A medical image analysis method and device based on fallopian tube anatomical segmentation

Abstract

The application relates to a medical image analysis method and device based on fallopian tube anatomical segmentation, which comprises the following steps: acquiring a laparoscope image, extracting blood flow direction and velocity information, and generating a blood flow direction graph; dividing the fallopian tube into three regions of an umbrella part, a ampulla part and an isthmus part, extracting and marking the structure edges of the three regions, generating an anatomical structure boundary graph, performing image fusion processing on the blood flow direction graph and the anatomical structure boundary graph, and identifying a blood flow abnormality concentration region; performing image segmentation and pathological feature identification on the blood flow abnormality concentration region, judging whether pathological tissues exist, if yes, marking the region, and generating a prompt information. Through the analysis of the blood flow direction and velocity and the anatomical segmentation of the fallopian tube, the position of a blood point can be accurately identified, the blood flow abnormality region can be found in time, and the doctor can make more accurate and safe decisions in the operation process.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of medical imaging, and in particular to a medical image analysis method and device based on anatomical segmentation of fallopian tubes. BACKGROUND

[0002] In laparoscopic surgery, especially surgical operations involving fallopian tubes, the occurrence of bleeding often requires rapid and accurate identification and treatment. As an important part of the female reproductive system, the fallopian tube has a complex anatomical structure, including the fimbria, ampulla, and isthmus, and the physiological characteristics and functions of each region are different. Especially during the operation, it is extremely important to accurately locate the bleeding points for different anatomical regions and make reasonable judgments based on blood flow maps, anatomical segmentation, and physiological characteristics.

[0003] However, traditional surgical methods rely too much on the experience of doctors and real-time visual observation, facing challenges such as time pressure and high operation difficulty. In emergency situations, doctors may not be able to quickly locate the bleeding points, leading to misdiagnosis or mistreatment, or even taking a "one-size-fits-all" approach, resulting in unnecessary tissue damage, loss of function, and even affecting the patient's fertility. Especially for different regions of the fallopian tube, there are significant differences in blood supply and pathological manifestations, and traditional treatment methods cannot accurately distinguish and individualize treatment. Therefore, how to use advanced image analysis technology to quickly locate bleeding points, analyze blood flow trajectories, and combine anatomical structures and pathological characteristics to assist doctors in making more accurate and reasonable treatment decisions has become an important issue to improve surgical efficiency and safety. SUMMARY

[0004] The purpose of the present application is to solve the problems of difficulty in quickly locating bleeding points in emergency situations, difficulty in accurately distinguishing the blood flow characteristics of different regions of the fallopian tube, and inability to judge in real time whether there are pathological tissues or foreign bodies left.

[0005] According to an aspect of the present application, a medical image analysis method based on anatomical segmentation of fallopian tubes is provided, comprising:

[0006] Obtaining a laparoscopic image, analyzing the blood distribution in the laparoscopic image, extracting the flow direction and velocity information of the blood, and generating a blood flow direction map;

[0007] Partitioning and identifying the fallopian tube structure in the laparoscopic image, dividing the fallopian tube into three regions: fimbria, ampulla, and isthmus, extracting and marking the structural edges of the three regions, and generating an anatomical structure boundary map, wherein the fimbria includes the connection segment of the fallopian tube and the ovary, the isthmus includes the connection part of the fallopian tube and the uterus, and the ampulla is the connection end of the fimbria and the isthmus;

[0008] image fusion processing is performed on the blood flow direction map and the anatomical structure boundary map to mark a contact boundary of blood and the three regions, analyze a contact relationship between a blood flow trajectory and a structure edge of the three regions, and identify an abnormal blood flow concentration region;

[0009] Image segmentation and pathological feature recognition are performed on the abnormal blood flow concentration region to determine whether pathological tissue exists, and if so, the region is marked and prompt information is generated.

[0010] Preferably, after the blood flow direction map is generated, a vector field model is constructed based on the blood flow direction and speed information, and blood point spatial coordinates are reversely calculated, specifically,

[0011] A speed gradient maximum value point is extracted in the blood flow direction map and is marked as a candidate blood point;

[0012] Flow field distribution of different combinations of the candidate blood points is simulated, and a blood point is determined through maximum likelihood estimation.

[0013] Preferably, after the blood point is determined, the method further comprises:

[0014] It is determined whether the blood point is in the ampulla;

[0015] If so, the bleeding speed and bleeding volume of the blood point are calculated, the size of the wound of the blood point is determined, and if the size of the wound is smaller than a preset wound size threshold, prompt information suggesting conservative treatment is output.

[0016] Preferably, the structure edges of the three regions are extracted and marked to generate the anatomical structure boundary map, including:

[0017] An anatomical segmentation of the oviduct in the laparoscope image is identified;

[0018] An edge detection algorithm or an edge detection model based on deep learning is applied to extract edges of each anatomical structure, and the edges are marked.

[0019] Preferably, the edge detection algorithm includes a Canny operator, a Sobel operator, an HED algorithm, or a CombineNet algorithm.

[0020] Preferably, the image fusion processing on the blood flow direction map and the anatomical structure boundary map includes:

[0021] The blood flow direction map and the anatomical structure boundary map are fused at a channel level through a multi-channel data fusion technology, and the fusion method includes weighted average, maximum value synthesis, or fusion filtering.

[0022] The fused image is subjected to boundary enhancement processing;

[0023] The relative position and contact of the blood flow and the boundary of the anatomical structure are analyzed by a spatial relationship modeling algorithm;

[0024] The fused image is output in the form of a multidimensional data graph.

[0025] Preferably, the identifying of the blood flow abnormal concentration area comprises:

[0026] Edge extraction is performed on the blood flow area, including pixel-level feature extraction on the blood flow area;

[0027] Based on the blood flow speed, flow amount and blood diffusion range, the concentration of blood flow in each blood flow area is calculated;

[0028] It is judged whether the concentration of blood flow exceeds a preset concentration threshold, and if so, the area is marked as a blood flow abnormal concentration area.

[0029] Preferably, the calculation of the concentration of blood flow in each blood flow area comprises: based on one or more of K-Means clustering algorithm, DBSCAN algorithm, Otsu algorithm, Canny edge detection algorithm, U-Net algorithm and CNN algorithm, the concentration of blood flow in the blood flow area is calculated.

