A thyroid nodule ultrasound image feature extraction method and system

Abstract

This invention relates to the technical field of ultrasound image feature extraction for thyroid nodules, specifically to a method and system for ultrasound image feature extraction of thyroid nodules. The method includes the following steps: acquiring an ultrasound image containing thyroid nodules; performing multi-scale image analysis on the ultrasound image to construct a multi-scale image representation of the ultrasound image; using the multi-scale image representation to identify low-echo regions in the ultrasound image, and combining feature analysis, judgment correction, and spatial relationship verification to determine the acoustic shadowing region in the ultrasound image; performing suppression processing on the acoustic shadowing region, wherein the suppression processing includes unifying the signal within the acoustic shadowing region and softening the boundary of the acoustic shadowing region to obtain an optimized image; identifying the contour of the thyroid nodule based on the optimized image; and quantifying the morphology and internal structure of the thyroid nodule based on its contour. This application can effectively identify and suppress acoustic shadowing regions in ultrasound images.

Description

Technical Field

[0001] This invention relates to the technical field of ultrasound image feature extraction for thyroid nodules, and specifically to a method and system for ultrasound image feature extraction for thyroid nodules. Background Technology

[0002] In thyroid ultrasound imaging diagnosis, computer-aided diagnostic systems are widely used to improve the objectivity and consistency of diagnosis. These systems typically preprocess ultrasound images using multi-scale filtering, followed by adaptive segmentation to delineate nodule boundaries and finally calculate quantitative indicators. However, in actual clinical applications, the pathological features of thyroid nodules are complex and diverse, with acoustic shadowing artifacts caused by calcification posing a significant challenge to the accuracy of image processing. For example, when an image of a nodule containing large calcifications is fed into the system, existing multi-scale filtering techniques, while sharpening the overall edge of the nodule to some extent, cannot effectively distinguish between the true boundary of the nodule and the pseudo-boundary caused by acoustic shadowing. The grayscale value of the acoustic shadowing region is extremely low, forming a strong contrast with the internal tissue of the nodule. This strong edge information can mislead the adaptive segmentation program. When searching for the path with the lowest energy or strongest gradient, the segmentation program easily misjudges the edge of the acoustic shadowing as part of the true boundary of the nodule, leading to a serious deviation in the final delineated nodule contour. For example, an unwarranted "depression" may appear in the acoustic shadowing region, or the acoustic shadowing region may be directly "cut out" from the nodule. This inaccurate segmentation result directly undermines the reliability of all subsequent feature extractions. Summary of the Invention

[0003] The purpose of this invention is to address the aforementioned shortcomings by proposing a method and system for extracting features from ultrasound images of thyroid nodules.

[0004] The present invention adopts the following technical solution: A method for extracting features from ultrasound images of thyroid nodules, the method comprising the following steps: Obtain ultrasound images containing thyroid nodules; Multi-scale image analysis and processing are performed on ultrasound images to construct a multi-scale image representation of ultrasound images; By utilizing multi-scale image representation, low-echo regions in ultrasound images are identified, and by combining feature analysis, judgment correction, and spatial relationship verification, acoustic shadowing regions in ultrasound images are determined. The sound shadow region is suppressed. The suppression process includes unifying the signal within the sound shadow region and softening the boundary of the sound shadow region to obtain an optimized image. Based on optimized images, the outline of thyroid nodules is identified; Based on the outline of thyroid nodules, the morphology and internal structure of thyroid nodules are quantified.

[0005] This technical solution can effectively identify and suppress acoustic shadowing regions in ultrasound images, thereby avoiding the interference of acoustic shadowing artifacts on the identification and feature quantification of thyroid nodules and improving the accuracy of diagnosis.

[0006] This application also discloses a thyroid nodule ultrasound image feature extraction system. Applying the above-mentioned thyroid nodule ultrasound image feature extraction method, the system includes: The acquisition module acquires ultrasound images containing thyroid nodules; The module performs multi-scale image analysis and processing on ultrasound images to construct a multi-scale image representation of ultrasound images. The acoustic shadow recognition module uses multi-scale image representation to identify low-echo regions in ultrasound images, and combines feature analysis, judgment correction and spatial relationship verification to determine the acoustic shadow regions in ultrasound images. The suppression module performs suppression processing on the sound shadow region. The suppression processing includes unifying the signal within the sound shadow region and softening the boundary of the sound shadow region to obtain an optimized image. The contour recognition module identifies the contours of thyroid nodules based on optimized images; The quantification module quantifies the morphology and internal structure of thyroid nodules based on their outlines.

[0007] This technical solution provides an integrated system that enables acoustic shadowing, suppression, contour recognition, and feature quantification of thyroid nodule ultrasound images, thereby improving the automation and accuracy of thyroid nodule diagnosis.

[0008] This application can provide reliable quantitative data based on accurate segmentation, effectively assisting doctors in making judgments, significantly improving the objectivity and consistency of ultrasound imaging diagnosis of thyroid nodules, and avoiding potential risks caused by misleading data in the prior art.

[0009] To further understand the features and technical content of the present invention, please refer to the following detailed description and drawings of the present invention. However, the drawings provided are for reference and illustration only and are not intended to limit the present invention. Attached Figure Description

[0010] Figure 1 This is a flowchart of a method for extracting features from ultrasound images of thyroid nodules according to the present invention; Figure 2 This is a schematic diagram of the structure of a thyroid nodule ultrasound image feature extraction system according to the present invention. Detailed Implementation

[0011] The following specific embodiments illustrate the implementation of the present invention. Those skilled in the art can understand the advantages and effects of the present invention from the content disclosed in this specification. The present invention can be implemented or applied through other different specific embodiments, and various details in this specification can also be modified and changed based on different viewpoints and applications without departing from the spirit of the present invention. Furthermore, the accompanying drawings of the present invention are for simple illustrative purposes only and are not depictions of actual dimensions; this is stated in advance. The following embodiments will further describe the relevant technical content of the present invention in detail, but the disclosed content is not intended to limit the scope of protection of the present invention.

[0012] This embodiment provides a method and system for extracting features from ultrasound images of thyroid nodules, combined with... Figure 1 and Figure 2 As shown.

[0013] refer to Figure 1 A method for extracting features from ultrasound images of thyroid nodules, the method comprising the following steps: Obtain ultrasound images containing thyroid nodules; Multi-scale image analysis and processing are performed on ultrasound images to construct a multi-scale image representation of ultrasound images; By utilizing multi-scale image representation, low-echo regions in ultrasound images are identified, and by combining feature analysis, judgment correction, and spatial relationship verification, acoustic shadowing regions in ultrasound images are determined. The sound shadow region is suppressed. The suppression process includes unifying the signal within the sound shadow region and softening the boundary of the sound shadow region to obtain an optimized image. Based on optimized images, the outline of thyroid nodules is identified; Based on the outline of thyroid nodules, the morphology and internal structure of thyroid nodules are quantified.

[0014] The term "ultrasound image" as used in this application refers to medical images acquired using ultrasound technology, typically presented as grayscale images, where different grayscale values ​​represent the intensity of ultrasound reflection from tissue. A "thyroid nodule" is an abnormal mass forming within the thyroid gland, appearing as an area with a different echogenicity than surrounding tissue on an ultrasound image. "Multi-scale image analysis processing" is an image processing technique that analyzes images at different resolutions or scales to obtain rich information such as texture and edges, thereby constructing a "multi-scale image representation." A "hypoechoic region" refers to an area with relatively low grayscale values ​​in an ultrasound image, potentially representing nodules, cysts, or acoustic shadowing. An "acoustic shadowing region" specifically refers to a low-echoic region formed by ultrasound attenuation or blocking behind hyperechoic structures (such as calcifications), typically with missing or extremely low internal signals and blurred boundaries. "Suppression processing" aims to eliminate or reduce the interference of acoustic shadowing regions on subsequent image analysis, including "unifying the signal within the acoustic shadowing region" to prevent it from presenting extremely low grayscale, and "softening the boundaries of the acoustic shadowing region" to make it transition naturally with surrounding tissue. "Optimized image" refers to an image processed with acoustic shadowing suppression, which more closely resembles the actual tissue structure. "Outline of thyroid nodule" refers to the boundary of the nodule on the ultrasound image; accurate identification of its outline is the basis for quantifying nodule characteristics. "Quantification" refers to the numerical evaluation of the nodule's morphology (such as size, shape, and edge regularity) and internal structure (such as echo homogeneity and presence of calcification).

[0015] There are several methods for acquiring ultrasound images containing thyroid nodules. For example, one can connect to an ultrasound diagnostic device to receive and store digital image data generated in real time by the ultrasound probe scanning the thyroid region. This image data can be in DICOM (Digital Imaging and Communication in Medicine) format or other common image formats such as JPEG and PNG. Another method is to import stored thyroid ultrasound image files from an existing medical imaging database. These files are usually pre-anonymized to protect patient privacy.

