.

[0007] To achieve the above objectives, the present invention aims to provide an ultrasound image quality assessment and automatic correction system for nerve blocks, comprising: An ultrasound image acquisition unit is used to control an ultrasound imaging device to acquire ultrasound images of the nerve block area in real time in B-mode. A similarity risk detection unit is used to process the ultrasound image, extract neural candidate region features and vascular candidate region features respectively, and generate a visual confusion risk score representing nerves and blood vessels based on the similarity relationship between the neural candidate region features and the vascular candidate region features. Based on whether the visual confusion risk score exceeds a preset risk threshold, high-risk confusion regions in the ultrasound image are identified. A multimodal verification triggering unit is used to trigger the ultrasound imaging device to switch from the currently running B-mode imaging to Doppler imaging mode when the visual confusion risk score exceeds a preset risk threshold, and to perform blood flow signal verification on the high-risk confusion area, thereby distinguishing between nerve tissue and vascular tissue. An adaptive correction unit is used to generate parameter adjustment instructions based on the results of the blood flow signal verification.

[0008] As a further improvement to this technical solution, the similarity risk detection unit includes a candidate region localization module, a feature extractor, and a similarity calculation and scoring module; The candidate region localization module is used to process the ultrasound image, identify and output at least one nerve candidate region and at least one blood vessel candidate region. The feature extractor is used to extract image features from the images of the neural candidate region and the blood vessel candidate region respectively, to obtain the first deep visual feature and the second deep visual feature; The similarity calculation and scoring module is used to take the first deep visual feature and the second deep visual feature as the first feature vector and the second feature vector, respectively, calculate the cosine similarity between them, and map the cosine similarity to the scoring interval through a preset nonlinear mapping function to generate the visual confusion risk score.

[0009] As a further improvement to this technical solution, the feature extractor is trained using a training set containing differential samples. The samples in the training set are classified based on objective indicators of the original ultrasound images, and include at least the following three categories: The first type of sample: In the original ultrasound image corresponding to it, the average image gradient amplitude between the nerve and blood vessel regions is higher than the first preset threshold, and the normalized mutual information between the two is lower than the second preset threshold. The second type of sample: In the original ultrasound image, the average image gradient amplitude between the nerve and blood vessel regions is lower than the first preset threshold, and the structural similarity index between the two is higher than the third preset threshold. The third type of sample: the overall signal-to-noise ratio of the original ultrasound image is lower than the fourth preset threshold.

[0010] As a further improvement to this technical solution, the training process of the feature extractor is optimized by minimizing a combined loss function, which includes: The first loss term, which is constructed based on the first feature vector and the second feature vector, is used to reduce the cosine similarity between the first feature vector and the second feature vector when training on the first type of samples; The second loss term, which is constructed based on the first feature vector and the second feature vector, is used to increase the cosine similarity between the first feature vector and the second feature vector when training on the second type of samples. The third loss term is used to increase the uncertainty measure of the feature extractor's prediction of whether the input image patch belongs to the nerve or blood vessel category when training on the third type of samples.

[0011] As a further improvement to this technical solution, the Doppler imaging modes include color Doppler mode, power Doppler mode, and spectral Doppler mode.

[0012] As a further improvement to this technical solution, the multimodal verification triggering unit includes an instruction generation and sending module and a verification analysis module; The instruction generation and sending module is used to perform the following operations when the confusion risk score exceeds a preset threshold: Save the current B-mode imaging parameters of the ultrasound imaging device, and send a mode switching command to the ultrasound imaging device through the device control interface to switch it to Doppler imaging mode, and collect blood flow signal data for a preset duration while maintaining the current imaging plane. The verification analysis module is used to analyze the spatial distribution and signal intensity of the blood flow signal data in the high-risk confusion area, determine whether there is a blood flow signal in the area that is higher than a preset intensity threshold, and mark the high-risk confusion area as vascular tissue or nerve tissue based on the judgment result.

[0013] As a further improvement to this technical solution, the instruction generation and sending module specifically sends the Doppler imaging mode switching instruction to the ultrasound imaging device as follows: Based on the preset risk range to which the visual confusion risk score belongs, different mode switching instructions are generated; When the visual confusion risk score is in the first preset risk range, the mode switching command indicates switching to the color Doppler mode; When the visual confusion risk score is in a second preset risk range that is higher than the first preset risk range, the mode switching command indicates that both color Doppler mode and power Doppler mode are enabled simultaneously.

