Fetal brain ultrasound image processing method and ultrasound device

Abstract

Embodiments of the present application provide a fetal brain ultrasound image processing method and an ultrasound device. The method comprises: determining a key point of a tissue structure on a median sagittal plane of a fetal brain; determining a standard line based on the key point of the tissue structure to determine an initial standard plane, wherein the initial standard plane is a plane passing through the standard line and perpendicular to the median sagittal plane; inputting the initial standard plane into a standard plane detection model to obtain a first detection result, wherein the first detection result comprises information about a probability that the initial standard plane is a first type of standard plane; adjusting the pose of the standard line based on the first detection result to obtain a new standard plane corresponding to an adjusted standard line; and determining a tissue structure measurement item of the fetal brain based on the new standard plane. This significantly reduces the requirements for the operator, reduces the scanning time of the ultrasound device, and effectively improves the accuracy and efficiency of the measurement item.

Description

Technical Field

[0001] This invention relates to the field of medical image processing technology, and more specifically to a method for processing fetal cranial ultrasound images, an ultrasound device, a storage medium, and a computer program product. Background Technology

[0002] In prenatal fetal screening, measurements of various structures within the fetal brain are crucial for assessing fetal growth and development. Typically, a standard view of the fetal brain is first determined, and then structural measurements are determined based on this standard view. Currently, obtaining this standard view clinically relies primarily on operators manually manipulating ultrasound equipment, relying on their own understanding of the intracranial structures. This traditional method has several significant drawbacks. Firstly, it demands extremely high levels of professional skill and experience from the operators. Differences in operating habits, experience levels, and understanding of intracranial structures among different operators can lead to inconsistent quality of the obtained standard view, thus affecting the accuracy and consistency of structural measurements. Secondly, because operators may need to scan repeatedly to determine the standard view, the scanning process is extremely time-consuming, significantly increasing the operator's workload. Prolonged examinations can also cause discomfort for pregnant women, negatively impacting their experience.

[0003] Therefore, how to accurately and quickly determine the tissue structure measurements of the fetal brain has become an urgent technical problem to be solved. Summary of the Invention

[0004] The present invention is proposed in view of the above-mentioned problems. According to one aspect of the present invention, a method for processing fetal cranial ultrasound images is provided, comprising:

[0005] To identify key structural points of the fetal brain in the midsagittal plane;

[0006] The standard line is determined based on the key points of the organizational structure to determine the initial standard plane, which is a plane that passes through the standard line and is perpendicular to the midsagittal plane;

[0007] An initial standard surface is input into the standard surface detection model to obtain a first detection result, wherein the first detection result includes information about the probability that the initial standard surface is a first type of standard surface;

[0008] Based on the first detection result, the pose of the standard line is adjusted to obtain the new standard plane corresponding to the adjusted standard line.

[0009] Based on the new standard plane, the tissue structure measurement items of the fetal brain were determined.

[0010] For example, the midsagittal plane is determined in the following way:

[0011] Based on three-dimensional ultrasound data of the fetal brain, the three-dimensional midline data of the fetal brain were determined.

[0012] Planar fitting was performed on the three-dimensional brain midline data to determine the midsagittal plane of the fetal brain.

[0013] For example, determining three-dimensional midline data of the fetal brain based on three-dimensional ultrasound data of the fetal brain includes:

[0014] Three-dimensional ultrasound data is input into a first detection model to obtain three-dimensional brain midline data, wherein the first detection model includes interconnected encoders and decoders.

[0015] For example, the three-dimensional ultrasound data includes multiple two-dimensional image data. Based on the three-dimensional ultrasound data of the fetal brain, the three-dimensional midline data of the fetal brain is determined, including:

[0016] Each two-dimensional image data is input into the second detection model to obtain the two-dimensional brain midline data in each two-dimensional image data;

[0017] Based on the sequence relationship of multiple two-dimensional image data, all two-dimensional brain midline data are merged into three-dimensional brain midline data.

[0018] For example, performing planar fitting on three-dimensional brain midline data to determine the midsagittal plane of the fetal brain includes:

[0019] Connectivity determination was performed on the three-dimensional brain midline data, and the largest connected component was identified from all connected components to serve as the brain midline region.

[0020] Plane fitting is performed based on the midline region of the brain to determine the fitted plane;

[0021] The fitted plane is adjusted to be vertical to obtain the midsagittal plane.

[0022] For example, identifying key structural points of the fetal brain in the midsagittal plane includes:

[0023] For the cavum septum pellucidum and cerebellum in the fetal brain, image segmentation was performed on the midsagittal plane to obtain the segmentation results;

[0024] Determine the two endpoints of the cavum septum pellucidum and the centroid of the cerebellum;

[0025] Calculate the distances between the centroid of the cerebellum and the two endpoints of the cavum septum pellucidum to determine the distal and proximal ends of the cavum septum pellucidum, respectively; wherein, the key points of the tissue structure include the distal end of the cavum septum pellucidum, and the key points of the tissue structure also include the centroid of the cerebellum and / or the proximal end of the cavum septum pellucidum.

[0026] Based on key points of the organizational structure, determine the benchmarks, including:

[0027] The line connecting the first point to the distal end of the cavum septum pellucidum is defined as the standard line, where the first point is the point on the line connecting the centroid of the cerebellum and the proximal end of the cavum septum pellucidum.

[0028] For example, the first type of standard plane includes a thalamic transverse section and / or a lateral ventricle transverse section, wherein for both the thalamic and lateral ventricle transverse sections, the first point is the proximal end of the cavum septum pellucidum; and / or

[0029] The first type of standard plane includes the cerebellar transverse section. For the cerebellar transverse section, the first point is the centroid of the cerebellum.

[0030] For example, adjusting the pose of a standard line includes:

[0031] After inputting the new standard surface into the standard surface detection model to obtain a new first detection result, the standard line is translated and / or rotated based on the new first detection result.

[0032] For example, the first type of standard plane includes a thalamic cross section. Based on the new standard plane, the tissue structure measurements of the fetal brain are determined, including:

[0033] For the thalamus in the fetal brain, image segmentation is performed on the final standard plane of the thalamus cross section to determine the thalamic region in the thalamus cross section;

[0034] Ellipse fitting was performed on the thalamic region, and the major axis, minor axis, and circumference of the fitted ellipse were calculated to represent the occipitofrontal diameter, biparietal diameter, and head circumference, respectively. The tissue structure measurements included the occipitofrontal diameter, biparietal diameter, and head circumference.

[0035] For example, the first type of standard plane includes a cerebellar cross section. Based on the new standard plane, the tissue structure measurements of the fetal brain are determined, including:

[0036] For the cerebellum in the fetal brain, the final standard plane of the cerebellar cross section is segmented to determine the cerebellum and cerebellomedullary cisterns in the cerebellar cross section;

[0037] Determine the bounding box of the cerebellum and the bounding box of the cerebellomedullary cistern;

[0038] Based on the bounding box of the cerebellum, the actual physical length of the transverse diameter of the cerebellum is determined; based on the bounding box of the cerebellomedullary cistern, the actual physical length of the cerebellomedullary cistern is determined.

[0039] Among them, the tissue structure measurement items include the actual physical length of the transverse diameter of the cerebellum and the actual physical length of the cerebellomedullary cistern.

[0040] For example, the first type of standard plane includes a transverse section of the lateral ventricle. Based on the new standard plane, the tissue structure measurements of the fetal brain are determined, including:

[0041] For the lateral ventricle in the fetal brain, the final standard plane of the transverse section of the lateral ventricle is segmented to obtain the lateral ventricle region in the transverse section of the lateral ventricle.

