Medical image segmentation method based on machine vision
By analyzing the parameters during the acquisition process, abnormal points were identified and corrected, solving the problems of irregular data and volume effect in medical image acquisition, and improving the efficiency and accuracy of image segmentation.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- NANJING AIKEMAN INFORMATION TECH CO LTD
- Filing Date
- 2026-01-29
- Publication Date
- 2026-04-21
AI Technical Summary
In existing technologies, irregular data and partial volume effects are easily generated during medical image acquisition, leading to a decrease in overall image clarity, especially in areas where multiple tissues are tightly adhered, which in turn leads to image segmentation failure.
By acquiring environmental and operational parameters during the acquisition process, time-series analysis is performed to extract stability features, calculate data acquisition coefficients, identify and extract pixel contours, analyze texture extension status and color difference values, determine abnormal points, and perform cyclic data acquisition to correct the image.
It improves image segmentation efficiency and accuracy, avoids repeated acquisition, and enhances image recognizability and segmentation quality.
Smart Images
Figure CN121904079A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of data analysis technology, and in particular to a medical image segmentation method based on machine vision. Background Technology
[0002] Medical image segmentation is a crucial step in medical image analysis, aiming to accurately extract target tissues or lesion regions from medical images to provide quantitative evidence for disease diagnosis, surgical planning, and efficacy evaluation. Traditional segmentation methods mainly rely on traditional machine vision algorithms such as threshold analysis, region growing, and active contour models. Their performance is highly dependent on manually designed features and prior knowledge, and their generalization ability is limited in scenarios with complex anatomical structures, weak boundaries, and uneven gray levels, often requiring a large amount of manual intervention. In recent years, with the breakthroughs in deep learning technology, semantic segmentation methods based on convolutional neural networks (CNNs), such as U-Net and its variants, have significantly improved the accuracy and automation of segmentation. These methods can automatically extract multi-level features through end-to-end learning and have shown superior performance on many public datasets.
[0003] Chinese Patent Publication No. CN119323587A discloses a medical image segmentation method, comprising: acquiring raw PET indirect imaging image data; preprocessing the images, including denoising, contrast enhancement, and gray-level normalization, to obtain a preprocessed PET image dataset; designing a segmentation middleware that integrates multiple segmentation algorithms, including threshold segmentation, region growing, and level set methods, and selecting appropriate segmentation algorithms for different types of lesion regions based on feature vector analysis results; generating labeled PET image data using simulation technology, constructing a training dataset by combining it with real case data, expanding the sample size using data augmentation technology, and generating a large-scale dataset for algorithm training; refining the segmentation results of U-Net using the segmentation middleware, focusing on the edges of lesion regions, and accurately locating lesion boundaries through edge detection algorithms to obtain refined segmentation results.
[0004] Chinese Patent Publication No. CN115222755A discloses a method and apparatus for medical image target segmentation based on medical imaging equipment. The method includes: generating medical image data to be processed based on the medical imaging equipment; performing artifact recognition processing on the medical image data to be processed; performing annotation processing on the medical image data to be processed; correcting artifacts in the medical image data to be processed based on the annotation results; confirming the boundaries in the corrected medical image data to be processed and performing a first target image segmentation process; extracting edge contours in the corrected medical image data to be processed using an image contour extraction algorithm and performing a second target image segmentation process; and fusing the first segmented target image and the second segmented target image proportionally to form a fused segmented target image. In this embodiment of the invention, the influence of artifacts in the image on medical image segmentation can be reduced, improving the accuracy of target segmentation in medical images.
[0005] However, the following problems still exist in the existing technology. The accuracy of data acquisition affects the segmentation efficiency and precision of medical images. Furthermore, the three-dimensional images in existing medical images are composed of continuous stacked two-dimensional slices. During the acquisition process, irregular data and partial volume effects are easily generated, which leads to a decrease in the overall image clarity. In particular, the problem is more prominent in areas where multiple tissues are tightly adhered, which in turn leads to image segmentation failure. Summary of the Invention
[0006] To address this issue, the present invention provides a medical image segmentation method based on machine vision, which solves the problem that irregular data and partial volume effects are easily generated during the acquisition process, leading to a decrease in the overall image clarity. In particular, the problem is more prominent in areas where multiple tissues are tightly adhered, thus causing image segmentation failure.
