Ultrasonic fetal image segmentation method
By using dual-line detection and parallel processing technology in ultrasonic fetal image segmentation method and combining preset image segmentation models for semantic segmentation, the problems of low image quality and low detection speed efficiency in the prior art are solved, and higher quality is achieved and faster fetal facial imaging and diagnosis.
Patent Information
- Application Number
- CN202411569976.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-11-05
- Publication Date
- 2025-02-11
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
The existing ultrasonic fetal image segmentation methods have problems such as low image quality and low detection speed efficiency, which leads to insufficient clear fetal facial imaging and difficult to provide real-time diagnostic information.
Using ultrasonic fetal image segmentation method, by acquiring ultrasonic fetal images and preprocessing, double-line detection is performed based on the fetal neck transparent layer features to obtain the reference area, input the image to the parallel processor for segmentation and classification processing, and use preset image segmentation The model performs semantic segmentation to obtain accurate image segmentation results.
It achieves more accurate and detailed fetal shape segmentation, improves fetal facial imaging effects, helps doctors to conduct prenatal diagnosis faster, and improves the real-time and accuracy of the detection.
Smart Images
Figure CN119418057A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of ultrasonic fetal image segmentation, and more particularly to an ultrasonic fetal image segmentation method. Background Art
[0002] Prenatal ultrasound diagnosis is one of the important means of discovering congenital defects. In prenatal diagnosis, 3D4D imaging technology of the fetal face can effectively present the complete face of the fetus, which is helpful for the prenatal diagnosis of various diseases. However, at this stage, during the fetal facial imaging process, the fetal face is easily blocked by other interfering objects, resulting in unclear fetal facial imaging. The main process of 3D4D imaging on the machine is to input the fetal volume data, which requires excessive participation of doctors and manual selection of regions of interest. 3D4D imaging based on the selected regions of interest requires doctors to have certain clinical experience, which is not easy for many new doctors. Secondly, it is necessary to manually select the region of interest and perform 3D4D imaging based on the selected region of interest. Although this method reduces the probability of interference blocking the fetal face to a certain extent, it has high requirements for image quality and is not ideal for machines with poor image quality. In addition, during the doctor's detection process, the intelligent ultrasound device can provide auxiliary functions for screening standard sections and measuring parameters. The overall detection algorithm is usually placed on the CPU or GPU for coarse-grained parallelization. In clinical applications, the computing speed of the CPU or GPU cannot keep up with the speed of the ultrasound device generating image frames, and frame-by-frame processing cannot be achieved, resulting in the loss of some detection sections, making it difficult to provide real-time and accurate diagnostic information. Therefore, the existing technology has the problem of low detection speed and efficiency. Summary of the invention
[0003] 1. Technical issues to be resolved
[0004] In view of the problems existing in the prior art, the present invention provides an ultrasound fetal image segmentation method to solve the technical problems mentioned in the background technology.
[0005] (II) Technical solution
[0006] To achieve the above object, the present invention provides the following technical solution: an ultrasonic fetal image segmentation method, comprising the following steps:
[0007] Step 1: acquiring an ultrasonic fetal image, acquiring a preset number of continuous frame sequences of fetal ultrasonic image frames, and acquiring an ultrasonic image including a fetal nuchal transparent layer, and sequentially performing preprocessing operations on the continuous frame sequences of the fetal ultrasonic image frames to obtain preprocessed image frames;
[0008] Step 2: Based on the fetal nuchal translucency feature, double straight line detection is performed on the ultrasound image to obtain a fetal nuchal translucency reference area, and the preprocessed image frame is sent to a first processor for parallel processing, and the first processor inputs the preprocessed image frame into a trained segmentation model to obtain a segmentation result of the preprocessed image frame;
[0009] Step three: Input the ultrasound fetal image into a preset image segmentation model to obtain an image segmentation result. The preset image segmentation model includes: a feature extraction layer, a pixel representation layer, a first semantic segmentation layer, a pixel semantic layer, a region representation layer, a pixel stitching layer, a pixel aggregation layer and a second semantic segmentation layer. Send the segmentation result to a second processor for parallel processing. The second processor inputs the preprocessed image frame into a trained classification model to obtain and display the final detection result of the fetal ultrasound image frame, and obtains the target contour of the fetal nuchal transparent layer based on the reference area.
