A method for gastric antrum identification and gastric antrum motility index calculation based on deep learning

Abstract

The present invention belongs to the field of medical imaging and relates to a method for gastric antrum identification and gastric antrum motility index calculation based on deep learning, comprising: collecting a standard gastric antrum contraction ultrasound video of a preset time length; the standard gastric antrum contraction ultrasound video refers to a video of the gastric antrum under a standard gastric antrum slice obtained by ultrasound; performing fuzzy removal processing on the standard gastric antrum contraction ultrasound video to obtain a first gastric antrum frame image sequence; performing artifact suppression processing on the first gastric antrum frame image sequence to obtain a second gastric antrum frame image sequence; tracking the gastric antrum area in the second gastric antrum frame image sequence to obtain a gastric antrum binary segmentation image; and calculating indicators such as the gastric antrum motility index and number of contractions of the object to be detected in the current state under bedside ultrasound images based on a deep learning algorithm system.

Description

Technical Field

[0001] The present invention relates to the field of medical imaging, and specifically discloses a method for gastric antrum recognition and gastric antrum motility index calculation based on deep learning. Background Art

[0002] Gastrointestinal function assessment is a critical step in patients receiving enteral nutrition support. With the development of ultrasound, trained medical personnel can now accurately and qualitatively and quantitatively assess gastrointestinal contents, gastric residual volume, and monitor gastrointestinal motility using a single-view antral approach. Currently, manual methods are often used in clinical practice to measure the cross-sectional area of ​​the antrum. These methods involve a doctor or nurse manually marking the major and minor axes of the antrum and approximating it using an ellipse, or manually tracing the antrum's boundaries to obtain the antrum area. However, this procedure requires high skill levels, requires precise marking, is prone to errors, and is time-consuming. Furthermore, this method is too passive and cannot provide real-time patient monitoring. Currently, medical resources are limited, and many hospitals lack the resources to designate a doctor or nurse to conduct real-time examinations of patients. Currently, doctors or nurses use a single-view antral approach. The specific method involves placing an ultrasound probe below the patient's xiphoid process, perpendicular to the abdomen, while simultaneously examining the antrum, superior mesenteric artery, left lobe of the liver, and abdominal aorta to locate the antrum. Ultrasound then displays the antrum's size, and the antrum's transverse and anteroposterior diameters are measured to calculate the antrum area. Therefore, there is an urgent need for a method and system that can monitor and evaluate the antral motility index in real time so that doctors can promptly determine the antral motility and blood flow changes of the person being tested.

[0003] In view of this, the present invention discloses a method for gastric antrum identification and gastric antrum motility index calculation based on deep learning. Based on the deep learning algorithm system, the gastric antrum motility index and number of contractions and other indicators reflecting gastrointestinal function in the patient's current state are calculated under the bedside ultrasound image, overcoming the current clinical problems of insufficient medical and nursing manpower, inaccurate manual counting, and inability to dynamically measure relevant indicators. Summary of the Invention

[0004] The object of the present invention is to provide a gastric antrum recognition method based on deep learning, and the specific scheme includes: collecting a standard gastric antrum contraction ultrasound video of a preset time length; the standard gastric antrum contraction ultrasound video refers to a video of the gastric antrum under a standard gastric antrum slice obtained by ultrasound; performing a blur removal process on the standard gastric antrum contraction ultrasound video to obtain a first gastric antrum frame image sequence; performing an artifact suppression process on the first gastric antrum frame image sequence to obtain a second gastric antrum frame image sequence; tracking the gastric antrum area in the second gastric antrum frame image sequence to obtain a gastric antrum binary segmentation image.

[0005] Furthermore, the acquisition of a standard gastric antrum contraction ultrasound video of a preset time length includes: placing an ultrasound probe under the xiphoid process of the object to be detected, and acquiring a gastric antrum slice image in the sagittal direction; using a target recognition algorithm to identify multiple key structures in the acquired gastric antrum slice image; the multiple key structures are used to identify the relative position of the gastric antrum in the abdominal cavity; based on the multiple key structures, determining whether the acquired gastric antrum slice image belongs to the standard gastric antrum slice image; when the gastric antrum slice image does not belong to the standard gastric antrum slice image, issuing an alarm and giving a probe adjustment direction until the acquired gastric antrum slice image belongs to the standard gastric antrum slice image; repeating the gastric antrum slice image acquisition operation until a video of a standard gastric antrum slice of a preset time length is acquired.