[0016] Various techniques can be employed to perform multi-scale image analysis and processing on ultrasound images to construct multi-scale image representations. One approach utilizes image pyramid structures such as Gaussian pyramids or Laplacian pyramids. Specifically, by performing continuous Gaussian smoothing and downsampling operations on the original ultrasound image, a series of images at different resolutions are generated, forming an image pyramid. Each pyramid level represents the image representation at different scales, with higher pyramid levels containing macroscopic information and lower pyramid levels containing detailed information. Another approach uses wavelet transform. By performing multi-level wavelet decomposition on the ultrasound image, it can be decomposed into different frequency sub-bands, each representing the image's features at a specific scale and direction. For example, Daubechies wavelets or Haar wavelets can be used to decompose the image, obtaining low-frequency components (approximate image) and high-frequency components (detail information), which together constitute the multi-scale representation of the image.

[0017] This study utilizes multi-scale image representation to identify hypoechoic regions in ultrasound images. By combining feature analysis, judgment correction, and spatial relationship verification, it identifies acoustic shadowing regions within the ultrasound images. When identifying hypoechoic regions, a global threshold-based method can be employed. For example, a fixed grayscale threshold can be set, marking all pixels with grayscale values ​​below this threshold as hypoechoic pixels. These pixels can then undergo connected component analysis, clustering spatially consecutive hypoechoic pixels into independent connected regions. In determining acoustic shadowing regions, all potential hypoechoic regions in the image can be identified first. Then, preliminary feature analysis can be performed on each hypoechoic region, such as calculating its average grayscale value, area, and shape. Next, based on these preliminary features and combined with pre-defined rules or models, the acoustic shadowing probability of each hypoechoic region can be assessed. For example, if a hypoechoic region has extremely low grayscale values ​​and an irregular shape, it is more likely to be preliminarily identified as an acoustic shadowing region.

[0018] Suppression processing is applied to the sound-shadow region. This suppression includes unifying the signal within the sound-shadow region and softening its boundaries to obtain an optimized image. Unifying the signal within the sound-shadow region can be achieved using region filling techniques. For example, the grayscale values ​​of all pixels within the identified sound-shadow region can be uniformly set to the average grayscale value of the surrounding non-sound-shadow tissue, or to a preset intermediate grayscale value. This eliminates artifacts caused by signal loss or extremely low grayscale levels within the sound-shadow region. Softening the boundaries of the sound-shadow region can employ image smoothing filtering techniques. For example, Gaussian smoothing can be applied to the boundary pixels of the sound-shadow region and the surrounding non-sound-shadow pixels, resulting in a smoother and more natural transition of grayscale values ​​at the boundaries, thereby eliminating sharp artifacts at the sound-shadow boundaries.

[0019] Based on an optimized image, the contours of thyroid nodules can be identified. One approach is edge detection-based methods. For example, the Canny or Sobel operator can be applied to the optimized image to obtain an edge map. Then, by connecting edge points or using morphological operations, the nodule's contour can be initially outlined. Another approach is to use a region growing method. Starting from a seed point inside the nodule, the region grows outwards based on pixel similarity (e.g., grayscale values, texture features) until dissimilar pixels are encountered or a preset stopping condition is met, thus identifying the complete region of the nodule and extracting its boundaries as the contour.

[0020] Based on the contours of thyroid nodules, the morphology and internal structure of thyroid nodules are quantified. When quantifying the morphology of thyroid nodules, geometric parameters such as area, perimeter, aspect ratio, roundness, and regularity can be calculated. For example, the area can be obtained by calculating the number of pixels enclosed by the contour, and the perimeter can be obtained by calculating the sum of the distances between adjacent pixels on the contour. When quantifying the internal structure of thyroid nodules, gray-level histograms and texture features of the regions within the contour can be analyzed. For example, the average gray value, gray-level standard deviation, entropy, and contrast within the nodule can be calculated to assess its echo uniformity and texture complexity.

[0021] This application proposes a method for extracting features from ultrasound images of thyroid nodules. Its core innovation lies in the introduction of precise identification and suppression of acoustic shadowing regions. Traditional methods, when processing ultrasound images of thyroid nodules containing calcifications, often suffer from inaccurate nodule contour segmentation due to acoustic shadowing artifacts, thus affecting subsequent quantification of morphological and internal structural features. For example, while existing multi-scale filtering techniques can sharpen the overall edge of the nodule to some extent, they cannot effectively distinguish between the true boundary of the nodule and the pseudo-boundary caused by acoustic shadowing. The extremely low grayscale value of the acoustic shadowing region contrasts sharply with the internal tissue of the nodule; this strong edge information can mislead the adaptive segmentation process. When searching for the path with the lowest energy or strongest gradient, the segmentation process easily misjudges the edge of the acoustic shadowing as part of the true boundary of the nodule, resulting in a significant deviation in the final nodule contour.

[0022] This application further proposes the following steps for identifying hypoechoic regions in ultrasound images using multi-scale image representation, and determining acoustic shadowing regions in ultrasound images by combining feature analysis, judgment correction, and spatial relationship verification: In multi-scale image representation, low-echo regions are identified by gray-scale thresholding. For each hypoechoic region, the brightness distribution features and morphological features of the hypoechoic region are calculated at the image level, and high-brightness calcified regions are identified. The first acoustic shadow determination value of the hypoechoic region is obtained. The high-brightness calcified region is composed of multiple high-brightness calcified points. High-brightness calcified points refer to points in the ultrasound image whose gray value is higher than a preset high threshold. The potential acoustic shadowing regions identified at the image level and their surrounding environment are mapped to a multi-scale image representation, and fine feature analysis is performed to obtain the second acoustic shadowing determination value of the low echo region. The fine feature analysis includes brightness histogram morphology analysis and gradient direction consistency analysis. The potential acoustic shadowing region is a region with acoustic shadowing artifact characteristics. When the difference between the first sound shadow judgment value and the second sound shadow judgment value is greater than the preset difference threshold, the iterative correction process is triggered to recalculate the first sound shadow judgment value and the second sound shadow judgment value until the difference between the first sound shadow judgment value and the second sound shadow judgment value is less than the preset difference threshold or the maximum number of iterations is reached. After iterative correction, the first sound shadow judgment value and the second sound shadow judgment value are weighted and fused to obtain the comprehensive sound shadow judgment value of the low echo region. Before determining the acoustic shadow region, the spatial relationship of the low echo region is verified to confirm whether there are high-brightness calcification points in front of the ultrasonic wave propagation path of the low echo region, and to verify whether the low echo region is located in the reflection area of ​​the ultrasonic wave propagation path of the high-brightness calcification point. Based on the comprehensive sound and shadow judgment value and the results of spatial relationship verification, the sound and shadow area is determined.

[0023] Specifically, in multi-scale image representation, identifying low-echo regions through grayscale thresholding involves setting one or more grayscale thresholds at different image scales. Pixels with grayscale values ​​below these thresholds are identified as low-echo pixels, thus forming preliminary low-echo regions. The purpose is to initially filter out areas in the image that may contain sound shadows.

[0024] Specifically, for each hypoechoic region, the brightness distribution characteristics and morphological features of the hypoechoic region are calculated at the image level, and high-brightness calcified regions are identified, resulting in a first acoustic shadowing determination value for the hypoechoic region. Brightness distribution characteristics may include average brightness value, brightness standard deviation, etc., used to characterize the overall brightness and uniformity of the region. Morphological features may include area, perimeter, roundness, aspect ratio, etc., used to describe the geometry of the region. High-brightness calcified regions refer to regions composed of multiple high-brightness calcified points. High-brightness calcified points are points in the ultrasound image with grayscale values ​​higher than a preset high threshold; these points are usually the physical cause of acoustic shadowing. The first acoustic shadowing determination value is a preliminary quantitative assessment of whether a hypoechoic region is an acoustic shadowing based on these macroscopic features.

[0025] In practical applications, potential acoustic shadowing regions identified at the image level and their surrounding environment are mapped to a multi-scale image representation for fine feature analysis, resulting in a second acoustic shadowing determination value for low-echo regions. Potential acoustic shadowing regions refer to areas exhibiting acoustic shadowing artifact characteristics. Fine feature analysis includes brightness histogram morphology analysis and gradient direction consistency analysis. Brightness histogram morphology analysis assesses the distribution pattern of pixel brightness within the region, such as the presence of bimodal or skewed distributions, which helps distinguish between acoustic shadowing and non-acoustic structures. Gradient direction consistency analysis assesses the consistency of the gradient direction of pixel brightness changes within the region; acoustic shadowing regions typically exhibit consistent gradient directions, while other structures may exhibit more complex gradient patterns. The second acoustic shadowing determination value provides a more in-depth quantitative assessment of whether a low-echo region is an acoustic shadowing region based on these microscopic and fine features.

[0026] Furthermore, when the difference between the first and second sound shadow judgment values ​​exceeds a preset difference threshold, an iterative correction process is triggered. This indicates a significant discrepancy between the judgment results at two different levels, necessitating adjustments to improve accuracy. During the iterative correction process, the first and second sound shadow judgment values ​​can be recalculated, for example, by adjusting feature weights, thresholds, or resampling regions, until the difference between the two judgment values ​​is less than the preset difference threshold or the maximum number of iterations is reached. The aim is to improve the robustness of the judgment by repeatedly optimizing the results at different levels to achieve consistency.

