Oral dilation distance automatic measuring method based on intranatal ultrasound image

Through the combination of the deep learning model UODNet and ALD method, the cervical dilation distance is automatically measured, which solves the subjectivity problem in ultrasound image evaluation in obstetrics, improves the accuracy and stability of the measurement, and reduces the burden on doctors.

CN118986407APending Publication Date: 2024-11-22GUANGZHOU LIAN MED TECH CO LTD
View PDF 6 Cites 0 Cited by

Patent Information

Application Number
CN202310558242.1
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2023-05-18
Publication Date
2024-11-22

AI Technical Summary

Technical Problem

In obstetrics, vaginal digital examination has subjective deviations and discomfort, and ultrasound image evaluation still cannot completely eliminate the subjective problem of cervical dilation distance judgment, affecting the accuracy of labor assessment.

Method used

Using deep learning-based methods, the UODNet model and ALD method are used to automatically measure the cervical dilation distance. By segmenting and feature extraction of ultrasound images, the automated measurement of the cervical dilation distance is achieved, which reduces the workload of doctors and enhances the robustness of measurement. sex.

Benefits of technology

It achieves good consistency with the doctor's manual measurement results, reduces subjective deviation, improves the accuracy and stability of the measurement of cervical dilation distance, and reduces the burden on doctors.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118986407A_ABST
    Figure CN118986407A_ABST
Patent Text Reader

Abstract

The invention discloses a method for automatically measuring the dilation distance of an oral uterus based on an intranatal ultrasound image. The method comprises the following steps: collecting the intranatal ultrasound image and the label annotation of the oral uterus from a hospital; further screening is carried out; a deep learning model UODNet is used for processing and calculating the input ultrasonic image, and the target position and segmentation information of the uterine orifice are obtained; and according to the segmentation information, an ALD (Angle-Line Dichomy) method is adopted to measure the parameter of the dilation distance of the uterine orifice. According to the method, the UOD automatic measurement method based on the birth time ultrasonic image is provided, the vacancy of the UOD in the birth time ultrasonic parameter measurement is filled up, and the birth time evaluation method based on ultrasound has more judgment information.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of integrated medicine and computer vision, and in particular to a method for automatically measuring cervical dilation distance based on intrapartum ultrasound images. Background Art

[0002] In clinical obstetrics, the main modes of delivery are vaginal delivery and cesarean section. Vaginal delivery, as a normal natural physiological process, has significant advantages over cesarean section. It is recognized by the World Health Organization and is encouraged and advocated for normal pregnant women. Correct and timely assessment of labor progress is crucial to promoting natural delivery and reducing cesarean section surgical intervention. In clinical obstetrics, the assessment of labor mainly adopts the traditional method of vaginal digital examination, but vaginal digital examination is affected by personal subjective factors and personal experience of medical staff, and there are certain deviations. At the same time, vaginal digital examination will cause greater discomfort to pregnant women and may also increase the risk of intrauterine infection.

[0003] Compared with the many disadvantages of the traditional vaginal digital examination mentioned above, intrapartum ultrasound technology can be non-invasive, and pregnant women do not feel any obvious discomfort, which is easy to accept. Many studies have shown that it can help medical staff to evaluate the progress of labor and predict the mode of delivery. By measuring the changes in ultrasound parameters and their values, such as the Angle of Descend (AOD), Head Symphysis Distance (HSD), Head Perineum Distance (HPD), Uterus Opening Distance (UOD), etc., it can reflect the progress of labor.

[0004] However, using ultrasound images to evaluate and measure the distance of cervical dilation cannot eliminate the subjective problem in the judgment of labor progress. Different doctors may have different judgments on the same ultrasound image based on their personal experience.

[0005] On the other hand, with the rapid development of deep learning, it is possible to build deep and complex models for specific tasks with the help of computer equipment. For graphic tasks, the perception accuracy of the model is at the pixel level, which is far better than that of the human eye. The previous ultrasound image data is sent to the model for training, and a model that meets the requirements is obtained after testing. When faced with a large number of ultrasound images again, they are input into the deep learning model for key anatomical structure positioning, classification, and segmentation, which can greatly reduce the workload of doctors and reduce the subjective drawbacks of previous manual measurements. The present invention discloses an automatic measurement method for cervical dilation distance based on intrapartum ultrasound images.