[0007] To achieve the above objectives, the present invention provides a medical image segmentation method based on machine vision, comprising: Acquire the acquisition environment parameters and operation parameters recorded during the acquisition process of the target medical image. The acquisition environment parameters include the real-time temperature of the acquisition device and the pressure applied to the acquisition site. The operation parameters include the moving speed of the acquisition device and the sequence of acquisition points. A time-series analysis is performed on the real-time temperature and pressure to extract stability features. The data acquisition coefficient is calculated by combining the acquisition speed and the distribution characteristics of the acquisition point sequence to determine the tendency of data acquisition to affect the medical image. In response to the strong influence of data acquisition, the vertical layer spacing of the target medical image is acquired in real time to determine the spacing influence characterization coefficient, the pixel contours of each vertical layer of the target medical image are identified and extracted, the inter-layer contour coherence coefficient is calculated, and the data blur influence feature value is calculated in combination with the data acquisition coefficient to analyze the synthesis state of the medical image. In response to the abnormal synthesis state of the medical image, the irregular boundary region of the target medical image is scanned, the texture extension state of the target medical image and the color difference value on both sides of the texture are analyzed, and the similarity of each vertical axis layer is compared to determine the abnormal point location. The abnormal locations are collected in a loop to correct the medical image.
[0008] Furthermore, the process of extracting stability features includes, A temperature time-domain curve is constructed based on the real-time temperature, and a pressure time-domain curve is constructed based on the pressure. Calculate the coefficients of variation for the temperature time-domain curve and the pressure time-domain curve, respectively; The average value of each of the aforementioned coefficients of variation is determined to be a stability characteristic.
[0009] Furthermore, the process of calculating the data acquisition coefficients includes, The ratio of the acquisition speed to the reference acquisition speed is determined as the first acquisition factor; The ratio of the dispersion of the collected point sequence to the baseline dispersion is determined as the second collection factor; The weighted sum of the first acquisition factor, the second acquisition factor, and the reciprocal of the stability feature is determined as the data acquisition coefficient.
[0010] Furthermore, the determination of the influence tendency of data acquisition on the medical image, wherein, If the data acquisition coefficient is greater than the data acquisition coefficient threshold, then the influence of data acquisition on the medical image is determined to be a strong influence tendency. If the data acquisition coefficient is less than or equal to the data acquisition coefficient threshold, then the influence of data acquisition on the medical image is determined to be weak.
[0011] Furthermore, the process of determining the spacing influence characterization coefficient includes, Obtain the actual interlayer spacing between all adjacent layers of the target medical image in the vertical direction; Calculate the absolute deviation between each actual interlayer spacing and the preset standard interlayer spacing; Calculate the variance of all the absolute deviation values and determine the variance as the spacing influence characterization coefficient.
[0012] Furthermore, the process of calculating the interlayer profile coherence coefficient includes, Determine the geometric center point of the target contour in each vertical axis layer image, and align the geometric center points of all vertical axis layers along the vertical axis; Calculate the edge matching degree of the target contour between every two adjacent vertical axis layers; Calculate the average value of the interlayer edge matching degree of all adjacent vertical axes, and determine the average value as the interlayer profile coherence coefficient.
[0013] Furthermore, the process of calculating the feature values affected by data fuzziness includes, The ratio of the data acquisition coefficient to the benchmark data acquisition coefficient is determined as the first fuzzy factor; The ratio of the spacing influence characterization coefficient to the reference spacing influence characterization coefficient is determined as the second fuzzy factor; The ratio of the reference interlayer profile continuity coefficient to the interlayer profile continuity coefficient is determined as the third fuzzy factor. The weighted sum of the first fuzzy factor, the second fuzzy factor, and the third fuzzy factor is determined to be the characteristic value of the data fuzziness influence.
[0014] Furthermore, the analysis of the synthetic state of the medical image, wherein, If the data fuzziness impact feature value is greater than the data fuzziness impact feature value threshold, then the synthesis state of the medical image is determined to be an abnormal synthesis state. If the data fuzziness impact feature value is less than or equal to the data fuzziness impact feature value threshold, then the synthetic state of the medical image is determined to be a normal synthetic state.
[0015] Furthermore, the process of analyzing the texture extension state of the medical image includes, Analyze the gradient change of texture width within the irregular boundary region to determine the main extension direction of the texture; The texture angle is determined based on the main extension direction; Extract the texture continuity features in each of the main extension directions and determine the texture continuity representation value; The texture continuity features include texture clustering and texture continuity.