[0027] After iterative correction, the first and second sound shadow judgment values ​​are weighted and fused to obtain a comprehensive sound shadow judgment value for the low-echo region. The weighted fusion can assign different weights to the two judgment values ​​based on their reliability or importance, thereby obtaining a more comprehensive and reliable comprehensive sound shadow judgment value.

[0028] Furthermore, before determining the acoustic shadow region, the spatial relationship of the low-echo region is verified. This includes confirming whether there are high-brightness calcification points ahead of the ultrasonic wave propagation path of the low-echo region, and verifying whether the low-echo region is located in the reflection area of ​​the ultrasonic wave propagation path of the high-brightness calcification point. The physical mechanism of acoustic shadows is that ultrasonic waves are blocked or reflected after encountering high-echo structures (such as calcification points), forming a low-echo region behind them. Therefore, this spatial relationship verification is a crucial step in confirming the authenticity of the acoustic shadow.

[0029] Finally, the sound shadow region is determined based on the comprehensive sound shadow judgment value and the results of spatial relationship verification. Only when the comprehensive judgment value supports the existence of sound shadow and the spatial relationship verification also conforms to the physical formation mechanism of sound shadow, is the low echo region finally determined as the sound shadow region.

[0030] This application's solution effectively addresses the accuracy limitations of traditional methods in sound and shadow recognition by introducing a comprehensive judgment mechanism involving multi-level, iterative, and spatial verification. First, low-echo regions are initially identified in multi-scale image representations through grayscale thresholding, providing a foundation for subsequent analysis. Second, brightness distribution and morphological features are calculated at the image level, and high-brightness calcified regions are identified, resulting in a first sound and shadow judgment value. This provides a macroscopic, physically based preliminary assessment. Simultaneously, potential sound and shadow regions are mapped to multi-scale image representations for refined feature analysis, including brightness histogram morphology analysis and gradient direction consistency analysis, yielding a second sound and shadow judgment value. This provides a microscopic, in-depth judgment based on image texture and structural changes. When significant differences exist between these two judgment values ​​at different levels, an iterative correction process is triggered. Through repeated adjustments and optimizations, the judgment results are made more consistent, thereby enhancing the robustness of the judgment. Subsequently, the corrected judgment values ​​are weighted and fused to obtain a more reliable comprehensive sound and shadow judgment value. More importantly, before finally determining the sound shadow region, spatial relationship verification was introduced. By confirming the physical association between the low-echo region and the high-brightness calcification point along the ultrasonic wave propagation path, the authenticity of the sound shadow was fundamentally verified. This series of steps worked together, progressing step by step, to ensure the accuracy and reliability of sound shadow recognition.

[0031] In some preferred embodiments, suppose an ultrasound image containing a thyroid nodule is acquired, in which a hypoechoic region suspected of acoustic shadowing exists. First, multi-scale image analysis processing is performed on the ultrasound image to construct its multi-scale image representation. In this multi-scale image representation, the hypoechoic region is initially identified through grayscale thresholding. For this hypoechoic region, its brightness distribution characteristics, such as average brightness value and brightness standard deviation, as well as its morphological characteristics, such as area and roundness, are calculated at the image level. Simultaneously, a high-brightness calcification point is identified in front of the ultrasound propagation path of the hypoechoic region. Based on these characteristics, the first acoustic shadowing judgment value of the hypoechoic region is calculated to be 0.7 (indicating a high probability of acoustic shadowing).

[0032] Subsequently, the hypoechoic region and its surrounding environment were mapped onto a multi-scale image representation for detailed feature analysis. For example, brightness histogram morphology analysis revealed that its brightness distribution exhibited typical sound shadow characteristics, and gradient direction consistency analysis also showed high directional consistency. Based on these detailed features, the second sound shadow determination value for the hypoechoic region was calculated to be 0.4 (indicating a low probability of sound shadow).

[0033] At this point, because the difference of 0.3 between the first acoustic shadowing judgment value (0.7) and the second acoustic shadowing judgment value (0.4) is greater than the preset difference threshold of 0.1, the system triggers an iterative correction process. During the iteration process, the system may adjust the feature weights or thresholds used to calculate the judgment values. For example, it may re-evaluate the weights of the brightness histogram morphology analysis to make it more focused on distinguishing acoustic shadowing from cystic lesions. After several iterations, the first acoustic shadowing judgment value may be adjusted to 0.6, and the second acoustic shadowing judgment value may be adjusted to 0.55. At this point, the difference of 0.05 is less than the difference threshold of 0.1, and the iteration stops.

[0034] After iterative correction, the adjusted first sound shadow judgment value of 0.6 and the second sound shadow judgment value of 0.55 are weighted and fused. For example, the first sound shadow judgment value of 0.4 and the second sound shadow judgment value of 0.6 are given a weight, resulting in a comprehensive sound shadow judgment value of 0.6*0.4+0.55*0.6=0.24+0.33=0.57.

[0035] Before finally determining the acoustic shadow region, spatial relationship verification was performed. The system confirmed that the previously identified high-brightness calcification point did indeed exist ahead of the ultrasonic wave propagation path of the low-echo region, and that the low-echo region was located in the reflection area of ​​the ultrasonic wave propagation path of the calcification point.

[0036] Ultimately, based on the sound-shadow comprehensive judgment value of 0.57 (assuming the preset sound-shadow judgment threshold is 0.5) and the results of spatial relationship verification (confirming the existence of a physical connection), the low-echo area was finally identified as a sound-shadow area. Through this series of refined and verified steps, the accuracy of sound-shadow recognition was ensured, avoiding misjudging other low-echo structures as sound shadows.

[0037] This application further proposes steps for identifying low-echo regions, including: The ultrasound image is segmented by grayscale threshold to identify all low-echo pixels and form a set of low-echo pixels, where low-echo pixels are pixels with grayscale values ​​lower than a preset grayscale threshold. Connectivity analysis is performed on the set of low-echo pixels to cluster spatially continuous low-echo pixels into independent connected regions. Evaluate the area of ​​each connected region. If the area of ​​a connected region is less than a preset minimum area threshold, exclude the corresponding connected region. If the area of ​​a connected region is greater than the minimum area threshold, calculate the distance between the geometric center of the corresponding connected region and the adjacent connected regions. When the distance between geometric centers is less than a preset merging distance threshold, and the projections of connected regions on the ultrasound image overlap, the boundaries of the spatially overlapping connected regions are divided according to the gray-scale change trend and morphological characteristics of the connected regions in order to identify mutually independent low-echo regions.

[0038] Specifically, grayscale thresholding refers to dividing pixels in an ultrasound image into hypoechoic pixels and non-hypoechoic pixels by setting a preset grayscale threshold. Hypoechoic pixels are defined as pixels with grayscale values ​​below the preset grayscale threshold; these pixels are collected to form a hypoechoic pixel set. The purpose of this step is to initially screen out pixels in the image that may represent hypoechoic regions.

[0039] Furthermore, connected component analysis is performed on the resulting set of low-echo pixels. Connected component analysis aims to identify and cluster spatially continuous low-echo pixels, thereby organizing them into independent connected regions. This process helps transform discrete pixels into candidate regions with a certain spatial range and shape.

[0040] Based on this, the area of ​​each identified connected region is evaluated. When the area of ​​a connected region is less than a preset minimum area threshold, the region is considered noise or an insignificant microstructure and is excluded. The purpose of this step is to filter out small noise points or artifacts in the image, ensuring that the regions processed subsequently have sufficient clinical significance. When the area of ​​a connected region is greater than the minimum area threshold, the distance between the geometric centers of the connected region and its adjacent connected regions is calculated, providing a basis for subsequent region merging or boundary delineation.

[0041] In a preferred embodiment, when the distance between the geometric centers of two connected regions is less than a preset merging distance threshold, and their projections on the ultrasound image overlap, it indicates that the two regions may belong to the same hypoechoic structure, or they are two closely adjacent but independent hypoechoic structures. In this case, to accurately identify the independent hypoechoic regions, it is necessary to perform fine boundary delineation of these spatially overlapping connected regions based on their grayscale variation trends and morphological characteristics. For example, the optimal segmentation line can be determined by analyzing the grayscale gradient, local texture changes, or shape features within the overlapping regions, thereby accurately separating them into independent hypoechoic regions.

[0042] This application's solution effectively addresses the fragmentation, incompleteness, or blurred boundaries that can occur in traditional simple grayscale thresholding when identifying hypoechoic regions by introducing connected component analysis, area evaluation, and boundary segmentation based on grayscale variation trends and morphological features. First, grayscale thresholding can initially identify all potential hypoechoic pixels. Second, connected component analysis clusters these discrete pixels into meaningful connected components, avoiding fragmentation. Subsequently, by evaluating the area of ​​connected components and excluding excessively small regions, noise and artifacts can be effectively filtered out, improving the accuracy of identification. More importantly, for connected components with close geometric center distances and overlapping projections, this solution no longer simply merges them or treats them as a single region, but further utilizes the grayscale variation trends and morphological features of the connected components for refined boundary segmentation. This segmentation mechanism can identify closely adjacent or partially overlapping independent hypoechoic structures, such as two very close but actually independent nodules or lesions, thereby avoiding misjudgment or omission and ensuring the integrity and independence of hypoechoic region identification.

