Ultrasonic imaging system and method for fetal measurement

Abstract

An ultrasound imaging system and method for fetal measurement generates at least two different types of volume data based on ultrasound echo signals. The volume data is three-dimensional data, and the different types of volume data have the same size and one-to-one spatial position correspondence. The at least two different types of volume data include at least two of a first type of volume data, a second type of volume data, and a third type of volume data. Furthermore, based on the respective characteristics of the different types of volume data, corresponding segmentation methods are used to segment the placenta. This solution is not only simple and feasible, saves computing resources, but also provides reliable and accurate segmentation results.

Description

Technical Field

[0001] The present invention relates to an ultrasonic imaging system, and in particular to an ultrasonic imaging system and method for fetal measurement. Background Art

[0002] Prenatal ultrasound examinations are an important tool for assessing fetal growth and development. Currently, fetal examinations are primarily performed using two-dimensional ultrasound images, requiring doctors to manually measure fetal parameters, such as placental parameters and umbilical cord length. This method is subjective, the accuracy of measurement results is easily affected by human factors, and work efficiency is low.

[0003] The development of three-dimensional ultrasound imaging technology has provided a new approach to improving the objectivity and efficiency of prenatal examinations. Using 3D ultrasound data, it is possible to automatically measure parameters related to the placenta and umbilical cord length, providing more objective evidence for assessing fetal development. However, existing 3D ultrasound prenatal examination methods still have certain deficiencies in automated measurement solutions, such as data segmentation, parameter extraction, and diagnosis, making them difficult to meet the needs of actual clinical applications. Summary of the Invention

[0004] In view of the above problems, the present invention provides an ultrasonic imaging system and method for fetal measurement, which are described in detail below.

[0005] According to a first aspect, an embodiment provides an ultrasound imaging system for fetal measurement, comprising:

[0006] An ultrasound probe, configured to transmit ultrasound waves to an area containing the placenta and receive returned ultrasound echoes to obtain ultrasound echo signals;

[0007] a transceiver control circuit, configured to control the ultrasonic probe to transmit ultrasonic waves and receive ultrasonic echoes; and

[0008] Processor for:

[0009] generating at least two different types of volume data based on the ultrasound echo signal, the volume data being three-dimensional data, the different types of volume data having the same size and corresponding spatial positions; the at least two different types of volume data comprising at least two of a first type of volume data, a second type of volume data, and a third type of volume data;

[0010] Based on different types of volume data, a segmentation strategy is selected from a first segmentation strategy and a second segmentation strategy, and the volume data is segmented using a hybrid segmentation network for placenta segmentation to obtain a segmentation result, wherein the hybrid segmentation network for placenta segmentation is used to process the input volume data to segment placenta volume data from the input volume data; wherein:

[0011] determining whether the at least two different types of volume data include the first type of volume data, and if so, selecting a first segmentation strategy and segmenting the volume data using the hybrid segmentation network for placenta segmentation; and if not, selecting a second segmentation strategy and segmenting the volume data using the hybrid segmentation network for placenta segmentation;

[0012] The selecting a first segmentation strategy and segmenting the volume data using the hybrid segmentation network for placenta segmentation includes: inputting the first type of volume data into the hybrid segmentation network for placenta segmentation to segment first type of placenta volume data from the first type of volume data; segmenting other types of volume data based on the spatial position of the first type of placenta volume data in the first type of volume data and the spatial position correspondence between the different types of volume data to obtain other types of placenta volume data, wherein the other types of volume data refer to volume data that is not the first type among the at least two different types of volume data; and using each type of placenta volume data obtained as the segmentation result;

[0013] The selecting of the second segmentation strategy and segmenting the volume data using the hybrid segmentation network for placenta segmentation comprises: inputting each type of volume data of the at least two different types into the hybrid segmentation network for placenta segmentation to segment the corresponding type of placenta volume data from each type of volume data; obtaining the spatial position of the placenta volume data obtained based on weighted calculation based on the obtained spatial position of all types of placenta volume data in the respective types of volume data; and segmenting the corresponding type of placenta volume data from each type of volume data based on the weighted calculation spatial position of the placenta volume data, and using the segmentation result as the segmentation result;

[0014] Calculate relevant parameters of the placenta based on the segmentation results.