[0006] Invention content information

[0007] The present invention provides an automatic measurement method for the cervical dilation distance based on an ultrasound image during labor. The step of manually measuring the cervical dilation distance on the ultrasound image by the doctor after the ultrasound examination during labor in the clinic is changed to automatic measurement using deep learning and ALD (Angle-Line Dichotomy) method on a computer device. The ultrasound image is input into the trained UODNet to obtain segmentation information, and then the cervical dilation distance is measured for the obtained segmentation information and the measurement result is output, so as to realize automatic measurement and reduce the workload of the doctor. In addition, considering that due to time and complexity constraints, clinicians often only measure the cervical dilation distance of a single ultrasound image during labor, the present invention also includes automatically measuring the average value of the cervical dilation distance of multiple ultrasound images of a single person, so that a more stable cervical dilation distance can be obtained, and the robustness of the measurement can be enhanced. In clinical practice, due to the different professional levels and experiences of the operating physicians, the measurement results of the cervical dilation distance for the same pregnant woman may be different, and there is a large subjectivity. The present invention can avoid the above disadvantages to a large extent, and the cervical dilation distance measured by the present invention has a strong robustness. In addition, the cervical dilation distance in the past required manual measurement by doctors, but the automated measurement of the present invention can reduce the burden on doctors to a great extent.

[0008] To achieve the above object, the present invention provides the following technical solution: a method for automatically measuring the cervical dilation distance based on ultrasound images during labor, the method comprising the following steps:

[0009] S1. Collect transperineal ultrasound images and label annotations of the cervical os;

[0010] S2, further screening the ultrasound images and selecting ultrasound images with clear anatomical structures;

[0011] S3, using UODNet to process and calculate the image obtained in S2, and obtain the cervical position and segmentation information of the image;

[0012] S4. Measure the cervical dilation distance using the ALD method based on the segmentation information of S3.

[0013] Furthermore, in step S1, the intrapartum ultrasound images obtained from the hospital are sorted, and the cervical region of the ultrasound images is annotated.

[0014] Furthermore, the intrapartum ultrasound images are screened according to the transverse and longitudinal cut types of the cervix, and the intrapartum ultrasound images with transverse cuts of the cervix are selected.

[0015] Furthermore, the intrapartum ultrasound images that meet the screening criteria are sent to UODNet for processing and calculation. The target position and segmentation information of the cervix are obtained, and the cervix dilation distance is measured. The network model structure of UODNet is a series of different downsampling convolutional layers (encoder part) and upsampling convolutional layers (decoder part). The downsampling convolutional layer extracts features from the ultrasound image for subsequent upsampling segmentation tasks; the encoder part of the UODNet network model structure uses PoolFormer to capture ultrasound image features. After the input ultrasound image (size is H×W) passes through PoolFormer, four feature maps of different sizes will be obtained, and their sizes are respectively Different feature maps capture different information. Shallow feature maps capture shallow features, deep feature maps capture deep features, and finally upsampling is used to perform feature fusion. The decoder part of the UODNet network model structure adopts a FPN-like structure, and its main task is to perform feature fusion on the image features captured by the encoder part for subsequent segmentation tasks. Specifically, it includes upsampling to obtain a feature map of the same size as the upper layer, then merging it, and then convolving it to the original number of channels. This iterative cycle eventually obtains a feature map of the same size as the original image, and the information obtained in the test phase is pixel-level segmentation information.

[0016] Furthermore, for each segmentation information output by the model UODNet, the centroid position of the cervical os area is first determined, and a new coordinate system is established with this as the center and parallel to the length and width of the image. Then, a rectangular frame that just frames the cervical os is determined for auxiliary distance calculation. Next, 0° and 90° are used as the initial starting points, and the two intersection points of the straight lines at 0° and 90° to the coordinate system and the cervical os are calculated, and their distances are calculated. The distances obtained at 45° and 135° are calculated in the same way, and it is determined whether the smaller value falls into the [0°-90°] interval or the [90°-180°] interval. This is repeated and updated iteratively. When the difference between the two cervical os measurements is less than 0.01, the calculation is stopped and the mean is output. This is the ALD method.

[0017] Furthermore, considering that due to time and complexity constraints, clinicians often only measure the cervical dilation distance of a single intrapartum ultrasound image, the present invention also includes automatically measuring the average cervical dilation distance of multiple ultrasound images of a single person, which may result in a more stable cervical dilation distance and enhance the robustness of the measurement.

[0018] Compared with the prior art, the present invention has the following advantages and effects:

[0019] The present invention proposes a deep learning-based ultrasound image cervical dilation distance (UOD) measurement framework. It is mainly for the segmentation and recognition of cervical anatomical structures and the measurement of cervical dilation distance. The method of the present invention measures the cervical dilation distance of 62 test cases and compares it with the manual measurement results of doctors. The results show that the method of the present invention has good consistency with clinicians in the measurement of cervical dilation distance. In addition, considering that due to time and complexity constraints, clinicians often only measure the cervical dilation distance of a single ultrasound image during delivery. The present invention also includes automatically measuring the average value of the cervical dilation distance of multiple ultrasound images of a single person, so that a more stable cervical dilation distance can be obtained, and the robustness of the measurement can be enhanced. In clinical practice, due to the different professional levels and experiences of operating physicians, the measurement results of the cervical dilation distance for the same pregnant woman may be different, and there is a large degree of subjectivity. The present invention can avoid the above disadvantages to a large extent, and the cervical dilation distance measured by the present invention has strong robustness. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] The drawings described herein are used to provide a further understanding of the present invention and constitute a part of this application. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention. In the drawings:

[0021] Figure 1 It is a flow chart of the method for automatically measuring the cervical dilation distance based on ultrasound images during labor disclosed by the present invention;

[0022] Figure 2 is an example of image data according to an embodiment of the present invention;

[0023] Figure 3 1 is a schematic diagram of the ALD method calculation flow of an embodiment of the present invention;

[0024] Figure 4 is a comparison diagram of segmentation and detection results in an embodiment of the present invention;

[0025] Figure 5 is a diagram showing the difference between automatic and manual measurement of cervical dilation distance in an embodiment of the present invention; DETAILED DESCRIPTION

[0026] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0027] Embodiment 1:

[0028] This embodiment discloses a method for automatically measuring the cervical dilation distance based on an ultrasound image during labor, which mainly includes the following steps:

[0029] S1. Collect transperineal ultrasound images and label annotations of the cervical os;

[0030] The data source in step S1 is collected by the radiologist. The equipment used is Philip's SONOS5500B ultrasound for intrapartum ultrasound image examination. Under the guidance of the radiologist, four students performed the task of labeling the cervical information, and after completion, the doctor checked it again to confirm that it was correct.

[0031] S2, further screening of ultrasound images;

[0032] Among them, images with transverse sections of the cervix and clearer anatomical information were selected.

[0033] S3, using UODNet to process and calculate the image obtained in S2, and obtain the cervical position and segmentation information of the image;

[0034] In step S3, the input is the intrapartum ultrasound image, and the output is the segmented region and category. Specifically, we use the UODNet model to process and calculate the input intrapartum ultrasound image. UODNet outputs the segmentation result of cervical dilation.

[0035] Specifically, UODNet uses PoolFormer to capture ultrasound image features. After the input ultrasound image (size is H×W) passes through PoolFormer, feature maps of different sizes will be obtained. Different feature maps capture different information. Shallow feature maps capture shallow features, and deep feature maps capture deep features. Then upsampling is used for feature fusion. Specifically, it includes upsampling to obtain feature maps of the same size as the upper feature map, merging them, and then convolving them to the original number of channels. Iterate the cycle to finally obtain a feature map of the same size as the original image and output pixel-level segmentation information.

[0036] All experiments of the present invention are carried out under the PyTorch framework, and the graphics card used is NVIDIA GeForce RTX 3090. The optimizer in the UODNet model uses the AdamW optimizer, with an initial learning rate of 0.0002 and a weight decay of 0.0001. In the early stage of training, a larger learning rate can make the model converge faster, while in the later stage of training, a larger learning rate will cause the model to oscillate and be difficult to converge. In order to stabilize the convergence of the model weight parameters, we use a learning rate decay strategy, a poly decay strategy, and a learning rate decay power parameter equal to 0.9. The training batch size is 8. Due to the small amount of data, the present invention also uses rotation, folding, and cropping to enhance data during the training process. It is worth noting that, considering that the shape of the cervix should ideally be a smooth ellipse, we have added a shape-based loss function, so that the better model performance is not only limited to the segmentation inside the cervix, but also the segmentation results on the cervix boundary are improved. In addition, considering that the area occupied by the background area and the cervix area is quite different, we introduce class imbalance so that the model pays more attention to the segmentation results of the cervix area. In order to facilitate subsequent adjustments, the training loss values ​​are recorded throughout the training process, and an iterative model is saved every 400 batches. After the training, it is tested to select the optimal model. In addition, this model does not require the size of the input intrapartum ultrasound image. In the preprocessing of UODNet, the input image is automatically reshaped to a pixel size of 512×512.

[0037] S4. Measure the cervical dilation distance using the ALD method based on the segmentation information of S3.

[0038] In step S4, the UOD is measured according to the segmentation result in S3, such as Figure 2 The figure shows a schematic diagram of cervical dilation segmentation. The Dice coefficient and Hausdorff distance Hd are both important indicators for measuring the segmentation results of the model. The Dice coefficient is sensitive to the internal filling of the segmentation, and the larger the Dice coefficient, the better. The Hausdorff distance Hd is sensitive to the boundary of the segmentation, and the smaller the Hausdorff distance Hd, the better. The commonly used segmentation model UNet has a segmentation dice coefficient of 92.45% and a Hausdorff distance Hd of 15.79, while the segmentation dice coefficient of UODNet is 94.77% and the Hausdorff distance Hd is 13.43. It can be seen that the UODNet used in the present invention has a better segmentation performance.