[0043] In some preferred embodiments, it is assumed that in the acquired thyroid ultrasound image, there are two closely adjacent hypoechoic nodules, which appear as two regions with low gray values ​​on the image, and due to the characteristics of ultrasound imaging, the boundary between these two regions may be slightly blurred or overlapped.

[0044] First, grayscale thresholding is performed on the ultrasound image to identify all pixels with grayscale values ​​below a preset threshold, forming a set of low-echo pixels. Next, connected component analysis is performed on this pixel set to preliminarily identify two or more connected regions, which may contain regions representing the two nodules as well as some minor noise regions.

[0045] Subsequently, the areas of these connected regions are evaluated. Regions with areas smaller than a preset minimum area threshold are excluded to filter out noise. For connected regions with areas larger than the minimum area threshold, such as regions representing two nodes, their respective geometric centers are calculated. This is based on the assumption that the distance between the geometric centers of the connected regions corresponding to these two nodes is less than a preset merging distance threshold, and that their projections on the image overlap.

[0046] At this point, this approach does not simply merge them into one region, but further analyzes the grayscale variation trends and morphological features of these two overlapping connected regions. For example, the system may detect a local inflection point where the grayscale value increases or decreases at the boundary between the two regions, or the two regions may have significantly different morphological contour features. Based on these features, the system performs precise boundary delineation at the inflection point or feature difference, thus successfully identifying these two closely adjacent but independent hypoechoic nodules as two independent hypoechoic regions. In this way, even under complex or blurry image conditions, the accuracy and independence of hypoechoic region identification can be ensured.

[0047] This application further proposes a step for calculating the brightness distribution features and morphological features of each low-echo region at the image layer level, identifying high-brightness calcified regions, and deriving the first acoustic shadowing value of the low-echo region, including: For each hypoecho region, calculate the average brightness value, brightness standard deviation, morphological characteristics of the hypoecho region, and determine whether there is a high-brightness calcified region in the upstream direction of the hypoecho region, where the upstream direction is the reverse path relative to the direction of ultrasonic wave propagation. The average brightness value and standard deviation of the low-echo region were analyzed to obtain the macroscopic brightness distribution of the low-echo region. The macroscopic brightness distribution is used to characterize the overall brightness level and brightness dispersion of the low-echo region at the image level. When the parameters corresponding to the macroscopic brightness distribution and morphological characteristics of multiple low-echo regions meet the same preset range, brightness profile analysis is performed on multiple low-echo regions to obtain the brightness change curve along the ultrasonic wave propagation direction. Based on the brightness change curve, identify whether there is a pattern of brightness decreasing sharply and then maintaining low brightness, and whether there is a pattern of brightness decreasing slowly and then maintaining medium brightness. Here, a sharp decrease in brightness is defined as a brightness change rate greater than a preset first brightness change threshold, and a slow decrease in brightness is defined as a brightness change rate less than a preset second brightness change threshold. The first brightness change threshold is greater than the second brightness change threshold. A low brightness pattern is defined as a brightness lower than the preset first brightness threshold, and a medium brightness pattern is defined as a brightness range between the preset second brightness threshold and the preset third brightness threshold. The first brightness threshold is less than the second brightness threshold. When a pattern of maintaining low brightness after a sharp drop in brightness is identified, and there are high-brightness calcification points in front of the ultrasonic wave propagation path in the low-echo region, the sound shadow probability score of the low-echo region is increased. The sound shadow probability score of the low-echo region is used to quantify the probability that the low-echo region will become a sound shadow region. When a pattern of slow decrease in brightness followed by maintenance of moderate brightness is identified, and there are no high-brightness calcification points in front of the ultrasonic wave propagation path in the low-echo region, the acoustic shadow probability score of the low-echo region is reduced. Calculate the first sound shadow determination value for the low echo region based on the sound shadow probability score.

[0048] Specifically, for each hypoechoic region, its average luminance value and luminance standard deviation are first calculated. These parameters together constitute the macroscopic luminance distribution of the hypoechoic region, used to preliminarily assess the overall luminance and uniformity of the region. Morphological characteristics can include the region's area, perimeter, roundness, aspect ratio, etc. These characteristics help distinguish hypoechoic regions of different shapes and sizes. The upstream direction refers to the reverse path of the ultrasound waves emitted from the probe, passing through the tissue to reach the hypoechoic region. Determining whether a high-luminance calcified region exists in this direction is to verify the physical premise of acoustic shadow formation.

[0049] The analysis of macroscopic brightness distribution aims to provide a global perspective to understand the overall brightness performance of hypoechoic regions at the image level. When multiple hypoechoic regions exhibit similarities in macroscopic brightness distribution and morphological features, brightness profile analysis is further performed for more refined differentiation. Brightness profile analysis involves extracting one or more brightness variation curves along the direction of ultrasound propagation. These curves reflect the dynamic process of energy attenuation as ultrasound passes through tissue.

[0050] In practical applications, analyzing the brightness change curves can identify two typical patterns: one is a pattern where brightness drops sharply and then remains low. This is usually related to the ultrasound waves being completely blocked by high-brightness calcification points in front, resulting in almost no echo signal in the area behind, manifesting as a sound shadow. A sharp drop in brightness means the rate of brightness change exceeds a preset first brightness change threshold, while the low brightness pattern means the brightness is below the preset first brightness threshold. The other pattern is a pattern where brightness drops slowly and then remains at a moderate level. This may indicate that the ultrasound waves are only partially attenuated or pass through other non-sound shadow structures, and there is still some echo signal in the area behind. A slow drop in brightness means the rate of brightness change is below a preset second brightness change threshold, while the moderate brightness pattern means the brightness range is between the preset second and third brightness thresholds. It is important to note that the first brightness change threshold is greater than the second brightness change threshold, and the first brightness threshold is less than the second brightness threshold.

[0051] This application's solution employs a multi-dimensional assessment of the sound shadow probability of low-echo regions by combining macroscopic brightness distribution, morphological characteristics, brightness profile analysis, and spatial relationship with high-brightness calcification points. When the brightness profile shows a sharp decrease and maintenance of low brightness, and high-brightness calcification points are present in front of the ultrasonic wave propagation path, this strongly indicates that the region is a sound shadow, thus its sound shadow probability score is increased. Conversely, if the brightness decreases slowly and maintains moderate brightness, and there are no high-brightness calcification points in front, the probability of the region being a sound shadow is low, and its sound shadow probability score is decreased. This comprehensive judgment mechanism makes the calculation of the first sound shadow determination value more accurate and reliable.

[0052] In some preferred embodiments, it is assumed that multiple hypoechoic regions are identified in the ultrasound image. First, for each hypoechoic region, its average luminance value, luminance standard deviation, and morphological features (e.g., area, roundness) are calculated. Simultaneously, the system checks for the presence of a high-luminance calcification region upstream of each hypoechoic region. For example, for a hypoechoic region A, its average luminance value is low, its luminance standard deviation is small, and a high-luminance calcification point is detected upstream. In this case, the system compares its macroscopic luminance distribution and morphological features with those of other hypoechoic regions. If multiple hypoechoic regions (e.g., regions A, B, and C) are found to have macroscopic luminance distribution and morphological feature parameters that meet a preset acoustic shadowing feature range, luminance profile analysis is performed on these regions. By extracting luminance change curves along the ultrasound propagation direction, it is found that the luminance curve of region A exhibits a pattern of sharp luminance decrease (e.g., the rate of luminance change exceeds a first luminance change threshold) followed by a period of extremely low luminance (below the first luminance threshold). Combined with the previously detected upstream high-luminance calcification point, the system significantly improves the acoustic shadowing probability score of region A. For region B, its brightness curve may show a pattern of slow brightness decrease (the rate of brightness change is less than the second brightness change threshold) followed by a sustained brightness level (between the second and third brightness thresholds), and no high-brightness calcification points are detected upstream. In this case, the system will lower the sound shadow probability score for region B. Finally, based on these adjusted sound shadow probability scores, a first sound shadow determination value is calculated for each low-echo region, thus providing a more accurate preliminary judgment for subsequent sound shadow region identification.

[0053] This application further proposes a step-by-step approach to map potential acoustic shadowing regions and their surrounding environment identified at the image level to a multi-scale image representation, perform fine feature analysis, and derive a second acoustic shadowing determination value for low-echo regions, including: A luminance histogram morphology analysis was performed on the potential sound shadow area and its surrounding environment to obtain the luminance histogram morphology analysis results; Gradient direction consistency analysis was performed on the potential sound shadow area and its surrounding environment to obtain the gradient direction consistency analysis results; When the brightness histogram morphology analysis results and gradient direction consistency analysis results show that the similarity between the potential acoustic shadow region and the adjacent non-acoustic shadow low echo structure meets the preset similarity threshold, multiple boundary brightness sampling lines are extracted along the blurred boundary between the potential acoustic shadow region and the adjacent non-acoustic shadow low echo structure. The adjacent non-acoustic shadow low echo structure refers to the low echo region adjacent to the potential acoustic shadow region in the ultrasound image, and the boundary brightness sampling line refers to the brightness change curve extracted along the set direction at the position of the blurred boundary. For the boundary brightness sampling line, calculate the brightness change rate in its extension direction, and calculate the second derivative of the brightness change rate to identify the inflection point of brightness change; For the boundary brightness sampling line, the texture complexity on both sides is calculated to obtain the texture abrupt change point. Here, texture complexity is a quantitative parameter used to characterize the degree of texture change in a local area, and texture abrupt change point refers to the sampling point where the difference in texture complexity on both sides of the boundary brightness sampling line changes abruptly. When the brightness change inflection point and the texture change abrupt point are spatially consistent, and the region on one side of the brightness change inflection point has the characteristics of sound shadow, while the region on the other side has the characteristics of non-sound shadow low echo structure, boundary division is performed at the brightness change inflection point to obtain the boundary division result. Based on the boundary division results, calculate the second acoustic shadow determination value for the low echo region.