[0030] Preferably, the image segmentation and pathological feature recognition of the blood flow abnormal concentration area comprises:

[0031] The blood flow abnormal concentration area is preliminarily extracted by using image segmentation technology, and the blood area is separated from other normal tissue areas;

[0032] The shape and boundary features of the blood area are extracted;

[0033] Using deep learning-based image classification, target detection and lesion detection methods, it is identified whether the blood flow abnormal concentration area belongs to cyst, adhesion or foreign matter, the abnormal blood flow area with pathological features is labeled, and prompt information including lesion type, disease diagnosis suggestion and treatment scheme suggestion is generated.

[0034] The application also provides a medical image analysis device based on fallopian tube anatomical segmentation, which applies the medical image analysis method based on fallopian tube anatomical segmentation.

[0035] An image acquisition and processing module is configured to acquire laparoscope images and perform preprocessing to generate blood flow direction maps and fallopian tube anatomical structure boundary maps;

[0036] An identification and analysis module is configured to analyze the blood flow direction map and the anatomical structure boundary map, locate the bleeding point through image segmentation and pathological feature recognition technology, evaluate the wound size and bleeding speed of the bleeding point, and determine whether pathological tissue exists;

[0037] A decision support module is configured to generate a corresponding treatment suggestion based on the analysis result of the identification and analysis module.

[0038] An interaction module is configured to display the analysis result in the form of an image, a chart or a sound.

[0039] The present application has the following beneficial effects: By acquiring a laparoscope image and analyzing blood flow direction and speed, combined with anatomical segmentation of the fallopian tube (umbrella part, isthmus part, ampulla part), the position of the bleeding point can be accurately identified. After the bleeding point is confirmed, the bleeding speed, flow rate and wound size of the bleeding point can be calculated to further evaluate the severity of the bleeding. If the wound is small, the system will automatically output a conservative treatment suggestion to avoid the doctor making an over-treatment decision in an emergency, thereby reducing the damage to the patient. Through image segmentation and pathological feature recognition of the abnormal blood flow area, it can be determined whether there is pathological tissue (such as cysts, adhesions or fertilized egg rupture residue). This function provides more accurate intraoperative diagnostic information for doctors, helping doctors avoid missing the lesion area or making mistakes.

[0040] Through image fusion, blood flow trajectory analysis and regional contact relationship modeling, the system can timely find the abnormal blood flow area to assist the doctor in making more accurate and safe decisions during the operation. Especially in emergency situations, the doctor can avoid making a "one-size-fits-all" treatment decision and reduce the risk of misdiagnosis.

[0041] The entire process can effectively improve the efficiency of intraoperative judgment, reduce the work pressure and operation time of the doctor, and improve the overall efficiency of the operation through the comprehensive application of image fusion, edge detection, deep learning and other technologies.

[0042] The present application not only can detect blood flow and tissue structure, but also can provide treatment suggestions based on the analysis result to support the doctor in making accurate decisions in emergency situations, reduce misoperation and risk, and further improve the quality and safety of medical services. BRIEF DESCRIPTION OF DRAWINGS

[0043] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiments or prior art description. Obviously, the drawings in the following description are only some embodiments of the present application, and those skilled in the art can also obtain other drawings according to these drawings without creative labor.

[0044] Figure 1 A logic block diagram of a medical image analysis method based on anatomical segmentation of fallopian tubes according to an embodiment of the present application. DETAILED DESCRIPTION

[0045] For the purpose of promoting a fuller understanding of the present application, reference is now made to the following description, taken in conjunction with the accompanying drawings. In the drawings, preferred embodiments of the present application are illustrated. However, it is understood that the present application can be carried out in a multitude of different forms and should not be construed as limited to the embodiments set forth in the description and / or drawings. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the application to those skilled in the art.

[0046] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used in the description of the application herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the application. As used herein, the term "and / or" includes any and all combinations of one or more of the associated listed items.

[0047] Reference is made to Figure 1 An embodiment of the present application provides a medical image analysis method based on anatomical segmentation of fallopian tubes, comprising:

[0048] S10, acquiring a laparoscope image, analyzing blood distribution in the laparoscope image, extracting blood flow direction and speed information, and generating a blood flow direction map.

[0049] In this step, it should be noted that the laparoscope image is collected by a laparoscope device, usually using high-resolution video images or static images. The image should have sufficient clarity to ensure accurate capture of key elements such as blood flow, anatomical structure and blood vessels. Identification and extraction of blood regions in the image usually relies on image processing techniques such as threshold segmentation, color space conversion (such as RGB to HSV space) to distinguish the color difference between blood and surrounding normal tissue. Edge detection algorithms (such as Canny operator, Sobel operator) are used to clearly define the boundaries of the blood region, ensuring that the blood distribution area is accurately extracted. Deep learning models (such as U-Net) can be used for automated segmentation of blood regions to further improve accuracy, especially in complex environments with small blood flow areas. In an optional embodiment, an optical flow method can be used to estimate the flow direction and speed of blood using pixel motion in video sequences or consecutive image frames. The optical flow method can calculate the motion vector of each pixel in the image, reflecting the direction and speed of blood flow.

[0050] Particle tracking algorithms can also be used to estimate blood flow velocity by tracking particles or fluid markers in the blood. By analyzing consecutive image frames, a velocity vector field of blood flow can be generated.

[0051] Local region analysis can also be used to select a region of interest (ROI) in the image and estimate blood flow trends by analyzing pixel changes within the region.

[0052] The extracted flow direction and velocity information form a vector field that accurately describes the blood flow path and velocity distribution.

[0053] Based on the extracted blood flow direction and velocity data, a blood flow direction map is generated, which reflects the blood flow path and velocity information, usually in the form of a vector map, each vector representing a direction and velocity.

[0054] This map can visually display the direction, velocity changes and concentration or dispersion areas of blood flow, providing data support for subsequent steps such as bleeding point localization and flow anomaly area analysis.

