Method, apparatus and storage medium for automatically identifying and locating perforator vessels

Abstract

The present invention relates to a method, device, and storage medium for automatically identifying and locating perforator vessels, which includes the following steps: determining a scanning range in the scanning direction from the target perforator, and extracting 3D image data within the scanning range; performing dimensionality reduction processing on the 3D image data to obtain slice-segmented image data; using the slice-segmented image data as the smallest analysis unit, screening the signal points of perforator vessels using a signal-to-noise ratio dynamic threshold method; 3D reconstructing the signal points of the connected perforator vessels in the 3D image data, and screening candidate perforator vessels based on the total score of the signal-to-noise ratio after reconstruction; and identifying the candidate perforator vessels based on their relative position, angle, and depth to the superficial layer and deep muscle layer. The present invention can realize automatic and fast location identification of perforator vessels.

Description

[Technical Field]

[0001] The present invention relates to the technical field of medical image processing, and in particular to a method for automatically identifying and locating perforator vessels. [Background technology]

[0002] Perforator flap transplantation is at the forefront of microsurgery and is an essential technique for superficial organ reconstruction (e.g., breast reconstruction), repair of traumatic or tumor defects, achieving optimal repair results while minimizing damage to the donor area.

[0003] A skin flap is a composite tissue fragment containing skin and subcutaneous tissue, including its own nutrient vessels. Skin flaps are typically transferred from one location to another in the body to repair defects caused by the above factors; this surgical procedure is called a skin flap transplant. Perforator vessels are blood vessels that penetrate the skin and the corresponding subcutaneous tissue, and their function is to supply blood to the perforated skin. A distinctive feature of perforator flaps is that the perforator vessels that supply the flap are completely dissected to the skin perforation point, preserving the main trunk vessels from which they originate, and leaving other tissues, such as muscles, that receive nutrients from the main trunk unaffected. Conventional flap resection techniques do not dissect the vessels to the skin perforation point, but instead severe the vessels at the main trunk, causing significant damage.

[0004] The main technical difficulty with perforator flap is that the perforator vessels vary widely, not only between different bodies but also between different sides of the same body. In addition, the diameter of the perforator vessels at their skin penetration points is very small (0.5-1mm). If the exact location is unknown, they are easily damaged, which means the surgery will fail. This increases the surgical risk and places a great deal of stress on the surgeon during the procedure.

[0005] Preoperative localization of the perforator vessel skin penetration points using visualization methods allows the surgeon to know the location of the perforator vessel skin penetration points before surgery, significantly reducing the risk of intraoperative dissection and the required time, thereby improving the success rate of surgery. Tomographic medical imaging examinations (e.g., enhanced CT angiography, CTA, enhanced MR angiography, MRA) are common methods. Tomographic perforator vessel localization technology is relatively mature, with high localization accuracy and the location information of the perforator vessels can be confirmed from different angles. Volume rendering (VR) technology is then used to mark the perforator vessel skin penetration points on the patient's body surface.

[0006] However, a major limitation of the above method is that the image analysis and localization tasks must be completed by the surgeon on the surgical team; domestic radiologists cannot substitute for them due to limited anatomical understanding or other reasons, and there are few radiologists who have mastered this technique. The time required for tomographic image analysis and marking of perforator vessels usually takes about 45 minutes, which greatly consumes the surgeon's energy and involves mechanical duplication of labor. Summary of the Invention

[0007] The technical problem to be solved by the present invention is to provide a method for automatically identifying and locating perforator vessels, which realizes automatic and fast location of perforator vessels.

[0008] The technical solution used to solve the technical problem of the present invention provides a method for automatically identifying and locating perforator vessels, the method comprising the following steps: Determine a scanning range in the scanning direction from the target perforator vein, and extract three-dimensional image data within the scanning range; performing dimension reduction processing on the three-dimensional image data to obtain layer-segmented image data; The slice-segmented image data is used as the minimum analysis unit, and the signal points of the perforator vessels are screened by the signal-to-noise ratio dynamic threshold method; 3D reconstructing the signal points of the connected perforator vessels in the 3D image data, and screening candidate perforator vessels based on the total score of signal-to-noise ratio after reconstruction; The perforator vessels are identified based on their relative position, inclination and depth to the superficial and deep muscle layers.