[0014] As a further improvement to this technical solution, the verification analysis module determines whether there is a blood flow signal in the region that is higher than a preset intensity threshold, specifically as follows: In the color Doppler mode or power Doppler mode, spectral Doppler mode sampling is initiated at at least one selected point within the high-risk confusion region; Based on the waveform obtained by the spectral Doppler sampling, at least one parameter among the peak flow velocity, average flow velocity, or pulsatility index is calculated, and the parameter is compared with the corresponding preset parameter threshold to verify the judgment result based on the spatial distribution and signal intensity of the blood flow signal.

[0015] As a further improvement to this technical solution, the adaptive correction unit is pre-set with at least two parameter optimization strategies, and executes the corresponding strategies based on the labeling results of the verification analysis module: When the labeling result is vascular tissue, the first parameter optimization strategy is executed. The parameter adjustment instruction is used to adjust at least one parameter among the time gain compensation, harmonic imaging state and dynamic range of the ultrasound imaging device. When the labeling result is neural tissue, a second parameter optimization strategy is executed. The parameter adjustment command is used to adjust at least one parameter among the probe scanning angle, focal position, and spatial composite imaging state of the ultrasound imaging device.

[0016] As a further improvement to this technical solution, the adaptive correction unit acquires new ultrasound images and recalculates the visual confusion risk score after each parameter adjustment command is executed, and calculates the change based on the visual confusion risk score before and after the command adjustment. The adaptive correction unit according to Perform the following actions: like If the value is greater than the first preset feedback threshold, the current parameter settings will be retained. like If the absolute value is less than or equal to the second preset feedback threshold, then a new parameter adjustment instruction is generated from the preset parameter optimization strategies, which is different from the current strategy. like If the value is less than the third preset feedback threshold, a rollback instruction is generated to restore the parameters to their pre-adjustment state.

[0017] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. This ultrasound image quality assessment and automatic correction system for nerve blocks introduces a deep learning-based automatic similarity risk detection and multimodal verification triggering mechanism, achieving a shift from subjective experience-based judgment to objective quantitative assessment, and from manual operation to intelligent linkage. The system can automatically identify and quantify the visual confusion risk between nerves and blood vessels, seamlessly triggering Doppler verification when the risk exceeds a threshold. This significantly reduces the risk of misjudgment due to human oversight or lack of experience, improving operational safety and efficiency.

[0018] 2. This ultrasound image quality assessment and automatic correction system for nerve blocks achieves intelligent closed-loop improvement of ultrasound image quality by establishing an adaptive parameter correction and feedback optimization loop based on verification results. The system can automatically adjust the optimal imaging parameters according to tissue type and continuously optimize based on quantitative feedback of image quality, ultimately outputting a more resolvable target image. This reduces the burden of manual adjustments for doctors and fundamentally improves the reliability and consistency of images. Attached Figure Description

[0019] Figure 1 This is a flowchart illustrating the overall process of the present invention. Detailed Implementation

[0020] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0021] Example: Please refer to Figure 1 As shown, an ultrasound image quality assessment and automatic correction system for nerve blocks is provided, including: Ultrasound image acquisition unit, which is used to control the ultrasound imaging device to acquire ultrasound images of the nerve block area in real time in B mode. The similarity risk detection unit is used to process ultrasound images, extract neural candidate region features and vascular candidate region features respectively, and generate a visual confusion risk score representing nerves and blood vessels based on the similarity relationship between the neural candidate region features and vascular candidate region features. Based on whether the visual confusion risk score exceeds a preset risk threshold, high-risk confusion regions in ultrasound images are identified. The similarity risk detection unit includes a candidate region localization module, a feature extractor, and a similarity calculation and scoring module; The candidate region localization module is used to process ultrasound images, identify and output at least one neural candidate region and at least one vascular candidate region; The feature extractor is used to extract image features from images of neural candidate regions and blood vessel candidate regions respectively, to obtain the first deep visual feature and the second deep visual feature; The feature extractor is preferably implemented using a convolutional neural network (CNN). Deep visual features refer to feature maps or feature vectors extracted from the middle or back layers of the CNN. Unlike the raw pixels of the input image or "shallow features" such as edges and textures obtained through traditional image processing, these deep features undergo multiple layers of nonlinear transformations within the network, enabling them to capture more semantic and high-level abstract information in the image, such as structural patterns of tissues and complex distributions of echo characteristics.