[0006] Furthermore, the multiple key structures include the gastric antrum, liver, and abdominal aorta, and whether the gastric antrum slice image is the standard gastric antrum slice image is determined by feature recognition and spatial coordinate relationship of the multiple key structures.

[0007] Furthermore, the deblurring process of the standard gastric antral contraction ultrasound video to obtain a first gastric antral frame image sequence includes: decomposing the standard gastric antral contraction ultrasound video frame by frame to obtain an initial gastric antral image sequence; dynamically segmenting multiple standard gastric antral slice images in the initial gastric antral image sequence to obtain multiple groups of gastric antral region blocks, wherein the gastric antral region blocks include near-field region blocks and far-field region blocks, and the scale of the near-field region blocks is smaller than the scale of the far-field region blocks; calculating the grayscale histogram of each block in the multiple groups of gastric antral region blocks to obtain multiple groups of grayscale histogram blocks; cropping and allocating each block in the multiple groups of grayscale histogram blocks based on a contrast limiting threshold to obtain multiple groups of cropped allocated blocks; equalizing each block in the multiple groups of cropped allocated blocks to obtain multiple groups of block cumulative distribution functions; mapping the multiple groups of gastric antral region blocks based on the multiple block cumulative distribution functions to obtain multiple mapped gastric antral images; and interpolating each pixel in the multiple mapped gastric antral images to obtain the first gastric antral frame image sequence.

[0008] Furthermore, the calculation formula for calculating the grayscale histogram of each block in the multiple groups of gastric antrum area blocks is:

[0009] ;

[0010] in, Indicates the The number of times the pixel with gray value g appears in the gray histogram of the antrum region block is the value of the gray histogram; i represents the horizontal coordinate variable of the antrum region block; j represents the vertical coordinate variable of the antrum region block; N represents the scale of the antrum region block; represents the Dirac function; Indicates the gastric antrum area block The grayscale value of the pixel in the i-th row and j-th column; represents the gray value variable;

[0011] The calculation formula for performing cropping and allocation on each block in the plurality of grayscale histogram blocks is:

[0012] ;

[0013] ;

[0014] ;

[0015] in, Indicates the The histogram obtained after clipping and allocating the gastric antrum area blocks is the value of the clipped allocation block; min means taking the minimum value; Indicates the maximum allowed pixel; Indicates the amount of cropping; L indicates the number of gray levels; max indicates the maximum value; represents the contrast limit threshold;

[0016] The expression of the block cumulative distribution function is:

[0017] ;

[0018] in, Indicates the Cumulative distribution function of gastric antrum area blocks; m represents the pixel variable of the cropped allocation block; Indicates the Histogram of cropped distribution blocks for each antrum region block.

[0019] Furthermore, the artifact suppression processing is performed on the first gastric antrum frame image sequence to obtain the second gastric antrum frame image sequence, including: using a three-dimensional wavelet basis function to process the continuous preset frame video of the first gastric antrum frame image sequence to obtain frequency band gastric antrum frame images of multiple frequency bands; the multiple frequency bands include low-frequency sub-bands and high-frequency sub-bands; locating artifacts in the main frequency band of gastric antrum contraction to obtain an artifact area; the main frequency band of gastric antrum contraction is characterized by high spatial frequency and medium temporal frequency; and soft threshold processing is performed on the frequency band gastric antrum frame image of the high-frequency sub-band based on the artifact area to obtain the second gastric antrum frame image sequence.

[0020] Furthermore, the expression of the three-dimensional wavelet basis function is:

[0021] ;

[0022] , Represents three-dimensional wavelet transform coefficients; represents a three-dimensional video signal, that is, a continuous preset frame video of a first gastric antrum frame image sequence; represents the three-dimensional wavelet basis function; 、 and Respectively represent the horizontal axis displacement parameter, vertical axis displacement parameter and time displacement parameter; x and y represent the x, y spatial domain; t represents the time domain; Represents the integration operation in three-dimensional space; Indicates the number of decomposition levels;

[0023] The calculation formula of the soft threshold processing is:

[0024] ;

[0025] ;

[0026] in, represents soft threshold; represents the noise standard deviation estimate; represents the natural logarithm function; represents the median of the absolute value; HHH represents the highest frequency sub-band.