[0016] Furthermore, the similarity of each vertical axis layer is compared to determine outlier locations, wherein... If the similarity is greater than the similarity threshold, then the location is determined to be a normal location. If the similarity is less than or equal to the similarity threshold, then the location is determined to be an anomaly. The similarity is the average of the similarity between the texture angle, the texture continuity representation value, and the color difference value.
[0017] Compared with existing technologies, this invention extracts stability features through time-series analysis of acquisition temperature and acquisition pressure, calculates data acquisition coefficients based on acquisition speed and acquisition point location, determines the influence tendency of data acquisition on the medical image, determines the spacing influence characterization coefficient and calculates the inter-layer contour coherence coefficient in response to the strong influence tendency of data acquisition, and calculates the data fuzziness influence feature value based on the data acquisition coefficients to analyze the synthesis state of the medical image. For abnormal synthesis states of the medical image, the similarity of each vertical axis layer is compared to identify abnormal points and correct the medical image. This invention optimizes the pre-process of image segmentation by analyzing abnormal reactions in the image data acquisition process and image construction process to improve image recognizability, image segmentation efficiency and accuracy.
[0018] In particular, by quantitatively analyzing the parameters during the acquisition process, this invention aims to analyze the influence of data acquisition on the generation of medical images. In practice, especially for 3D medical images, since 3D images are synthesized from a series of continuous 2D slices (vertical axis layers), multiple data captures are required. During the data capture process, any non-ideal conditions may disrupt the consistency of data between layers, leading to geometric distortion, uneven resolution, or structural blurring in the final 3D model. For example, the temperature and pressure of the acquisition equipment can cause abnormal changes in the acquisition position, thus affecting the data obtained in the medical image. Furthermore, if the acquisition speed is mismatched or acquisition points are missed, it will increase the success rate and completeness of the synthesized medical image and reduce the efficiency of medical image segmentation. Based on this, this invention considers pre-analyzing the parameters during the acquisition process and calculating the influence tendency of data acquisition coefficients, providing a data basis for subsequent targeted analysis of the synthesis state of medical images, thereby achieving a forward-looking prediction of acquisition quality and improving the efficiency and accuracy of subsequent image segmentation.
[0019] In particular, for data acquisition coefficients with a strong tendency to influence the image, the synthesis state of the medical image is analyzed by calculating the data fuzziness influence feature value. In reality, the final quality of a 3D medical image is not only determined by the acquired data, but is also affected by the imaging process. Even if the medical image synthesis has the ability to self-correct, and some abnormal data can be improved through its internal model, some data may still cause deviations in the synthesis of the medical image due to abnormalities in the synthesis spacing and contour alignment. If the acquired data has a strong tendency to influence the image, it will further aggravate this deviation, leading to partial or even complete failure of the synthesis of the medical image. Based on this, the present invention considers calculating the data fuzziness influence feature value to quantify the impact of the two steps of acquisition and synthesis on the synthesis of medical images, providing a data basis for subsequent targeted corrections to improve the efficiency and accuracy of subsequent image segmentation.
[0020] In particular, for abnormal synthesis states of medical images, abnormal points are accurately located in sensitive areas such as irregular boundaries of the image by comparing the similarity of each vertical axis layer in terms of texture direction, continuity, and color features. In practice, traditional methods directly process low-quality images, which not only reduces processing speed but also wastes resources. Based on this, the present invention feeds back the spatial coordinates of abnormal points to the control end, triggering directional and cyclic data acquisition of the original anatomical site, thereby obtaining high-quality supplementary data, which is ultimately used for local correction and repair of the initial image. This fundamentally improves the data quality required for image segmentation, avoids repeated acquisition of all data, and also improves the segmentation efficiency of subsequent medical images. Attached Figure Description
[0021] Figure 1 This is a schematic diagram illustrating the steps of a machine vision-based medical image segmentation method according to an embodiment of the invention. Figure 2 A logic block diagram illustrating the tendency of data acquisition to influence the medical image in an embodiment of the invention; Figure 3 A logic block diagram illustrating the synthesis state of the medical image in an embodiment of the invention; Figure 4 A logic block diagram for comparing the similarity of each vertical axis layer to determine the location of anomalies in an embodiment of the invention. Detailed Implementation
[0022] To make the objectives and advantages of the present invention clearer, the present invention will be further described below with reference to embodiments; it should be understood that the specific embodiments described herein are merely for explaining the present invention and are not intended to limit the present invention.