[0042] Determine the bounding box of the lateral ventricle region;

[0043] The actual physical width of the lateral ventricle region is determined based on the bounding box of the lateral ventricle region.

[0044] Among them, the fetal brain tissue structure measurement includes the actual physical width of the lateral ventricle region.

[0045] According to another aspect of the present invention, an ultrasound device is also provided, comprising a processor and a memory, wherein the memory stores computer program instructions, which, when executed by the processor, are used to perform the fetal cranial ultrasound image processing method as described above.

[0046] According to another aspect of the present invention, a storage medium is also provided, on which program instructions are stored, which, when executed, are used to perform the fetal cranial ultrasound image processing method as described above.

[0047] According to another aspect of the present invention, a computer program product is also provided, comprising computer program instructions which, when executed, are used to perform the fetal cranial ultrasound image processing method as described above.

[0048] According to the above technical solution, a standard line can be determined first, followed by the initial standard plane corresponding to the standard line. Then, a new standard plane is obtained by automatically adjusting the pose of the standard line based on the first detection result. The tissue structure measurements of the fetal brain can then be determined on this new standard plane. This significantly reduces the requirements for operators, shortens the scanning time of ultrasound equipment, and effectively improves the accuracy and efficiency of measurement. It plays a positive role in ensuring the healthy development of the fetus and the safety of mother and child, while also promoting the development of prenatal ultrasound examination technology towards intelligence and precision.

[0049] The above description is merely an overview of the technical solution of the present invention. In order to better understand the technical means of the present invention and to implement it in accordance with the contents of the specification, and in order to make the above and other objects, features and advantages of the present invention more apparent and understandable, specific embodiments of the present invention are described below. Attached Figure Description

[0050] The above and other objects, features, and advantages of the present invention will become more apparent from the more detailed description of the embodiments of the invention in conjunction with the accompanying drawings. The drawings are provided to further illustrate the embodiments of the invention and form part of the specification. They are used together with the embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings, the same reference numerals generally represent the same parts or steps.

[0051] Figure 1 A schematic flowchart of a fetal cranial ultrasound image processing method according to an embodiment of this application is shown;

[0052] Figure 2 A schematic diagram of a median sagittal plane image according to an embodiment of this application is shown;

[0053] Figure 3 A schematic diagram showing an image of a thalamic region in a cross-section of the thalamus according to an embodiment of this application is illustrated.

[0054] Figure 4 A schematic diagram showing an image of the cerebellar region in a cross-section of the cerebellum according to an embodiment of this application is illustrated.

[0055] Figure 5 A schematic diagram showing an image of the cerebellomedullary cistern region in a cross-section of the cerebellum according to an embodiment of this application is illustrated.

[0056] Figure 6 A schematic diagram showing an image of the lateral ventricle region in a cross-section of the lateral ventricle according to an embodiment of this application is provided; and

[0057] Figure 7 A schematic block diagram of an ultrasonic device according to an embodiment of this application is shown. Detailed Implementation

[0058] To make the objectives, technical solutions, and advantages of the present invention more apparent, exemplary embodiments according to the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are merely a part of the embodiments of the present invention, and not all of the embodiments of the present invention. It should be understood that the present invention is not limited to the exemplary embodiments described herein. Based on the embodiments of the present invention described herein, all other embodiments obtained by those skilled in the art without inventive effort should fall within the protection scope of the present invention.

[0059] To at least partially solve the above-mentioned technical problems, an embodiment of the present invention provides a method for processing fetal cranial ultrasound images. Figure 1 A schematic flowchart of a fetal cranial ultrasound image processing method 100 according to an embodiment of this application is shown. Figure 1As shown, method 100 may include, but is not limited to, the following steps S120 to S160.

[0060] Step S120: Determine the key points of the tissue structures on the midsagittal plane of the fetal brain.

[0061] Key structural points of the fetal brain are pixels in the midsagittal plane that are significant and stable in terms of structure, such as points at the edges or center of structures. Specifically, key structural points may include the centroid of the cerebellum, the endpoints of the cavum septum pellucidum, etc.

[0062] For example, machine learning algorithms such as semantic segmentation networks like U-Net, SegNet, DeepLab, or support vector machines (SVM) can be used to segment the midsagittal plane of an image, thereby outputting segmentation results of the cranial tissue structure. Keypoint extraction algorithms can be used to extract corresponding keypoints from the segmentation results of the fetal cranial tissue structure.

[0063] Step S130: Determine the standard line based on the key points of the tissue structure to determine the initial standard plane, wherein the initial standard plane is a plane that passes through the standard line and is perpendicular to the midsagittal plane.

[0064] For example, after determining the key points of the organizational structure, standard lines can be determined on the median sagittal plane based on these key points. A standard line can be determined based on at least two key points. For instance, the key points of the organizational structure may include three points: A, B, and C. The line connecting points A and B can be determined as standard line 1, and similarly, the line connecting points B and C can be determined as standard line 2. Alternatively, the line connecting points A and B can be determined as line 1, and the line connecting point C to any point on line 1 can be determined as standard line 3. Or, the key points of the organizational structure may include four points: A, B, C, and D. The line connecting points A and B can be determined as line 2, the line connecting points C and D can be determined as line 3, and the line connecting any point on line 2 and any point on line 3 can be determined as standard line 4. It can be understood that the determined standard lines lie on the median sagittal plane. The plane passing through the standard lines and perpendicular to the median sagittal plane can be determined as the initial standard plane. It should be noted that one standard line can uniquely determine one initial standard plane.

[0065] For example, in three-dimensional space, the equation of a plane perpendicular to the median sagittal plane and passing through the standard line can be determined geometrically based on the direction and position of the standard line and the normal vector of the median sagittal plane, thus obtaining the initial standard plane. Specifically, in three-dimensional space, the direction vector of the standard line and the normal vector of the median sagittal plane are known. Assume a point on the standard line has coordinates (x0, y0, z0), the direction vector of the standard line is (a, b, c), and the normal vector of the median sagittal plane is (m, n, p). Since the initial standard plane is perpendicular to the median sagittal plane, its normal vector is perpendicular to both the direction vector of the standard line and the normal vector of the median sagittal plane. Based on the point normal form of the plane, the normal vector of the initial standard plane can be calculated using the cross product of vectors, thereby determining the equation of the initial standard plane and obtaining it.

[0066] Step S140: Input the initial standard surface into the standard surface detection model to obtain a first detection result, wherein the first detection result includes information about the probability that the initial standard surface is a first type of standard surface.

[0067] For example, the standard surface detection model can be a trained semantic segmentation network. When training the standard surface detection model, a large amount of image data containing different types of standard surfaces, such as thalamic cross-sections, lateral ventricle cross-sections, and cerebellar cross-sections, is collected, and these image data are labeled to indicate the standard surface type of each image. The model is then trained using this labeled data, enabling it to learn the features of different types of standard surfaces. By learning from a large number of known standard surface images, the trained model can classify and evaluate initial input standard surfaces.

[0068] Inputting the initial standard plane into a trained standard plane detection model outputs a first detection result. This first detection result includes information about the probability that the initial standard plane is a first-type standard plane. The first-type standard plane can include thalamic cross-sections, lateral ventricle cross-sections, and cerebellar cross-sections, etc. In one embodiment, the first detection result might indicate a probability of 0.7 for the initial standard plane being a thalamic cross-section, 0.2 for the initial standard plane being a lateral ventricle cross-section, and 0.1 for the initial standard plane being a cerebellar cross-section, etc. This probability information allows for a preliminary assessment of the similarity between the initial standard plane and the target standard plane.