[0009] Specifically, performing dimension reduction processing on the three-dimensional image data to obtain layer-segmented image data includes the following steps: Merging n consecutive images in the 3D image data to obtain signal-enhanced 2D image data; Mark the boundary of the skin surface and the boundary between the subcutaneous fat and muscle in the 2D image data, and set the subcutaneous fat area surrounded by the two boundary lines as the target analysis area; The target analysis region is divided into layers, and a plurality of continuous images of layer-divided image data at the subcutaneous depth are obtained. When the n consecutive images in the three-dimensional image data are merged, the pixel value of the merged image is determined to be the smallest pixel value at the same position.

[0010] The boundary coordinates between the air and the skin surface are obtained by a machine learning algorithm, and the boundary coordinates between the subcutaneous fat and the muscle are obtained from the rate of change of the pixel average value.

[0011] The step of screening the signal points of the perforator vessels by using the signal-to-noise ratio dynamic thresholding method with the layer-by-layer segmented image data as the smallest analysis unit specifically includes the following steps: converting the layer-segmented image data of the same subcutaneous depth into a two-dimensional array, determining the minimum or maximum pixel value of several pixel points at different depths at the same skin position in the two-dimensional array as a candidate blood vessel signal point, and determining the minimum or maximum pixel value of a nearby position as a background reference value, and calculating the relative signal-to-noise ratio of the candidate blood vessel signal point; Points whose pixel values are smaller than a pixel threshold are selected as a set, and points whose relative signal-to-noise ratios exceed the signal-to-noise ratio threshold are selected from the set as signal points of perforator vessels, and the abscissas corresponding to the signal points of the perforator vessels are marked.

[0012] The relative signal-to-noise ratio of the candidate vessel signal points is calculated by ((X4+X5 / 2)*(X2+X3+X4+X5)) / (X0+X1), where X n is the average value of pixels that are n pixels away from the candidate blood vessel signal point, and X0 is the pixel value of the candidate blood vessel signal point.

[0013] The three-dimensional reconstruction of the signal points of the connected perforator vessels in the three-dimensional image data is specifically as follows: Project the abscissas corresponding to the signal points of the perforator vessels onto the two-dimensional image data, merge the points adjacent to the signal points of the perforator vessels in the two-dimensional image data, and calculate the continuity, spatial progression and signal-to-noise ratio of the merged signal points of the perforator vessels by summing; The two-dimensional image data is fused to form a three-dimensional space, a merged point in the two-dimensional image data is placed in the three-dimensional space, and the merged point in the two-dimensional image data is merged with a point adjacent to the merged point in the three-dimensional space, and a signal-to-noise ratio after merging is calculated by summing.

[0014] After step (5), a total score for each perforator vessel is further calculated based on the signal-to-noise ratio and vessel mass, and the perforator vessel with the highest total score is taken as the final perforator vessel.

[0015] The technical solution used to solve the technical problem of the present invention is a computer device, which includes a storage device and a processor, a computer program stored in the storage device, and when the computer program is executed by the processor, the processor performs the steps of the method for automatically identifying and locating perforator vessels.

[0016] The technical solution used to solve the technical problem of the present invention is a readable storage medium having an executable computer program stored therein, which, when executed by a processor, causes the processor to perform the steps of the method for automatically identifying and locating perforator vessels described above. [Effects of the Invention]

[0017] By using the above technical solution, the present invention has the following advantages and positive effects compared to existing technologies: The present invention reduces 3D image data to 2D and searches for perforator signal points in 1D images, thereby improving processing speed and enabling the task of locating perforator vessels to be completed within 5 minutes. The method of the present invention uses the artificial judgment result of a doctor who is familiar with the technology of locating perforator vessels in tomographic medical images as the gold standard, and the results of automatic perforator vessel location basically match the gold standard, with an identification rate of dominant perforator vessels exceeding 99%.