[0055] Further, after generating the blood flow direction map, it further includes: constructing a vector field model based on the blood flow direction and velocity information, and inversely calculating the spatial coordinates of the bleeding point, specifically including:

[0056] S11, extracting the maximum value point of velocity gradient in the blood flow direction map and marking it as a candidate bleeding point.

[0057] In this step, it should be noted that in the blood flow direction map, the velocity gradient represents the intensity of flow rate change, reflecting the area of blood flow change. Areas with high velocity gradients are usually related to potential bleeding points, as bleeding areas often cause abnormal changes in local blood flow.

[0058] Gradient calculation methods such as Sobel operator and Laplacian operator are used to extract the intensity of flow velocity change. By calculating the local gradient of each point in the blood flow direction map, the area with the most dramatic velocity change can be found.

[0059] Based on the maximum value point of velocity gradient, these areas are marked as candidate bleeding points. These areas may be where blood flow changes dramatically, usually indicating that blood is flowing from the bleeding point and may be related to the location of the bleeding point.

[0060] After marking the candidate bleeding points, these points can be further screened and verified to improve the accuracy of the final bleeding point localization.

[0061] S12, simulate the flow field distribution of different candidate bleeding point combinations and determine the bleeding point by maximum likelihood estimation.

[0062] In this step, it is necessary to simulate the flow field distribution of the candidate bleeding points using numerical simulation methods. By simulating the combination of candidate bleeding points, the blood flow patterns of these points are calculated. The simulation results provide blood flow paths and flow trends under different bleeding point assumptions.

[0063] Flow field simulation can be based on fluid dynamics models, combined with the physical properties of blood flow (such as blood viscosity, flow rate, blood vessel shape, etc.) for accurate calculation.

[0064] The maximum likelihood estimation (MLE) method is used to determine the most likely bleeding point based on the simulated flow field distribution and the actual observed blood flow information. The MLE method finds the bleeding point that best fits the data by evaluating the matching degree between different candidate points and the actual blood flow.

[0065] Through maximum likelihood estimation, candidate areas that are unlikely to be bleeding points can be effectively filtered out, improving the accuracy of the final determination of bleeding points.

[0066] After determining the bleeding point, it also includes:

[0067] S13, determine whether the bleeding point is in the ampulla.

[0068] In this step, it is necessary to determine whether the candidate bleeding point is located in the ampulla of the fallopian tube based on the aforementioned anatomical segmentation information of the fallopian tube. The ampulla, as the middle segment of the fallopian tube, has a relatively thick diameter and certain blood flow and tissue recovery capacity. Therefore, if the bleeding point is located in this area and the wound is small, conservative treatment can be used to avoid unnecessary surgical intervention. This strategy can reduce harm to patients and comply with the best practices of treatment, especially in emergency situations, rapid and accurate diagnosis can help doctors make more appropriate decisions. By analyzing the laparoscopic image, the blood flow direction map is matched with the anatomical structure boundary map to determine whether the candidate bleeding point is located in the defined area of the ampulla.

[0069] The specific judgment criteria can be based on anatomical annotations or algorithm rules, such as using edge detection-based segmentation algorithms to accurately divide the various regions of the fallopian tube, thereby achieving accurate ampulla positioning.

[0070] S14, if yes, calculate the bleeding speed and bleeding volume of the bleeding point, judge the wound size of the bleeding point, and if the wound size is less than the preset wound size threshold, output a prompt information suggesting conservative treatment.

[0071] In this step, it is necessary to estimate the blood velocity and the amount of bleeding using the aforementioned velocity information in combination with a fluid dynamics model. In combination with the aforementioned blood flow direction and velocity data, the size of the wound at the bleeding point is estimated. The wound size reflects the severity of the bleeding, and a larger wound generally means a large amount of bleeding, which may require immediate intervention.

[0072] According to the distribution characteristics of blood flow, the area or other related indicators of the wound at the bleeding point are calculated using image segmentation and morphological processing techniques. If the size of the wound is judged to be smaller than the preset threshold (i.e., a smaller bleeding point), a conservative treatment suggestion is output according to clinical experience and preset rules. Such treatment usually includes drug control, local hemostasis, etc., to avoid excessive intervention or unnecessary surgical operation.

[0073] The threshold of wound size, bleeding velocity, etc. should be based on medical data and clinical experience, and can be calibrated through historical case data.

[0074] After the bleeding point is determined, the system should generate corresponding treatment suggestions based on the doctor's operation decision-making process, such as whether surgery is needed, whether conservative treatment is taken, etc.

[0075] S20, partition identification is performed on the fallopian tube structure in the laparoscope image, the fallopian tube is divided into three regions of the umbrella part, the ampulla part and the isthmus part, the structure edges of the three regions are extracted and labeled, and an anatomical structure boundary map is generated, wherein the umbrella part includes the connecting segment of the fallopian tube and the ovary, the isthmus part includes the connecting part of the fallopian tube and the uterus, and the ampulla part is the connecting end of the umbrella part and the isthmus part.

[0076] In this step, it is necessary to perform partition identification on the fallopian tube structure in the laparoscope image, in order to accurately locate the various anatomical regions of the fallopian tube, and thus provide reliable anatomical basis for subsequent blood flow analysis and bleeding point positioning. The specific description is as follows:

[0077] Umbrella part: The umbrella part is located at the distal end of the fallopian tube and is the part connecting the ovary. Its end is umbrella-shaped, expanding outward, close to the position of the ovary. The umbrella part is relatively wide and has finger-like tentacles. In image processing, the identification of the umbrella part helps to distinguish its junction with the ovary or other tissues.

[0078] Isthmus part: It is the narrowest part of the fallopian tube, connecting the uterine angle. The diameter of the isthmus part is smaller, and compared with the umbrella part and the ampulla part, the lumen of the isthmus part is narrow and has less elasticity. Due to the narrowness of this part, any abnormal bleeding will quickly appear in the local area, and accurate identification of the isthmus part is particularly important for diagnosis.

[0079] Amphora: The ampulla is located between the umbrella and isthmus, belonging to the middle segment of the fallopian tube. It is located at the wider part of the middle segment of the fallopian tube. The structure of the ampulla is relatively wide and the lumen is relatively large. This area is relatively thick and has a high blood flow, so the severity of bleeding may be higher. Because of its obvious characteristics, accurate positioning helps quickly identify the bleeding point and abnormal area.