[0010] The present invention is further configured such that before the step of inputting the ultrasound fetal image into a preset image segmentation model, it also includes: obtaining a fetal image training set, the image training set including an initial image and a label image; inputting the initial image into an initial image segmentation model to obtain an initial image segmentation result, the initial image segmentation model including: a feature extraction layer, a pixel representation layer, a first semantic segmentation layer, a pixel semantic layer, a region representation layer, a pixel splicing layer, a pixel aggregation layer and a second semantic segmentation layer; and determining a preset image segmentation model based on the initial image segmentation result and the label image.
[0011] The present invention is further configured such that the image training set of the fetus is obtained, and the image training set includes the steps of an initial image and a label image, specifically including: obtaining three-dimensional volume data of the fetus; performing two-dimensional slicing on the three-dimensional volume data to obtain an initial image; and segmenting and annotating the initial image to obtain a label image.
[0012] The present invention is further configured such that the segmentation result is sent to a second processor for parallel processing, and the second processor inputs the preprocessed image frame into a trained classification model to obtain and display the final detection result of the fetal ultrasound image frame, including: sending the segmentation result to a third processor for parallel processing, and the third processor makes a preliminary judgment on whether the preprocessed image frame is a standard section image according to a preset standard; if it is preliminarily judged to be a standard section image, the preprocessed image frame is sent to the second processor, and the second processor inputs the preprocessed image frame into a trained classification model to obtain and display the final detection result of the fetal ultrasound image frame.
[0013] The present invention is further configured such that the double straight line detection of the ultrasound image comprises: detecting two high echo band characteristic areas of the fetal nuchal transparent layer from the ultrasound image to obtain two parallel baselines; and obtaining the reference area based on the two parallel baselines.
[0014] The present invention is further configured such that the detecting of the two high echo band characteristic areas of the fetal nuchal transparent layer from the ultrasound image comprises: enhancing the characteristics of the two high echo band characteristic areas of the fetal nuchal transparent layer on the ultrasound image to obtain a high echo band characteristic image; and detecting the two high echo band characteristic areas based on the characteristic image.
[0015] The present invention is further configured as follows: inputting the preprocessed image frame into a trained segmentation model to obtain a segmentation result of the preprocessed image frame, including: determining a plurality of candidate blocks according to the preprocessed image frame; inputting each of the candidate blocks into a trained segmentation model respectively to obtain a segmentation result of the preprocessed image frame; and making a preliminary judgment on whether the preprocessed image frame is a standard section image according to a preset standard, including: judging whether the candidate block has the tissue of the standard section image corresponding to the preprocessed image frame according to the segmentation result and the preset standard; if so, preliminarily judging the preprocessed image frame as a standard section image.
[0016] The present invention is further configured such that each of the candidate blocks is input into a trained segmentation model respectively to obtain a segmentation result of the preprocessed image frame, including: taking each channel corresponding to each of the candidate blocks as a thread respectively to obtain multi-threaded image data; and inputting the multi-threaded image data into a trained segmentation model respectively to obtain a segmentation result of the preprocessed image frame.