[0027] Furthermore, the tracking of the gastric antrum area in the second gastric antrum frame image sequence to obtain a gastric antrum binary segmentation image includes: segmenting the second gastric antrum frame image sequence through a first time window to obtain multiple groups of gastric antrum frame image subsequences; using an image segmentation network to extract the gastric antrum boundary point set for the first frame image of each group of gastric antrum frame image subsequences to obtain a network boundary point set; using an optical flow field to track the gastric antrum boundary point set for the remaining frame images of each group of gastric antrum frame image subsequences to obtain an optical flow boundary point set; fusing the network boundary point set and the optical flow boundary point set through Kalman filtering to obtain an optimal boundary; using a spline curve to fit the optimal boundary to segment the gastric antrum area of ​​the current frame, and outputting the gastric antrum binary segmentation image.

[0028] The present invention also provides a method for calculating the antral motility index of the antral gastric identification method based on deep learning according to any of the above-mentioned items, comprising: determining the cross-sectional area size of the antral gastric binary segmentation image, and calculating the actual physical area of ​​the antral gastric antrum frame by frame to obtain a antral gastric area sequence; dividing the antral gastric area sequence by a second time window to obtain multiple antral gastric area subsequences; determining the minimum and maximum values ​​in each antral gastric area subsequence, and taking the minimum value as the contraction peak and the maximum value as the stretch trough to obtain a motion sequence; taking the alternating contraction peak and stretch trough as the number of contractions, performing periodic judgment on the motion sequence to obtain the total number of contractions of the object to be detected in a preset time length; and determining the contraction number antral motility index based on the total number of contractions.

[0029] Furthermore, the calculation of the actual physical area of ​​the gastric antrum includes: counting the area and position of the connected region based on the binary segmentation image of the gastric antrum; taking the connected region with the largest area as the gastric antrum region, and extracting the contour of the gastric antrum region; counting the total number of pixels in the gastric antrum region to determine the gastric antrum pixel area; and determining the actual physical area of ​​the gastric antrum based on the gastric antrum pixel area.

[0030] The present invention has the following advantages and beneficial effects:

[0031] The present invention collects the gastric antral motility index and contraction frequency of the person being measured by fixing the ultrasound probe on the person being measured, and compares the collected ultrasound image with the standard image in real time to correct the position and angle of the ultrasound probe, thereby obtaining a standard gastric antral short-axis section. The collected data is transmitted to the background and various patient data are calculated through a deep learning algorithm, thereby improving the accuracy of data calculation and avoiding human errors.

[0032] The present invention can make real-time judgment and early warning of non-standard gastric antrum video images by judging and identifying gastric antrum slice images acquired by ultrasound.

[0033] The present invention can automatically segment the gastric antrum area and extract parameters of the gastric antrum cross-sectional area by tracking and identifying the gastric antrum.

[0034] The present invention can identify gastric antrum contraction and relaxation that cannot be identified by the naked eye through continuous recording of gastric antrum cross-sectional area in time series and a peak detection algorithm. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] Figure 1 This is an exemplary flow chart of a gastric antrum recognition method based on deep learning proposed by the present invention;

[0036] Figure 2 Schematic diagram of a standard gastric antrum slice in the present invention. DETAILED DESCRIPTION

[0037] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions of the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Generally, the components of the embodiments of the present invention described and shown in the drawings herein can be arranged and designed in various different configurations.

[0038] Figure 1 This is an exemplary flow chart of a gastric antrum recognition method based on deep learning proposed by the present invention. Figure 1 As shown in FIG, the gastric antrum recognition method based on deep learning includes:

[0039] Collect a standard gastric antral contraction ultrasound video of a preset time length; the standard gastric antral contraction ultrasound video refers to a video of the gastric antrum obtained by ultrasound under a standard gastric antral slice. The preset time length refers to the pre-set time length for obtaining a standard gastric antrum video. For example, the preset time length can be 6 minutes, that is, 6 minutes of standard gastric antral contraction ultrasound video is collected. The standard gastric antral slice can refer to the abdominal aorta and the left lobe of the liver as the gastric antrum section marks, and the gastric antrum is located in the triangle formed by the liver and the abdominal aorta. Figure 2 As shown, I represents the gastric antrum, II represents the left lobe of the liver, III represents the pancreas, IV represents the superior mesenteric artery, and V represents the abdominal aorta.