[0080] By extracting and labeling the structural edges of the three regions, an accurate anatomical structure boundary map can be generated, laying the foundation for subsequent blood flow direction map and flow abnormality area identification.

[0081] In one specific embodiment, extracting and labeling the structural edges of the three regions to generate an anatomical structure boundary map includes:

[0082] S21, identify the anatomical segmentation of the fallopian tube in the laparoscopic image.

[0083] S22, apply edge detection algorithms or deep learning-based edge detection models to extract the edges of each anatomical structure and label the edges. The edge detection algorithm includes Canny operator, Sobel operator, HED algorithm or CombineNet algorithm.

[0084] In this embodiment, it should be noted that before anatomical segmentation, the laparoscopic image is first pre-processed, including denoising, enhancing contrast and adjusting brightness, etc. to ensure image quality and facilitate subsequent structure recognition.

[0085] By applying segmentation methods based on anatomical features such as color, texture and shape, different parts of the fallopian tube in the laparoscopic image are identified. The anatomical segmentation process divides the fallopian tube into three regions: the umbrella, the ampulla and the isthmus.

[0086] A trained deep learning model (such as a convolutional neural network CNN) can be used for anatomical segmentation, or classical image processing techniques can be used for feature extraction. Deep learning models can accurately identify and segment these anatomical structures by learning from a large amount of labeled data.

[0087] Canny operator is a classic edge detection algorithm that can effectively detect edge information in images. Its advantage is that it can better extract detailed information in images, especially suitable for use in areas with sharp changes in image edges.

[0088] Sobel operator extracts edges by calculating the gradient of image gray values, which is suitable for detecting areas with large pixel changes in images, especially for line-type structure recognition.

[0089] HED (Holistically-Nested Edge Detection) is a deep learning-based edge detection algorithm that can effectively identify edges in complex backgrounds, making it particularly suitable for complex medical images. This algorithm uses a multi-level convolutional neural network (CNN) for end-to-end edge detection, allowing for more detailed and accurate edge information.

[0090] CombineNet is another deep learning-based edge detection method that combines the outputs of multiple neural network models to enhance the accuracy and robustness of edge detection, making it suitable for complex medical image analysis.

[0091] By applying the above edge detection algorithms, the system will extract the edges of the fallopian tube anatomical region in the laparoscopic image. The extracted edges will be marked to form clear boundaries. This boundary information will serve as an important basis for the subsequent fusion of blood flow direction maps and anatomical structure boundary maps.

[0092] After extracting the edges, different colors or symbols will be used to mark the edges, clearly distinguishing between different regions (fimbria, ampulla, and isthmus). This will help doctors quickly identify key anatomical structures in laparoscopic images, improving diagnostic efficiency.

[0093] S30, image fusion processing is performed on the blood flow direction map and the anatomical structure boundary map to mark the contact boundaries between blood and the three regions, analyze the contact relationship between the blood flow trajectory and the structure edges of the three regions, and identify the blood flow abnormal concentration area. In this step, it needs to be noted that the purpose of image fusion is to combine the blood flow direction map and the anatomical structure boundary map to form a fused image, so as to facilitate the subsequent analysis of the relationship between blood flow trajectory and anatomical structure. Through image fusion, the contact and interaction between blood flow and fallopian tube anatomical structure (fimbria, ampulla, and isthmus) can be more clearly displayed, helping doctors better locate blood points and related abnormalities.

[0094] Image fusion is to simultaneously display blood flow direction and structure boundary information in the same image, allowing doctors to intuitively understand the spatial relationship between blood flow and fallopian tube anatomical structure, especially in complex medical scenarios, which helps to quickly locate blood flow abnormal areas and provides a basis for clinical decision-making.

[0095] After image fusion processing, the contact boundaries between blood flow areas and anatomical structures can be marked on the fused image. These boundaries reflect the interaction between blood and the three regions of the fallopian tube. Through automated algorithms or manual annotation, doctors can quickly identify blood flow abnormal areas, especially areas with concentrated bleeding.

[0096] The markers of the contact boundary can be marked with different colors or dotted frames for easy visual identification by doctors. For example, the blood flow area can be marked with red, the anatomical boundary area with blue or green, and the intersection part highlighted with yellow or other eye-catching colors.

[0097] In an optional embodiment, the image fusion processing of the blood flow direction map and the anatomical structure boundary map includes:

[0098] S31, by multi-channel data fusion technology, the blood flow direction map and the anatomical structure boundary map are fused at the channel level, and the fusion method includes weighted average, maximum synthesis or fusion filtering. In this step, it needs to be explained that the blood flow direction map and the anatomical structure boundary map are usually derived from different image channels. In order to fuse the two images, multi-channel data fusion technology can be used to combine blood flow information and anatomical structure boundary information in one image channel. The fusion method includes: weighted average fusion, the blood flow information and the structure boundary information of each pixel point are weighted and averaged, so that the weight of the two information sources in the fused image can be adjusted according to the needs. Maximum synthesis, for each pixel, select the information with larger value in the blood flow direction map and the anatomical structure boundary map, and retain the more important information source. Fusion filtering, using filter to filter and fuse two images, smoothing the noise in the fusion process, while retaining the boundary and flow direction information.

[0099] S32, the boundary enhancement processing of the fused image, the fused image may need to further enhance its edge information. Through edge detection technology such as Canny algorithm or Sobel operator, the structure boundary in the fused image is further strengthened, so that the anatomical structure (umbrella, ampulla, isthmus) and the boundary of blood flow are more obvious.

[0100] S33, by spatial relationship modeling algorithm to analyze the relative position and contact situation of blood flow and anatomical structure boundary, through spatial relationship modeling algorithm (such as image-based spatial analysis, geometric model construction, etc.), the relative position and contact situation between blood flow trajectory and anatomical structure boundary can be analyzed. This process can help to determine whether the blood flows along the anatomical structure and to determine the abnormal concentration area. During the analysis process, attention is focused on whether the blood flow has strong contact with the edge of the anatomical structure in some areas. For example, whether the blood is concentrated near the ampulla, which may indicate that there is a more serious bleeding in this area, which needs to be paid attention to by doctors. The flow direction map can reflect the flow rate and direction of blood flow, and by comparing with the anatomical structure boundary, it can be judged whether the flow is hindered or guided by the anatomical structure. If the blood flow stagnates or suddenly turns in some areas, it may mean the existence of bleeding point.