[0039] In this embodiment, the UOD measurement calculation process based on the ALD method is as follows: Figure 3As shown in the figure, according to the cervical segmentation results obtained by UODNet in S3, UOD measurement is performed through the following steps. First, the centroid position of the cervical region is determined, and a new coordinate system is established with this as the center and parallel to the length and width of the image. Then, a rectangular frame that just frames the cervical region is determined to assist in distance calculation. Then, 0° and 90° are used as the initial starting points, and the two intersection points of the straight lines with the coordinate system at 0° and 90° are calculated, and their distances are calculated. The distances obtained by 45° and 135° are calculated in the same way, and it is determined whether the smaller value falls into the [0°-90°] interval or the [90°-180° interval]. Then, the clamping line is updated, and so on. The calculation and update are iterated downward. When the difference between the two cervical measurement values ​​is less than 0.01, the calculation is stopped and the mean value is output.

[0040] In this embodiment, the training data set consists of 176 intrapartum ultrasound images of cervical transection, divided into a training set and a validation set ratio of 9:1. In order to prevent the model from overfitting and enhance the reliability of the test, 62 ultrasound images that did not appear in the training set were used in the test part. The reliability of the present invention was tested by comparing the automatic measurement results of UODNet and ALD method with the manual measurement results of radiologists. Figure 4 The figure shows the segmentation and measurement results of three cervical dilations (#1, #2, and #3) by UNet and UODNet. UODNet can detect and segment objects (e.g., #2 and #3) more accurately. Figure 4 As shown in the figure, for #1, UNet and UODNet can measure the cervical dilation distance very well. In the case of #2, UNet only segmented a small part of the central area of ​​cervical dilation, and the calculation result of cervical dilation distance in this case had a large deviation, while UODNet segmented a relatively complete cervical dilation area very well. For #3, Unet lacked some boundary information, resulting in a small cervical dilation distance, while UODNet achieved significantly better results at the boundary.

[0041] In summary, this embodiment provides a method for automatically measuring the cervical dilation distance based on intrapartum ultrasound images. Figure 5 In the 62 test cases, the results automatically calculated by UODNet and ALD methods are only around the manual measurement results, which shows that the cervical dilation distance results calculated by UODNet are consistent with the manual measurement results of doctors. According to statistics, the average absolute error of the cervical dilation distance automatic measurement method based on intrapartum ultrasound images provided by this embodiment on 62 test cases is 9.82% compared with the manual measurement value, which has strong reliability.

Claims

1. A method for automatically measuring the cervical dilation distance based on intrapartum ultrasound images, characterized in that: The method for automatically measuring the cervical dilation distance of an ultrasound image during labor comprises the following steps: S1. Collect transperineal ultrasound images and label annotations of the cervical os; S2, further screening the ultrasound images and selecting ultrasound images with clear anatomical structures; S3, using UODNet to process and calculate the image obtained in S2, and obtain the cervical position and segmentation information of the image; S4. Measure the cervical dilation distance using the ALD method based on the segmentation information of S3.

2. The method for automatically measuring the cervical dilation distance based on ultrasound images during labor according to claim 1, characterized in that: The image data in step S1 comes from intrapartum ultrasound images with different cervical dilation distances.

3. The method for automatically measuring the cervical dilation distance based on ultrasound images during labor according to claim 1 or 2, characterized in that: The step S2 selects an intrapartum ultrasound image of a cervical transection.

4. The method for automatically measuring the cervical dilation distance based on ultrasound images during labor according to any one of claims 1 to 3, characterized in that: The network model structure of UODNet in step S3 includes an encoder and a decoder.

5. The method for automatically measuring the cervical dilation distance based on ultrasound images during labor according to claim 4, characterized in that: The UODNet network model structure encoder adopts PoolFormer.

6. The method for automatically measuring the cervical dilation distance based on ultrasound images during labor according to claim 4 or 5, characterized in that: The UODNet network model structure decoder adopts a FPN-like structure.

7. The method for automatically measuring the cervical dilation distance based on ultrasound images during labor according to any one of claims 1 to 6, characterized in that: For the segmentation information output by UODNet, the mean value after the clamping reaches the threshold calculated by the ALD method is the cervical dilation distance of the image.

Citation Information

Patent Citations

  • Device and method for detecting expansion degree of cervix mouth during labor

    CN109770928A

  • Intelligent cervix measuring method and device

    CN110393530A

  • Method, system and equipment for measuring labor progress angle based on ultrasonic image and medium

    CN115526835A

  • Method and device for measuring uterine orifice diameter of ultrasonic image based on convolutional neural network

    CN116664660A

  • Ultrasound imaging apparatus and method of controlling the same

    US20210219950A1