[0040] In some embodiments, acquiring a standard gastric antral contraction ultrasound video of a preset time length includes:

[0041] An ultrasound probe is placed beneath the xiphoid process of the subject to be examined, and a sagittal slice image of the gastric antrum is acquired. The subject may be a human being whose gastrointestinal function is being evaluated. The antral slice orientation refers to the image of the gastric antrum acquired by the ultrasound probe. An object recognition algorithm is used to identify multiple key structures in the acquired antral slice image; these key structures are used to identify the relative position of the gastric antrum within the abdominal cavity. The object recognition algorithm may be a YOLOv8s deep learning model. The multiple key structures include the gastric antrum, liver, and abdominal aorta. By identifying the features and spatial coordinate relationships of these multiple key structures, it is determined whether the gastric antrum slice image is a standard gastric antrum slice image. For example, the anatomical structure recognition of a standard section uses the YOLOv8s deep learning model to detect three key structures in the B-ultrasound image: the gastric antrum (a hypoechoic tubular structure located at the inferior edge of the liver); the liver (a homogeneous region with moderate to high echoes, serving as a reference for antral positioning); and the abdominal aorta (an echoless circular / tubular structure). The image is then determined to be a standard section by identifying the features of these three key structures and their spatial coordinate relationships. Based on the multiple key structures, determine whether the acquired gastric antrum slice image belongs to the standard gastric antrum slice image. When the gastric antrum slice image does not belong to the standard gastric antrum slice image, an alarm is issued and the probe adjustment direction is given until the acquired gastric antrum slice image belongs to the standard gastric antrum slice image. For example, an edge computing module, a probe posture sensor, and an audio-visual alarm unit are used to achieve that when an unqualified image is acquired, a red frame alarm appears on the screen, and specific prompts for adjusting the probe direction and angle are given to ensure that each frame of the gastric antrum contraction video acquired for 6 minutes is qualified. Repeat the acquisition operation of the gastric antrum slice image until a video of a standard gastric antrum slice of a preset time length is acquired.

[0042] Deblurring is performed on the standard gastric antral contraction ultrasound video to obtain a first gastric antral frame image sequence. The first gastric antral frame image sequence is a sequence of multiple deblurred ultrasound video frame images. For example, a CLAHE algorithm can be used to enhance contrast and resolve local blurring issues in the ultrasound image.

[0043] In some embodiments, the blur removal process includes: decomposing the standard gastric antrum contraction ultrasound video frame by frame to obtain an initial gastric antrum image sequence.

[0044] Dynamically segment multiple standard antral slice images in the initial antral image sequence to obtain multiple groups of antral region blocks. The antral region blocks include near-field region blocks and far-field region blocks. The scale of the near-field region blocks is smaller than that of the far-field region blocks. For example, the antral ultrasound video is decomposed into static images frame by frame, and each frame is processed separately. The standard antral images are dynamically segmented. In the near-field region (depth <6cm), 8x8 small blocks are used to enhance the fine structure of the gastric wall. In the far-field region (depth ≥6cm), the 16x16 block is switched to suppress deep noise.

[0045] Calculate each block in multiple groups of gastric antral area blocks separately The value of the grayscale histogram ,in, , and obtain multiple groups of grayscale histogram blocks. The calculation formula for calculating the grayscale histogram of each block in the multiple groups of gastric antrum area blocks is:

[0046] ;

[0047] in, Indicates the The number of times the pixel with gray value g appears in the gray histogram of the antrum region block is the value of the gray histogram; i represents the horizontal coordinate variable of the antrum region block; j represents the vertical coordinate variable of the antrum region block; N represents the scale of the antrum region block; represents the Dirac function, (if , then δ=1, otherwise δ=0); Represents the gastric antrum area block The grayscale value of the pixel in the i-th row and j-th column; Represents a grayscale value variable.

[0048] Based on the contrast limit threshold, each block in the multiple groups of grayscale histogram blocks is cropped and allocated to obtain multiple groups of crop allocation blocks. The contrast limit threshold refers to a pre-set value for crop allocation, usually =2.0∼3.0. Clipping allocation means clipping the histogram values ​​that exceed the maximum allowed pixel value and evenly distributing the clipping amount to all gray levels. The calculation formula for clipping allocation for each block in multiple groups of gray histogram blocks is:

[0049] ;

[0050] ;

[0051] ;

[0052] in, Indicates the The histogram obtained after clipping and allocating the gastric antrum area blocks is the value of the clipped allocation block; min means taking the minimum value; Indicates the maximum allowed pixel; Indicates the amount of cropping; L indicates the number of gray levels; max indicates the maximum value; Indicates the contrast limit threshold.