[0101] S34, output the fused image in the form of multi-dimensional data graph.

[0102] Further, identifying the blood flow abnormal concentration area includes:

[0103] Edge extraction of the blood flow area includes:

[0104] S35, pixel-level feature extraction of the blood flow area can extract edge and region features in the following ways:

[0105] Edge detection algorithm: for example, Canny edge detection algorithm, Sobel operator, HED algorithm (Holistically-Nested Edge Detection), etc., applied to the image of the blood flow area to obtain the details of the blood flow edge.

[0106] Deep learning model: based on U-Net or other convolutional neural network (CNN) structure, through training model to segment the image at pixel level, to more accurately extract the blood flow area and its edge.

[0107] S36, based on the blood flow velocity, flow volume and blood diffusion range, calculate the concentration of blood flow in each blood flow area. It can help to identify the area where blood flow is more concentrated as a potential abnormal blood flow area.

[0108] Calculation method includes:

[0109] Flow velocity: according to the blood flow direction and speed information of each pixel point in the image, calculate the intensity of blood flow.

[0110] Flow volume: according to the flow rate of blood and the volume of the area it passes through, estimate the blood flow volume.

[0111] Blood diffusion range: according to the range of blood diffusion, calculate the spatial distribution of blood flow.

[0112] Based on these calculation results, the concentration value of the blood flow area can be generated.

[0113] S37, judge whether the concentration of blood flow exceeds the preset concentration threshold, if yes, mark the area as blood flow abnormal concentration area.

[0114] Optionally, the calculation of the concentration of blood flow in each blood flow region comprises: calculating the concentration of blood flow in the blood flow region based on one or more of a K-Means clustering algorithm, a DBSCAN algorithm, an Otsu algorithm, a Canny edge detection algorithm, a U-Net algorithm, and a CNN algorithm. The K-Means clustering algorithm can cluster the pixel points of the blood flow region, and based on the center position and distribution of each cluster, the blood flow concentration of each cluster region can be calculated. The DBSCAN algorithm uses a density clustering method to identify regions where blood flow is more concentrated, which can better handle blood flow regions with irregular shapes or noise effects. The Otsu algorithm is used for global threshold segmentation of images, which can be used to automatically determine the concentration threshold in the blood flow region to distinguish abnormal regions. The Canny edge detection algorithm helps to extract the edges of the blood flow region, which is further used for concentration analysis. The U-Net algorithm uses a deep learning network to accurately segment the blood region, thereby calculating the concentration of each region. The CNN algorithm uses the feature learning ability of the convolutional neural network to identify abnormal patterns of blood flow and calculate the concentration.

[0115] Once it is determined that the blood flow concentration exceeds the threshold, the region can be marked as an abnormal blood flow concentration region through image marking technology.

[0116] In this embodiment, the calculation of the "blood flow concentration" described above, in combination with the speed field, diffusion area, and pixel density of the blood region, is implemented in the following way:

[0117] 1. Concentration calculation based on clustering algorithm (K-Means or DBSCAN):

[0118] The blood pixel region extracted from the blood flow direction map is used to construct a feature vector, including: position coordinates (x, y), flow speed, flow direction angle, color channel (such as RGB or HSV value).

[0119] Apply K-Means or DBSCAN to cluster the blood region to obtain several blood sub-region clusters;

[0120] Let the number of pixels in each sub-region be , and the area be , then the flow concentration of the sub-region is defined as:

[0121]

[0122] The concentration of the overall blood flow region can be defined as the maximum value or weighted average value of in all clusters:

[0123] or

[0124] 2. Based on edge detection algorithm (Canny) and diffusion range judgment:

[0125] Use Canny algorithm to detect the edge of blood area, extract the closed boundary;

[0126] Calculate the area A surrounded by the boundary, the number of blood pixels is N, and define the concentration as:

[0127]

[0128] If the concentration C is greater than the empirically set threshold, it is determined to be an abnormal blood flow concentration area.

[0129] 3. Based on deep learning model (U-Net or CNN) to extract the concentration area:

[0130] Use pre-trained or self-defined trained U-Net network to perform semantic segmentation on the blood area in the image, output probability map , where P∈[0,1], represents the probability of each pixel being a blood area;

[0131] Get binary image B(x,y) by threshold segmentation, and count the continuous connected regions;

[0132] Calculate the average probability density or area to pixel ratio of each region, i.e. concentration:

[0133]

[0134] 4. Otsu algorithm as a reference for concentration threshold judgment:

[0135] Use Otsu method to perform gray automatic threshold segmentation on the blood area image, determine the best segmentation threshold;

[0136] To automatically determine the concentration threshold of blood "abnormal concentration" .

[0137] Because the blood distribution of abnormal bleeding area will show local high speed, small diffusion range, high pixel density and other phenomena in space, therefore the concentration C defined above can reflect the potential bleeding points and lesion sites, thereby assisting in diagnosis.

[0138] Further, the boundary shape of the region can be analyzed by morphological processing or region growing method to confirm whether the region is an actual abnormal blood flow area.

[0139] S40. Image segmentation and pathological feature recognition of the blood flow abnormality concentration area, determine whether there is pathological tissue, if so, mark the area, and generate a prompt message.

[0140] In this step, it is necessary to explain the process of image segmentation and pathological feature recognition of the blood flow abnormality concentration area, mainly including the following key links:

[0141] S41. Image segmentation, using image segmentation techniques to preliminarily extract the blood flow abnormality concentration area, separating the blood area from other normal tissue areas. The purpose of image segmentation is to separate the blood flow abnormality area from other normal tissue areas, providing clear regional boundaries for pathological feature recognition.

[0142] Common image segmentation methods include:

[0143] Threshold segmentation: separate the blood area from the background in the image by setting a threshold.