[0054] Specifically, luminance histogram morphology analysis aims to characterize the overall luminance characteristics of potential sound-shadow areas and their surrounding environment by analyzing the luminance distribution features of pixels within a region, such as peaks, valleys, skewness, and kurtosis. Its purpose is to preliminarily assess the luminance uniformity and distribution pattern of the region. Gradient direction consistency analysis, on the other hand, calculates the gradient vectors of pixels within a region and analyzes the direction consistency of these gradient vectors to reflect the edge strength and directionality of the region. Its purpose is to identify the internal structure and boundary sharpness of the region.

[0055] Specifically, when the above two analysis results indicate a high similarity between the potential shadow region and the adjacent non-shadow low-echo structure, and this similarity meets a preset similarity threshold, it suggests that the boundary between them may be somewhat blurred, requiring more refined analysis. In this case, the system extracts multiple boundary brightness sampling lines along this blurred boundary. A boundary brightness sampling line can be understood as collecting a series of pixel brightness values ​​along a specific path in the transition region between the potential shadow region and the adjacent non-shadow low-echo structure, forming a brightness variation curve.

[0056] In practical applications, for each boundary brightness sampling line, the rate of change of brightness along its extension direction is calculated, i.e., the rate at which the brightness value changes with spatial location. Further, the second derivative of the rate of change of brightness is calculated, and by analyzing the zero-crossing points or extreme points of the second derivative, inflection points of brightness change can be identified. Brightness change inflection points typically correspond to locations where the brightness change trend in the image changes significantly, such as the transition from a gradual change to a sharp change, or from a decreasing trend to an increasing trend. Simultaneously, the texture complexity on both sides of the boundary brightness sampling line is also calculated. Texture complexity is a parameter that quantifies the texture detail and degree of change within a local region, and can be calculated, for example, using methods such as the gray-level co-occurrence matrix (GLCM) and local binary mode. Texture abrupt change points refer to locations where the difference in texture complexity on both sides of the boundary brightness sampling line changes significantly, which usually indicates the boundary between different texture regions.

[0057] When the inflection point of brightness change and the abrupt change point of texture are highly consistent in spatial location, this provides strong evidence that the location is the true boundary between the potential shadow region and the adjacent non-shadow hypoechoic structure. Furthermore, if the region on one side of the inflection point or abrupt change point exhibits characteristics unique to shadows, such as low brightness and low texture complexity, while the region on the other side exhibits brightness distribution and texture characteristics unique to non-shadow hypoechoic structures, then precise boundary delineation can be performed at that location. Therefore, based on the final boundary delineation result, the second shadow determination value of the hypoechoic region can be calculated more accurately.

[0058] This application's scheme performs a preliminary assessment of potential sound-shadow regions and their surrounding environment by combining brightness histogram morphology analysis and gradient direction consistency analysis. When these analyses indicate the existence of a fuzzy boundary between the potential sound-shadow region and adjacent non-sound-shadow low-echo structures, the scheme no longer relies solely on macroscopic features but delves deeper into the microscopic level. By extracting boundary brightness sampling lines and performing second-order derivative analysis of their brightness change rates, it accurately captures the inflection points of brightness changes. Simultaneously, by calculating the texture complexity on both sides of the boundary and identifying texture abrupt change points, the scheme can verify the existence of the boundary from another dimension. It is precisely because of the high spatial consistency between the brightness change inflection points and texture abrupt change points, as well as the clear distinction of the features of the regions on both sides, that the system can overcome the limitations of traditional methods in accurately delineating fuzzy boundaries, thereby achieving precise boundary delineation of potential sound-shadow regions. This multi-dimensional and refined analysis method effectively improves the accuracy and reliability of the second sound-shadow determination value.

[0059] In some preferred embodiments, it is assumed that when processing a thyroid ultrasound image, the system initially identifies a potential acoustic shadowing region adjacent to a cystic structure that also appears as hypoechoic. At the image level, the macroscopic brightness distribution and gradient direction consistency analysis results of the two regions show a high degree of similarity, making their boundary appear blurred.

[0060] At this point, the proposed solution is activated. First, a brightness histogram morphology analysis and gradient direction consistency analysis are performed on the potential sound-shadow region and its surrounding environment. If the analysis results indicate that the similarity between the two meets a preset similarity threshold, the system will extract multiple boundary brightness sampling lines along the fuzzy transition zone between the two regions.

[0061] For these sampling lines, the system calculates the rate of change of brightness along their extension direction and performs second derivative calculations to identify the inflection points of brightness change. For example, on a certain sampling line, the brightness drops sharply from the relatively stable brightness of the cyst region and maintains extremely low brightness in the shadow region after a certain point; this point is identified as the inflection point of brightness change.

[0062] Simultaneously, the system also calculates the texture complexity on both sides of these sampling lines. For example, cystic regions may exhibit a relatively uniform texture, while acoustic shadow regions, due to their artifact characteristics, may have a smoother texture or a specific noise pattern. When the difference in texture complexity on both sides changes abruptly, that point is identified as a texture abrupt change point.

[0063] If the brightness change inflection point and the texture abrupt change point are highly consistent in spatial location, and the regional characteristics on one side of this point (such as extremely low brightness and smooth texture) conform to the typical characteristics of a sound shadow, while the regional characteristics on the other side (such as medium-low brightness and uniform texture) conform to the typical characteristics of a cyst, then the system will perform precise boundary delineation at this consistent point. Through this refined boundary delineation, potential sound shadow regions can be accurately distinguished from adjacent cystic structures, thereby calculating a more accurate second sound shadow determination value.

[0064] When the difference between the first sound shadow judgment value and the second sound shadow judgment value is still greater than the difference threshold after reaching the maximum number of iterations, the method further includes the following steps: When the difference between the first and second sound shadow judgment values ​​is still greater than the difference threshold after reaching the maximum number of iterations, the low echo region is marked as the region to be reviewed. The original pixel data of the ultrasound image of the area to be reviewed is extracted, and a comprehensive judgment is made by combining the analysis results corresponding to the first and second acoustic shadow judgment values ​​to generate a feature set containing multi-dimensional information. The feature set of multi-dimensional information includes morphological features, brightness features and gradient direction features. Analyze the feature set of multi-dimensional information, identify whether there are contradictory feature manifestations, and obtain contradictory features; When contradictory features are identified, the sound and shadow judgment values ​​of the area to be reviewed are dynamically adjusted according to the type of contradictory features, providing a reference for manual review.

[0065] Specifically, after the iterative correction process reaches the maximum number of iterations, if the difference between the first and second sound shadow judgment values ​​still exceeds the preset difference threshold, it indicates that the sound shadow attributes of the low-echo region have a high degree of uncertainty. At this time, the low-echo region will be automatically marked by the system as a region to be reviewed, indicating that further manual intervention or deeper analysis is needed.

[0066] Furthermore, to assist in judging these areas to be reviewed, the system extracts the raw pixel data of the ultrasound image of the area. The raw pixel data contains the most basic and comprehensive image information, avoiding information loss that may be introduced during preprocessing or feature extraction. Simultaneously, it combines the analysis results corresponding to the previously calculated first and second acoustic shadow judgment values, such as brightness distribution characteristics, morphological features, brightness histogram morphological analysis results, and gradient direction consistency analysis results, to make a comprehensive judgment. Through this comprehensive judgment, a feature set containing multi-dimensional information can be generated. This multi-dimensional feature set aims to comprehensively characterize various visual and structural characteristics of the area to be reviewed, specifically including morphological features (such as the shape, size, and edge smoothness of the area), brightness features (such as average brightness, brightness standard deviation, and brightness profile), and gradient direction features (such as edge direction and texture direction consistency).

[0067] Based on this, the system will conduct in-depth analysis of the feature set of this multi-dimensional information to identify whether there are contradictory features. For example, a region may exhibit sound and shadow characteristics in the macroscopic brightness distribution, but be more similar to non-sound and shadow regions in the microscopic texture or gradient direction. This inconsistency is identified as contradictory features.

[0068] When contradictory features are identified, the system dynamically adjusts the acoustic shadowing determination value of the area to be reviewed based on the type of contradictory feature. For example, if the brightness feature strongly points to acoustic shadowing, but the morphological feature is atypical, the system may add an uncertainty weight to the acoustic shadowing determination value, or adjust the determination value to an intermediate range based on the severity of the contradiction. The purpose of this dynamic adjustment is to provide more valuable reference for subsequent manual review, helping doctors or professionals to more accurately understand the complexity of the area and make a final judgment.