[0053] Each block in the multiple groups of cropped allocation blocks is equalized to obtain the cumulative distribution function of the multiple groups of blocks. The expression of the block cumulative distribution function is:

[0054] ;

[0055] in, Indicates the Cumulative distribution function of gastric antrum area blocks; m represents the pixel variable of the cropped allocation block; Indicates the Histogram of cropped distribution blocks for each antrum region block.

[0056] Based on the multi-block cumulative distribution function, multiple groups of gastric antrum area blocks are mapped to obtain multiple mapped gastric antrum images. The calculation formula for the mapping is:

[0057] ;

[0058] in, Represents the mapping formula; Indicates rounding the input value to the nearest integer.

[0059] Interpolation is performed on each pixel in the plurality of mapped gastric antrum images to obtain the first gastric antrum frame image sequence, so as to avoid discontinuity between blocks and retain continuity of the gastric antrum edge.

[0060] Perform artifact suppression processing on the first gastric antrum frame image sequence to obtain a second gastric antrum frame image sequence. The second gastric antrum frame image sequence refers to a sequence of images obtained after performing artifact suppression processing on the images in the first gastric antrum frame image sequence. In some embodiments, performing artifact suppression processing on the first gastric antrum frame image sequence to obtain the second gastric antrum frame image sequence includes: processing continuous preset frame videos of the first gastric antrum frame image sequence using a three-dimensional wavelet basis function to obtain frequency band gastric antrum frame images of multiple frequency bands; the multiple frequency bands include low-frequency sub-bands and high-frequency sub-bands. Continuous preset frame videos refer to the time length of the pre-set processed frame videos. For example, 15 consecutive frames of video can be regarded as a three-dimensional signal (space (x, y) + time t), decomposed into low-frequency (LLL) and high-frequency sub-bands (HLL, LHL, LLH, HHH). The expression of the three-dimensional wavelet basis function is:

[0061] ;

[0062] in, Represents three-dimensional wavelet transform coefficients; represents a three-dimensional video signal, that is, a continuous preset frame video of a first gastric antrum frame image sequence; represents the three-dimensional wavelet basis function; 、 and Respectively represent the horizontal axis displacement parameter, vertical axis displacement parameter and time displacement parameter; x and y represent the x, y spatial domain; t represents the time domain; Represents the integration operation in three-dimensional space; Indicates the number of decomposition levels.

[0063] Artifacts in a main frequency band of gastric antral contraction are located to obtain an artifact region; the main frequency band of gastric antral contraction is characterized by high spatial frequency and medium temporal frequency (eg, HHL / LHL / HLH).

[0064] Based on the artifact area, a soft threshold process is performed on the gastric antrum frame image of the high frequency sub-band to obtain a second gastric antrum frame image sequence. The calculation formula of the soft threshold process is:

[0065] ;

[0066] ;

[0067] in, represents soft threshold; represents the noise standard deviation estimate; represents the natural logarithm function; represents the median of the absolute value; HHH represents the highest frequency sub-band.

[0068] The gastric antrum region in the second gastric antrum frame image sequence is tracked to obtain a gastric antrum binary segmentation image. The gastric antrum binary segmentation image refers to an image in which the gastric antrum region is white and the background is black. In some embodiments, the tracking of the gastric antrum region in the second gastric antrum frame image sequence to obtain a gastric antrum binary segmentation image includes: segmenting the second gastric antrum frame image sequence through a first time window to obtain multiple groups of gastric antrum frame image subsequences. The first time window refers to the time length for segmenting the second gastric antrum frame image sequence. The gastric antrum frame image subsequence refers to an image sequence of multiple second time window lengths obtained by segmenting the second gastric antrum frame image according to the first time window. For example, the first time window can be 6 frames of image.

[0069] For the first frame of each antral image subsequence, an image segmentation network is used to extract the antral boundary point set, resulting in a network boundary point set. The first frame refers to the first frame of the subsequence. For example, the first frame of the first subsequence can be the first frame of an ultrasound video. The network boundary point set is the set of boundary points extracted by segmenting the antral image using a U-Net segmentation network.