[0017] (III) Beneficial effects
[0018] Compared with the prior art, the present invention provides an ultrasound fetal image segmentation method, which has the following beneficial effects:
[0019] The present invention obtains an ultrasound fetal image and then inputs the ultrasound fetal image into a preset image segmentation model to obtain an image segmentation result. The preset image segmentation model includes: a feature extraction layer, a pixel representation layer, a first semantic segmentation layer, a pixel semantic layer, a region representation layer, a pixel splicing layer, a pixel aggregation layer, and a second semantic segmentation layer. The present invention inputs the ultrasound fetal image into the preset image segmentation model, and can perform a coarse segmentation of the image output by the pixel representation layer through the first semantic segmentation layer, and perform a fine segmentation of the image output by the pixel aggregation layer through the second semantic segmentation layer, so as to obtain an accurate image segmentation result. Compared with the existing method that requires manual selection of the region of interest, the above method of the present invention can more accurately and finely segment the shape of the fetus, and use the image segmentation result of the preset image segmentation model for 3D4D fetal facial imaging, thereby improving the effect of fetal facial imaging and helping doctors to perform prenatal diagnosis faster and better. The continuous frame sequence of the acquired fetal ultrasound image frames is preprocessed in sequence, and the preprocessed image frames are sent to the first processor for segmentation processing, and the second processor classifies the segmented image frames to obtain the final detection results of the standard section or non-standard section. In addition, an ultrasound image containing the fetal neck transparent layer is obtained, and a double straight line detection is performed on the ultrasound image to obtain the fetal neck transparent layer reference area, and the target contour of the fetal neck transparent layer is obtained based on the reference area. The method automatically segments the NT contour, improves the degree of automation, and at the same time, by performing double straight line detection on the NT area, the NT position can be relatively accurately located, thereby obtaining the NT reference area. On the one hand, the interference of the area outside the baseline on the subsequent calculation can be eliminated, thereby saving the subsequent calculation time, and on the other hand, a relatively accurate basis is provided for the subsequent construction of the NT target contour, thereby increasing the accuracy of the calculation. BRIEF DESCRIPTION OF THE DRAWINGS
[0020] Figure 1 is an overall flow chart of the ultrasound fetal image segmentation method of the present invention; DETAILED DESCRIPTION
[0021] Embodiment 1:
[0022] like Figure 1 As shown, the present invention discloses a method for segmenting an ultrasonic fetal image, comprising the following steps:
[0023] Step 1: acquiring an ultrasonic fetal image, acquiring a preset number of continuous frame sequences of fetal ultrasonic image frames, and acquiring an ultrasonic image including a fetal nuchal transparent layer, and sequentially performing preprocessing operations on the continuous frame sequences of the fetal ultrasonic image frames to obtain preprocessed image frames;
[0024] Step 2: Based on the fetal nuchal translucency feature, double straight line detection is performed on the ultrasound image to obtain a fetal nuchal translucency reference area, and the preprocessed image frame is sent to a first processor for parallel processing, and the first processor inputs the preprocessed image frame into a trained segmentation model to obtain a segmentation result of the preprocessed image frame;
[0025] Step three: Input the ultrasound fetal image into a preset image segmentation model to obtain an image segmentation result. The preset image segmentation model includes: a feature extraction layer, a pixel representation layer, a first semantic segmentation layer, a pixel semantic layer, a region representation layer, a pixel stitching layer, a pixel aggregation layer and a second semantic segmentation layer. Send the segmentation result to a second processor for parallel processing. The second processor inputs the preprocessed image frame into a trained classification model to obtain and display the final detection result of the fetal ultrasound image frame, and obtains the target contour of the fetal nuchal transparent layer based on the reference area.
[0026] In a further embodiment of the present invention, before the step of inputting the ultrasound fetal image into a preset image segmentation model, it also includes: obtaining a fetal image training set, the image training set including an initial image and a label image; inputting the initial image into the initial image segmentation model to obtain an initial image segmentation result, the initial image segmentation model including: a feature extraction layer, a pixel representation layer, a first semantic segmentation layer, a pixel semantic layer, a region representation layer, a pixel splicing layer, a pixel aggregation layer and a second semantic segmentation layer; determining a preset image segmentation model according to the initial image segmentation result and the label image; the step of obtaining a fetal image training set, the image training set including an initial image and a label image, specifically includes: obtaining three-dimensional volume data of the fetus; performing segmentation on the three-dimensional volume data The method comprises: performing two-dimensional slicing to obtain an initial image; segmenting and annotating the initial image to obtain a label image; sending the segmentation result to a second processor for parallel processing, which inputs the preprocessed image frame into a trained classification model, and obtains and displays the final detection result of the fetal ultrasound image frame, including: sending the segmentation result to a third processor for parallel processing, which performs a preliminary determination on whether the preprocessed image frame is a standard section image according to a preset standard by the third processor; if the preliminary determination is that it is a standard section image, sending the preprocessed image frame to the second processor, which inputs the preprocessed image frame into a trained classification model by the second processor, and obtains and displays the final detection result of the fetal ultrasound image frame.