[0023] Preferred embodiments of the present invention will now be described with reference to the accompanying drawings. Those skilled in the art should understand that these embodiments are merely illustrative of the technical principles of the present invention and are not intended to limit the scope of protection of the present invention.
[0024] Please see Figure 1 , Figure 1 This is a schematic diagram illustrating the steps of a machine vision-based medical image segmentation method according to an embodiment of the invention. The machine vision-based medical image segmentation method of the present invention includes: Step S1: Obtain the acquisition environment parameters and operation parameters recorded during the acquisition process of the target medical image. The acquisition environment parameters include the real-time temperature of the acquisition device and the pressure applied to the acquisition site. The operation parameters include the moving speed of the acquisition device and the sequence of acquisition points. Step S2: Perform time-series analysis on the real-time temperature and pressure, extract stability features, and calculate the data acquisition coefficient by combining the acquisition speed and the distribution characteristics of the acquisition point sequence, in order to determine the influence tendency of data acquisition on the medical image. Step S3: In response to the strong influence of data acquisition, the vertical layer spacing of the target medical image is acquired in real time to determine the spacing influence characterization coefficient, the pixel contours of each vertical layer of the target medical image are identified and extracted, the inter-layer contour coherence coefficient is calculated, and the data blur influence feature value is calculated in combination with the data acquisition coefficient to analyze the synthesis state of the medical image. Step S4: For the abnormal synthesis state of the medical image, scan the irregular boundary area of the target medical image, analyze the texture extension state of the target medical image and the color difference value on both sides of the texture, and compare the similarity of each vertical axis layer to determine the abnormal point location. Step S5: Circularly collect data from the abnormal locations to correct the medical image.
[0025] Specifically, there are no restrictions on the method of obtaining real-time temperature. For example, it can be obtained through a temperature sensor installed on the probe of the acquisition device, or it can be obtained through infrared thermal imaging. Any reasonable method will suffice, and will not be elaborated further.
[0026] Specifically, there are no restrictions on the method of obtaining pressure. For example, it can be obtained through a pressure sensor installed on the probe of the acquisition device. Of course, it can also be obtained through other methods, as long as they are reasonable. This will not be elaborated further.
[0027] Specifically, there are no restrictions on the method of obtaining the movement speed. For example, it can be captured by a high-definition camera or by a speed sensor deployed on the probe of the acquisition device. Any reasonable method will suffice, and will not be elaborated further.
[0028] Specifically, there are no restrictions on the method of obtaining the sequence of collection points. For example, it can be captured by a high-definition camera. Of course, it can also be obtained by other reasonable methods, which will not be elaborated here.
[0029] Specifically, the spacing between vertical axis layers can be calculated from the data or obtained by recognizing the image; as long as it is reasonable, it will not be elaborated further.
[0030] It is understandable that the vertical layer is a sequence of two-dimensional images continuously acquired along the Z-axis.
[0031] Specifically, the process of extracting stability features includes, A temperature time-domain curve is constructed based on the real-time temperature, and a pressure time-domain curve is constructed based on the pressure. Calculate the coefficients of variation for the temperature time-domain curve and the pressure time-domain curve, respectively; The average value of each of the aforementioned coefficients of variation is determined to be a stability characteristic.
[0032] Specifically, the standard deviation and mean of the temperature time-domain curves are determined as the coefficient of variation for temperature; the standard deviation and mean of the pressure time-domain curves are determined as the coefficient of variation for pressure. The standard deviation measures the dispersion of data points relative to their mean, which represents the average level of data during the acquisition period. Dividing the standard deviation by the mean makes the coefficient of variation a dimensionless relative indicator that can directly and fairly compare the fluctuations of physical quantities of different dimensions and orders of magnitude, such as temperature and pressure. It can be understood that the larger the value, the more drastic the relative fluctuation of the corresponding parameter. This non-steady state of the acquisition environment increases the possibility of introducing random or systematic errors into the data acquisition process, thereby increasing the risk of artifacts, distortions, or inconsistencies in the final medical images.
[0033] Specifically, the process of calculating the data acquisition coefficients includes, The ratio of the acquisition speed to the reference acquisition speed is determined as the first acquisition factor; The ratio of the dispersion of the collected point sequence to the baseline dispersion is determined as the second collection factor; The weighted sum of the first acquisition factor, the second acquisition factor, and the reciprocal of the stability feature is determined as the data acquisition coefficient.