[0070] For each subsequence of gastric antral image frames, the optical flow field is used to track the gastric antral boundary point set to obtain an optical flow boundary point set. The remaining frames refer to the frames in the subsequence other than the first frame. For example, for a subsequence with a first time window of 6 frames, the remaining frames of the first subsequence are the second through sixth frames. The optical flow boundary point set refers to the set of boundary points obtained by updating the boundary points through optical flow calculation.

[0071] The network boundary point set and the optical flow boundary point set are fused by Kalman filtering to obtain an optimal boundary. The optimal boundary point refers to the fused boundary point.

[0072] The optimal boundary is fitted using a spline curve, and the gastric antrum region of the current frame is obtained by real-time segmentation, and the gastric antrum binary segmentation image is output.

[0073] The present invention also provides a method for calculating the gastric antrum motility index, comprising: determining the cross-sectional area of ​​the gastric antrum binary segmentation image, and calculating the actual physical area of ​​the gastric antrum frame by frame to obtain the gastric antrum area sequence , where S represents the antral area sequence, whose elements include multiple antral areas, the first antral area , second gastric antrum area and T antral area etc., where T represents the total number of frames. The actual physical area refers to the true area of ​​the gastric antrum under actual conditions, and the gastric antrum area sequence refers to a sequence of gastric antrum areas sorted by time. In some embodiments, calculating the actual physical area of ​​the gastric antrum includes: counting the areas and locations of connected regions based on the binary segmented image of the gastric antrum. A connected region refers to a set of pixels in the binary image that have the same pixel value and are connected through adjacent pixels. The connected region with the largest area is defined as the gastric antrum region, and the contour of the gastric antrum region is extracted.

[0074] The total number of pixels in the gastric antrum region (white pixels in the mask image, with a value of 255) is counted to determine the gastric antrum pixel area, which is used as a preliminary measure of the gastric antrum area. The calculation formula for the gastric antrum pixel area is:

[0075] ;

[0076] in, represents the pixel area of ​​the gastric antrum; x and y represent the horizontal and vertical axis variables of the binary segmentation image, respectively; W and H represent the width and height of the binary segmentation image, respectively; mask represents the mask area, which is 255.

[0077] Based on the gastric antrum pixel area, the actual physical area of ​​the gastric antrum is determined. The calculation formula of the actual physical area is:

[0078] ;

[0079] Among them, actual_area represents the actual physical area; pixel_area represents the pixel area of ​​the gastric antrum; Indicates standard physical area; Represents the standard pixel area, and dividing the two gives the pixel-square conversion coefficient.

[0080] The gastric antrum area sequence is divided into multiple gastric antrum area subsequences by a second time window. The second time window refers to the time period for dividing the gastric antrum area sequence, and the division can be performed based on experience or actual conditions. The gastric antrum area subsequences are sequences consisting of gastric antrum areas in multiple time periods of the second time window length obtained by dividing the gastric antrum area sequence.

[0081] The minimum and maximum values ​​in each antral area subsequence are determined, and the minimum value is used as the contraction peak, and the maximum value as the expansion trough, to obtain a motion sequence. The contraction peak refers to the minimum contraction value in each antral area subsequence during antral motion. The expansion trough refers to the maximum expansion value in each antral area subsequence during antral motion.

[0082] The alternating contraction peaks and stretch troughs are regarded as the number of contractions. The period of the motion sequence is determined to obtain the total number of contractions of the object to be tested within a preset time length:

[0083] ;

[0084] in, Indicates the number of gastric antral contractions within a preset time length (e.g., 6 minutes); min indicates the minimum value; Indicates the contraction peak; Indicates stretching the trough.

[0085] Based on the total number of contractions, the antral motility index (MI) is determined. Taking a 6-minute video as an example, the antral motility index (MI) calculation requires splitting the 6-minute antral contraction video into three 2-minute videos, and recording the number of antral contractions per 2 minutes as the antral contraction frequency (ACF); continuously measuring the maximum antral relaxation ( ) and minimum shrinkage ( ) area, and the antral contraction amplitude .

[0086] =ACF x ACA;

[0087] 1. Calculate the frequency of gastric antral contraction in segments ( ):

[0088] For each 2-minute segment U (U=1,2,3), = Number of contractions during the Uth 2-minute segment.

