Automatic nipple positioning method for three-dimensional volumetric breast ultrasound imaging

Abstract

The present invention relates to the technical field of three-dimensional ultrasound imaging, in particular to an automatic nipple positioning method for three-dimensional volumetric breast ultrasound imaging. The present invention uses the following technical solution: determining a two-dimensional coronal plane image having obvious nipple features from two-dimensional coronal plane images parsed from three-dimensional volumetric breast ultrasound data, identifying a plurality of nipple position coordinates from the two-dimensional coronal plane image, removing outlier data from the nipple position coordinates, and finally taking an average to obtain final nipple position coordinates. The present invention has the following beneficial effects: accurate coordinates of nipple positions can be directly calculated from three-dimensional volumetric breast ultrasound data without requiring operators to determine nipple positions on the basis of experiences, thereby effectively improving the efficiency and accuracy of positioning nipple positions, and improving the efficiency of breast examination.

Description

An automatic nipple localization method for three-dimensional volume detection of breast ultrasound Technical Field

[0001] This invention relates to the field of three-dimensional ultrasound detection technology, and in particular to an automatic nipple localization method for three-dimensional volume detection of breast ultrasound. Background Technology

[0002] During breast examination, to ensure a more complete scan of breast tissue, it is typically divided into four quadrants: upper outer, upper inner, lower outer, and lower inner. Horizontal and vertical lines are added with the nipple as the center to define these quadrants. Breast diseases usually occur within these four quadrants; therefore, accurately locating the nipple and using this information to determine the region where breast disease may occur is of significant clinical importance.

[0003] Traditional nipple localization based on three-dimensional volumetric ultrasound breast data requires doctors to manually mark the nipple position, which relies heavily on the experience of clinicians and requires doctors to judge the coronal images frame by frame, which is very time-consuming. Summary of the Invention

[0004] The purpose of this invention is to provide an automatic nipple localization method for three-dimensional volume detection of breast ultrasound, specifically a method that can automatically determine the nipple position coordinates in three-dimensional volume data of breast ultrasound.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: an automatic nipple localization method for three-dimensional volume detection of breast ultrasound, comprising the following steps:

[0006] S01. First, extract several frames of two-dimensional coronal images from the acquired three-dimensional volume data of breast ultrasound, mark the frames in sequence, and denote the number of frames of the two-dimensional coronal images as N. Then, remove the two-dimensional coronal images that do not have nipple features, leaving two-dimensional coronal images with frame numbers N1 to N2.

[0007] S02. Using the trained nipple target detection model, nipple detection is performed on the two-dimensional coronal images with frame numbers N1~N2 obtained in step S01. Two-dimensional coronal images with obvious nipple features are selected from them. The frame numbers of the two-dimensional coronal images with obvious nipple features are denoted as T1~T2, and the number of frames of the two-dimensional coronal images with obvious nipple features is denoted as n.

[0008] S03. Perform nipple position detection on the two-dimensional coronal images with frame numbers T1~T2 obtained in step S02, obtaining n coordinate points, denoted as (x1, y1), (x2, y2), ..., (x...). n y n ), and for x1, x2, ... x nAbnormal data is removed, and at the same time, y1, y2, ... y n Abnormal data is removed from (x1, y1), (x2, y2), ..., (x n y n ) Filter out the values ​​that have passed through x1, x2, ... x n and y1, y2, ... y n The coordinates of the points where both x and y exist after removing outlier data are denoted as (x'1, y'1), (x'2, y'2), ..., (x'1, y'1). m y' m ).

[0009] S04. Based on the coordinates (x'1, y'1), (x'2, y'2), ... (x'2) obtained in step S03, m y' m The final output coordinates (x, y) of the nipple position are obtained by calculating the mean. a y a ), where x a = (x'1+x'2+……+x') m ) / m,y a =(y'1+y'2+……+y' m ) / m.