[0069] This application's solution effectively addresses the problem that traditional iterative methods may not fully converge or have insufficient judgment ability in certain ambiguous areas in complex ultrasound images by introducing a mechanism for marking and depth analysis of low-echo regions that still have uncertainty after iterative correction. Specifically, when the difference between the first and second acoustic shadow judgment values ​​still exceeds a preset difference threshold after reaching the maximum number of iterations, the system no longer forces an uncertain automatic judgment, but instead marks the low-echo region as a region to be reviewed, thereby avoiding potential misjudgments.

[0070] Furthermore, by extracting the original pixel data of the region to be verified and combining it with the analysis results corresponding to the existing first and second sound shadow judgment values, a feature set containing multi-dimensional information including morphological features, brightness features, and gradient direction features can be constructed. This process aims to re-examine the region from a more comprehensive and detailed perspective, compensating for the limitations that a single judgment value may have. Subsequently, this multi-dimensional feature set is analyzed to identify whether there are contradictory feature representations. For example, a region may conform to sound shadow characteristics in terms of brightness but be atypical in terms of morphology. Identifying such contradictory features is key to understanding the complexity of the region.

[0071] Finally, based on the identified types of contradictory features, the sound and shadow judgment values ​​for the areas to be reviewed are dynamically adjusted. This adjustment does not directly provide a final judgment, but rather provides richer and more instructive information for human review by quantifying uncertainties or highlighting contradictory points. Thus, this solution effectively transfers complex situations that are difficult for machines to accurately judge to human decision-makers, significantly improving the accuracy and reliability of sound and shadow recognition.

[0072] In some preferred embodiments, assuming that during acoustic shadowing region identification of a thyroid ultrasound image, a hypoechoic region has been corrected after the maximum number of iterations, but the difference (0.45) between its first acoustic shadowing determination value (e.g., 0.75) and second acoustic shadowing determination value (e.g., 0.30) is still greater than a preset difference threshold (e.g., 0.20), then the hypoechoic region will be automatically marked by the system as a region to be reviewed.

[0073] Subsequently, the system extracts the original ultrasound pixel data of the area to be reviewed and combines it with previously calculated brightness distribution features, morphological features, brightness histogram morphological analysis results, and gradient direction consistency analysis results to generate a feature set containing multi-dimensional information. For example, this feature set may show that the average brightness of the area is low, which is consistent with the macroscopic features of sound shadow (brightness features), but its edge morphology is relatively regular and does not have the blurry or irregular edges of typical sound shadow (morphological features). At the same time, its internal gradient direction consistency is also high, which does not match the messy texture of typical sound shadow (gradient direction features).

[0074] By analyzing this feature set, the system identifies contradictions between brightness features and morphological and gradient direction features: brightness indicates sound shadow, but morphology and texture do not fully support it. Based on the type of these contradictory features, the system dynamically adjusts the sound shadow determination value for the area to be reviewed. For example, the system might adjust the sound shadow determination value to an intermediate value (such as 0.5) and add a note stating "Brightness features support sound shadow, but morphological and texture features contradict each other," thus providing doctors with a more balanced and cautionary reference, prompting them to pay special attention to these contradictory points during the final judgment and conduct manual review.

[0075] The above-mentioned method for extracting features from ultrasound images of thyroid nodules, after dynamically adjusting the acoustic shadowing determination value of the area to be reviewed, highlights the area to be reviewed on the ultrasound image, and includes the dynamically adjusted acoustic shadowing determination value and contradictory features of the area to be reviewed, prompting the doctor to perform manual review.

[0076] Specifically, highlighting areas requiring review refers to visually identifying areas on ultrasound images that require special attention from the physician. This may include, but is not limited to, drawing colored outlines around the boundaries of the areas to be reviewed, or applying a semi-transparent color overlay to the area to visually distinguish it from surrounding normal tissue or treated areas. The aim is to guide the physician's attention, enabling them to quickly locate areas where the system's assessment is uncertain.

[0077] The accompanying dynamically adjusted acoustic shadowing value and contradictory features of the area to be reviewed can be understood as visually presenting the acoustic shadowing value obtained after complex analysis and adjustment by the system, along with the specific contradictory features that cause uncertainty in the system's judgment, in the vicinity of the area to be reviewed on the ultrasound image, in the form of text, charts, or floating prompts. For example, a label could be displayed next to the area to be reviewed, containing information such as "Acoustic shadowing value: 0.65" and "Contradictory features: Inconsistent brightness distribution and morphological features." The purpose is to provide doctors with the key information needed for decision-making, helping them understand the basis of the system's judgment and the source of uncertainty.

[0078] In practical applications, prompting doctors for manual review involves the system identifying the area to be reviewed and displaying relevant information. This is achieved through various means, such as sound alerts, flashing icons, pop-up dialog boxes, or explicit text instructions displayed on the user interface, proactively reminding doctors to conduct a manual review and final confirmation of that area. The aim is to ensure that all complex or uncertain cases requiring manual intervention are handled promptly and effectively, avoiding any omissions.

[0079] This application's solution effectively overcomes the limitations of the aforementioned basic solutions in information presentation by visually presenting the areas to be reviewed identified by the system and their related analysis results on the ultrasound image. Because the areas to be reviewed are visually highlighted, doctors can quickly and accurately locate the specific areas requiring attention, avoiding time-consuming searches in complex images. Simultaneously, by displaying dynamically adjusted acoustic shadowing judgment values ​​and contradictory features, doctors can directly obtain the quantitative basis for the system's judgment and the specific reasons for uncertainty. This allows doctors to not only see the results during manual review but also understand the underlying logic, enabling a more comprehensive and in-depth assessment of the area's nature. Furthermore, the mechanism that proactively prompts doctors to conduct manual review ensures that these critical cases requiring human intervention are not overlooked, thereby improving the rigor and safety of the diagnosis.

[0080] In some preferred embodiments, assuming that when processing a thyroid ultrasound image, during the process of identifying acoustic shadow regions, for a certain hypoechoic region, the difference between its first and second acoustic shadow determination values ​​still exceeds a preset difference threshold after reaching the maximum number of iterations, this hypoechoic region is marked as a region to be reviewed. The system further extracts the original pixel data of this region to be reviewed and performs a comprehensive analysis combining its morphological features, brightness features, and gradient direction features, identifying a contradictory relationship between the brightness distribution characteristics and morphological features of the region, i.e., contradictory features.

[0081] According to the scheme of this application, after the system dynamically adjusts the acoustic shadowing judgment value of the area to be reviewed (for example, adjusting it to 0.65), it will highlight the area to be reviewed on the ultrasound image with a red dashed outline. Simultaneously, an information box will pop up next to the area, clearly displaying "Acoustic Shadowing Judgment Value: 0.65" and "Contradictory Features: Brightness Distribution and Morphological Features are Inconsistent." Furthermore, a flashing "Please Manually Review" button or prompt message will appear in a prominent position on the image interface, reminding the doctor to examine the area within the red dashed box in detail. Through these intuitive visual and textual prompts, the doctor can quickly locate the complex area and, combined with the detailed analysis results provided by the system, make a final professional judgment, thereby avoiding potential misjudgments and improving the accuracy and efficiency of diagnosis.

[0082] The steps for highlighting the area to be reviewed on an ultrasound image include: Based on the spatial location and extent of the region to be reviewed in the ultrasound image, corresponding region identification information is generated; The region identification information is overlaid on the ultrasound image at the position corresponding to the region to be reviewed, in order to distinguish the region to be reviewed from the surrounding tissue region; Generate prompt information, which includes the dynamically adjusted sound and shadow judgment values ​​and contradictory characteristics of the area to be reviewed; The prompts are displayed in association with the corresponding areas to be reviewed, prompting doctors to conduct manual reviews.

[0083] The region identification information can be understood as a visual marker designed to clearly indicate the specific location and boundaries of the region to be verified on the ultrasound image. This information can be generated based on the pixel coordinates, geometry (e.g., polygon, ellipse, or rectangle) of the region to be verified, and its dimensions in the image. For example, a line that perfectly matches the outline of the region to be verified, or a semi-transparent color layer covering the region, can be generated.

[0084] Furthermore, the area identification information is overlaid on the ultrasound image at the position corresponding to the area to be examined, aiming to visually distinguish the area from the background or other normal tissue. This overlay display can be done in various ways, for example, by changing the color, brightness, transparency, or texture of the area identification information to give it sufficient contrast on the original ultrasound image, making it easy for doctors to identify.

[0085] Furthermore, the prompt information is a crucial component in providing doctors with additional diagnostic information. This prompt information not only includes the sound and shadow judgment values ​​of the area to be reviewed, dynamically adjusted by the system based on contradictory characteristics, but also details the contradictory characteristics that led to this adjustment. For example, if the contradictory characteristic is "inconsistency between brightness distribution and morphological characteristics," the prompt information will explicitly point this out.