[0069] Step S150: Based on the first detection result, the pose of the standard line is adjusted to obtain the new standard surface corresponding to the adjusted standard line.

[0070] The pose of the standard line can be adjusted once or multiple times based on the first detection result to obtain a new standard surface. Compared with obtaining the first detection result based on the initial standard surface, the probability that the new standard surface represented by the first detection result obtained by inputting the new standard surface into the standard surface detection model is a specific first-type standard surface increases. In other words, by adjusting the pose of the standard line, the corresponding standard surface obtained based on the adjusted standard line becomes closer and closer to, or even reaches, a specific first-type standard surface.

[0071] In one embodiment, the pose adjustment of the standard line can be achieved as follows: After inputting a new standard plane into the standard plane detection model to obtain a new first detection result, the standard line is translated and / or rotated based on the new first detection result. For example, suppose the position of the standard line in three-dimensional space is represented by the coordinates (x0, y0, z0) of a point on it. Since the standard line is located on the median sagittal plane, each translation operation can move a certain distance in any two of the three directions (x, y, z), such as the x and y directions. The coordinates of this point on the initial standard plane after the translation can be (x0+Δx, y0+Δy, z0). In this embodiment, the median sagittal plane is parallel to the z direction. After each translation, the new standard plane corresponding to the adjusted standard line is input into the standard plane detection model again to obtain a new first detection result. For example, based on the probability information output by the standard plane detection model, the standard line can be translated in the direction of increasing probability to obtain a new standard plane. If the probability of the initial standard plane being identified as a thalamic cross-section is 0.6, and the probability increases to 0.7 after shifting the standard line along a certain direction and then re-inputting the corresponding standard cross-section into the standard plane detection model, then this shift direction can be determined to be favorable for approaching the target standard plane. The standard line is continuously shifted in this direction until the first detection result output by inputting the corresponding new standard plane into the standard plane detection model meets the preset conditions. The shift operation is easy to implement and less prone to errors.

[0072] Optionally, adjusting the pose of the initial standard plane may also include performing a rotation operation. For example, the initial standard plane may be rotated about a line perpendicular to the midsagittal plane as an axis.

[0073] Alternatively, translation and rotation operations can be performed on the initial standard plane simultaneously, which can further improve the accuracy of the final standard plane.

[0074] Preset conditions can include the probability reaching a set probability threshold. For example, if the probability of the target's first type of standard surface in the first detection result is greater than 0.95, the preset condition can be considered met. Preset conditions can also include the number of pose adjustments to the standard line exceeding a set iteration threshold. For example, if the maximum number of iterations is set to 10, the adjustment ends when the number of pose adjustments (i.e., the number of iterations) reaches 10, regardless of whether the probability reaches the probability threshold. When the preset condition is met, the standard surface corresponding to the adjusted standard line at this point is determined as the final standard surface.

[0075] In the above embodiments, the standard plane corresponding to the adjusted standard line is directly used as the final standard plane. Alternatively, operable controls can be provided to the user to adjust the standard plane again in response to the user's manual adjustment operation on the standard plane corresponding to the adjusted standard line. The standard plane obtained by the manual adjustment operation can be used as the final standard plane. The operable controls may include controls for translation and rotation operations, respectively.

[0076] Step S160: Based on the new standard plane, determine the tissue structure measurement items of the fetal brain.

[0077] For example, after determining a new standard plane, some important tissue structures can be extracted from the new standard plane using methods such as image segmentation. Then, the structural parameters of the corresponding tissue structures can be obtained using methods such as automatic measurement. These structural parameters include, for example, the length and circumference of the tissue structure in a certain direction, specifically including, for example, the occipitofrontal diameter, biparietal diameter, and head circumference. The obtained structural parameters are used as data for the tissue structure measurement of the fetal brain.

[0078] According to the above technical solution, a standard line can be determined first, followed by the initial standard plane corresponding to the standard line. Then, a new standard plane is obtained by automatically adjusting the pose of the standard line based on the first detection result. The tissue structure measurements of the fetal brain can then be determined on this new standard plane. This significantly reduces the requirements for operators, shortens the scanning time of ultrasound equipment, and effectively improves the accuracy and efficiency of measurement. It plays a positive role in ensuring the healthy development of the fetus and the safety of mother and child, while also promoting the development of prenatal ultrasound examination technology towards intelligence and precision.

[0079] For example, the median sagittal plane is determined by the following steps S111 and S112.

[0080] Step S111: Based on the three-dimensional ultrasound data of the fetal brain, determine the three-dimensional midline data of the fetal brain.

[0081] For example, three-dimensional ultrasound data of the fetal brain can be acquired using an ultrasound probe, which can include, for example, a two-dimensional probe, a volumetric probe, or an area array probe. A two-dimensional probe acquires two-dimensional image information by scanning line by line, and then combines multiple two-dimensional images to construct three-dimensional ultrasound data. Two-dimensional probes are relatively inexpensive, but their acquisition efficiency and spatial information accuracy are inferior. Volumetric probes can directly acquire ultrasound data in three-dimensional space, offering high acquisition efficiency and accurate spatial information, and can quickly acquire complete volumetric data of the fetal brain. Area array probes can acquire clearer and smoother ultrasound images, making them particularly suitable for scenarios requiring high image quality. In practice, the appropriate probe can be selected based on the specific situation. For example, when fetal brain malformation is suspected and it is necessary to observe the complex internal structures of the brain, an area array probe is preferred. For scenarios requiring high acquisition speed, such as when the pregnant woman experiences frequent fetal movements and needs to acquire data quickly, a volumetric probe is preferred. In practice, for example, during an ultrasound examination, the ultrasound probe is placed in a suitable position on the pregnant woman's abdomen to scan the fetal brain, thereby acquiring three-dimensional ultrasound data of the fetal brain. Three-dimensional ultrasound data contains structural information about the fetal brain in three-dimensional space and can be presented in the form of three-dimensional images.

[0082] For example, a trained semantic segmentation network can be used to extract three-dimensional brain midline data from three-dimensional ultrasound data of the fetal brain. The three-dimensional brain midline data can be represented as multiple three-dimensional brain midlines. Each of these three-dimensional brain midlines specifically comprises multiple discrete pixels.

[0083] Step S112: Perform planar fitting on the three-dimensional brain midline data to determine the midsagittal plane of the fetal brain.

[0084] For example, the midsagittal plane appears as a brighter section than the surrounding area in a 3D image of the fetal brain presented using 3D ultrasound data. The midline of the brain can be considered as the intersection of the midsagittal plane with other sections; therefore, the set of midlines constitutes the midsagittal plane. Thus, after acquiring 3D midline data, the corresponding plane equation can be determined using plane fitting methods such as least squares, thereby determining the midsagittal plane of the fetal brain based on this plane equation.

[0085] By determining three-dimensional brain midline data and performing planar fitting on this data, the midsagittal plane containing the brain midline is determined. This not only accelerates the determination of the midsagittal plane but also ensures its accuracy. Consequently, it guarantees the accuracy of tissue structure measurements.

[0086] For example, step S111, which determines the three-dimensional midline data of the fetal brain based on three-dimensional ultrasound data of the fetal brain, may include:

[0087] Step S111a: Input the three-dimensional ultrasound data into the first detection model to obtain three-dimensional brain midline data, wherein the first detection model includes interconnected encoders and decoders.