[0010] Specifically, in step S01, when removing two-dimensional coronal images that do not contain papillary features, the following method is used:

[0011] S11. For the N frames of parsed two-dimensional coronal images, calculate the variance of their pixel values, arrange them in order of frame number, and select the frame number from which the variance of pixel values ​​begins to plateau, denoted as N. D The frame number is the value of N1, that is, N1 = N. D .

[0012] S12. Next, based on the acquisition depth H of the acquired breast ultrasound three-dimensional volume data and the number of frames N of the two-dimensional coronal image, calculate the number of frames represented by the unit length E=N / H, and obtain the actual nipple length F from the acquired breast ultrasound three-dimensional volume data, thus obtaining the value of N2: N2=N1+E*F.

[0013] Specifically, in step S03, x1, x2, ... x n When removing outlier data, it involves all x1, x2, ... x n Perform Z-score calculation to remove outliers that deviate from the overall trend based on the standard deviation; for y1, y2, ... y n When removing outlier data, it involves all y1, y2, ... y nPerform Z-score calculation to remove outliers that deviate from the overall trend based on the standard deviation.

[0014] The beneficial effects of this invention are as follows: by identifying two-dimensional coronal images with obvious nipple features from the two-dimensional coronal images parsed from the three-dimensional volume data of breast ultrasound, and then identifying multiple nipple position coordinates from them, abnormal data are removed from the nipple position coordinates to obtain the final nipple position coordinates, thereby improving the efficiency and accuracy of nipple location. Detailed Implementation

[0015] Example 1: The present invention adopts the following technical solution: an automatic nipple localization method for three-dimensional volume detection of breast ultrasound, comprising the following steps:

[0016] S01. First, extract several frames of two-dimensional coronal images from the acquired three-dimensional volume data of breast ultrasound, mark the frames in sequence, and denote the number of frames of the two-dimensional coronal images as N. Then, remove the two-dimensional coronal images that do not have nipple features, leaving two-dimensional coronal images with frame numbers N1 to N2.

[0017] S02. Using the trained nipple target detection model, nipple detection is performed on the two-dimensional coronal images with frame numbers N1~N2 obtained in step S01. Two-dimensional coronal images with obvious nipple features are selected from them. The frame numbers of the two-dimensional coronal images with obvious nipple features are denoted as T1~T2, and the number of frames of the two-dimensional coronal images with obvious nipple features is denoted as n.

[0018] S03. Perform nipple position detection on the two-dimensional coronal images with frame numbers T1~T2 obtained in step S02, obtaining n coordinate points, denoted as (x1, y1), (x2, y2), ..., (x...). n y n ), and for x1, x2, ... x n Abnormal data is removed, and at the same time, y1, y2, ... y n Abnormal data is removed from (x1, y1), (x2, y2), ..., (x n y n ) Filter out the values ​​that have passed through x1, x2, ... x n and y1, y2, ... y n The coordinates of the points where both x and y exist after removing outlier data are denoted as (x'1, y'1), (x'2, y'2), ..., (x'1, y'1). m y' m ).

[0019] S04. Based on the coordinates (x'1, y'1), (x'2, y'2), ... (x'2) obtained in step S03,m y' m The final output coordinates (x, y) of the nipple position are obtained by calculating the mean. a y a ), where x a = (x'1+x'2+……+x') m ) / m,y a =(y'1+y'2+……+y' m ) / m.

[0020] Specifically, in step S01, when removing two-dimensional coronal images that do not contain papillary features, the following method is used:

[0021] S11. For the N frames of parsed two-dimensional coronal images, calculate the variance of their pixel values, arrange them in order of frame number, and select the frame number from which the variance of pixel values ​​begins to plateau, denoted as N. D The frame number is the value of N1, that is, N1 = N. D .

[0022] S12. Next, based on the acquisition depth H of the acquired breast ultrasound three-dimensional volume data and the number of frames N of the two-dimensional coronal image, calculate the number of frames represented by the unit length E=N / H, and obtain the actual nipple length F from the acquired breast ultrasound three-dimensional volume data, thus obtaining the value of N2: N2=N1+E*F.