[0144] Edge detection: use Canny edge detection, Sobel operator, etc. to extract the edge of the blood area to ensure that the abnormal area is clear and identifiable.

[0145] Deep learning method: use deep learning models (such as U-Net) to learn global and local features of the image to accurately segment the blood flow abnormality area.

[0146] S42. Extract the shape and boundary features of the blood area. Extract the pathological features of the blood flow abnormality area, which usually includes:

[0147] Morphological features: such as the size, shape, and boundary smoothness of the area, which help identify the likelihood of abnormal tissue.

[0148] Texture features: based on methods such as gray level co-occurrence matrix (GLCM) to extract texture features of the image to determine whether the area has malignant or pathological features.

[0149] Color features: by analyzing the color difference between blood and surrounding tissue, identify whether there is abnormal blood or lesion area.

[0150] Application of deep learning models: use trained deep neural networks (such as convolutional neural networks CNN, ResNet, VGG, etc.) to classify the segmented area and determine whether it has pathological features (such as cysts, adhesions, foreign bodies, etc.).

[0151] In this step, the pathological features of the blood flow abnormality area are extracted, including:

[0152] 1. Extract the shape features of the blood area:

[0153] Segment the blood region into binary image B(x, y), for each connected region R i Calculate the following features:

[0154] Area:

[0155] Perimeter (Boundary Length): Obtain the boundary line using edge detection algorithm (e.g. Canny), the perimeter is the total number of edge pixels.

[0156] Compactness: Used to evaluate whether the shape of the bleeding region is approximately regular:

[0157]

[0158] where is the area of the region, is the perimeter, The closer to 1 indicates the closer to a circle.

[0159] Irregularity can be defined as:

[0160]

[0161] The higher indicates the more irregular the bleeding boundary, usually indicating pathological abnormalities.

[0162] 2. Extract boundary features (Blood Edge Sharpness):

[0163] Calculate the boundary strength by image gradient. Let the image gray level be I(x, y), then the gradient modulus:

[0164]

[0165] Take the boundary pixel set Ei, calculate the average boundary gradient strength:

[0166]

[0167] Higher values usually indicate clear boundaries, while blurred boundaries may indicate diffuse bleeding or lesion areas.

[0168] 3. Pathological feature extraction method:

[0169] Integrate the above shape and boundary features, combined with the speed, density, diffusion direction of blood flow, to construct the pathological feature index set as shown in Table One:

[0170]

[0171] Table One

[0172] In clinical medicine, the pathological bleeding area is usually characterized by irregular bleeding shape, fuzzy boundary, high local blood concentration, abnormal flow speed, wide blood diffusion, and disordered direction.

[0173] The present application quantifies the above-mentioned phenomena into identifiable feature indicators using image processing means, and realizes automatic identification and positioning of the pathological bleeding area.

[0174] S43, using deep learning-based image classification, target detection, and lesion detection methods, to identify whether the blood flow abnormality concentration area belongs to cyst, adhesion, or foreign matter, to label the abnormal blood flow area with pathological features, and to generate prompt information including lesion type, disease diagnosis suggestion, and treatment scheme suggestion.

[0175] Lesion detection: by analyzing the pathological features of the blood flow abnormality area, it is determined whether the area has pathological tissues (such as cysts, adhesions, foreign matter, etc.). In this process, the deep learning model will determine whether the area meets certain pathological features based on the annotation data during training.

[0176] Classification and labeling: once the abnormal pathological area is identified, the system can label the area and generate corresponding prompt information, including:

[0177] Lesion type: such as whether it is a cyst, adhesion, etc.

[0178] Disease diagnosis suggestion: based on the identification result, the system can generate a preliminary diagnosis suggestion to help doctors determine the pathological condition.

[0179] Treatment scheme suggestion: based on the diagnosis result, a treatment scheme suggestion is generated, such as whether further examination is needed, whether surgery is needed, etc.

[0180] In this step, to identify whether the blood flow abnormality concentration area belongs to cyst, adhesion, foreign matter, etc., a deep learning-based image recognition model is used for image classification, target detection, and lesion identification. This method includes the following steps:

[0181] 1. Model selection and structure design

[0182] The present application can use one of the following network structures:

[0183] (1) Image classification model: ResNet-50

[0184] Input: blood flow concentration area image segment (size uniformed as 224x224 RGB) segmented from laparoscope image.

[0185] Output: classification result, belonging to one of [normal, cyst, adhesion, foreign matter, other lesions].

[0186] Loss function: Cross-Entropy Loss

[0187]

[0188] where C is the number of classes, is the true label, is the model prediction probability.

[0189] (2) Object detection model: YOLOv5 or Faster R-CNN

[0190] Input: Complete laparoscopic image or blood flow abnormal area image;

[0191] Output: Position coordinates [x, y, w, h] of each target box, class label (cyst, adhesion, foreign body, etc.);

[0192] Loss function:

[0193] Classification loss (Cross-Entropy);

[0194] Bounding box regression loss (GIoU or SmoothL1);

[0195] Total loss is:

[0196]

[0197] represents the classification loss function, which measures the model's recognition accuracy for blood abnormal areas as cysts, adhesions, or foreign body lesions, etc., using loss functions such as Cross-Entropy Loss or Focal Loss;

[0198] represents the bounding box regression loss function, which measures the deviation between the predicted lesion area bounding box and the true bounding box, using loss functions such as IoU Loss or GIoU Loss;

[0199] is the loss weight coefficient, used to control the contribution proportion of classification loss and regression loss to the total loss, usually set through hyperparameter tuning or automatic weighting learning strategy during training.

[0200] (3) Lesion detection model: U-Net or DeepLabV3+

[0201] Input: Laparoscopic image or blood flow abnormal area image;

[0202] Output: Pixel-level segmentation map, labeling lesion area (using labels such as "adhesion area", "foreign body area").

[0203] Loss function: Dice loss + BCE (Binary Cross Entropy):

[0204]

[0205] Where P is the predicted region and G is the ground truth region.

[0206] 2. Data Preparation and Training Process

[0207] Data acquisition: Collect labeled laparoscopic surgical image data, indicating the lesion area and type;

[0208] Preprocessing includes image enhancement, normalization, and data augmentation (rotation, scaling, cropping).