[0027] Embodiment 2:
[0028] The ultrasound fetal image segmentation method comprises the following steps:
[0029] Step 1: acquiring an ultrasonic fetal image, acquiring a preset number of continuous frame sequences of fetal ultrasonic image frames, and acquiring an ultrasonic image including a fetal nuchal transparent layer, and sequentially performing preprocessing operations on the continuous frame sequences of the fetal ultrasonic image frames to obtain preprocessed image frames;
[0030] Step 2: Based on the fetal nuchal translucency feature, double straight line detection is performed on the ultrasound image to obtain a fetal nuchal translucency reference area, and the preprocessed image frame is sent to a first processor for parallel processing, and the first processor inputs the preprocessed image frame into a trained segmentation model to obtain a segmentation result of the preprocessed image frame;
[0031] Step three: Input the ultrasound fetal image into a preset image segmentation model to obtain an image segmentation result. The preset image segmentation model includes: a feature extraction layer, a pixel representation layer, a first semantic segmentation layer, a pixel semantic layer, a region representation layer, a pixel stitching layer, a pixel aggregation layer and a second semantic segmentation layer. Send the segmentation result to a second processor for parallel processing. The second processor inputs the preprocessed image frame into a trained classification model to obtain and display the final detection result of the fetal ultrasound image frame, and obtains the target contour of the fetal nuchal transparent layer based on the reference area.
[0032] In a further embodiment of the present invention, the double straight line detection of the ultrasound image includes: detecting two high echo band characteristic areas of the fetal nuchal transparent layer from the ultrasound image to obtain two parallel baselines; and obtaining the reference area based on the two parallel baselines.
[0033] In a further embodiment of the present invention, detecting two high echo band characteristic areas of the fetal nuchal translucency layer from the ultrasound image includes: enhancing the features of the two high echo band characteristic areas of the fetal nuchal translucency layer on the ultrasound image to obtain a high echo band characteristic image; and detecting the two high echo band characteristic areas based on the characteristic image.
[0034] In a further embodiment of the present invention, the step of inputting the preprocessed image frame into a trained segmentation model to obtain a segmentation result of the preprocessed image frame comprises: determining a plurality of candidate blocks according to the preprocessed image frame; inputting each of the candidate blocks into a trained segmentation model to obtain a segmentation result of the preprocessed image frame; the step of making a preliminary judgment on whether the preprocessed image frame is a standard section image according to a preset standard comprises: judging whether the candidate block has the tissue of the standard section image corresponding to the preprocessed image frame according to the segmentation result according to a preset standard; if so, preliminarily judging the preprocessed image frame as a standard section image; the step of inputting each of the candidate blocks into a trained segmentation model to obtain a segmentation result of the preprocessed image frame comprises: taking each channel corresponding to each of the candidate blocks as a thread to obtain multi-threaded image data; and inputting the multi-threaded image data into the trained segmentation model to obtain a segmentation result of the preprocessed image frame.
[0035] The above-mentioned embodiments only express several implementation methods of the present invention, and the descriptions thereof are relatively specific and detailed, but they cannot be listed as limiting the scope of the invention patent. It should be pointed out that, for ordinary technicians in this field, several variations and improvements can be made without departing from the concept of the present invention, and these all belong to the protection scope of the present invention. Therefore, the protection scope of the patent of the present invention shall be subject to the attached claims.