[0089] 2. Calculate the antral contraction amplitude in segments ( ):

[0090] ;

[0091] is the Uth 2-minute antral contraction amplitude, is the number of complete cycles in the Uth 2-minute period, is the period number, For the The maximum diastolic area in a cycle, For the The minimum shrinkage area in a cycle.

[0092] 3. Calculate the antral motility index (MI):

[0093] ;

[0094] MI is the final antral motility index.

[0095] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Those skilled in the art will readily appreciate that various modifications and variations of the present invention are possible. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present invention shall be included within the scope of protection of the present invention.

Claims

1. A gastric antrum recognition method based on deep learning, characterized in that: include: Collecting a standard gastric antral contraction ultrasound video of a preset length of time; the standard gastric antral contraction ultrasound video refers to a video of the gastric antrum obtained by ultrasound under a standard gastric antral slice; The standard gastric antrum contraction ultrasound video is subjected to blur removal processing to obtain a first gastric antrum frame image sequence, including: Decomposing the standard gastric antrum contraction ultrasound video frame by frame to obtain an initial gastric antrum image sequence; Dynamically segmenting multiple standard gastric antrum slice images in the initial gastric antrum image sequence to obtain multiple groups of gastric antrum region blocks; the gastric antrum region blocks include near-field region blocks and far-field region blocks; the scale of the near-field region blocks is smaller than the scale of the far-field region blocks; the near-field region refers to the area with a depth of less than 6 cm, and uses 8x8 small blocks to enhance the fine structure of the gastric wall; the far-field region refers to the area with a depth of ≥6 cm, and switches to 16x16 blocks to suppress deep noise; Calculating the grayscale histogram of each block in the multiple groups of gastric antrum area blocks respectively to obtain multiple groups of grayscale histogram blocks; Based on the contrast limiting threshold, each block in the multiple groups of grayscale histogram blocks is cropped and allocated respectively to obtain multiple groups of cropped and allocated blocks; Equalizing each block in the multiple groups of cropped allocation blocks respectively to obtain the cumulative distribution function of the multiple groups of blocks; Based on the multi-block cumulative distribution function, multiple groups of gastric antrum area blocks are mapped respectively to obtain multiple mapped gastric antrum images; interpolating each pixel in the plurality of mapped gastric antrum images to obtain the first gastric antrum frame image sequence; performing artifact suppression processing on the first gastric antrum frame image sequence to obtain a second gastric antrum frame image sequence; The gastric antrum region in the second gastric antrum frame image sequence is tracked to obtain a gastric antrum binary segmentation image.

2. The gastric antrum recognition method based on deep learning according to claim 1, characterized in that: The acquisition of a standard gastric antral contraction ultrasound video of a preset time length includes: Place the ultrasound probe under the xiphoid process of the subject to be examined, and obtain a sagittal slice image of the gastric antrum; Using a target recognition algorithm, identifying multiple key structures in the acquired gastric antrum slice image; the multiple key structures are used to identify the relative position of the gastric antrum in the abdominal cavity; determining, based on the multiple key structures, whether the acquired gastric antrum slice image belongs to the standard gastric antrum slice image; When the gastric antrum slice image does not belong to the standard gastric antrum slice image, an alarm is issued and a probe adjustment direction is given until the acquired gastric antrum slice image belongs to the standard gastric antrum slice image; The operation of acquiring gastric antrum slice images is repeated until a video of a standard gastric antrum slice of a preset time length is acquired.

3. The gastric antrum recognition method based on deep learning according to claim 2, characterized in that: The multiple key structures include the gastric antrum, liver, and abdominal aorta. Whether the gastric antrum slice image is the standard gastric antrum slice image is determined by feature recognition and spatial coordinate relationship of the multiple key structures.

4. The gastric antrum recognition method based on deep learning according to claim 1, characterized in that: The calculation formula for calculating the grayscale histogram of each block in the multiple groups of gastric antrum area blocks is: ； in, Indicates the The number of times the pixel with gray value g appears in the gray histogram of the antrum region block is the value of the gray histogram; i represents the horizontal coordinate variable of the antrum region block; j represents the vertical coordinate variable of the antrum region block; N represents the scale of the antrum region block; represents the Dirac function; Indicates the gastric antrum area block The grayscale value of the pixel in the i-th row and j-th column; represents the gray value variable; The calculation formula for performing cropping and allocation on each block in the plurality of grayscale histogram blocks is: ； ； ； in, Indicates the The histogram obtained after clipping and allocating the gastric antrum area blocks is the value of the clipped allocation block; min means taking the minimum value; Indicates the maximum allowed pixel; Indicates the amount of cropping; L indicates the number of gray levels; max indicates the maximum value; represents the contrast limit threshold; The expression of the block cumulative distribution function is: ； in, Indicates the Cumulative distribution function of gastric antrum area blocks; m represents the pixel variable of the cropped allocation block; Indicates the Histogram of the cropped distribution blocks of the gastric antrum area blocks; The calculation formula of the mapping is: ； in, Represents the mapping formula; Indicates rounding the input value to the nearest integer.

