Three-dimensional volume data-based fetal face imaging method

Abstract

The present invention relates to the technical field of ultrasound imaging, in particular to a three-dimensional volume data-based fetal face imaging method. The present invention uses the following technical solution: after first performing frame extraction processing on selected sequential two-dimensional images of the face of a fetus, then calculating mean values and variances of the two-dimensional images; removing a problematic image of which the mean value or the variance is lower than a set threshold; using a trained segmentation network model to perform segmentation and recognition processing to obtain a reference profile; and after performing fitting to obtain a reference curve, using the reference curve to segment the sequential two-dimensional images. The beneficial effects of the present invention are: by means of removing problematic images such as entirely black images or incomplete sector images and performing frame extraction processing, the number of two-dimensional images to be subjected to segmentation and recognition processing is reduced, and segmentation of the sequential two-dimensional images related to the face of the fetus is performed by using the reference profile, so that occlusions in front of the face of the fetus can be effectively removed. Moreover, the present invention has the advantages of requiring a small amount of computation and having a high three-dimensional imaging rate.

Description

A fetal facial imaging method based on three-dimensional volume data Technical Field

[0001] The present invention relates to the technical field of ultrasonic imaging, and in particular to a fetal face imaging method based on three-dimensional volume data. Background Art

[0002] With the rapid development of ultrasound technology, three-dimensional ultrasound has been widely used in clinical examinations due to its advantages such as low acquisition difficulty, fast display speed, and fast computing speed. When examining a fetus in the womb, it is necessary to observe the fetus's face in detail. During three-dimensional ultrasound imaging, fluid accumulation or certain postures of the fetus can cause obstructions to appear in different positions in front of the fetus's face, preventing doctors and parents from directly observing the details of the fetus's face. Currently, a commonly used method is to use a pre-trained model to identify and process two-dimensional ultrasound image frames of the fetus's face, removing parts that do not belong to facial features to achieve the purpose of removing obstructions on the fetus's face. However, if there are too many two-dimensional ultrasound image frames of the fetus's face, performing model recognition processing on all of them will greatly reduce the rate of three-dimensional imaging. Furthermore, since the fetus often moves during the examination, a low imaging rate will result in poor three-dimensional imaging results, and higher requirements will be placed on hardware processing equipment. Summary of the Invention

[0003] The purpose of the present invention is to provide a fetal face imaging method based on three-dimensional volume data, specifically to provide an imaging method that can accurately and quickly eliminate obstructions in front of the fetal face in three-dimensional volume data and completely segment the fetal face.

[0004] To achieve the above object, the present invention adopts the following technical solution: a fetal face imaging method based on three-dimensional volume data, comprising the following steps:

[0005] S01. Acquire three-dimensional volumetric data of a fetus, and select a sequence of two-dimensional images constituting the fetal face from the three-dimensional volumetric data of the fetus, where the number of two-dimensional images contained in the sequence is n; wherein the directions of the sequence of two-dimensional images are all parallel to the midsagittal plane of the fetal face.

[0006] S02. Perform frame extraction processing on n two-dimensional images at intervals of one image to obtain k two-dimensional images.

[0007] S03. Calculate the mean and variance of the k two-dimensional images respectively, and eliminate the two-dimensional images whose mean is less than the set threshold T1 and whose variance is less than the set threshold T2. The number of eliminated two-dimensional images is m, and the remaining km two-dimensional images are obtained.

[0008] S04. Use the trained Unet segmentation network model to segment and recognize the km two-dimensional images after the frame extraction processing in step S03, and segment the facial area from the km two-dimensional images.

[0009] S05. Binarize the km 2D images that segment the facial area, traverse each contour in each 2D image, retain the contour with the largest area in each 2D image, select the contour with the largest distance in the Y-axis direction from the contours with the largest area in the km 2D images as the reference contour, extract points on the left half of the reference contour, and fit it to obtain a reference curve.

[0010] S06 , using the reference curve, processing the n two-dimensional images obtained in step S01 , and blacking out all pixels in the n two-dimensional images that are directly to the left, above the left, and below the left of the reference curve.

[0011] S07 , replacing the sequence of two-dimensional images selected in step S01 from the three-dimensional volume data of the fetus with the n two-dimensional images processed using the reference curve, thereby generating three-dimensional volume data of the fetus after removing obstructions.

[0012] The beneficial effects of the present invention are as follows: by eliminating problem images such as pure black images or incomplete fan-shaped images, and by using frame extraction processing, the number of two-dimensional images to be segmented and recognized is reduced, and the largest contour in the image to be segmented and recognized is used as the reference contour to segment the sequence of two-dimensional images of the fetal face, thereby achieving effective removal of obstructions in front of the fetal face, while having the advantages of low computational complexity and high three-dimensional imaging rate. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] Figure 1 is a flow chart of a fetal face imaging method according to an embodiment;

[0014] FIG2 is an image obtained after visualization of the three-dimensional volume data of the fetus' face that has not been processed by the method of this embodiment;

[0015] FIG3 is a two-dimensional image of the contour (i.e., the reference contour) with the largest distance in the Y-axis direction selected in step S05;

[0016] FIG4 is a two-dimensional image after blackening processing is performed on the reference curve obtained by using the reference profile of FIG2 in step S06;

[0017] FIG5 is an image obtained by visualizing the three-dimensional volume data after removing the obstruction in front of the face using the method of this embodiment. Implementation Method

[0018] Example 1, referring to FIG1 , a fetal face imaging method based on three-dimensional volume data comprises the following steps:

[0019] S01. Acquire three-dimensional volumetric data of a fetus, and select a sequence of two-dimensional images constituting the fetal face from the three-dimensional volumetric data of the fetus, where the number of two-dimensional images contained in the sequence is n; wherein the directions of the selected sequence of two-dimensional images are all parallel to the midsagittal plane of the fetal face.