[0088] During the initial model training, three-dimensional ultrasound training data of the fetal brain can be collected first. This data includes ultrasound data from different fetuses at different gestational weeks to improve the generalization ability of subsequent model training. After collecting the training data, noise in the data can be removed, as this noise may interfere with model training and lead to inaccurate three-dimensional brain midline data extracted by the model. To reduce interference factors such as speckle noise in the training data, denoising processing can be performed. Denoising methods can include Gaussian filtering, median filtering, etc. Appropriate denoising methods can be selected based on the noise characteristics of the training data, or multiple denoising methods can be combined to achieve better denoising results. Professional medical image processing software can be used to finely annotate the collected three-dimensional ultrasound data, i.e., the brain midline data in the training data. The annotation results will serve as supervisory information for subsequent model training, guiding the model to learn the features of the brain midline data. To increase the diversity and robustness of the training data, data augmentation techniques, including rotation, translation, scaling, and flipping, can be applied. These transformations of the original training data generate a large amount of training data with different angles, positions, and sizes, enabling the model to learn the characteristics of brain midline data under various conditions and improving its adaptability to different 3D ultrasound data. Furthermore, noise simulation can be performed, adding different types and intensities of noise to the original training data to further enhance the model's robustness to noise. During data augmentation, the characteristics of the brain midline data are preserved to avoid information loss due to over-augmentation. After model training is complete, it can be used for 3D brain midline data extraction.

[0089] For example, the first detection model may be a trained semantic segmentation network. In one embodiment, the semantic segmentation network may include interconnected encoders and decoders.

[0090] First, the 3D ultrasound data is input into the encoder. The encoder may include an input layer, a convolutional layer, and a downsampling layer connected in sequence. The input layer of the encoder can perform preprocessing on the 3D image presented by the 3D ultrasound data of the fetal brain, such as standardization and cropping. Standardization normalizes the pixel values ​​of the image to a specific range, for example, normalizing the pixel values ​​to the range [0, 1]. Standardization ensures the consistency of the input data, giving data from different sources a uniform scale and distribution characteristics. Cropping removes redundant parts of the data that are unrelated to the brain midline, focusing on key areas for efficient subsequent processing and accurate analysis.

[0091] An encoder's convolutional layers can include multiple convolutional blocks. Each convolutional block can contain two or more three-dimensional convolutional layers. Three-dimensional convolutional layers can perform convolution operations on the input data in three-dimensional space, extracting features from the data in different directions. Through three-dimensional convolution operations, feature information at different scales and in different directions can be captured from the data, laying a foundation for subsequent midline brain recognition.

[0092] The downsampling layer of the encoder, also known as the pooling layer, progressively reduces the resolution of the feature map while significantly increasing the receptive field. Pooling methods include max pooling and average pooling. Max pooling takes the maximum value of pixels in the neighborhood as the pooling result, highlighting important features in the image. Average pooling calculates the average value of pixels in the neighborhood, smoothing the image and reducing the impact of noise. In practical applications, an appropriate pooling method can be selected based on the model's needs, or max pooling and average pooling can be used alternately. The downsampling layer summarizes local region features through pooling operations, reducing the amount of data and computational complexity, while retaining the main feature information in the image, enabling the model to grasp key image features at a more macroscopic level.

[0093] The encoder inputs a downsized feature map into the decoder. The decoder can consist of an upsampling layer, a convolutional layer, and an output layer connected in sequence. The upsampling layer progressively restores the resolution of the feature map, making it the same size as the original input image. Upsampling methods include deconvolution and nearest neighbor interpolation.

[0094] Similar to the encoder, the decoder's convolutional layers can also contain multiple convolutional blocks for further feature extraction and refinement. In this process, the model performs deeper and more detailed feature extraction and fusion on the restored resolution feature map, comprehensively considering feature information at different levels, and continuously optimizing the accuracy of midline segmentation.

[0095] The output layer of the decoder can be a convolutional layer, used to convert the feature map output by the decoder's convolutional layer into the final segmentation result. The number of output channels of this convolutional layer can be the same as the number of segmentation categories, with each channel outputting a probability map for one category. This probability distribution clearly shows the likelihood of each pixel belonging to a different category. Segmentation categories can include the brain midline and non-brain midline. Optionally, after the output layer, a normalized exponential function (Softmax) or a sigmoid function can be used to convert the probability map into the final segmentation label map. The Softmax function is suitable for multi-class classification problems, normalizing the output probability values ​​so that the sum of the probabilities of each category is always 1, thus accurately obtaining the probability distribution of each pixel belonging to different categories, providing a clear basis for accurate brain midline segmentation. The Sigmoid function is suitable for binary classification problems, mapping the output values ​​to between 0 and 1, determining the pixel category by setting a reasonable threshold, effectively distinguishing between the brain midline and non-brain midline.

[0096] Therefore, using the first detection model to identify the brain midline in three-dimensional space can capture more spatial and contextual information when processing three-dimensional ultrasound data, thereby improving the accuracy and robustness of segmentation, i.e., obtaining more accurate three-dimensional brain midline data. This, in turn, ensures the accuracy of measurements of fetal brain tissue structure.

[0097] Alternatively, the feature maps of the encoder and the decoder, which have the same resolution, can be concatenated using skip connections and then fused through a convolutional layer. Skip connections allow feature information from different levels in the encoder to be passed to the decoder, providing the decoder with more comprehensive contextual information, effectively improving segmentation accuracy, and avoiding information loss and misjudgment.

[0098] Optionally, attention modules can be introduced after the convolutional layers of the encoder and decoder, respectively. These attention modules can include Squeeze-Excitation (SE) modules, Convolutional Block Attention (CBAM) modules, etc. The SE module obtains global information about the feature map by performing global average pooling, then calculates the weights for each channel through two fully connected layers, and finally weights the feature map. The CBAM module calculates attention weights simultaneously in both the channel and spatial dimensions, enabling it to focus on feature regions more comprehensively. Therefore, by introducing an attention mechanism, the first detection model can focus more on important feature regions, improving segmentation accuracy.

[0099] Optionally, a multi-scale feature pyramid can be constructed, fusing feature maps of different scales through upsampling and downsampling operations to capture richer contextual information. Feature maps of different scales contain different levels of information. Small-scale feature maps have high resolution and can present clearer details, helping to accurately depict the subtle contours of the brain's midline; large-scale feature maps have a large receptive field and can provide global information, helping the first detection model determine the position and direction of the brain's midline within the overall cranial structure. Through multi-scale feature fusion, the model can fully utilize this rich information, significantly improving the accuracy and comprehensiveness of segmentation.

[0100] Optionally, residual connections can be introduced into the convolutional blocks of both the encoder and decoder convolutional layers, allowing the input to be directly added to the output via an identity mapping. As network depth increases, gradients gradually vanish during backpropagation, making model training and convergence difficult. Introducing residual connections effectively alleviates the vanishing gradient problem in deep networks, enabling gradients to propagate more smoothly backpropagate, accelerating model convergence, and improving the model's generalization ability and performance stability.

[0101] Optionally, auxiliary loss functions can be added to multiple intermediate layers in the encoder and decoder. Deep supervision guides the model to progressively optimize feature representations during training. These intermediate layers are those other than the input and output layers. Deep supervision allows the model to focus not only on the final output during training but also on the feature representations of the intermediate layers, prompting the model to learn more deeply the intrinsic features and patterns of the data. This significantly improves the accuracy and robustness of segmentation, making the model more adaptable to complex and varied ultrasound images.