[0209] Training methods:

[0210] Use the Adam or SGD optimizer;

[0211] Learning rate scheduling;

[0212] Batch size is typically set to 16~32;

[0213] Use GPUs to accelerate training (such as NVIDIA A100);

[0214] Evaluation metrics: The model performance is evaluated using standards such as mAP (mean Average Precision), IoU (Intersection over Union), and F1-score.

[0215] 3. Model Application and Hint Generation

[0216] After training, the model is deployed to the image recognition and analysis module to automatically identify whether areas of abnormal blood flow in newly input laparoscopic images belong to pathological tissue and generate diagnostic information, specifically including:

[0217] Lesion type: Classification or detection label output by the model (e.g., "foreign body", "adhesion");

[0218] Diagnostic recommendations are provided based on the location of the lesion and the confidence level of the model, such as "chronic tubal adhesions may be present";

[0219] Treatment options: Based on the lesion area and confidence level, suggestions may include "conservative treatment is recommended" or "further laparoscopic intervention is recommended".

[0220] 4. Combining medical principles

[0221] According to common medical knowledge:

[0222] Adhesion: usually manifested as structural boundary blur, irregular connection between tissues, segmentation models such as U-Net can effectively extract;

[0223] Cyst: local fluid filling, regular outline, ResNet or YOLO can easily identify its appearance;

[0224] Foreign body: high contrast, abnormal texture, classification network can quickly learn its edge features.

[0225] Further, it further comprises: S44, the generation of prompt information, labeling the pathological region on the image, and generating text or voice prompt information. The prompt information can include: the region may be a pathological region, and further diagnosis is recommended.

[0226] If there is a specific type of lesion (such as cyst, adhesion), the doctor may be prompted to take the next step.

[0227] The application also provides a medical image analysis device based on fallopian tube anatomical segmentation, which applies the medical image analysis method based on fallopian tube anatomical segmentation as described above, comprising:

[0228] An image acquisition and processing module is used to acquire laparoscopic images and perform preprocessing to generate blood flow direction maps and fallopian tube anatomical structure boundary maps;

[0229] An identification and analysis module is used to analyze the blood flow direction maps and anatomical structure boundary maps, locate the bleeding points through image segmentation and pathological feature recognition technology, evaluate the wound size and bleeding speed indicators of the bleeding points, and determine whether there are pathological tissues;

[0230] A decision support module: based on the analysis results of the identification and analysis module, generate corresponding treatment suggestions;

[0231] An interactive module is used to display the analysis results in the form of images, charts or sounds.

[0232] In this embodiment, it should be noted that the image acquisition and processing module uses a laparoscope to acquire images to ensure that high-resolution laparoscopic images are obtained. The acquired images need to have sufficient clarity and details for subsequent processing. In the image preprocessing process, denoising, contrast enhancement, standardization of size, etc. may be required to improve the accuracy and robustness of subsequent image analysis. After image preprocessing, blood flow direction maps and fallopian tube anatomical structure boundary maps are generated to provide basic data for subsequent analysis.

[0233] The identification and analysis module comprehensively analyzes the generated blood flow direction maps and fallopian tube anatomical structure boundary maps. The specific steps include:

[0234] Blood flow analysis: By analyzing the speed, direction, and spread of blood flow, potential bleeding points are identified, and abnormal concentration areas of blood flow are further analyzed.

[0235] Anatomical structure analysis: Using image segmentation and edge detection techniques, different anatomical regions of the fallopian tube are extracted and labeled, such as the fimbria, ampulla, and isthmus, providing a basis for subsequent pathological feature recognition.

[0236] Pathological feature recognition: Based on deep learning or traditional machine learning methods, pathological tissues such as cysts and adhesions are identified and analyzed to assess the nature of abnormal areas and determine whether further treatment is needed.

[0237] Decision support module: Based on the results of the recognition and analysis module, this module generates treatment recommendations. Treatment recommendations may include:

[0238] Conservative treatment recommendations: When the wound is small, blood flow is not significant, or pathological tissues are mild, the system may recommend conservative treatment.

[0239] Further examination or surgery: When larger wounds, abnormal blood flow, or severe pathological tissues are found, the system may recommend further examination or surgical treatment.

[0240] Personalized support: The system generates personalized treatment recommendations based on the patient's specific circumstances (such as age, medical history, etc.) to provide more accurate decision support.

[0241] Interactive module: The analysis results are delivered to doctors or patients through images, charts, or sound. Interactive methods may include:

[0242] Image display: By highlighting bleeding points, pathological areas, and related anatomical structures, doctors can visually see problem areas.

[0243] Chart display: Charts showing blood flow speed, flow, concentration, and other data help doctors better understand bleeding conditions.

[0244] Voice prompt: If necessary, the system can also provide voice prompts to remind doctors of key areas of concern or disease types in real time.

[0245] User interaction experience: The interactive module also has a simple and easy-to-use user interface that supports doctors to perform interactive operations, view and analyze different results, and make further diagnoses or decisions.

[0246] The technical solutions of steps S10-S40 of the embodiment are implemented, the position of the bleeding point can be accurately identified by acquiring the laparoscope image and analyzing the blood flow direction and speed, combined with the anatomical segmentation of the fallopian tube (umbrella part, isthmus part, ampulla part). After the bleeding point is confirmed, the bleeding speed, flow rate and wound size of the bleeding point can be calculated to further evaluate the severity of the bleeding. If the wound is small, the system will automatically output a conservative treatment suggestion to avoid the doctor making an over-treatment decision in an emergency, thereby reducing the damage to the patient. Through image segmentation and pathological feature recognition of the abnormal blood flow area, it can be judged whether there is pathological tissue (such as cyst, adhesion or fertilized egg rupture residue). This function provides more accurate intraoperative diagnostic information for doctors, helping doctors avoid missing the lesion area or making mistakes.

[0247] Through image fusion, blood flow trajectory analysis and regional contact relationship modeling, the system can timely find the abnormal blood flow area, assist the doctor to make more accurate and safe decisions during the operation. Especially in emergency situations, it avoids the doctor making a "one-size-fits-all" treatment decision and reduces the risk of misdiagnosis.