[0018] In the following, in order to more clearly explain the embodiments of the present invention or the technical solutions in the prior art, drawings necessary for explaining the embodiments or the prior art will be briefly introduced. Of course, the drawings in the following description are only some embodiments of the present invention, and those skilled in the art can derive other drawings from these drawings without any creative work. [Brief explanation of the drawings]

[0019] [Figure 1] FIG. 1 is a flow diagram of the first embodiment of the present invention. [Figure 2] FIG. 2 is a schematic diagram of an analysis range of two-dimensional image data in the first embodiment of the present invention. [Figure 3] FIG. 3 is a schematic diagram showing how two-dimensional image data is converted into a rectangular shape in the first embodiment of the present invention. [Figure 4] FIG. 4 is a schematic diagram showing how two-dimensional image data is converted into one-dimensional image data in the first embodiment of the present invention. [Figure 5]FIG. 5 is a schematic diagram showing the output result of single-layer two-dimensional image data in the first embodiment of the present invention. [Figure 6] FIG. 6 is a schematic diagram of the result after merging two-dimensional image data of different layers at consecutive points in three-dimensional space in the first embodiment of the present invention. [Figure 7] FIG. 7 is a schematic diagram showing the distinction between subcutaneous veins and perforator vessels in the first embodiment of the present invention. [Figure 8] FIG. 8 is a schematic diagram of a preliminary comparison of the calculation results of the first embodiment of the present invention with the gold standard. [Figure 9] FIG. 9 is a schematic diagram of a human-computer interaction interface according to a second embodiment of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0020] The present invention will be further described below with reference to specific examples. It should be understood that these examples are only used to explain the present invention and do not limit the scope of the present invention. After reading the contents of the present invention, a person skilled in the art can make various changes and modifications to the present invention, and it should be understood that equivalent embodiments thereof are included in the scope of the claims of the present invention.

[0021] An embodiment of the present invention relates to a method for automatically identifying and locating perforator vessels, which can automatically analyze medical image data, automatically calculate and identify target perforator vessels, sort them by dominance, and return their location information, which is useful for clinical and scientific research applications that require preoperative location of perforator vessels or other perforator vessels. As shown in Figure 1, the method includes the following steps: determine a scanning range in the scanning direction from the target perforator, and extract 3D image data within the scanning range; perform dimensionality reduction processing on the 3D image data to obtain slice-segmented image data; use the slice-segmented image data as the minimum analysis unit to screen for perforator vessel signal points using a signal-to-noise ratio dynamic threshold method; perform 3D reconstruction of the signal points of the connected perforator vessels in the 3D image data, and screen for candidate perforator vessels based on the total score of the signal-to-noise ratio after reconstruction; identify the candidate perforator vessels based on their relative positions, inclinations, and depths to the superficial layer and deep muscle layer, and return the coordinates of the perforator vessels in the 3D coordinate system.

[0022] The method automatically reads and analyzes data in normalized tomographic medical image files (e.g., DICOM), automatically identifies the muscle penetration points of perforator vessels in the data and their progression paths in the subcutaneous soft tissue, scores and sorts the perforator vessels based on information such as signal strength and progression direction, and outputs the location information of the perforator vessels, as well as their scores and sorting status, in the calculation results.

[0023] 1) Read and analyze the data from the normalized tomography file (read a multidimensional array containing the z, y, and x coordinates from the original image plus the pixel values at those coordinates).

[0024] 2) Limiting the analysis range in the z-axis. The scanning range in the z-axis direction is limited based on the target perforator, and the corresponding data is extracted. Taking the deep inferior epigastric artery (DIEA) as an example, the original image covers the human body from the chest to the knee, and the analysis range required for the DIEA vessel is from the navel to the perineum. Therefore, based on the existing image features (the navel appears as a depression in the cross-sectional image), the boundary of the body surface is first identified, and the location where the depression appears on the boundary of the body surface is identified as the navel. The DICOM image from the navel to the pubic bone is then used for analysis.

[0025] 3) After appropriately merging adjacent horizontal cross-sectional images in the z-axis, 2D image data is acquired and analyzed. Because the subcutaneous perforator vessels travel continuously, the perforation signals in adjacent cross-sectional images are generally adjacent and essentially continuous or fused after merging. The number of horizontal cross-sectional images in the z-axis after merging = the number of horizontal cross-sectional images within the range taken in the previous step / n (n = the number of adjacent merged images). The merging logic in this step is to use the point with the smallest image value at the same position in the adjacent n images as the pixel value of the merged image. Subsequent analysis is performed on each merged horizontal cross-sectional image (2D), converting the 3D image into a 2D image for analysis.