[0022] The feature extractor is trained using a training set containing differentially expressed samples, which are classified based on objective metrics from the original ultrasound images and include at least the following three categories: The first type of sample: In the original ultrasound image corresponding to it, the average image gradient amplitude between the nerve and blood vessel regions is higher than the first preset threshold, and the normalized mutual information between the two is lower than the second preset threshold. The second type of sample: In the original ultrasound image, the average image gradient amplitude between the nerve and blood vessel regions is lower than the first preset threshold, and the structural similarity index between the two is higher than the third preset threshold. The third type of sample: the overall signal-to-noise ratio of the original ultrasound image is lower than the fourth preset threshold.

[0023] The first type of sample is easily distinguishable. It has clear boundaries (high gradient) and large statistical differences in internal echo texture (low mutual information). For example, a blood vessel with uniform echo is adjacent to a honeycomb-like nerve. During training, the model is instructed to "push" the feature vectors of the two images apart in space when faced with such clear images, thereby outputting a lower similarity score, corresponding to a low risk score.

[0024] The second type of sample is easily confused. These samples have blurred boundaries (low gradient) and are visually highly similar in shape and texture (high SSIM). For example, small, circular, low-echoic nerves and veins appear in cross-section. During training, the model is instructed to capture potential, subtle differences even in visually similar regions, or to output high-similarity features when indistinguishable. This ensures the system's sensitivity to truly high-risk scenarios.

[0025] The third category of samples consists of low-quality samples. These images are generally noisy, with weak signals of useful tissue, making it difficult to distinguish any structures. During training, by maximizing the entropy (uncertainty) of the model's category prediction, the system can more cautiously assess the risk when encountering low-quality images, avoiding misjudgments due to noise interference.

[0026] The first preset threshold corresponds to the average image gradient magnitude index, typically between [5, 30] (depending on image size and normalization method). A larger value indicates sharper edges. In this embodiment, the 70th percentile of the average gradient magnitude of all "neurovascular region pairs" in the dataset is used. For example, if the calculated value is 15.6, then the first preset threshold is 15.6 (values ​​higher are considered "clear boundaries"). The second preset threshold corresponds to the normalized mutual information index, ranging from [0, 1]. Values ​​closer to 0 indicate stronger statistical independence between the two regions. In this embodiment, the 30th percentile of the NMI values ​​for all region pairs in the dataset is used. For example, if the value is 0.25, then the second preset threshold is 0.25 (values ​​lower are considered "large statistical differences"). The third preset threshold corresponds to the structural similarity index, ranging from [0, 1]. Values ​​closer to 1 indicate greater structural similarity. In this embodiment, the 80th percentile of the SSIM values ​​for all region pairs in the dataset is used. For example, if the value is 0.85, then: the third preset threshold is 0.85 (values ​​higher than this are considered "highly similar in structure"). The fourth preset threshold corresponds to the image signal-to-noise ratio (SNR), which is usually greater than 0 (dB). Clinical ultrasound images typically have an SNR in the range of 5-30 dB. In this embodiment, the 20th percentile of the overall SNR of all images in the dataset is used. For example, if the value is 9.5 dB, then: the fourth preset threshold is 9.5 dB (values ​​lower than this are considered "low-quality images").

[0027] The similarity calculation and scoring module uses the first deep visual feature and the second deep visual feature as the first feature vector and the second feature vector, respectively, calculates the cosine similarity between them, and maps the cosine similarity to a scoring interval through a preset nonlinear mapping function to generate a visual confusion risk score. .

[0028] Visual confusion risk score The generation of the score is an end-to-end computational process from the original image to the quantized score, and the specific steps are as follows: Step 1: Extraction of candidate region image patches A candidate region localization module (e.g., using an object detection model) processes the input ultrasound image and outputs the bounding box coordinates of neural and vascular candidate regions. Based on these coordinates, the system crops the corresponding image patches from the original image.

[0029] Step 2: Deep Feature Vector Extraction The cropped neural and vascular candidate region image patches are input into a pre-trained feature extractor (such as a CNN). The network outputs two fixed-dimensional feature vectors: The feature vector F of the neural candidate region n (First deep visual feature) Feature vector F of blood vessel candidate region v (Second deep visual feature) Step 3: Calculate cosine similarity Calculate two eigenvectors F n and F v The cosine similarity S between them is used to measure their directional similarity in the feature space. The calculation formula is:

[0030] in Represents the vector dot product. S represents the modulus (L2 norm) of the vector. The value of S ranges from [-1, 1], and the closer the value is to 1, the more similar the features of the two vectors are.