[0102] For example, the three-dimensional ultrasound data may include multiple two-dimensional image data. In this embodiment, step S111, based on the three-dimensional ultrasound data of the fetal brain, determines the three-dimensional brain midline data of the fetal brain, which may include steps S111b and S111c.

[0103] Step S111b: Input each two-dimensional image data into the second detection model to obtain the two-dimensional brain midline data in each two-dimensional image data.

[0104] The second detection model can also be a semantic segmentation network. Similar to the first detection model, it can also adopt an interconnected encoder-decoder network structure. When processing two-dimensional image data, the input image first undergoes convolution and downsampling operations in the encoder part to extract features at different levels. Then, in the decoder part, upsampling and convolution operations are used to restore the image resolution, and features can be refined by combining the encoder features from skip connections, finally outputting two-dimensional brain midline data.

[0105] Step S111c: Based on the sequence relationship of multiple two-dimensional image data, merge all two-dimensional brain midline data into three-dimensional brain midline data.

[0106] For example, after obtaining the two-dimensional brain midline data from each two-dimensional image data, these two-dimensional brain midline data are merged into three-dimensional brain midline data according to the sequence relationship of multiple two-dimensional image data, such as their spatial position order.

[0107] Specifically, based on image coordinate information, corresponding two-dimensional brain midline data can be stitched together in three-dimensional space to obtain complete and accurate three-dimensional brain midline data. Assume the three-dimensional ultrasound data includes 100 two-dimensional images. Each two-dimensional image data is processed by a second detection model to obtain corresponding two-dimensional brain midline data. Based on the sequential positional relationship of these two-dimensional image data in three-dimensional space, these two-dimensional brain midline data can be stitched together to form three-dimensional brain midline data.

[0108] Therefore, when three-dimensional ultrasound data includes multiple two-dimensional image data, brain midline identification can be performed in two-dimensional space using a two-dimensional network model, and the identified two-dimensional brain midline data can be merged into three-dimensional brain midline data. The above method has simpler data processing and lower computational cost.

[0109] Alternatively, three-dimensional brain midline data can also be extracted manually. In one specific embodiment, firstly, by manually operating, for example, rotating the ultrasound probe, a cross-section of the brain midline that can be clearly displayed on the ultrasound device's screen is located. Then, using this cross-section as a reference, a series of parallel brain midline cross-sections are extracted along a specific direction. The brain midline is manually marked on some or all of these cross-sections, using methods such as rectangles, tracing, or erasing. Specifically, the rectangle tool can be used to select the brain midline; the tracing tool can be used to trace along the brain midline outline; and the eraser tool can be used to remove unwanted portions, retaining only the brain midline itself. Using the above manual methods, three-dimensional brain midline data can also be accurately determined.

[0110] For example, step S112, which involves performing planar fitting on the three-dimensional brain midline data to determine the midsagittal plane of the fetal brain, may include steps S112a to S112c.

[0111] Step S112a: Perform connected component judgment on the three-dimensional brain midline data, and determine the largest connected component from all connected components as the brain midline region.

[0112] In three-dimensional space, the three-dimensional brain midline data determined in step S111 can include multiple connected regions, each corresponding to a different tissue structure in the fetal brain. In the three-dimensional brain midline data, the data corresponding to the brain midline region of the fetal brain differs from the other data. Connected region analysis can be performed on the three-dimensional brain midline data to obtain all connected regions. The size, shape, and other features of each connected region can be further analyzed. The largest connected region can be determined from all connected regions and used as the brain midline region.

[0113] Step S112b: Perform plane fitting based on the brain midline region to determine the fitted plane.

[0114] For example, plane fitting methods such as the least squares method can be used to fit the plane equation corresponding to the midline region of the brain, thereby determining the corresponding plane. Assume there is a series of point coordinates (x1, y1, z1), (x2, y2, z2), ..., (x... n y n , z n Based on the least squares method, by constructing an objective function that minimizes the sum of squared distances from these points to the fitted plane, the parameters of the plane equation corresponding to the plane are determined.

[0115] Step S112c: Adjust the fitted plane to the vertical direction to obtain the midsagittal plane.

[0116] For example, an adjustment matrix can be calculated to align the fitted plane vertically, rather than tilted. This vertical direction is understood to be the direction of gravity. The adjustment matrix can be used to adjust the fitted plane to obtain the midsagittal plane.

[0117] In the above technical solution, the midline region of the brain is determined by connecting regions, improving the accuracy of the midline region with relatively low computational cost. Combined with this, the midsagittal plane is obtained by performing planar fitting and orientation adjustment on the midline region, making the determined midsagittal plane more accurate. This ensures the accuracy of the measured tissue structure measurements.

[0118] For example, step S120, which determines the key points of tissue structures in the midsagittal plane of the fetal brain, may include steps S121 to S123.

[0119] Step S121: For the cavum septum pellucidum and cerebellum in the fetal brain, perform image segmentation on the midsagittal plane to obtain the segmentation results.

[0120] For example, segmentation methods such as deep learning models like Unet, SegNet, and DeepLab, or machine learning models like SVM, can be used to segment the midsagittal plane. Taking the Unet model as an example, when segmenting the midsagittal plane image, the input image undergoes a series of convolution and pooling operations to extract features. Then, features are fused through upsampling and skip connections with the corresponding layers of the encoder, and finally, the segmentation result is output, separating the septum pellucidum and cerebellum from the midsagittal plane image.

[0121] Step S122: Determine the two endpoints of the cavum septum pellucidum and the centroid of the cerebellum.

[0122] Figure 2 A schematic diagram of a median sagittal plane image according to one embodiment of this application is shown. Figure 2 As shown, the septum pellucidum generally appears as a strip in the midsagittal plane. For example, it can be thinned by gradually reducing the area occupied by the septum pellucidum in the midsagittal plane from its edge inwards using thinning algorithms and edge detection algorithms. After thinning the septum pellucidum to approximately a line, the two endpoints of this line can be determined as the two endpoints of the septum pellucidum, i.e. Figure 2 Endpoints A and B are shown in the circle. Alternatively, the center point of the area can be determined first, and then the two endpoints can be determined based on the distance between the points on the edge of the area and the center point.

[0123] For example, the centroid of the cerebellum can be calculated by weighted averaging the coordinates of pixels within the cerebellar region. Figure 2 End point C in the circle shown.

[0124] Step S123: Calculate the distance between the centroid of the cerebellum and the two endpoints of the cavum septum pellucidum to determine the distal end and proximal end of the cavum septum pellucidum, respectively; wherein, the key points of the tissue structure include the distal end of the cavum septum pellucidum, and the key points of the tissue structure also include the centroid of the cerebellum and / or the proximal end of the cavum septum pellucidum.

[0125] For example, the distances between the centroid of the cerebellum and the two endpoints of the cavum septum pellucidum are calculated using a distance formula. The two calculated distances are compared; the endpoint corresponding to the larger distance is called the distal end of the cavum septum pellucidum, and the endpoint corresponding to the smaller distance is called the proximal end. Figure 2 In the illustrated embodiment, endpoint A is referred to as the distal end of the transparent cavity, and endpoint B is referred to as the proximal end of the transparent cavity.

[0126] The above method can accurately determine the key points of the fetal brain's tissue structure, laying a solid foundation for subsequent standard surface determination.

[0127] Alternatively, the orientation of the median sagittal plane can be determined by the distribution of key points in the tissue structure. For example, when the centroid of the cerebellum is located above the proximal end of the cavum septum pellucidum, the median sagittal plane is determined to be reversed. In this case, the median sagittal plane can be flipped up and down to orthogonalize it.