[0026] 4) Limit the analysis range in the two-dimensional image, i.e., mark the boundary between air and the skin surface and the boundary between subcutaneous fat and muscle in the two-dimensional image data, and define the area surrounded by the two boundary lines as the target analysis area.

[0027] 4.1 Upper boundary of the analysis area: In this step, we define the upper boundary as skin, and mark the boundary between the air and the skin surface using the watershed algorithm.

[0028] 4.2 Lower Boundary of Analysis Area: a) The 2D image is cut into rectangular strips, and the body surface shape is used as the edge. The image values of a 10-pixel high pixel are read. These values are then input into a child image with a height of 10 and a width equal to the full length of the original image on the x-axis. The child image is then reconstructed to convert the irregular shape into a rectangle. b) Determining the Lower Boundary of the Analysis Area: The abdominal cavity and the area other than the body surface are similar and primarily dark in color. The pixel values of the subcutaneous tissue in the target analysis area are relatively high. A threshold is set on the cut-out rectangular child image to determine the lower boundary of the analysis area. The average grayscale value of each of the cut-out rectangular images with a height of 10 pixels is calculated. When the proportion of points with high pixel values and closer to white reaches the threshold, it is determined that the image has entered the muscle area and reached the lower boundary of the analysis area. Figure 2 shows the analysis area of 2D image data in this embodiment. In Figure 2, A is the upper boundary of the analysis area, and B is the lower boundary of the analysis area. The area between the two is the target analysis area.

[0029] 5) The 2D image within the analysis range is divided into strips, and each strip is analyzed as a 1D array. Using the method described in 4.2 a), the 2D image is divided into strips with a height of 10 pixels. See FIG. 3. At this time, the 2D image is converted into a 2D array, which is then converted into a 1D array. As shown in FIG. 4, the data structure of the 1D array is [value corresponding to abscissa 1, value corresponding to abscissa 2, ...], where the element number in the array corresponds to the abscissa, and the value corresponding to the abscissa is the minimum pixel value (i.e., the point with the strongest signal and closest to black) among the 10 pixel points corresponding to the same abscissa. The signal-to-noise ratio (SNR) of "strong signal" points is calculated. In this embodiment, the SNR is the ratio of the pixel value of the point itself to the pixel values of its surrounding points. The pixel values of blood vessels are low, close to 0 (dark in the image), while the pixel values of the surrounding points are high (closer to white in the image). Therefore, the greater the difference between the signal point and the surrounding points, the better it matches the characteristics of blood vessels and the higher the score. The logic for calculating the signal-to-noise ratio in this process is ((X4+X5 / 2)*(X2+X3+X4+X5)) / (X0+X1), where X n is the average value of pixels that are n pixels away from the signal point of the perforator vein, and X0 is the pixel value of the signal point of the perforator vein.

[0030] 6) Find "strong signal" points (points closest to black in the figure and with the smallest pixel values) from the one-dimensional image data. In this embodiment, "strong signal" points are perforator vessel signal points, which can be found by the following method: Select points with pixel values less than 60 to form a set, and further select points from the set whose relative signal-to-noise ratio is greater than the median cutoff of the set as perforator vessel signal points, and mark the abscissas corresponding to the perforator vessel signal points. If the ordinates corresponding to multiple perforator vessel signal points are too close, they can be merged appropriately.

[0031] The above steps 5) and 6) are repeated until each two-dimensional image is divided into layers and the calculation is completed.

[0032] 7) At the location of a "strong signal" point in the 2D image, merge the adjacent points and add up the signal-to-noise ratios of these points. That is, project the abscissa corresponding to the signal point of the perforator vessel onto the 2D image data, merge the points adjacent to the signal point of the perforator vessel in the 2D image data, and calculate the sum of the signal-to-noise ratios of these points. This will obtain the continuity, spatial progression, and signal-to-noise ratio of the merged signal point of the perforator vessel. Figure 5 is a schematic diagram of the output result of a single-layer 2D image.

[0033] 8) The points in the 2D image from the previous step are put into a 3D space, adjacent points are merged, and the signal intensities are summed. Because the signals from the same perforator vein appear in different scan sections but are adjacent in location, this step fuses the 2D images to form a 3D space, places the previously identified points in the 3D space, determines the distance, merges adjacent points, and sums the signal intensities (see Figure 6).