[0248] The whole process can effectively improve the efficiency of intraoperative judgment, reduce the work pressure and operation time of the doctor, and improve the overall efficiency of the operation through the comprehensive application of image fusion, edge detection, deep learning and other technologies.

[0249] The present application can not only detect blood flow and tissue structure, but also provide treatment suggestions based on the analysis results, support doctors to make accurate decisions in emergency situations, reduce misoperation and risk, and further improve the quality and safety of medical services.

[0250] The above-described embodiments only express several embodiments of the present application, and the description is more specific and detailed, but it should not be understood as limiting the scope of the patent application. It should be noted that for ordinary skilled in the art, without departing from the concept of the present application, a number of modifications and improvements can be made, which are within the scope of protection of the present application. Therefore, the protection scope of the present application patent should be subject to the appended claims.

Claims

1. A medical image analysis method based on fallopian tube anatomical segmentation, characterized in that, include: Acquire laparoscopic images, analyze the blood distribution in the laparoscopic images, extract the blood flow direction and velocity information, and generate a blood flow direction map; The fallopian tube structure in the laparoscopic image is divided into three regions: fimbriae, ampullae, and isthmus. The structural edges of the three regions are extracted and marked to generate an anatomical boundary map. The fimbriae include the connection between the fallopian tube and the ovary, the isthmus includes the connection between the fallopian tube and the uterus, and the ampullae is the connection between the fimbriae and the isthmus. Image fusion processing is performed on the blood flow direction map and the anatomical structure boundary map to mark the contact boundaries between blood and the three regions, analyze the contact relationship between the blood flow trajectory and the structural edges of the three regions, and identify areas of abnormal blood flow concentration. The image segmentation and pathological feature recognition are performed on the area of ​​concentrated abnormal blood flow to determine whether there is pathological tissue. If so, the area is marked and a prompt message is generated.

2. The medical image analysis method based on fallopian tube anatomical segmentation according to claim 1, characterized in that, After generating the blood flow direction map, the process further includes: constructing a vector field model based on the blood flow direction and velocity information, and inversely calculating the spatial coordinates of the bleeding points. Specifically, Extract the points with maximum velocity gradients from the blood flow pattern and mark them as candidate bleeding points; The flow field distribution of different combinations of the candidate bleeding points is simulated, and the bleeding points are determined by maximum likelihood estimation.

3. The medical image analysis method based on fallopian tube anatomical segmentation according to claim 2, characterized in that, After determining the bleeding point through maximum likelihood estimation, the process also includes: Determine whether the bleeding point is located in the ampulla of Vater; If so, calculate the bleeding rate and amount of bleeding at the bleeding point, determine the size of the wound at the bleeding point, and if the wound size is less than a preset wound size threshold, output a prompt message suggesting conservative treatment.

4. The medical image analysis method based on fallopian tube anatomical segmentation according to claim 1, characterized in that, The step of extracting and marking the structural edges of the three regions to generate an anatomical boundary map includes: The anatomical segments of the fallopian tubes in the laparoscopic images are identified; An edge detection algorithm or a deep learning-based edge detection model is applied to extract the edges of each anatomical structure and mark the edges.

5. The medical image analysis method based on fallopian tube anatomical segmentation according to claim 4, characterized in that, The edge detection algorithms include: Canny operator, Sobel operator, HED algorithm, or CombineNet algorithm.

6. The medical image analysis method based on fallopian tube anatomical segmentation according to claim 1, characterized in that, The image fusion processing of the blood flow direction map and the anatomical structure boundary map includes: The blood flow direction map and the anatomical structure boundary map are fused at the channel level using multi-channel data fusion technology. The fusion methods include weighted averaging, maximum value synthesis, or fusion filtering. Perform boundary enhancement processing on the fused image; The relative position and contact between blood flow and anatomical structure boundaries are analyzed using spatial relationship modeling algorithms. Output the fused image as a multidimensional data graph.

7. The medical image analysis method based on fallopian tube anatomical segmentation according to claim 1, characterized in that, The identified areas of concentrated abnormal blood flow include: Edge extraction of the blood flow region includes: pixel-level feature extraction of the blood flow region; Based on blood flow velocity, flow rate, and blood diffusion range, the concentration of blood flow in each blood flow region is calculated; Determine whether the concentration of blood flow exceeds a preset concentration threshold. If so, mark the area as an area of ​​abnormally concentrated blood flow.

8. The medical image analysis method based on fallopian tube anatomical segmentation according to claim 7, characterized in that, The calculation of the concentration of blood flow in each blood flow region includes: calculating the concentration of blood flow in the blood flow region based on one or more of the following algorithms: K-Means clustering algorithm, DBSCAN algorithm, Otsu algorithm, Canny edge detection algorithm, U-Net algorithm, and CNN algorithm.

9. The medical image analysis method based on fallopian tube anatomical segmentation according to claim 1, characterized in that, Image segmentation and pathological feature identification of the aforementioned areas of concentrated abnormal blood flow include: Image segmentation technology is used to initially extract the areas of abnormal blood flow, separating the blood areas from other normal tissue areas; Extract the shape and boundary features of the blood region; Using deep learning-based image classification, target detection, and lesion detection methods, the system identifies whether the areas of abnormal blood flow concentration belong to cysts, adhesions, or foreign bodies. Abnormal blood flow areas with pathological characteristics are labeled, and prompts including lesion type, disease diagnosis suggestions, and treatment plan suggestions are generated.

10. A medical image analysis device based on fallopian tube anatomical segmentation, employing the medical image analysis method based on fallopian tube anatomical segmentation as described in any one of claims 1-9, characterized in that, include: The image acquisition and processing module is used to acquire laparoscopic images and perform preprocessing to generate blood flow direction maps and fallopian tube anatomical structure boundary maps. The identification and analysis module is used to analyze the blood flow direction map and the anatomical structure boundary map. Through image segmentation and pathological feature recognition technology, it locates bleeding points, assesses the wound size and bleeding rate of the bleeding points, and determines whether pathological tissue exists. Decision support module: Based on the analysis results of the identification and analysis module, it generates corresponding treatment suggestions; Interactive modules are used to display analysis results in the form of images, charts, or sound.

Images