[0034] Specifically, the baseline acquisition speed is calculated in advance. The historical acquisition speeds corresponding to several successful medical image segmentation operations are obtained in advance, and the average of each historical acquisition speed is determined as the baseline acquisition speed.
[0035] Specifically, the baseline dispersion is calculated in advance. The historical dispersion corresponding to several successful medical image segmentation operations is obtained in advance, and the mean of each historical dispersion is determined as the baseline dispersion.
[0036] Specifically, the sum of the weight coefficients of the first acquisition factor, the second acquisition factor, and the reciprocal of the stability feature is 1. When configuring the weights, considering that the direct contact with the operating equipment will have a significant impact on the data, the weight coefficients of the first acquisition factor and the second acquisition factor are both set to 0.3, and the weight coefficient of the reciprocal of the stability feature is set to 0.4.
[0037] Specifically, by quantitatively analyzing parameters during the acquisition process, this invention aims to analyze the impact of data acquisition on medical image generation. In practice, especially for 3D medical images, which are synthesized from a series of continuous 2D slices (vertical axis layers), multiple data captures are required. During the data capture process, any non-ideal conditions can disrupt the consistency of data between layers, leading to geometric distortion, uneven resolution, or structural blurring in the final 3D model. For example, the temperature and pressure of the acquisition equipment can cause abnormal changes in the acquisition position, thus affecting the data obtained in the medical image. Furthermore, if the acquisition speed is mismatched or acquisition points are missed, it will increase the success rate and completeness of the synthesized medical image and reduce the efficiency of medical image segmentation. Based on this, this invention considers pre-analyzing the parameters during the acquisition process and calculating the impact tendency of data acquisition coefficients, providing a data basis for subsequent targeted analysis of the synthesis state of medical images. This enables a forward-looking prediction of acquisition quality and improves the efficiency and accuracy of subsequent image segmentation.
[0038] Please see Figure 2 , Figure 2 This is a logic block diagram illustrating the tendency of data acquisition to influence the medical image according to an embodiment of the invention. Specifically, in determining the tendency of data acquisition to influence the medical image, wherein... If the data acquisition coefficient is greater than the data acquisition coefficient threshold, then the influence of data acquisition on the medical image is determined to be a strong influence tendency. If the data acquisition coefficient is less than or equal to the data acquisition coefficient threshold, then the influence of data acquisition on the medical image is determined to be weak.
[0039] Specifically, the data acquisition coefficient threshold represents a boundary that the acquired data will affect the medical image synthesis. It is calculated in advance by acquiring several historical data acquisition coefficients that caused medical image segmentation failures due to data acquisition errors. The product of the mean of each historical data acquisition coefficient and the acquisition accuracy is determined as the data acquisition coefficient threshold. The acquisition accuracy is determined within the interval [0.8, 1.0]. In practice, in order to improve the accuracy of data acquisition, the acquisition accuracy is determined to be 0.9.
[0040] Specifically, the process of determining the spacing influence characterization coefficient includes, Obtain the actual interlayer spacing between all adjacent layers of the target medical image in the vertical direction; Calculate the absolute deviation between each actual interlayer spacing and the preset standard interlayer spacing; Calculate the variance of all the absolute deviation values and determine the variance as the spacing influence characterization coefficient.
[0041] Specifically, the preset standard interlayer spacing is a predetermined data that can be obtained from the factory technical specifications of the medical image acquisition equipment or the pre-set scanning protocol. All data obtained are authorized data. Of course, those skilled in the art can also use other methods to obtain the required data, as long as it is reasonable, which will not be elaborated here.
[0042] Understandably, the larger the variance, the worse the uniformity of the interlayer spacing. The composite image in the vertical direction may produce blur or geometric distortion due to interlayer misalignment or uneven spacing.
[0043] Specifically, the process of calculating the interlayer profile coherence coefficient includes, Determine the geometric center point of the target contour in each vertical axis layer image, and align the geometric center points of all vertical axis layers along the vertical axis; Calculate the edge matching degree of the target contour between every two adjacent vertical axis layers; Calculate the average value of the interlayer edge matching degree of all adjacent vertical axes, and determine the average value as the interlayer profile coherence coefficient.