[0086] Therefore, associating the prompts with the corresponding areas to be reviewed ensures that doctors can immediately access relevant judgment values ​​and contradictory characteristics when reviewing these areas, thus providing comprehensive reference information for manual review. This association can be achieved by displaying text boxes or pop-ups near the area identification information, or through color coding.

[0087] This application's solution transforms the spatial location and extent of the area to be reviewed into region identification information and overlays it onto the ultrasound image, enabling doctors to intuitively identify areas requiring special attention. Simultaneously, by generating and displaying prompts containing dynamically adjusted acoustic shadowing judgment values ​​and contradictory features, this application's solution provides doctors with detailed evidence of the system's internal judgment, including uncertainties encountered during the identification process (i.e., contradictory features) and adjustments made based on these uncertainties. This mechanism allows doctors, during manual review, not only to see the area to be reviewed but also to understand why the system marked it as requiring review and the system's latest assessment of its acoustic shadowing probability, thereby assisting doctors in making more accurate diagnoses.

[0088] The aforementioned area identification information is used to display the outline or area coverage of the area to be reviewed in ultrasound images.

[0089] In this context, area identification information can be understood as a visual cue, its purpose being to clearly mark areas on the ultrasound image that require manual review by the physician. Specifically, this display information can take several forms. For example, it can depict the boundaries of the area to be reviewed, forming an outline, i.e., "outline marking"; or it can fill the entire area to be reviewed with color, semi-transparent coverage, or texture overlay, creating a clear visual contrast with the surrounding normal tissue area, i.e., "area coverage marking." Both of these marking methods aim to intuitively guide the physician's attention, ensuring that the physician can quickly and accurately locate the areas identified by the system that require further confirmation.

[0090] This application's solution, by clearly defining the specific form of area identification information, makes the display of areas to be reviewed on ultrasound images more standardized and intuitive. When the system identifies areas requiring manual review, it effectively highlights these areas from the complex background of the ultrasound image by generating clear outline markers or area coverage markers. This clear visual distinction helps doctors quickly identify and understand which areas the system judges to have uncertain or contradictory characteristics, thus providing clear visual guidance for subsequent manual review.

[0091] refer to Figure 2 A thyroid nodule ultrasound image feature extraction system, applying the above-mentioned thyroid nodule ultrasound image feature extraction method, the system comprising: The acquisition module acquires ultrasound images containing thyroid nodules; The module performs multi-scale image analysis and processing on ultrasound images to construct a multi-scale image representation of ultrasound images. The acoustic shadow recognition module uses multi-scale image representation to identify low-echo regions in ultrasound images, and combines feature analysis, judgment correction and spatial relationship verification to determine the acoustic shadow regions in ultrasound images. The suppression module performs suppression processing on the sound shadow region. The suppression processing includes unifying the signal within the sound shadow region and softening the boundary of the sound shadow region to obtain an optimized image. The contour recognition module identifies the contours of thyroid nodules based on optimized images; The quantification module quantifies the morphology and internal structure of thyroid nodules based on their outlines.

[0092] Specifically, the acquisition module can be understood as an input interface or data acquisition unit in the system. Its purpose is to receive and acquire ultrasound image data containing thyroid nodules from external devices (such as ultrasound diagnostic instruments) or storage media (such as PACS systems). In practical applications, the acquisition module can be a hardware interface or a software interface for data transmission with the image source.

[0093] The construction module refers to performing multi-scale image analysis and processing on the acquired ultrasound images and constructing a multi-scale image representation of the ultrasound images. Multi-scale image representation can be understood as the way an image is presented at different resolutions or levels of detail. Its purpose is to capture image features from multiple perspectives, providing more comprehensive information for subsequent sound and shadow recognition. For example, multi-scale image representations can be constructed using methods such as Gaussian pyramids, Laplacian pyramids, or wavelet transforms.

[0094] Furthermore, the acoustic shadowing recognition module utilizes the constructed multi-scale image representation and identifies acoustic shadowing regions in ultrasound images based on features such as brightness distribution, brightness variation trends, and internal structural uniformity in specific areas. Acoustic shadowing regions are common artifacts in ultrasound images, and the accuracy of their identification directly affects the accuracy of subsequent nodule contour recognition. This module aims to improve the robustness of acoustic shadowing recognition by comprehensively analyzing multi-dimensional features.

[0095] In addition, the suppression module performs suppression processing on the sound shadow regions identified by the sound shadow recognition module. Specifically, the suppression processing includes unifying the signal within the sound shadow region and softening the boundaries of the sound shadow region. Its purpose is to eliminate the negative impact of sound shadows on image quality and subsequent analysis, resulting in a clearer and more realistic optimized image. For example, unifying the signal within the sound shadow region can be achieved through local mean filtering or median filtering, while softening the boundaries can be accomplished through Gaussian smoothing or morphological operations.

[0096] Building upon this, the contour recognition module identifies the contours of thyroid nodules based on the optimized image after sound and shadow suppression processing. The quality of the optimized image directly affects the accuracy of contour recognition; therefore, this module can more accurately delineate the boundaries of the nodules, providing a reliable foundation for subsequent quantitative analysis.

[0097] Ultimately, the quantification module quantifies the morphology and internal structure of thyroid nodules based on the contours identified by the contour recognition module. The quantification results can include various parameters such as nodule size, shape, edge features, and internal echo homogeneity. The aim is to provide doctors with objective and quantitative diagnostic evidence, assisting them in assessing the benign or malignant nature of thyroid nodules.

[0098] This application's solution automates and systematizes the process by breaking down each step of the thyroid nodule ultrasound image feature extraction method into independent system modules. The acquisition module handles data input, providing the foundation for subsequent processing; the construction module, through multi-scale analysis, lays the groundwork for in-depth image understanding and feature extraction; the acoustic shadowing recognition module utilizes multi-scale information to accurately locate and identify acoustic shadowing regions in the ultrasound image, a crucial step in artifact removal; the suppression module processes the identified acoustic shadowing regions, effectively eliminating their interference with subsequent analysis and generating clearer, more accurate optimized images; the contour recognition module, based on the optimized image, more accurately delineates the boundaries of the thyroid nodules; finally, the quantification module, based on accurate contour information, performs quantitative analysis of the nodule's morphology and internal structure, providing objective data support for clinical diagnosis. This modular design not only improves processing efficiency but also enhances the system's maintainability and scalability.

[0099] In some preferred embodiments, the thyroid nodule ultrasound image feature extraction system can be deployed on a hospital radiology workstation. When a doctor acquires an ultrasound image containing a thyroid nodule using ultrasound equipment, the acquisition module automatically receives these images from the ultrasound equipment or PACS (Picture Archiving and Communication System). Subsequently, the construction module performs multi-scale analysis on the image to generate a multi-scale image representation. The acoustic shadowing recognition module uses these representations, combined with features such as brightness distribution, variation trends, and internal structural uniformity, to automatically identify acoustic shadowing regions in the image. The suppression module then processes these acoustic shadowing regions, for example, by unifying the internal signals of the acoustic shadowing through local mean filtering and softening the boundaries through Gaussian smoothing to generate an optimized image. Next, the contour recognition module, based on the optimized image, uses a combination of edge detection and region growing algorithms to accurately identify the contour of the thyroid nodule. Finally, the quantization module calculates quantitative indicators such as the nodule's aspect ratio, area, perimeter, and internal echo uniformity based on the identified contour, and displays these results on the user interface for the doctor's reference, thereby assisting the doctor in the diagnosis and evaluation of thyroid nodules.

[0100] The content disclosed above is only a preferred and feasible embodiment of the present invention, and is not intended to limit the scope of protection of the present invention. Therefore, all equivalent technical changes made based on the content of the present invention specification and drawings are included within the scope of protection of the present invention. Furthermore, the elements therein can be updated as technology develops.

Claims

1. A method for extracting features from ultrasound images of thyroid nodules, characterized in that, The method includes the following steps: Obtain ultrasound images containing thyroid nodules; Multi-scale image analysis and processing are performed on ultrasound images to construct a multi-scale image representation of ultrasound images; By utilizing multi-scale image representation, low-echo regions in ultrasound images are identified, and by combining feature analysis, judgment correction, and spatial relationship verification, acoustic shadowing regions in ultrasound images are determined. The sound shadow region is suppressed. The suppression process includes unifying the signal within the sound shadow region and softening the boundary of the sound shadow region to obtain an optimized image. Based on optimized images, the outline of thyroid nodules is identified; Based on the outline of thyroid nodules, the morphology and internal structure of thyroid nodules are quantified. 2.The method of claim 1, wherein, The steps for identifying hypoechoic regions in ultrasound images using multi-scale image representation, and combining feature analysis, judgment correction, and spatial relationship verification, to determine the acoustic shadowing regions in ultrasound images include: In multi-scale image representation, low-echo regions are identified by gray-scale thresholding. For each hypoechoic region, the brightness distribution features and morphological features of the hypoechoic region are calculated at the image level, and high-brightness calcified regions are identified. The first acoustic shadow determination value of the hypoechoic region is obtained. The high-brightness calcified region is composed of multiple high-brightness calcified points. High-brightness calcified points refer to points in the ultrasound image whose gray value is higher than a preset high threshold. The potential acoustic shadowing regions identified at the image level and their surrounding environment are mapped to a multi-scale image representation, and fine feature analysis is performed to obtain the second acoustic shadowing determination value of the low echo region. The fine feature analysis includes brightness histogram morphology analysis and gradient direction consistency analysis. The potential acoustic shadowing region is a region with acoustic shadowing artifact characteristics. When the difference between the first sound shadow judgment value and the second sound shadow judgment value is greater than the preset difference threshold, the iterative correction process is triggered to recalculate the first sound shadow judgment value and the second sound shadow judgment value until the difference between the first sound shadow judgment value and the second sound shadow judgment value is less than the preset difference threshold or the maximum number of iterations is reached. After iterative correction, the first sound shadow judgment value and the second sound shadow judgment value are weighted and fused to obtain the comprehensive sound shadow judgment value of the low echo region. Before determining the acoustic shadow region, the spatial relationship of the low echo region is verified to confirm whether there are high-brightness calcification points in front of the ultrasonic wave propagation path of the low echo region, and to verify whether the low echo region is located in the reflection area of ​​the ultrasonic wave propagation path of the high-brightness calcification point. Based on the comprehensive sound and shadow judgment value and the results of spatial relationship verification, the sound and shadow area is determined.