5. The gastric antrum recognition method based on deep learning according to claim 1, characterized in that: The performing artifact suppression processing on the first gastric antrum frame image sequence to obtain a second gastric antrum frame image sequence includes: Processing the continuous preset frame videos of the first gastric antrum frame image sequence using a three-dimensional wavelet basis function to obtain frequency band gastric antrum frame images of multiple frequency bands; the multiple frequency bands include low-frequency sub-bands and high-frequency sub-bands; The artifacts in the main frequency band of gastric antrum contraction are located to obtain the artifact area; A soft threshold process is performed on the gastric antrum frame image of the high frequency sub-band based on the artifact region to obtain a second gastric antrum frame image sequence.

6. The gastric antrum recognition method based on deep learning according to claim 5, characterized in that: The expression of the three-dimensional wavelet basis function is: ； in, Represents three-dimensional wavelet transform coefficients; represents a three-dimensional video signal, that is, a continuous preset frame video of a first gastric antrum frame image sequence; represents the three-dimensional wavelet basis function; 、 and Respectively represent the horizontal axis displacement parameter, vertical axis displacement parameter and time displacement parameter; x and y represent the x, y spatial domain; t represents the time domain; Represents the integration operation in three-dimensional space; Indicates the number of decomposition levels; The calculation formula of the soft threshold processing is: ； ； in, represents soft threshold; represents the noise standard deviation estimate; represents the natural logarithm function; represents the median of the absolute value; HHH represents the highest frequency sub-band.

7. The gastric antrum recognition method based on deep learning according to claim 1, characterized in that: Tracking the gastric antrum region in the second gastric antrum frame image sequence to obtain a gastric antrum binary segmentation image includes: Segmenting the second gastric antrum frame image sequence through the first time window to obtain multiple groups of gastric antrum frame image subsequences; For the first frame image of each group of gastric antral frame image subsequence, the image segmentation network is used to extract the gastric antral boundary point set to obtain a network boundary point set; the first frame image refers to the first frame image of the subsequence; For the remaining frame images of each group of gastric antrum frame image subsequences, the gastric antrum boundary point set is tracked using the optical flow field to obtain an optical flow boundary point set; the remaining frame images refer to the frame images other than the first frame image in the subsequence; fusing the network boundary point set and the optical flow boundary point set through Kalman filtering to obtain an optimal boundary; The optimal boundary is fitted using a spline curve to segment the gastric antrum region of the current frame, and the gastric antrum binary segmentation image is output.

8. The method for calculating the gastric antral motility index of the gastric antral recognition method based on deep learning according to any one of claims 1 to 7, characterized in that: include: Determine the cross-sectional area of ​​the gastric antrum binary segmentation image, and calculate the actual physical area of ​​the gastric antrum frame by frame to obtain a gastric antrum area sequence; Dividing the gastric antrum area sequence by a second time window to obtain a plurality of gastric antrum area subsequences; Determine the minimum and maximum values ​​in each antral area subsequence, and use the minimum value as the contraction peak and the maximum value as the expansion trough to obtain a motion sequence; Taking the alternating contraction peaks and stretch troughs as the number of contractions, performing period determination on the motion sequence, and obtaining the total number of contractions of the subject to be tested within a preset time length; Based on the total number of contractions, the contraction number antral motility index was determined.

9. The method for calculating the gastric antral motility index according to claim 8, characterized in that: The calculation of the actual physical area of ​​the gastric antrum includes: Based on the gastric antrum binary segmentation image, counting the area and position of the connected region; The connected area with the largest area is taken as the gastric antrum area, and the contour of the gastric antrum area is extracted; Counting the total number of pixels in the gastric antrum region to determine the gastric antrum pixel area; Based on the gastric antrum pixel area, the actual physical area of ​​the gastric antrum is determined.

Images