[0023] Specifically, in step S03, x1, x2, ... x n When removing outlier data, it involves all x1, x2, ... x n Perform Z-score calculation to remove outliers that deviate from the overall trend based on the standard deviation; for y1, y2, ... y n When removing outlier data, it involves all y1, y2, ... y n Z-score calculation is performed to remove outliers that deviate from the overall trend based on the standard deviation. Z-score calculation, also known as standard score calculation, identifies outliers in a dataset based on its standard deviation, thereby removing them and improving the accuracy of nipple location detection.

[0024] Of course, the above are only preferred embodiments of the present invention and are not intended to limit the scope of application of the present invention. Therefore, any equivalent changes made to the principle of the present invention should be included within the protection scope of the present invention.

Claims

1. An automatic nipple localization method for three-dimensional volume detection of breast ultrasound, characterized in that, Includes the following steps: S01. First, extract several frames of two-dimensional coronal images from the acquired three-dimensional volume data of breast ultrasound, mark the frame number in sequence, and record the number of frames of the two-dimensional coronal images as N. Then, remove the two-dimensional coronal images that do not have nipple features, leaving two-dimensional coronal images with frame numbers N1 to N2. S02. Use the trained nipple target detection model to perform nipple detection on the two-dimensional coronal images with frame numbers N1~N2 obtained in step S01, and select two-dimensional coronal images with obvious nipple features. The frame numbers of the two-dimensional coronal images with obvious nipple features are denoted as T1~T2, and the number of frames of the two-dimensional coronal images with obvious nipple features is denoted as n. S03. Perform nipple position detection on the two-dimensional coronal images with frame numbers T1~T2 obtained in step S02, obtaining n coordinate points, denoted as (x1, y1), (x2, y2), ..., (x...). n y n ), and for x1, x2, ... x n Abnormal data is removed, and at the same time, y1, y2, ... y n Abnormal data is removed from (x1, y1), (x2, y2), ..., (x n y n ) Filter out the values ​​that have passed through x1, x2, ... x n and y1, y2, ... y n The coordinates of the points where both x and y exist after removing outlier data are denoted as (x'1, y'1), (x'2, y'2), ..., (x'1, y'1). m y' m ); S04. Based on the coordinates (x'1, y'1), (x'2, y'2), ... (x'2) obtained in step S03, m y' m The final output coordinates (x, y) of the nipple position are obtained by calculating the mean. a y a ), where x a = (x'1+x'2+……+x') m ) / m,y a =(y'1+y'2+……+y' m ) / m.

2. The method for automatic nipple localization in three-dimensional volume detection of breast ultrasound according to claim 1, characterized in that, In step S01, when removing two-dimensional coronal images that do not contain papillary features, the following method is specifically used: S11. For the N frames of parsed two-dimensional coronal images, calculate the variance of their pixel values, arrange them in order of frame number, and select the frame number from which the variance of pixel values ​​begins to plateau, denoted as N. D The frame number is the value of N1, that is, N1 = N. D ; S12. Next, based on the acquisition depth H of the acquired breast ultrasound three-dimensional volume data and the number of frames N of the two-dimensional coronal image, calculate the number of frames represented by the unit length E=N / H, and obtain the actual nipple length F from the acquired breast ultrasound three-dimensional volume data, thus obtaining the value of N2: N2=N1+E*F.

3. The method for automatic nipple localization in three-dimensional volume detection of breast ultrasound according to claim 1, characterized in that, In step S03, x1, x2, ... x n When removing outlier data, it involves all x1, x2, ... x n Perform Z-score calculation to remove outliers that deviate from the overall trend based on the standard deviation; for y1, y2, ... y n When removing outlier data, it involves all y1, y2, ... y n Perform Z-score calculation to remove outliers that deviate from the overall trend based on the standard deviation.