[0034] It should be noted that each threshold value in the above steps can be calibrated with the radiological data of an actual patient, and after adjustment, optimal calculation results can be obtained.

[0035] 9) Distinguishing between subcutaneous veins and perforator vessels: Signals with high signal-to-noise ratios in adjacent locations may represent not only the target perforator vessels, but also subcutaneous veins. While both have high signal intensities in the above process, the main difference between them is that perforator vessels continue to travel deep and eventually contact the surface of the muscle, while subcutaneous veins travel parallel to the fat layer and do not contact deep tissue (see Figure 7, where C represents the perforator vessel and D represents the venous signal). Therefore, in this process, the venous signal is eliminated by distinguishing between subcutaneous veins and perforator vessels using logic based on the slope of the signal obtained in the previous process relative to the deep muscle layer and logic based on signal depth.

[0036] 10) After marking the perforator vessels, they are comprehensively scored to obtain the optimal perforator vessel. After vein signals are eliminated, there may be multiple perforator vessels. To ensure the success rate of the surgery, in this embodiment, each perforator vessel is scored, and the optimal perforator vessel is selected as the final perforator vessel. When scoring in this process, the signal-to-noise ratio of the perforator vessels and other logic favorable to determining the quality of the identified vessels (e.g., outer diameter, branching) are combined to comprehensively score multiple perforator vessels. The scoring may be performed by weighted summation. After the score is calculated, the location of the candidate perforator vessel with the highest score is marked, and this location is identified as the optimal perforator vessel.

[0037] 11) The coordinates of the optimal muscle penetration points of the perforator vessels are converted into the coordinate system of the body surface. Specifically, the spatial position information of the muscle penetration points of the perforator vessels is projected onto the body surface, and a coordinate system is constructed with an arbitrary mark point within the target area of the skin on the body surface as the origin. An algorithm is used to calculate the coordinates of the projection point in the body surface coordinate system, making it easier for the user to mark the position information on the patient's body surface, and providing support for intraoperative design.

[0038] The perforator vessels obtained by the above method basically meet the gold standard, with a recognition rate of over 99%. The surgeon can review the images of the automatically identified perforator vessels, confirm the results, and select the most suitable candidate perforator vessels according to the requirements of the surgical design. Here, the gold standard is the artificial judgment result of a physician who is familiar with the technology of identifying perforator vessels in medical tomography images.

[0039] A preliminary test was conducted using data from 20 actual cases (12 cases were perforator vessels below the abdominal wall and 8 cases were perforator vessels of the descending branch of the lateral femoral artery). The calculation results of this embodiment were compared with the average judgment results of three clinicians with more than 5 years of experience in identifying artificial perforator vessels. The matching rate of the top 3 perforator vessels was 98.33%, the matching rate of the top 5 perforator vessels was 96%, and the matching rate of the top 10 perforator vessels was 90.50% (see Figure 8).

[0040] It is easy to understand that the present invention reduces the dimension of three-dimensional image data to two dimensions and searches for the signal points of the perforator vessels in one-dimensional images, thereby improving the processing speed and enabling the task of locating the perforator vessels to be completed within five minutes, which significantly reduces the work time for locating the perforator vessels compared to the original 45 minutes, and improves work efficiency.

[0041] A second embodiment of the present invention relates to a computer system including a storage device and a processor, wherein a computer program is stored in the storage device, and when the computer program is executed by the processor, the processor performs the steps of the method for automatically identifying and locating perforator vessels in the first embodiment. This embodiment can be implemented on many mobile computing platforms to realize mobile or remote operation of the identification process (see FIG. 9).

[0042] A third embodiment of the present invention relates to a readable storage medium having an executable computer program stored therein, the executable computer program, when executed by a processor, causing the processor to perform the steps of the method for automatically identifying and locating perforator vessels of the first embodiment.