[0128] For example, step S130, which determines the standard line based on key points of the organizational structure, may include step S131.

[0129] Step S131: The line connecting the first point and the distal end of the septum pellucidum is determined as the standard line, wherein the first point is the point on the line connecting the centroid of the cerebellum and the proximal end of the septum pellucidum.

[0130] For example, the selection of the first point can differ for different types of standard planes. When the first type of standard plane includes a thalamic cross section and / or a lateral ventricle cross section, the first point can be the proximal end of the cavum septum pellucidum. For example, let the coordinates of the proximal end of the cavum septum pellucidum be (x... p y p The coordinates of the distal end of the transparent cavity are (x d y d ), will (x p y p ) and (x d y dThe line connecting the two points is defined as the standard line. In this case, the key points of the tissue structure include the distal end and the proximal end of the cavum septum pellucidum. When the first type of standard plane includes a cerebellar cross section, the first point can be the centroid of the cerebellum. Let the coordinates of the centroid of the cerebellum be (x... c y c The coordinates of the distal end of the transparent cavity are (x d y d ), will (x c y c ) and (x d y d The lines connecting the points are used as the standard lines. In this case, the key points of the tissue structure include the distal end of the septum pellucidum and the centroid of the cerebellum. In this way, points on the connecting lines can be flexibly selected to determine the standard lines according to the needs of different types of standard planes, which improves the versatility and adaptability of the method, reduces the number of pose adjustments in step S150, and obtains the final standard plane more quickly, thereby improving the efficiency of determining tissue structure measurement items.

[0131] As mentioned earlier, when the first type of standard plane is a transverse section of the thalamus or the lateral ventricle, the first point can be the proximal end of the cavum septum pellucidum. When the first type of standard plane is a transverse section of the cerebellum, the first point can be the centroid of the cerebellum. The initial standard plane determined based on this first point is closer to the true first type of standard plane, in other words, closer to the final standard plane. Therefore, the number of pose adjustments in step S150 can be significantly reduced, thereby reducing the computational load and increasing the speed of determining tissue structure measurements.

[0132] According to different embodiments of the present invention, in addition to the proximal end of the septum pellucidum and the centroid of the cerebellum as described in the above examples, other points on the line connecting the centroid of the cerebellum and the proximal end of the septum pellucidum can also be used as first points. For example, any point on this line can be used as the first point to determine the initial standard plane. Finally, the final standard plane is determined by the pose adjustment in the subsequent step S150.

[0133] By selecting a first point on the line connecting the centroid of the cerebellum and the proximal end of the cavum septum pellucidum using the above method, and combining this with the distal end of the cavum septum pellucidum to form a standard line, an initial standard plane corresponding to a single standard line is uniquely determined based on the perpendicular relationship between the standard plane and the midsagittal plane. This provides a reliable guarantee for determining the final standard plane and the measurement items on the final standard plane in the subsequent steps. The initial standard plane determined above is close to the final standard plane, which can reduce the number of pose adjustments in step S150 to some extent, thereby reducing the amount of calculation and increasing the speed of determining tissue structure measurement items.

[0134] For example, the first type of standard plane may include a thalamic cross section. Step S160, based on the new standard plane, determines the tissue structure measurements of the fetal brain, which may include steps S161 and S162.

[0135] Step S161: For the thalamus in the fetal brain, perform image segmentation on a new standard plane of the thalamus cross section to determine the thalamic region in the thalamus cross section.

[0136] For example, semantic segmentation networks such as Unet, SegNet, and DeepLab can be used to segment the new standard plane of the thalamus cross section. Taking the Unet model as an example, when segmenting the final standard plane image of the thalamus cross section, the input image undergoes a series of convolution and pooling operations to extract features. Then, features are fused through upsampling and skip connections with the corresponding layers of the encoder, and finally, the segmentation result is output, separating the thalamus region from the final standard plane image of the thalamus cross section. Figure 3 A schematic diagram showing an image of a thalamic region in a cross-section of the thalamus according to an embodiment of this application is illustrated. Figure 3 As shown, the white, oval-shaped region is the thalamus region.

[0137] Optionally, after segmenting the thalamic region, image processing methods such as connected component judgment, threshold analysis, and morphological analysis can be used to further optimize and adjust the edges of the thalamic region.

[0138] Step S162: Perform ellipse fitting on the thalamic region and calculate the major axis, minor axis and circumference of the fitted ellipse to represent the occipitofrontal diameter, biparietal diameter and head circumference, respectively. The tissue structure measurement items include the occipitofrontal diameter, biparietal diameter and head circumference.

[0139] After segmentation in step S161, ellipse fitting of the thalamic region can be performed using methods such as least squares. For example, let the pixel coordinates in the thalamic region be (x... i y i For each point i = 1, 2, ..., n, the least squares method constructs an objective function that minimizes the sum of squared distances from these points to the fitted ellipse, thus determining the ellipse's parameters. The image lengths of the major axis, minor axis, and circumference of the fitted ellipse are calculated, and the actual physical lengths and corresponding gestational ages are calculated based on the corresponding pixel-to-physical ratio. Assuming a pixel-to-physical ratio of 1:0.1 mm and an image length of 100 pixels for the major axis of the fitted ellipse, the actual physical length of the occipital-frontal diameter (OFD) is 100 × 0.1 = 10 mm. The pixel-to-physical ratio can be set appropriately based on image clarity, etc. Similarly, the actual physical lengths of the biparietal diameter (BPD) and head circumference (HC) can be calculated. Furthermore, based on the correspondence between the occipital-frontal diameter, BPD, and head circumference and gestational age, the corresponding actual gestational age can be determined.

[0140] The calculated occipital-frontal diameter, biparietal diameter, and head circumference are more accurate, providing more precise diagnostic data for fetal screening. This helps to detect fetal brain development abnormalities in a timely and accurate manner, reducing the risk of misdiagnosis and missed diagnosis.

[0141] For example, the first type of standard plane may include a cerebellar cross section. Step S160, based on the new standard plane, determining the tissue structure measurements of the fetal brain may further include steps S163 to S165.

[0142] Step S163: For the cerebellum in the fetal brain, perform image segmentation on the new standard plane of the cerebellar cross section to determine the cerebellum and cerebellomedullary cisterns in the cerebellar cross section. Figure 4 A schematic diagram of a cerebellar region image in a cross-section of the cerebellum according to an embodiment of this application is shown. Figure 5 A schematic diagram of the cerebellomedullary cistern region in a cross-section of the cerebellum according to an embodiment of this application is shown. Figure 4 and Figure 5 As shown in the figure, the white areas represent the cerebellum and the cerebellomedullary cistern, respectively.

[0143] Step S163, which determines the cerebellum and cerebellomedullary cistern in the cerebellar cross section, is similar to step S161, which determines the thalamic region, and will not be repeated here.

[0144] Step S164: Determine the bounding box of the cerebellum and the bounding box of the cerebellomedullary cistern.

[0145] For example, after determining the regions occupied by the cerebellum and the cerebellomedullary cistern in the image, the bounding boxes of the cerebellum and the cerebellomedullary cistern can be calculated by traversing the coordinates of pixels in the cerebellar and cerebellomedullary cistern regions respectively. Taking the cerebellum as an example, the coordinate range of the bounding box is determined based on the minimum and maximum coordinate values ​​in the horizontal and vertical directions of the cerebellar region, respectively. The maximum coordinate value (X... max Y max The coordinates corresponding to ) are taken as the first vertex of the bounding box, and the minimum coordinate (X) is used as the first vertex. min Y min The coordinates corresponding to the first vertex are taken as the second vertex of the bounding box, where the bounding box is a rectangle and the first and second vertices are diagonal vertices.