[0031] Step 4: Nonlinear mapping to risk score The cosine similarity S is converted into a more intuitive and range-fixed risk score R through a pre-defined non-linear mapping function. This embodiment uses a variant of the sigmoid function for calculation, as shown in the following formula:

[0032] In the formula, exp is an exponential function; k (for example, a value of 10) is a scaling factor that controls the steepness of the function curve and determines the sensitivity of changes in similarity S to changes in rating R. S0 (for example, a value of 0.7) is the center offset. When S=S0, R=50. It defines the baseline similarity for "moderate risk".

[0033] This function maps S from the theoretical interval [-1,1] to the scoring interval of R [0,100].

[0034] Calculation example: If , , ,but This is considered high-risk; the rules for determining the risk range are given in the instruction generation and sending module below.

[0035] Step 5: Scoring Output Final output This is a visual confusion risk score, which can be directly compared with preset risk thresholds or ranges to trigger subsequent processes.

[0036] Furthermore, the training process of the feature extractor is optimized by minimizing a combined loss function, which includes: The first loss term, which is constructed based on the first feature vector and the second feature vector, is used to reduce the cosine similarity between the first feature vector and the second feature vector when training on the first type of samples. The second loss term, which is constructed based on the first feature vector and the second feature vector, is used to increase the cosine similarity between the first feature vector and the second feature vector when training on the second type of samples. The third loss term is used to increase the uncertainty measure of the feature extractor's prediction of whether the input image patch belongs to the nerve or blood vessel category when training on the third type of samples.

[0037] Specifically, the first loss term, L1, is applied to the first class of samples (easily distinguishable samples). It employs the concept of contrastive loss, aiming to differentiate neural and vascular features. The formula is: , where m1 is a set negative value or a small positive value (such as -0.5 or 0.2), and the training objective is to optimize the similarity S to be less than m1.

[0038] The second loss term, L2, is applied to the second class of samples (easily confused samples). Its purpose is to teach the model that even for visually similar nerves and blood vessels, their feature representations should be preserved or approximated to reflect their underlying confusion. The formula is: , where m2 is a large positive value (e.g., 0.8), and the training objective is to optimize S to be greater than m2.

[0039] The third loss term, L3, is applied to the third class of samples (low-quality samples). It aims to increase the uncertainty of the model's prediction of tissue categories in low-quality images, preventing the model from making overly confident erroneous judgments under noise interference. This loss term is defined as the negative entropy of the class prediction probability distribution p: , its p i To predict which input image patch belongs to the first... The probability of a class (nerve or blood vessel). Minimizing L3 is equivalent to maximizing entropy H(p), even with a more uniform prediction distribution and higher uncertainty.

[0040] For the first type of sample, the total loss is For the second type of sample, the total loss is For the third type of sample, the total loss is The total loss function is the weighted sum of all terms: ,in The weighting coefficient for the first loss term; The weighting coefficient for the second loss term; represents the weight coefficient of the third loss term; during training, based on the category of the input sample, only the corresponding loss term is activated for calculation and gradient backpropagation.

[0041] The multimodal verification trigger unit is used to trigger the ultrasound imaging device to switch from the currently running B-mode imaging to Doppler imaging mode when the visual confusion risk score exceeds a preset risk threshold, and to verify the blood flow signal in the high-risk confusion area, thereby distinguishing between nerve tissue and vascular tissue. Doppler imaging modes include color Doppler mode, power Doppler mode, and spectral Doppler mode; Among them, the color Doppler mode is used to detect and display the direction and velocity information of blood flow in high-risk confusion areas based on average frequency shift, so as to identify arterial vessels; the power Doppler mode is used to detect low-velocity blood flow signals in high-risk confusion areas with higher sensitivity than the color Doppler mode, so as to assist in the identification of venous vessels.

[0042] Spectral Doppler mode is used for point-to-point quantification and precise identification, and can provide the following functions: Point-of-sight analysis: It is not displayed on the entire plane, but rather analyzed within a tiny sampling volume specified by the operator (or automatically by the system). This point is precisely aligned with the previously identified "high-risk confounding area".