[0044] Specifically, there is no limitation on the method of determining the geometric center point. For example, for the target contour that has been identified in each vertical axis layer image, the spatial coordinate set of all contour pixels is extracted. The coordinates of the geometric center are obtained by calculating the average of the horizontal coordinates and the average of the vertical coordinates of all points in the coordinate set. Of course, those skilled in the art can also use other methods to determine it, as long as they are reasonable, which will not be elaborated here.
[0045] Specifically, the edge matching degree is calculated as follows: For each edge pixel of the upper contour, find the nearest edge pixel within a preset search range along the vertical axis at the corresponding position in the lower image; The ratio of the found edge pixel data to a preset number is determined as the edge matching degree; The preset search range is predetermined by obtaining the distribution locations of several images of the same category in advance to determine the distribution range, and then defining the distribution range as the preset search range. The preset quantity is predetermined. The historical quantities corresponding to several images of the same category are obtained in advance, and the average value of each historical quantity is determined as the preset quantity.
[0046] Specifically, the process of calculating the influence of data fuzziness on feature values includes, The ratio of the data acquisition coefficient to the benchmark data acquisition coefficient is determined as the first fuzzy factor; The ratio of the spacing influence characterization coefficient to the reference spacing influence characterization coefficient is determined as the second fuzzy factor; The ratio of the reference interlayer profile continuity coefficient to the interlayer profile continuity coefficient is determined as the third fuzzy factor. The weighted sum of the first fuzzy factor, the second fuzzy factor, and the third fuzzy factor is determined to be the characteristic value of the data fuzziness influence.
[0047] Specifically, the baseline data acquisition coefficients are the data acquisition coefficients corresponding to the baseline acquisition speed, baseline dispersion, and the reciprocal of the baseline stability feature. The reciprocal of the baseline stability feature is pre-calculated by acquiring historical stability features corresponding to several successful medical image segmentation operations in advance, and determining the mean of the reciprocals of each historical stability feature as the reciprocal of the baseline stability feature.
[0048] Specifically, the baseline spacing influence characterization coefficient is calculated in advance. Several historical spacing influence characterization coefficients that have been completed and segmented are obtained in advance, and the mean of each historical spacing influence characterization coefficient is determined as the baseline spacing influence characterization coefficient.
[0049] Specifically, the baseline inter-layer profile continuity coefficient is calculated in advance. Several historical inter-layer profile continuity coefficients that have been segmented are obtained in advance, and the average value of each historical inter-layer profile continuity coefficient is determined as the baseline inter-layer profile continuity coefficient.
[0050] Specifically, the sum of the weight coefficients of the first fuzzy factor, the second fuzzy factor, and the third fuzzy factor is 1. When configuring the weights, considering that the image synthesis step at this stage has a significant impact on image anomalies, the weight coefficient of the first fuzzy factor is determined to be 0.2, and the weight coefficients of the second fuzzy factor and the third fuzzy factor are both 0.4.
[0051] Specifically, for data acquisition coefficients with a strong tendency to influence the image, the synthesis state of the medical image is analyzed by calculating the data fuzziness influence feature value. In reality, the final quality of a 3D medical image is not only determined by the acquired data but is also affected by the imaging process. Even if the medical image synthesis has the ability to self-correct and some abnormal data can be improved through its internal model, some data may still cause deviations in the synthesis of the medical image due to abnormalities in the synthesis spacing and contour alignment. If the acquired data has a strong tendency to influence the image, this deviation will be further aggravated, leading to partial or even complete failure of the medical image synthesis. Based on this, the present invention considers calculating the data fuzziness influence feature value to quantify the impact of the two steps of acquisition and synthesis on the synthesis of medical images, providing a data basis for subsequent targeted corrections to improve the efficiency and accuracy of subsequent image segmentation.
[0052] Please see Figure 3 , Figure 3 This is a logic block diagram illustrating the analysis of the composite state of the medical image according to an embodiment of the invention. Specifically, the analysis of the composite state of the medical image includes... If the data fuzziness impact feature value is greater than the data fuzziness impact feature value threshold, then the synthesis state of the medical image is determined to be an abnormal synthesis state. If the data fuzziness impact feature value is less than or equal to the data fuzziness impact feature value threshold, then the synthetic state of the medical image is determined to be a normal synthetic state.