[0146] Step S165: Determine the actual physical length of the transverse diameter of the cerebellum based on the bounding box of the cerebellum; determine the actual physical length of the cerebellomedullary cistern based on the bounding box of the cerebellomedullary cistern.

[0147] For example, the length of the row of pixels with the largest number of pixels belonging to the cerebellum in the bounding box of the cerebellum is determined as the image length of the transverse diameter of the cerebellum, and the length of the row of pixels with the largest number of pixels belonging to the cerebellomedullary cistern in the bounding box of the cerebellomedullary cistern is determined as the image length of the cerebellomedullary cistern.

[0148] It should be noted that the bounding boxes of the cerebellum and the cerebellomedullary cistern may contain pixels that do not belong to the cerebellum or cerebellomedullary cistern regions. The grayscale values ​​of these pixels differ significantly from those of the pixels in the cerebellum and cerebellomedullary cistern regions. The length of the row of pixels in the bounding box that contains the most pixels belonging to the cerebellum can be determined as the image length of the transverse diameter of the cerebellum (TCD). For example, in the bounding box of the cerebellum, the number of pixels belonging to the cerebellum in each row is counted, and the row with the most pixels is found. The total length of all pixels belonging to the cerebellum in that row is the image length of the transverse diameter of the cerebellum. Here, the "row" mentioned above can be either a row or a column of the image; in other words, the row of pixels can be either a row of pixels or a column of pixels, depending on the number of pixels. The method for determining the image length of the cerebellomedullary cistern is similar to that described above and will not be repeated here.

[0149] The actual physical length of the cerebellar transverse diameter can be calculated based on the image length and pixel physical ratio of the cerebellar transverse diameter. For example, assuming a pixel physical ratio of 1:0.1mm and an image length of 50 pixels for the cerebellar transverse diameter, the actual physical length of the cerebellar transverse diameter is 50 × 0.1 = 5mm.

[0150] The actual physical length of the cerebellomedullary cistern can be calculated based on the image length and pixel physical ratio. Calculating the actual physical length of the cerebellomedullary cistern is similar to calculating the actual physical length of the cerebellar transverse diameter in step S166, and will not be repeated here for simplicity.

[0151] It is understood that in the above embodiments, the tissue structure measurements include the actual physical length of the cerebellar transverse diameter and the actual physical length of the cerebellomedullary cistern. The actual physical lengths of the cerebellar transverse diameter and the cerebellomedullary cistern calculated by the above scheme are more accurate, providing more precise diagnostic data for fetal screening, helping to detect fetal brain development abnormalities in a timely and accurate manner, and reducing the risk of misdiagnosis and missed diagnosis.

[0152] For example, the first type of standard plane may include a transverse section of the lateral ventricle. Step S160, based on the new standard plane, determining the tissue structure measurements of the fetal brain may further include steps S166 to S168.

[0153] Step S166: For the lateral ventricle in the fetal brain, perform image segmentation on the new standard plane of the transverse section of the lateral ventricle to obtain the lateral ventricle region in the transverse section of the lateral ventricle. Figure 6 A schematic diagram showing an image of the lateral ventricle region in a cross-section according to an embodiment of this application is illustrated. Figure 6 As shown in the figure, the white area is the lateral ventricle region.

[0154] Step S167: Determine the bounding box of the lateral ventricle region.

[0155] Step S168: Determine the actual physical width of the lateral ventricle region based on the bounding box of the lateral ventricle region.

[0156] For example, the length corresponding to the row of pixels with the largest number of pixels belonging to the lateral ventricle region within the bounding box of the lateral ventricle region is determined as the image width of the lateral ventricle region. Based on the image width of the lateral ventricle region and the pixel physical ratio, the actual physical width of the lateral ventricle region is calculated. The fetal cranial structure measurement includes the actual physical length of the width of the lateral ventricle region.

[0157] Steps S166 to S168 are similar to the implementation methods of steps S163 to S165 described above, and will not be repeated here for the sake of brevity.

[0158] It should be noted that the above steps S161 to S162, S163 to S165 and S166 to S168 belong to the measurement item determination steps under different standard surface types. There is no order in which they are executed, and they are only related to the standard surface type being determined.

[0159] According to another aspect of the present invention, an ultrasonic device is also provided. Figure 7 A schematic block diagram of an ultrasonic device 700 according to one embodiment of this application is shown. Figure 7 As shown, the ultrasound device 700 may include a processor 710 and a memory 720, wherein the memory 720 stores computer program instructions, which are executed by the processor 710 to perform the fetal cranial ultrasound image processing method 100 as described above.

[0160] According to another aspect of the present invention, a storage medium is also provided, on which program instructions are stored, which, when executed, are used to perform the fetal cranial ultrasound image processing method as described above. The storage medium may, for example, include a storage component of a tablet computer, a hard disk of a personal computer, a read-only memory (ROM), an erasable programmable read-only memory (EPROM), a portable compact disc read-only memory (CD-ROM), a USB memory, or any combination of the above storage media. The computer-readable storage medium may be any combination of one or more computer-readable storage media.

[0161] According to another aspect of the present invention, a computer program product is also provided, comprising computer program instructions which, when executed, are used to perform the fetal cranial ultrasound image processing method as described above.

[0162] Those skilled in the art can understand the specific implementation schemes and beneficial effects of the above-mentioned ultrasound equipment, storage medium, and computer program products by reading the relevant descriptions of the fetal cranial ultrasound image processing methods. For the sake of brevity, they will not be described in detail here.

[0163] Although exemplary embodiments have been described herein with reference to the accompanying drawings, it should be understood that the above exemplary embodiments are merely illustrative and are not intended to limit the scope of the invention. Various changes and modifications can be made therein by those skilled in the art without departing from the scope and spirit of the invention. All such changes and modifications are intended to be included within the scope of the invention as claimed in the appended claims.

[0164] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0165] In the several embodiments provided in this application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another device, or some features may be ignored or not executed.

[0166] Numerous specific details are set forth in the specification provided herein. However, it will be understood that embodiments of the invention may be practiced without these specific details. In some instances, well-known methods, structures, and techniques have not been shown in detail so as not to obscure the understanding of this specification.

[0167] Similarly, it should be understood that, in order to streamline the invention and aid in understanding one or more of the various aspects of the invention, features of the invention are sometimes grouped together in a single embodiment, figure, or description thereof in the description of exemplary embodiments of the invention. However, this approach should not be construed as reflecting an intention that the claimed invention requires more features than are expressly recited in each claim. Rather, as reflected in the corresponding claims, its inventive point lies in solving the corresponding technical problem with fewer features than all of those in a single disclosed embodiment. Therefore, the claims following the detailed description are hereby expressly incorporated into that detailed description, wherein each claim itself is a separate embodiment of the invention.

[0168] Those skilled in the art will understand that, apart from the mutual exclusion of features, all features disclosed in this specification (including the accompanying claims, abstract, and drawings) and all processes or units of any method or apparatus so disclosed can be combined in any combination. Unless otherwise expressly stated, each feature disclosed in this specification (including the accompanying claims, abstract, and drawings) may be replaced by an alternative feature that serves the same, equivalent, or similar purpose.

[0169] Furthermore, those skilled in the art will understand that although some embodiments described herein include certain features but not others included in other embodiments, combinations of features from different embodiments are intended to be within the scope of the invention and form different embodiments. For example, in the claims, any of the claimed embodiments can be used in any combination.