[0043] It provides a quantitative waveform: it outputs a velocity-time spectrum, with time on the horizontal axis and blood flow velocity on the vertical axis, and the width of the waveform representing the velocity distribution. This is the "fingerprint" of blood flow.

[0044] This enables the following: Distinguishing between true blood flow and motion artifacts: Color Doppler ultrasound may produce a color signal resembling blood flow (flicker artifact) due to tissue movement (such as probe jitter or conduction of adjacent arterial pulsations). Spectral Doppler can clearly show whether there is a real blood flow signal with specific spectral characteristics at that point. Artifacts typically appear as irregular, non-periodic low-frequency signals.

[0045] Accurately distinguishing between arteries and veins: Arterial spectra are pulsatile, with high velocity during systole and low velocity or even reverse during diastole; venous spectra are continuous, smooth fluctuations or change with respiration. Objective differentiation can be achieved by calculating parameters such as the pulsatility index (PI) and resistance index (RI).

[0046] Provides absolute velocity values: Provides specific values ​​such as peak flow velocity (PSV) and average flow velocity (TAMV), which can be compared with normal anatomical values ​​to help determine the nature of blood vessels.

[0047] The multimodal verification triggering unit includes an instruction generation and sending module and a verification analysis module; The instruction generation and sending module is used to perform the following operations when the confusion risk score exceeds a preset threshold: Save the current B-mode imaging parameters of the ultrasound imaging device, and send a mode switching command to the ultrasound imaging device through the device control interface to switch it to Doppler imaging mode, and collect blood flow signal data for a preset duration while maintaining the current imaging plane. The verification analysis module is used to analyze the spatial distribution and signal intensity of blood flow signal data in high-risk confusion areas, determine whether there are blood flow signals in the area that are higher than a preset intensity threshold, and mark the high-risk confusion areas as vascular tissue or nerve tissue based on the judgment results.

[0048] In the instruction generation and transmission module, the specific steps for sending the Doppler imaging mode switching instruction to the ultrasound imaging device are as follows: Based on the preset risk range to which the visual confusion risk score belongs, different mode switching instructions are generated; When the visual confusion risk score is in the first preset risk range, the mode switching command indicates that the mode should be switched to the color Doppler mode. When the visual confusion risk score is in the second preset risk range, which is higher than the first preset risk range, the mode switching command indicates that both the color Doppler mode and the power Doppler mode are enabled simultaneously.

[0049] The preset risk ranges specifically include a low-risk range, a medium-risk range (the first preset risk range), and a high-risk range (the second preset risk range). Low-risk zone: Within this range, the system considers the risk of confusion to be low and does not trigger Doppler verification.

[0050] Medium-risk zone (first preset risk zone): The system recognizes a certain risk of confusion and triggers color Doppler mode for blood flow verification, primarily identifying typical arterial blood flow.

[0051] High-risk zone (second preset risk zone): The system considers the risk of confusion to be high, possibly involving low-velocity blood flow or complex situations, and therefore simultaneously triggers both color Doppler and power Doppler modes to improve the detection sensitivity for low-velocity venous blood flow.

[0052] In the verification and analysis module, it is determined whether there is a blood flow signal in the region that is higher than a preset intensity threshold. Specifically: In either color Doppler mode or power Doppler mode, initiate spectral Doppler mode sampling at at least one selected location within the high-risk confusion area; Based on the waveform obtained by spectral Doppler sampling, at least one parameter among peak flow velocity, average flow velocity, or pulsatility index is calculated, and the parameter is compared with the corresponding preset parameter threshold to verify the judgment result based on the spatial distribution and signal intensity of blood flow signal.

[0053] Specifically, the input is velocity-time spectrum data (a sequence of velocity values ​​changing over time) obtained by spectral Doppler sampling at the center of a high-risk confusion area or the point where the blood flow signal is strongest.

[0054] Perform the following steps: Calculate one or more of the following standard hemodynamic parameters during a complete cardiac cycle: Peak flow rate: The highest velocity value within one cycle.

[0055] End-diastolic flow rate: the lowest velocity value before the end of a cycle (usually corresponding to the R-wave trigger point of the cardiac cycle).

[0056] Average flow velocity: the area under the velocity curve over one cycle divided by time.

[0057] Pulsatility index: a key indicator reflecting vascular resistance and pulsatility.

[0058] The calculated parameters are compared with physiologically reasonable preset parameter thresholds.