Claims

1. 1. A method for automatically identifying and locating perforator vessels, comprising: determining a scanning range in a scanning direction from the target perforator vein, and extracting three-dimensional image data within the scanning range; performing a dimension reduction process on the three-dimensional image data to obtain layer-divided image data; a step of screening signal points of perforator vessels by a signal-to-noise ratio dynamic threshold method using the layer-divided image data as a minimum analysis unit; 3D reconstruction of the signal points of the connected perforator vessels in the 3D image data, and screening candidate perforator vessels by the total score of signal-to-noise ratio after reconstruction; determining perforator vessels from the relative position, inclination, and depth of candidate perforator vessels with respect to the superficial and deep muscle layers; 1. A method for automatically identifying and locating perforator vessels, comprising:

2. Specifically, the step of performing dimension reduction processing on the three-dimensional image data to obtain layer-divided image data includes the following steps: Merging n consecutive images of the three-dimensional image data to obtain two-dimensional image data with enhanced signals; Marking the boundary line of the skin surface and the boundary line between the subcutaneous fat and muscle in the two-dimensional image data, and setting the subcutaneous fat area surrounded by the two boundary lines as the target analysis area; Segmenting the target analysis region into layers to obtain a plurality of continuous subcutaneous depth layer-segmented image data; 2. The method of claim 1, further comprising:

3. The method for automatically identifying and locating perforator vessels according to claim 2, characterized in that when n consecutive images in the three-dimensional image data are merged, the pixel value of the merged image is determined to be the point with the smallest pixel value at the same position.

4. The method for automatically identifying and locating perforator vessels according to claim 2, wherein the boundary coordinates between the air and the skin surface are obtained by a machine learning algorithm, and the coordinates of the boundary line between the subcutaneous fat and the muscle are obtained from the rate of change of pixel average values.

5. The step of screening the signal points of the perforator vessels by the signal-to-noise ratio dynamic threshold method using the layer-by-layer divided image data as the minimum analysis unit is specifically as follows: converting the layer-divided image data of the same subcutaneous depth into a two-dimensional array, and taking the minimum or maximum pixel value of several pixel points at different depths at the same skin position in the two-dimensional array as a candidate blood vessel signal point, and taking the minimum or maximum pixel value of a nearby position as a background reference value, and calculating a relative signal-to-noise ratio of the candidate blood vessel signal point; Selecting points whose pixel values are less than a pixel threshold as a set, and further selecting points whose relative signal-to-noise ratios exceed the signal-to-noise ratio threshold from the set as perforator vessel signal points, and marking the abscissas corresponding to the perforator vessel signal points; 2. The method of claim 1, further comprising:

6. When calculating the relative signal-to-noise ratio of the candidate vessel signal points, ((X ４ +X ５ / 2) * (X ２ +X ３ +X ４ +X ５ )) / (X ０ +X １ ) where X ｎ is the average value of pixels that are n pixels away from the candidate blood vessel signal point, and X ０ 6. The method of claim 5, wherein σ is the pixel value of the candidate vessel signal point.

7. The step of three-dimensionally reconstructing signal points of the connected perforator vessels in the three-dimensional image data specifically includes: Projecting the abscissas corresponding to the signal points of the perforator vessels onto the two-dimensional image data, merging the points adjacent to the signal points of the perforator vessels in the two-dimensional image data, and calculating the continuity, spatial progression and signal-to-noise ratio of the merged signal points of the perforator vessels by summing them; fusing the two-dimensional image data to form a three-dimensional space, placing a merged point in the two-dimensional image data into the three-dimensional space, merging the merged point in the two-dimensional image data with a point adjacent to the merged point in the three-dimensional space, and calculating a merged signal-to-noise ratio by summing; 2. The method of claim 1, further comprising:

8. 2. The method for automatically identifying and locating perforator vessels according to claim 1, further comprising the step of: after step (5), calculating a total score for each perforator vessel based on the signal-to-noise ratio and vessel mass; and selecting the perforator vessel with the highest total score as the final perforator vessel.

9. 9. A computer system comprising: a storage device and a processor; a computer program stored in the storage device; and, when the computer program is executed by the processor, the processor performs steps of the method for automatically identifying and locating perforator vessels according to any one of claims 1 to 8.

10. 10. A computer-readable storage medium having an executable computer program stored therein, the executable computer program, when executed by a processor, causing the processor to perform the steps of the method for automatically identifying and locating perforator vessels according to any one of claims 1 to 8.

Images