[0053] Specifically, the data blur effect feature value threshold characterizes a boundary where image segmentation fails due to synthesis. To determine this in advance, several historical data blur effect feature values of segments that failed to segment are obtained in advance. The product of the mean of each historical data blur effect feature value and the blur precision is determined as the data blur effect feature value threshold. The blur precision is selected within the interval [0.8, 1.0]. In practice, in order to improve the segmentation success rate, the blur precision is determined to be 0.9.
[0054] Specifically, the process of analyzing the texture extension state of the medical image includes, Analyze the gradient change of texture width within the irregular boundary region to determine the main extension direction of the texture; The texture angle is determined based on the main extension direction; Extract the texture continuity features in each of the main extension directions and determine the texture continuity representation value; The texture continuity features include texture clustering and texture continuity.
[0055] Specifically, the main extension direction is the direction in which the texture width gradient decreases.
[0056] Specifically, there are no restrictions on the calculation methods for texture clustering and texture continuity. For example, in implementation, texture clustering can be calculated by the local binary mode entropy value, and texture continuity can be determined by the gradient direction consistency ratio.
[0057] Specifically, the summation of texture clustering and texture continuity is determined as the texture continuity representation value; It is understandable that texture clustering represents the quantity and degree of clustering of textures. The more numerous and the higher the degree of clustering, the more discontinuous the texture is. Texture continuity represents the degree of breakage or blurring of the texture. The higher the texture continuity, the more discontinuous the texture is, the greater the extension of the texture, and the greater the probability that the location is an abnormal position.
[0058] Please see Figure 4 , Figure 4 This is a logic block diagram illustrating the comparison of similarity across vertical axis layers to determine outlier locations, as per an embodiment of the invention. Specifically, the comparison of similarity across vertical axis layers to determine outlier locations includes... If the similarity is greater than the similarity threshold, then the location is determined to be a normal location. If the similarity is less than or equal to the similarity threshold, then the location is determined to be an anomaly. Specifically, there are no restrictions on the method of calculating similarity. For example, it can be calculated using the cosine similarity method. Of course, those skilled in the art can also calculate it according to the actual situation, as long as it is reasonable. This will not be elaborated further.
[0059] The similarity is the average of the similarity of the texture angle, the texture continuity coefficient, and the color difference value.
[0060] Specifically, the similarity threshold is pre-calculated. In practice, in order to improve the image availability, the similarity threshold is set at 90%. Of course, those skilled in the art can also determine it according to the actual situation, as long as it is reasonable, which will not be elaborated here.
[0061] Specifically, for abnormal synthesis states of medical images, in sensitive areas such as irregular boundaries of the image, abnormal points are accurately located by comparing the similarity of each vertical axis layer in terms of texture direction, continuity, and color features. In practice, traditional methods directly process low-quality images, which not only reduces processing speed but also wastes resources. Based on this, the present invention feeds back the spatial coordinates of abnormal points to the control end, triggering directional and cyclic data acquisition of the original anatomical site, thereby obtaining high-quality supplementary data, which is ultimately used for local correction and repair of the initial image. This fundamentally improves the data quality required for image segmentation, avoids repeated acquisition of all data, and also improves the segmentation efficiency of subsequent medical images.
[0062] The technical solution of the present invention has been described above with reference to the preferred embodiments shown in the accompanying drawings. However, it will be readily understood by those skilled in the art that the scope of protection of the present invention is obviously not limited to these specific embodiments. Without departing from the principles of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will all fall within the scope of protection of the present invention.
[0063] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.
Claims
1. A medical image segmentation method based on machine vision, characterized in that, include: Acquire the acquisition environment parameters and operation parameters recorded during the acquisition process of the target medical image. The acquisition environment parameters include the real-time temperature of the acquisition device and the pressure applied to the acquisition site. The operation parameters include the moving speed of the acquisition device and the sequence of acquisition points. A time-series analysis is performed on the real-time temperature and pressure to extract stability features. The data acquisition coefficient is calculated by combining the acquisition speed and the distribution characteristics of the acquisition point sequence to determine the tendency of data acquisition to affect the medical image. In response to the strong influence of data acquisition, the vertical layer spacing of the target medical image is acquired in real time to determine the spacing influence characterization coefficient, the pixel contours of each vertical layer of the target medical image are identified and extracted, the inter-layer contour coherence coefficient is calculated, and the data blur influence feature value is calculated in combination with the data acquisition coefficient to analyze the synthesis state of the medical image. In response to the abnormal synthesis state of the medical image, the irregular boundary region of the target medical image is scanned, the texture extension state of the target medical image and the color difference value on both sides of the texture are analyzed, and the similarity of each vertical axis layer is compared to determine the abnormal point location. The abnormal locations are collected in a loop to correct the medical image.