3. The method for extracting features from ultrasound images of thyroid nodules as described in claim 2, characterized in that, The steps for identifying low-echo regions include: The ultrasound image is segmented by grayscale threshold to identify all low-echo pixels and form a set of low-echo pixels, where low-echo pixels are pixels with grayscale values ​​lower than a preset grayscale threshold. Connectivity analysis is performed on the set of low-echo pixels to cluster spatially continuous low-echo pixels into independent connected regions. Evaluate the area of ​​each connected region. If the area of ​​a connected region is less than a preset minimum area threshold, exclude the corresponding connected region. If the area of ​​a connected region is greater than the minimum area threshold, calculate the distance between the geometric center of the corresponding connected region and the adjacent connected regions. When the distance between geometric centers is less than a preset merging distance threshold, and the projections of connected regions on the ultrasound image overlap, the boundaries of the spatially overlapping connected regions are divided according to the gray-scale change trend and morphological characteristics of the connected regions in order to identify mutually independent low-echo regions.

4. The method for extracting features from ultrasound images of thyroid nodules as described in claim 2, characterized in that, The steps for calculating the brightness distribution features and morphological features of each low-echo region at the image level, identifying high-brightness calcified regions, and obtaining the first acoustic shadowing value for the low-echo region include: For each hypoecho region, calculate the average brightness value, brightness standard deviation, morphological characteristics of the hypoecho region, and determine whether there is a high-brightness calcified region in the upstream direction of the hypoecho region, where the upstream direction is the reverse path relative to the direction of ultrasonic wave propagation. The average brightness value and standard deviation of the low-echo region were analyzed to obtain the macroscopic brightness distribution of the low-echo region. The macroscopic brightness distribution is used to characterize the overall brightness level and brightness dispersion of the low-echo region at the image level. When the parameters corresponding to the macroscopic brightness distribution and morphological characteristics of multiple low-echo regions meet the same preset range, brightness profile analysis is performed on multiple low-echo regions to obtain the brightness change curve along the ultrasonic wave propagation direction. Based on the brightness change curve, identify whether there is a pattern of brightness decreasing sharply and then maintaining low brightness, and whether there is a pattern of brightness decreasing slowly and then maintaining medium brightness. Here, a sharp decrease in brightness is defined as a brightness change rate greater than a preset first brightness change threshold, and a slow decrease in brightness is defined as a brightness change rate less than a preset second brightness change threshold. The first brightness change threshold is greater than the second brightness change threshold. A low brightness pattern is defined as a brightness lower than the preset first brightness threshold, and a medium brightness pattern is defined as a brightness range between the preset second brightness threshold and the preset third brightness threshold. The first brightness threshold is less than the second brightness threshold. When a pattern of maintaining low brightness after a sharp drop in brightness is identified, and there are high-brightness calcification points in front of the ultrasonic wave propagation path in the low-echo region, the sound shadow probability score of the low-echo region is increased. The sound shadow probability score of the low-echo region is used to quantify the probability that the low-echo region will become a sound shadow region. When a pattern of slow decrease in brightness followed by maintenance of moderate brightness is identified, and there are no high-brightness calcification points in front of the ultrasonic wave propagation path in the low-echo region, the acoustic shadow probability score of the low-echo region is reduced. Calculate the first sound shadow determination value for the low echo region based on the sound shadow probability score. 5.The method of claim 2, wherein, The steps of mapping potential acoustic shadowing regions and their surrounding environment identified at the image level to a multi-scale image representation, performing fine feature analysis, and deriving a second acoustic shadowing determination value for the low-echo region include: A luminance histogram morphology analysis was performed on the potential sound shadow area and its surrounding environment to obtain the luminance histogram morphology analysis results; Gradient direction consistency analysis was performed on the potential sound shadow area and its surrounding environment to obtain the gradient direction consistency analysis results; When the brightness histogram morphology analysis results and gradient direction consistency analysis results show that the similarity between the potential acoustic shadow region and the adjacent non-acoustic shadow low echo structure meets the preset similarity threshold, multiple boundary brightness sampling lines are extracted along the blurred boundary between the potential acoustic shadow region and the adjacent non-acoustic shadow low echo structure. The adjacent non-acoustic shadow low echo structure refers to the low echo region adjacent to the potential acoustic shadow region in the ultrasound image, and the boundary brightness sampling line refers to the brightness change curve extracted along the set direction at the position of the blurred boundary. For the boundary brightness sampling line, calculate the brightness change rate in its extension direction, and calculate the second derivative of the brightness change rate to identify the inflection point of brightness change; For the boundary brightness sampling line, the texture complexity on both sides is calculated to obtain the texture abrupt change point. Here, texture complexity is a quantitative parameter used to characterize the degree of texture change in a local area, and texture abrupt change point refers to the sampling point where the difference in texture complexity on both sides of the boundary brightness sampling line changes abruptly. When the brightness change inflection point and the texture change abrupt point are spatially consistent, and the region on one side of the brightness change inflection point has the characteristics of sound shadow, while the region on the other side has the characteristics of non-sound shadow low echo structure, boundary division is performed at the brightness change inflection point to obtain the boundary division result. Based on the boundary division results, calculate the second acoustic shadow determination value for the low echo region. 6.The method of claim 2, wherein, If the difference threshold is still not met after reaching the maximum number of iterations, the method further includes the following steps: When the difference between the first and second sound shadow judgment values ​​is still greater than the difference threshold after reaching the maximum number of iterations, the low echo region is marked as the region to be reviewed. The original pixel data of the ultrasound image of the area to be reviewed is extracted, and a comprehensive judgment is made by combining the analysis results corresponding to the first and second acoustic shadow judgment values ​​to generate a feature set containing multi-dimensional information. The feature set of multi-dimensional information includes morphological features, brightness features and gradient direction features. Analyze the feature set of multi-dimensional information, identify whether there are contradictory feature manifestations, and obtain contradictory features; When contradictory features are identified, the sound and shadow judgment values ​​of the area to be reviewed are dynamically adjusted according to the type of contradictory features, providing a reference for manual review.

7. The method for extracting features from ultrasound images of thyroid nodules as described in claim 6, characterized in that, After dynamically adjusting the acoustic shadowing judgment value of the area to be reviewed, the area to be reviewed is highlighted on the ultrasound image, along with the dynamically adjusted acoustic shadowing judgment value and contradictory features of the area to be reviewed, prompting the doctor to perform manual review. 8.The method of claim 7, wherein, The steps for highlighting the area to be reviewed on an ultrasound image include: Based on the spatial location and extent of the region to be reviewed in the ultrasound image, corresponding region identification information is generated; The region identification information is overlaid on the ultrasound image at the position corresponding to the region to be reviewed, in order to distinguish the region to be reviewed from the surrounding tissue region; Generate prompt information, which includes the dynamically adjusted sound and shadow judgment values ​​and contradictory characteristics of the area to be reviewed; The prompts are displayed in association with the corresponding areas to be reviewed, prompting doctors to conduct manual reviews. 9.The method of claim 8, wherein, Region identification information is used to outline or cover regions in ultrasound images.

10. A thyroid nodule ultrasound image feature extraction system, using the thyroid nodule ultrasound image feature extraction method as described in claim 1, characterized in that, The system includes: The acquisition module acquires ultrasound images containing thyroid nodules; The module performs multi-scale image analysis and processing on ultrasound images to construct a multi-scale image representation of the ultrasound images. The acoustic shadow recognition module uses multi-scale image representation to identify low-echo regions in ultrasound images, and combines feature analysis, judgment correction and spatial relationship verification to determine the acoustic shadow regions in ultrasound images. The suppression module performs suppression processing on the sound shadow region. The suppression processing includes unifying the signal within the sound shadow region and softening the boundary of the sound shadow region to obtain an optimized image. The contour recognition module identifies the contours of thyroid nodules based on optimized images; The quantification module quantifies the morphology and internal structure of thyroid nodules based on their outlines.

Images