[0059] Example of conditional logic: If the peak flow velocity is greater than the peak flow velocity threshold (e.g., >15 cm / s) and the pulsatility index is greater than the pulsatility index threshold (e.g., >1.0), it is identified as an artery; this indicates high-velocity, pulsatile blood flow.

[0060] If the peak flow velocity is less than the peak flow velocity threshold and the pulsatility index is close to 0 or very low, and the average flow velocity is greater than 0, or if power Doppler shows a continuous signal but a flat spectrum, then it is confirmed as a vein. If the peak flow velocity is extremely low and the waveform is chaotic and non-periodic, or the pulsatility index is meaningless, it is judged as an artifact / noise.

[0061] Based on the above verification results, the system ultimately labels the high-risk confusion area as "arterial blood vessels", "venous blood vessels", or confirms it as "nerve tissue (no effective blood flow)".

[0062] The adaptive correction unit generates parameter adjustment instructions based on the results of blood flow signal verification, controls the ultrasound imaging equipment to adjust scanning parameters, thereby reducing the risk score of visual confusion between nerves and blood vessels in subsequent ultrasound images.

[0063] The adaptive calibration unit has at least two preset parameter optimization strategies and executes the corresponding strategy based on the labeling results of the verification analysis module: When the labeling result is vascular tissue, the first parameter optimization strategy is executed. The parameter adjustment command is used to adjust at least one parameter among the time gain compensation, harmonic imaging state and dynamic range of the ultrasound imaging equipment. When the labeling result is neural tissue, the second parameter optimization strategy is executed. The parameter adjustment command is used to adjust at least one parameter among the probe scanning angle, focal position, and spatial compound imaging state of the ultrasound imaging equipment.

[0064] Specifically, First parameter optimization strategy (for tissues labeled as blood vessels): Adjusting time gain compensation (TGC): Reduces the gain at the corresponding depth behind the blood vessel to suppress the "posterior echo enhancement" artifact caused by the anechoic blood, and prevents it from masking adjacent nerves or tissues.

[0065] Enable harmonic imaging: Imaging using tissue harmonic frequency signals can effectively reduce noise and side lobe artifacts from blood vessels, making the vessel wall interface clearer.

[0066] Reduce dynamic range: Compress the grayscale display range of the image to make the echo-free cavity of the blood vessel and the echo contrast with the surrounding tissue relatively soft, reduce visual impact, and facilitate observation of the blood vessel wall and surrounding structures.

[0067] Second parameter optimization strategy (for tissues labeled as neurons): Adjust the probe scanning angle: Fine-tune the incident angle of the ultrasound beam to make it as perpendicular as possible to the direction of the nerve bundle in order to obtain stronger nerve peritunic echoes and enhance the visualization of nerve structures.

[0068] Adjusting the focal position: Set the emission focal point of the ultrasound beam to match the depth of the nerve structure, thereby obtaining the highest spatial resolution at that depth and making the internal bundle structure of the nerve clearer.

[0069] Enabling or enhancing spatial composite imaging: Scanning the same structure from multiple different angles and averaging the images can significantly reduce echo loss artifacts caused by neural anisotropy, making the nerve appear more uniform and continuous throughout the imaging area.

[0070] After each parameter adjustment command is executed, the adaptive correction unit acquires new ultrasound images and recalculates the visual confusion risk score. It then calculates the change ΔR based on the visual confusion risk score before and after the adjustment. Based on ΔR, the adaptive correction unit performs the following actions: If ΔR is greater than the first preset feedback threshold, the current parameter settings will be retained. If the absolute value of ΔR is less than or equal to the second preset feedback threshold, then select another strategy different from the current strategy from the preset parameter optimization strategies to generate a new parameter adjustment instruction; If ΔR is less than the third preset feedback threshold, a rollback instruction is generated to restore the parameters to their pre-adjustment state.

[0071] Specifically, after executing a parameter adjustment command, the adaptive correction unit does not immediately terminate but instead initiates a feedback control loop. The specific process is as follows: (a) controlling the ultrasound equipment to image according to the new parameters and acquire new ultrasound images; (b) calling the similarity risk detection unit to recalculate the visual confusion risk score of the new images. (c) Calculate the change in the effect of this adjustment. (d) The score before adjustment; The parameter is compared with a preset set of feedback control thresholds, and the next action is determined based on the comparison result: whether to accept the current parameter, try other optimization strategies, or revert to the previous adjustment. The first feedback threshold T1 is defined as the critical value for determining whether the correction is significantly effective. At that time, it was believed that parameter adjustments significantly reduced the risk of confusion. T1 is usually set to a small positive value (e.g., +5). The rationale is that, considering the natural fluctuations in images, only when the score decreases by more than this threshold can it be confidently attributed to parameter optimization rather than random noise.