Claims
1. A method for segmenting an ultrasonic fetal image, characterized in that: The following steps are involved: Step 1: acquiring an ultrasonic fetal image, acquiring a preset number of continuous frame sequences of fetal ultrasonic image frames, and acquiring an ultrasonic image including a fetal nuchal transparent layer, and sequentially performing preprocessing operations on the continuous frame sequences of the fetal ultrasonic image frames to obtain preprocessed image frames; Step 2: Based on the fetal nuchal translucency feature, double straight line detection is performed on the ultrasound image to obtain a fetal nuchal translucency reference area, and the preprocessed image frame is sent to a first processor for parallel processing, and the first processor inputs the preprocessed image frame into a trained segmentation model to obtain a segmentation result of the preprocessed image frame; Step three: Input the ultrasound fetal image into a preset image segmentation model to obtain an image segmentation result. The preset image segmentation model includes: a feature extraction layer, a pixel representation layer, a first semantic segmentation layer, a pixel semantic layer, a region representation layer, a pixel stitching layer, a pixel aggregation layer and a second semantic segmentation layer. Send the segmentation result to a second processor for parallel processing. The second processor inputs the preprocessed image frame into a trained classification model to obtain and display the final detection result of the fetal ultrasound image frame, and obtains the target contour of the fetal nuchal transparent layer based on the reference area.
2. The method for segmenting an ultrasonic fetal image according to claim 1, wherein: Before the step of inputting the ultrasound fetal image into the preset image segmentation model, it also includes: obtaining a fetal image training set, the image training set includes an initial image and a label image; inputting the initial image into the initial image segmentation model to obtain an initial image segmentation result, the initial image segmentation model includes: a feature extraction layer, a pixel representation layer, a first semantic segmentation layer, a pixel semantic layer, a region representation layer, a pixel splicing layer, a pixel aggregation layer and a second semantic segmentation layer; and determining a preset image segmentation model according to the initial image segmentation result and the label image.
3. The method for segmenting an ultrasonic fetal image according to claim 1, wherein: The step of obtaining a fetal image training set, wherein the image training set includes an initial image and a label image, specifically includes: obtaining three-dimensional volume data of the fetus; performing two-dimensional slicing on the three-dimensional volume data to obtain an initial image; and segmenting and annotating the initial image to obtain a label image.
4. The method for segmenting an ultrasonic fetal image according to claim 1, wherein: The method of sending the segmentation result to a second processor for parallel processing, which inputs the preprocessed image frame into a trained classification model, obtains and displays the final detection result of the fetal ultrasound image frame, including: sending the segmentation result to a third processor for parallel processing, which makes a preliminary judgment on whether the preprocessed image frame is a standard section image according to a preset standard by the third processor; if it is preliminarily judged to be a standard section image, sending the preprocessed image frame to the second processor, which inputs the preprocessed image frame into a trained classification model, obtains and displays the final detection result of the fetal ultrasound image frame.
5. The method for segmenting an ultrasonic fetal image according to claim 1, wherein: The double straight line detection on the ultrasound image comprises: detecting two high echo band characteristic areas of the fetal nuchal transparent layer from the ultrasound image to obtain two parallel baselines; and obtaining the reference area based on the two parallel baselines.
6. The method for segmenting an ultrasonic fetal image according to claim 5, wherein: The detecting of two high echo band characteristic areas of the fetal nuchal translucency layer from the ultrasound image comprises: enhancing the characteristics of the two high echo band characteristic areas of the fetal nuchal translucency layer on the ultrasound image to obtain a high echo band characteristic image; and detecting the two high echo band characteristic areas based on the characteristic image.
7. The method for segmenting an ultrasonic fetal image according to claim 6, wherein: The step of inputting the preprocessed image frame into a trained segmentation model to obtain a segmentation result of the preprocessed image frame includes: determining a plurality of candidate blocks according to the preprocessed image frame; inputting each of the candidate blocks into a trained segmentation model to obtain a segmentation result of the preprocessed image frame; and making a preliminary judgment on whether the preprocessed image frame is a standard section image according to a preset standard, including: judging whether the candidate block has the tissue of the standard section image corresponding to the preprocessed image frame according to the segmentation result and the preset standard; if so, preliminarily judging the preprocessed image frame as a standard section image.
8. The method for segmenting an ultrasonic fetal image according to claim 3, wherein: The step of inputting each of the candidate blocks into a trained segmentation model to obtain a segmentation result of the preprocessed image frame includes: taking each channel corresponding to each of the candidate blocks as a thread to obtain multi-threaded image data; and inputting the multi-threaded image data into a trained segmentation model to obtain a segmentation result of the preprocessed image frame.