[0170] The various component embodiments of the present invention can be implemented in hardware, or as software modules running on one or more processors, or a combination thereof. Those skilled in the art will understand that microprocessors or digital signal processors (DSPs) can be used in practice to implement some or all of the steps of the method for determining the standard planes and measurement items of the fetal cranium according to embodiments of the present invention. The present invention can also be implemented as a program (e.g., a computer program and computer program product) for performing part or all of the methods described herein. Such programs implementing the present invention can be stored on a computer-readable medium or can be in the form of one or more signals. Such signals can be downloaded from an Internet website, provided on a carrier signal, or provided in any other form.

[0171] It should be noted that the above embodiments are illustrative of the invention and not restrictive, and that those skilled in the art can devise alternative embodiments without departing from the scope of the appended claims. In the claims, any reference signs placed between parentheses should not be construed as limiting the claims. The word "comprising" does not exclude the presence of elements or steps not listed in the claims. The word "a" or "an" preceding an element does not exclude the presence of a plurality of such elements. The invention can be implemented by means of hardware comprising several different elements and by means of a suitably programmed computer. The use of the words first, second, and third, etc., does not indicate any order. These words can be interpreted as names.

[0172] The above description is merely a specific embodiment of the present invention or an explanation of that embodiment. The scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. The scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for processing fetal cranial ultrasound images, characterized in that, include: To identify key structural points of the fetal brain in the midsagittal plane; Based on the key points of the tissue structure, a standard line is determined to determine an initial standard plane, wherein the initial standard plane is a plane that passes through the standard line and is perpendicular to the median sagittal plane; The initial standard surface is input into the standard surface detection model to obtain a first detection result, wherein the first detection result includes information about the probability that the initial standard surface is a first type of standard surface; Based on the first detection result, the pose of the standard line is adjusted to obtain a new standard surface corresponding to the adjusted standard line. Based on the new standard plane, the tissue structure measurement items of the fetal brain are determined.

2. The method as described in claim 1, characterized in that, The median sagittal plane is determined in the following way: Based on the three-dimensional ultrasound data of the fetal brain, the three-dimensional midline data of the fetal brain were determined. Planar fitting was performed on the three-dimensional brain midline data to determine the midsagittal plane of the fetal brain.

3. The method as described in claim 2, characterized in that, The determination of three-dimensional midline data of the fetal brain based on three-dimensional ultrasound data of the fetal brain includes: The three-dimensional ultrasound data is input into a first detection model to obtain the three-dimensional brain midline data, wherein the first detection model includes interconnected encoders and decoders.

4. The method as described in claim 2, characterized in that, The three-dimensional ultrasound data includes multiple two-dimensional image data. The three-dimensional ultrasound data based on the fetal brain, which determines the three-dimensional midline data of the fetal brain, includes: Each two-dimensional image data is input into the second detection model to obtain the two-dimensional brain midline data in each two-dimensional image data; Based on the sequence relationship of the multiple two-dimensional image data, all two-dimensional brain midline data are merged into the three-dimensional brain midline data.

5. The method as described in claim 2, characterized in that, The step of performing planar fitting on the three-dimensional brain midline data to determine the midsagittal plane of the fetal brain includes: The three-dimensional brain midline data is subjected to connected component determination, and the largest connected component is determined from all connected components to serve as the brain midline region. Plane fitting is performed based on the midline region of the brain to determine the fitted plane; The fitted plane is adjusted to be vertical to obtain the median sagittal plane.

6. The method as described in claim 1, characterized in that, The determination of key points of tissue structure in the midsagittal plane of the fetal brain includes: For the cavum septum pellucidum and cerebellum in the fetal brain, image segmentation is performed on the midsagittal plane to obtain segmentation results; Determine the two endpoints of the septum pellucidum and the centroid of the cerebellum; Calculate the distances between the centroid of the cerebellum and the two endpoints of the cavum septum pellucidum to determine the distal and proximal ends of the cavum septum pellucidum, respectively; wherein the key points of the tissue structure include the distal end of the cavum septum pellucidum, and the key points of the tissue structure also include the centroid of the cerebellum and / or the proximal end of the cavum septum pellucidum; The determination of the standard line based on the key points of the organizational structure includes: The line connecting the first point to the distal end of the septum pellucidum is defined as the standard line, wherein the first point is a point on the line connecting the centroid of the cerebellum and the proximal end of the septum pellucidum.

7. The method as described in claim 6, characterized in that, The first type of standard plane includes a thalamic cross section and / or a lateral ventricle cross section, wherein for the thalamic cross section and the lateral ventricle cross section, the first point is the proximal end of the cavum septum pellucidum; and / or The first type of standard plane includes a cerebellar cross section, and for the cerebellar cross section, the first point is the centroid of the cerebellum.

8. The method according to any one of claims 1 to 7, characterized in that, The pose adjustment of the standard line includes: After inputting the new standard surface into the standard surface detection model to obtain a new first detection result, the standard line is translated and / or rotated based on the new first detection result.

9. The method as described in claim 1, characterized in that, The first type of standard plane includes a thalamic cross section, and the determination of the tissue structure measurements of the fetal brain based on the new standard plane includes: For the thalamus in the fetal brain, the final standard plane of the thalamus cross section is image segmented to determine the thalamic region in the thalamus cross section; An ellipse is fitted to the thalamic region, and the major axis, minor axis, and circumference of the fitted ellipse are calculated to represent the occipitofrontal diameter, biparietal diameter, and head circumference, respectively. The tissue structure measurements include the occipitofrontal diameter, the biparietal diameter, and the head circumference.

10. The method as described in claim 1, characterized in that, The first type of standard plane includes a cerebellar cross section. The determination of the tissue structure measurements of the fetal brain based on the new standard plane includes: For the cerebellum in the fetal brain, the final standard plane of the cerebellar cross section is image segmented to determine the cerebellum and cerebellomedullary cisterns in the cerebellar cross section; Determine the bounding box of the cerebellum and the bounding box of the cerebellomedullary cistern; Based on the bounding box of the cerebellum, the actual physical length of the transverse diameter of the cerebellum is determined; based on the bounding box of the cerebellomedullary cistern, the actual physical length of the cerebellomedullary cistern is determined. The tissue structure measurement items include the actual physical length of the cerebellar transverse diameter and the actual physical length of the cerebellomedullary cistern.

11. The method as described in claim 1, characterized in that, The first type of standard plane includes a transverse section of the lateral ventricle. The determination of the tissue structure measurements of the fetal brain based on the new standard plane includes: For the lateral ventricle in the fetal brain, the final standard plane of the transverse section of the lateral ventricle is image segmented to obtain the lateral ventricle region in the transverse section of the lateral ventricle; Determine the bounding box of the lateral ventricle region; Based on the bounding box of the lateral ventricle region, the actual physical width of the lateral ventricle region is determined; The tissue structure measurement of the fetal brain includes the actual physical width of the lateral ventricle region.

12. An ultrasonic device, comprising a processor and a memory, wherein, The memory stores computer program instructions, which, when executed by the processor, are used to perform the fetal cranial ultrasound image processing method as described in any one of claims 1 to 11.

13. A storage medium storing program instructions that, when executed, perform the fetal cranial ultrasound image processing method as described in any one of claims 1 to 11.

14. A computer program product comprising computer program instructions, characterized in that, The computer program instructions, when executed, are used to perform the fetal cranial ultrasound image processing method as described in any one of claims 1 to 11.

Images