[0020] S02: Perform frame extraction on n two-dimensional images at intervals of one image to obtain k two-dimensional images. By performing frame extraction, the number of two-dimensional images that need to be segmented and recognized can be further reduced, thereby increasing the imaging rate.

[0021] S03. Calculate the mean and variance of each of the k 2D images. Eliminate 2D images whose mean is less than a set threshold T1 and whose variance is less than a set threshold T2. The number of eliminated 2D images is m, resulting in km remaining 2D images. When the mean of a 2D image is less than the set threshold, the image is considered pure black. When the variance of a 2D image is less than the set threshold T2, the image is considered incomplete. Both types of images are problematic and therefore unnecessary for subsequent baseline contour calculation. Therefore, they are eliminated.

[0022] S04. Use the trained Unet segmentation network model to segment and recognize the km two-dimensional images after the frame extraction processing in step S03, and segment the facial area from the km two-dimensional images.

[0023] S05. Binarize the km 2D images that segment the facial area, traverse each contour in each 2D image, retain the contour with the largest area in each 2D image, select the contour with the largest distance in the Y-axis direction from the contours with the largest area in the km 2D images as the reference contour, extract points on the left half of the reference contour, and fit it to obtain a reference curve.

[0024] S06. Using the reference curve, process the n two-dimensional images obtained in step S01, and black out all pixel points in the area directly to the left, above the left, and below the left of the reference curve in the n two-dimensional images. Specifically, when performing the black out process, if the coordinates of the two-dimensional image are viewed, the top point of the reference contour is D1 and the bottom point is D2. Then, starting from D1, draw an upper edge line perpendicular to the X-axis upward, and starting from D2, draw a lower edge line perpendicular to the X-axis downward. The upper edge line, the reference contour, and the lower edge line are connected to form a dividing line. All pixel points on the left side of the dividing line are all pixel points in the area directly to the left, above the left, and below the left of the reference curve.

[0025] S07 , replacing the sequence of two-dimensional images selected in step S01 from the three-dimensional volume data of the fetus with the n two-dimensional images processed using the reference curve, thereby generating three-dimensional volume data of the fetus after removing obstructions.

[0026] This embodiment also uses the aforementioned processing method to process three-dimensional volumetric data of a fetus. Figure 2 shows the image obtained by visualizing the three-dimensional volumetric data of the fetus's face without processing using the method of this embodiment. Figure 3 shows the two-dimensional image of the contour with the largest distance in the Y-axis direction (i.e., the reference contour) selected in step S05. Figure 4 shows the two-dimensional image after blacking out the reference curve obtained using the reference contour in Figure 2 in step S06. Figure 5 shows the image obtained by visualizing the three-dimensional volumetric data after removing obstructions in front of the face using the method of this embodiment. Clearly, more details of the fetus's face are visible after removing the obstructions.

[0027] Of course, the above are only preferred embodiments of the present invention and are not intended to limit the scope of use of the present invention. Therefore, any equivalent changes based on the principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A fetal facial imaging method based on three-dimensional volume data, characterized in that: The steps include: S01, obtaining three-dimensional volume data of a fetus, and selecting a sequence of two-dimensional images constituting the face of the fetus from the three-dimensional volume data of the fetus, wherein the number of two-dimensional images contained in the sequence of two-dimensional images is n; wherein the directions of the sequence of two-dimensional images are all parallel to the midsagittal plane of the face of the fetus; S02, performing frame extraction processing on n two-dimensional images at intervals of one image to obtain k two-dimensional images; S03, respectively calculating the mean and variance of k two-dimensional images, eliminating the two-dimensional images whose mean is less than the set threshold T1 and whose variance is less than the set threshold T2, and recording the number of eliminated two-dimensional images as m, to obtain the remaining km two-dimensional images; S04, using the trained Unet segmentation network model to segment and recognize the km pieces of two-dimensional images after the frame extraction processing in step S03, and segmenting the facial area from the km pieces of two-dimensional images; S05, binarizing the km two-dimensional images from which the facial area is segmented, traversing each contour in each two-dimensional image, retaining the contour with the largest area in each two-dimensional image, and selecting the contour with the largest distance in the Y-axis direction from the contours with the largest area in the km two-dimensional images as the reference contour, extracting points on the left half of the reference contour, and fitting to obtain a reference curve; S06, using the reference curve, processing the n two-dimensional images obtained in step S01, and blacking out all the pixel points in the n two-dimensional images in the area directly to the left, above the left, and below the left of the reference curve; S07 , replacing the sequence of two-dimensional images selected in the three-dimensional volume data of the fetus in step S01 with the n two-dimensional images processed by the reference curve, so as to generate the three-dimensional volume data of the fetus after removing the obstruction.

Images