[0072] The second feedback threshold T2 is defined as the critical value for determining whether the correction effect is insignificant or uncertain. When the current parameter adjustment has not produced a clear effect, it is considered that T2 is a positive value close to zero (e.g., +3). Its physical meaning is that this range covers fluctuations in the score caused by noise or small changes. Triggering a strategy switch is to avoid getting trapped in local optima by trying different optimization directions (e.g., switching from optimizing "contrast" to optimizing "resolution") to find a better solution.

[0073] The third feedback threshold T3 is defined as the critical value for determining whether the correction has a negative effect. When At this point, it is believed that adjusting the current parameters actually worsens image quality and increases the risk score. T3 is typically set to a negative value (e.g., -10). Its setting is more "strict" than T1, allowing for small, insignificant effects. However, a significant drop in quality cannot be tolerated. Once triggered, the system will perform a "rollback" to restore the system to its stable state before the adjustment.

[0074] Furthermore, if ΔR satisfies This indicates that the parameter adjustment has a positive effect but has not reached a significant standard. The system can be set to retain the current parameter settings and end the current calibration cycle.

[0075] These three thresholds satisfy the relationship The specific value can be determined by conducting numerous simulation and adjustment experiments on typical equipment and typical anatomical sites, and then statistically analyzing the distribution of ΔR. In this embodiment, T1 is taken as the 25th percentile of the positive distribution of ΔR, T3 is taken as the 25th percentile of the negative distribution, and T2 is set as the maximum value of the scoring fluctuation range caused by the noise of the measurement system.

[0076] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely preferred examples and are not intended to limit the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the claimed invention.

Claims

1. An ultrasound image quality assessment and automatic correction system for nerve blocks, characterized in that, include: An ultrasound image acquisition unit is used to control an ultrasound imaging device to acquire ultrasound images of the nerve block area in real time in B-mode. A similarity risk detection unit is used to process the ultrasound image, extract neural candidate region features and vascular candidate region features respectively, and generate a visual confusion risk score representing nerves and blood vessels based on the similarity relationship between the neural candidate region features and the vascular candidate region features. Based on whether the visual confusion risk score exceeds a preset risk threshold, high-risk confusion regions in the ultrasound image are identified. A multimodal verification triggering unit is used to trigger the ultrasound imaging device to switch from the currently running B-mode imaging to Doppler imaging mode when the visual confusion risk score exceeds a preset risk threshold, and to perform blood flow signal verification on the high-risk confusion area, thereby distinguishing between nerve tissue and vascular tissue. An adaptive correction unit is used to generate parameter adjustment instructions based on the results of the blood flow signal verification.

2. The ultrasound image quality assessment and automatic correction system for nerve block according to claim 1, characterized in that: The similarity risk detection unit includes a candidate region localization module, a feature extractor, and a similarity calculation and scoring module; The candidate region localization module is used to process the ultrasound image, identify and output at least one nerve candidate region and at least one blood vessel candidate region. The feature extractor is used to extract image features from the images of the neural candidate region and the blood vessel candidate region respectively, to obtain the first deep visual feature and the second deep visual feature; The similarity calculation and scoring module is used to take the first deep visual feature and the second deep visual feature as the first feature vector and the second feature vector, respectively, calculate the cosine similarity between them, and map the cosine similarity to the scoring interval through a preset nonlinear mapping function to generate the visual confusion risk score.

3. The ultrasound image quality assessment and automatic correction system for nerve blocks according to claim 2, characterized in that: The feature extractor is trained using a training set containing differentially expressed samples, which are classified based on objective metrics of the original ultrasound images and include at least the following three categories: The first type of sample: In the original ultrasound image corresponding to it, the average image gradient amplitude between the nerve and blood vessel regions is higher than the first preset threshold, and the normalized mutual information between the two is lower than the second preset threshold. The second type of sample: In the original ultrasound image, the average image gradient amplitude between the nerve and blood vessel regions is lower than the first preset threshold, and the structural similarity index between the two is higher than the third preset threshold. The third type of sample: the overall signal-to-noise ratio of the original ultrasound image is lower than the fourth preset threshold.