2. The medical image segmentation method based on machine vision according to claim 1, characterized in that, The process of extracting stability features includes, A temperature time-domain curve is constructed based on the real-time temperature, and a pressure time-domain curve is constructed based on the pressure. Calculate the coefficients of variation for the temperature time-domain curve and the pressure time-domain curve, respectively; The average value of each of the aforementioned coefficients of variation is determined to be a stability characteristic.
3. The medical image segmentation method based on machine vision according to claim 1, characterized in that, The process of calculating the data acquisition coefficients includes: The ratio of the acquisition speed to the reference acquisition speed is determined as the first acquisition factor; The ratio of the dispersion of the collected point sequence to the baseline dispersion is determined as the second collection factor; The weighted sum of the first acquisition factor, the second acquisition factor, and the reciprocal of the stability feature is determined as the data acquisition coefficient.
4. The medical image segmentation method based on machine vision according to claim 1, characterized in that, The determination of the influence tendency of data acquisition on the medical image, wherein... If the data acquisition coefficient is greater than the data acquisition coefficient threshold, then the influence of data acquisition on the medical image is determined to be a strong influence tendency. If the data acquisition coefficient is less than or equal to the data acquisition coefficient threshold, then the influence of data acquisition on the medical image is determined to be weak.
5. The machine vision-based medical image segmentation method according to claim 1, characterized in that, The process of determining the spacing influence characterization coefficient includes, Obtain the actual interlayer spacing between all adjacent layers of the target medical image in the vertical direction; Calculate the absolute deviation between each actual interlayer spacing and the preset standard interlayer spacing; Calculate the variance of all the absolute deviation values and determine the variance as the spacing influence characterization coefficient.
6. The medical image segmentation method based on machine vision according to claim 1, characterized in that, The process of calculating the interlayer profile coherence coefficient includes: Determine the geometric center point of the target contour in each vertical axis layer image, and align the geometric center points of all vertical axis layers along the vertical axis; Calculate the edge matching degree of the target contour between every two adjacent vertical axis layers; Calculate the average value of the interlayer edge matching degree of all adjacent vertical axes, and determine the average value as the interlayer profile coherence coefficient.
7. The medical image segmentation method based on machine vision according to claim 1, characterized in that, The process of calculating the feature values affected by data fuzziness includes, The ratio of the data acquisition coefficient to the benchmark data acquisition coefficient is determined as the first fuzzy factor; The ratio of the spacing influence characterization coefficient to the reference spacing influence characterization coefficient is determined as the second fuzzy factor; The ratio of the reference interlayer profile continuity coefficient to the interlayer profile continuity coefficient is determined as the third fuzzy factor. The weighted sum of the first fuzzy factor, the second fuzzy factor, and the third fuzzy factor is determined to be the characteristic value of the data fuzziness influence.
8. The medical image segmentation method based on machine vision according to claim 1, characterized in that, The analysis of the synthetic state of the medical image, wherein, If the data fuzziness impact feature value is greater than the data fuzziness impact feature value threshold, then the synthesis state of the medical image is determined to be an abnormal synthesis state. If the data fuzziness impact feature value is less than or equal to the data fuzziness impact feature value threshold, then the synthetic state of the medical image is determined to be a normal synthetic state.
9. The medical image segmentation method based on machine vision according to claim 1, characterized in that, The process of analyzing the texture extension state of the medical image includes, Analyze the gradient change of texture width within the irregular boundary region to determine the main extension direction of the texture; The texture angle is determined based on the main extension direction; Extract the texture continuity features in each of the main extension directions and determine the texture continuity representation value; The texture continuity features include texture clustering and texture continuity.
10. The medical image segmentation method based on machine vision according to claim 9, characterized in that, The similarity of each vertical axis layer is compared to determine the locations of outliers. If the similarity is greater than the similarity threshold, then the location is determined to be a normal location. If the similarity is less than or equal to the similarity threshold, then the location is determined to be an anomaly. The similarity is the average of the similarity between the texture angle, the texture continuity representation value, and the color difference value.
Citation Information
Patent Citations
Medical image target segmentation method and device based on medical imaging equipment
CN115222755A
Medical image segmentation method
CN119323587A