4. The ultrasound image quality assessment and automatic correction system for nerve blocks according to claim 3, characterized in that: The training process of the feature extractor is optimized by minimizing a combined loss function, which includes: The first loss term, which is constructed based on the first feature vector and the second feature vector, is used to reduce the cosine similarity between the first feature vector and the second feature vector when training on the first type of samples; The second loss term, which is constructed based on the first feature vector and the second feature vector, is used to increase the cosine similarity between the first feature vector and the second feature vector when training on the second type of samples. The third loss term is used to increase the uncertainty measure of the feature extractor's prediction of whether the input image patch belongs to the nerve or blood vessel category when training on the third type of samples.

5. The ultrasound image quality assessment and automatic correction system for nerve block according to claim 4, characterized in that: The Doppler imaging modes include color Doppler mode, power Doppler mode, and spectral Doppler mode.

6. The ultrasound image quality assessment and automatic correction system for nerve block according to claim 5, characterized in that: The multimodal verification triggering unit includes an instruction generation and sending module and a verification analysis module; The instruction generation and sending module is used to perform the following operations when the confusion risk score exceeds a preset threshold: Save the current B-mode imaging parameters of the ultrasound imaging device, and send a mode switching command to the ultrasound imaging device through the device control interface to switch it to Doppler imaging mode, and collect blood flow signal data for a preset duration while maintaining the current imaging plane. The verification analysis module is used to analyze the spatial distribution and signal intensity of the blood flow signal data in the high-risk confusion area, determine whether there is a blood flow signal in the area that is higher than a preset intensity threshold, and mark the high-risk confusion area as vascular tissue or nerve tissue based on the judgment result.

7. The ultrasound image quality assessment and automatic correction system for nerve block according to claim 6, characterized in that: In the instruction generation and sending module, sending the Doppler imaging mode switching instruction to the ultrasound imaging device specifically involves: Based on the preset risk range to which the visual confusion risk score belongs, different mode switching instructions are generated; When the visual confusion risk score is in the first preset risk range, the mode switching command indicates switching to the color Doppler mode; When the visual confusion risk score is in a second preset risk range that is higher than the first preset risk range, the mode switching command indicates that both color Doppler mode and power Doppler mode are enabled simultaneously.

8. The ultrasound image quality assessment and automatic correction system for nerve block according to claim 7, characterized in that: In the verification analysis module, determining whether there is a blood flow signal in the region that is higher than a preset intensity threshold specifically involves: In the color Doppler mode or power Doppler mode, spectral Doppler mode sampling is initiated at at least one selected point within the high-risk confusion region; Based on the waveform obtained by the spectral Doppler sampling, at least one parameter among the peak flow velocity, average flow velocity, or pulsatility index is calculated, and the parameter is compared with the corresponding preset parameter threshold to verify the judgment result based on the spatial distribution and signal intensity of the blood flow signal.

9. The ultrasound image quality assessment and automatic correction system for nerve block according to claim 8, characterized in that: The adaptive correction unit has at least two preset parameter optimization strategies, and executes the corresponding strategies based on the labeling results of the verification analysis module: When the labeling result is vascular tissue, the first parameter optimization strategy is executed. The parameter adjustment instruction is used to adjust at least one parameter among the time gain compensation, harmonic imaging state and dynamic range of the ultrasound imaging device. When the labeling result is neural tissue, a second parameter optimization strategy is executed. The parameter adjustment command is used to adjust at least one parameter among the probe scanning angle, focal position, and spatial composite imaging state of the ultrasound imaging device.

10. The ultrasound image quality assessment and automatic correction system for nerve block according to claim 9, characterized in that: After each parameter adjustment command is executed, the adaptive correction unit acquires new ultrasound images and recalculates the visual confusion risk score, calculating the change based on the visual confusion risk score before and after the command adjustment. The adaptive correction unit is based on Perform the following actions: like If the value is greater than the first preset feedback threshold, the current parameter settings will be retained. like If the absolute value is less than or equal to the second preset feedback threshold, then a new parameter adjustment instruction is generated from the preset parameter optimization strategies, which is different from the current strategy. like If the value is less than the third preset feedback threshold, a rollback instruction is generated to restore the parameters to their pre-adjustment state.

Images