[0015] In one embodiment, the at least two different types of volume data include the second type of volume data; and the processor is further configured to:

[0016] inputting the second type of volume data into a hybrid segmentation network for umbilical cord segmentation, wherein the hybrid segmentation network for umbilical cord segmentation is used to process the input volume data to segment umbilical cord volume data from the input volume data;

[0017] A skeleton line is extracted from the segmented umbilical cord body data, and a center line is extracted based on the extracted skeleton line; and relevant parameters of the umbilical cord are calculated based on the extracted center line, wherein the relevant parameters of the umbilical cord include the umbilical cord length.

[0018] In one embodiment, the hybrid segmented network includes a primary network, an intermediate layer, and a secondary network;

[0019] The primary network is used to perform a coarse segmentation on a target area in the input volume data and output a probability map with a low resolution relative to the original volume data, wherein the target area is a placenta area or an umbilical cord area; wherein the primary network includes a first encoder, a first decoder and a first skip connection; the first encoder is used to gradually extract features of the volume data and reduce the spatial resolution of the volume data; the first encoder includes a plurality of convolutional layers and pooling layers, the convolutional layers are used to capture features of the volume data in three-dimensional space, and the pooling layers are used to downsample so that the spatial resolution of the volume data gradually becomes smaller; the first decoder is used to gradually restore the spatial resolution of the volume data and perform feature fusion, the first decoder includes a plurality of deconvolutional layers and upsampling layers, the deconvolutional layers perform deconvolution operations, and the upsampling layers are used to perform spatial upsampling; the first skip connection exists between the first encoder and the first decoder, and is used to fuse the outputs of different levels in the first encoder with the inputs of corresponding levels in the first decoder;

[0020] The intermediate layer is used to generate candidate spatial regions using an adaptive threshold algorithm for the probability map, and extract corresponding high-resolution sub-volume data from the original volume data based on the candidate spatial regions;

[0021] The secondary network is used to refine the edge of the target area in the sub-volume data and output the target volume data, the target area and the target volume data are the placenta area and the placenta volume data, respectively, or the umbilical cord area and the umbilical cord volume data, respectively; the secondary network includes a second encoder, a second decoder and a second jump connection, the second encoder includes multiple residual dense blocks, the second decoder includes multiple upper layers, the second jump connection is a jump connection based on the convolutional attention feature fusion module, connecting the second encoder and the second decoder; wherein, the output of the residual dense block of the previous level is used as the input of the residual dense block of the next level, and as the input of the jump connection based on the convolutional attention feature fusion module of this layer, the jump connection based on the convolutional attention feature fusion module is used to learn the importance weight of each channel, and adaptively adjust the weight of the feature channel, and then dynamically adjust the importance of the different spatial positions of the features in the second encoder, and then perform channel-weighted splicing with the features of the second decoder.

[0022] In one embodiment, the convolution layer uses depthwise separable convolution, including depthwise convolution and pointwise convolution. The depthwise convolution performs convolution on each channel separately, and the pointwise convolution is used to combine the output channels of the depthwise convolution.

[0023] In one embodiment, the processor is further configured to: after obtaining the segmentation result, perform post-processing on the segmentation result; the post-processing includes at least one of the following:

[0024] Mark the connected regions in the segmentation results to obtain multiple different connected regions; sort the spatial volumes of the connected regions, and select the connected region with the largest spatial volume as the processed segmentation result;

[0025] The segmentation result is filled using the region growing algorithm, the area outside the placenta is filled, and the placenta and the holes inside the placenta are retained to obtain the processed segmentation result.

[0026] In one embodiment, the first type of volume data is tissue type volume data;

[0027] The second type of volume data is velocity type volume data, and the third type of volume data is energy type volume data; or the second type of volume data is energy type volume data, and the third type of volume data is velocity type volume data.

[0028] In one embodiment, the processor calculates relevant parameters of the placenta based on the segmentation results, including: calculating the volume of the placenta and / or the blood supply parameters of the placenta; the blood supply parameters of the placenta include one or more of the vascularization index, blood flow index and vascularization blood flow index.

[0029] According to a second aspect, an embodiment provides an ultrasound imaging system for fetal measurement, comprising:

[0030] An ultrasound probe, configured to transmit ultrasound waves to an area containing the placenta and receive returned ultrasound echoes to obtain ultrasound echo signals;

[0031] a transceiver control circuit, configured to control the ultrasonic probe to transmit ultrasonic waves and receive ultrasonic echoes; and

[0032] Processor for:

[0033] generating at least two different types of volume data based on the ultrasound echo signal, the volume data being three-dimensional data, the different types of volume data having the same size and corresponding spatial positions; the at least two different types of volume data including first type of volume data, and the at least two different types of volume data further including at least one of second type of volume data and third type of volume data;

[0034] Inputting the first type of volume data into a hybrid segmentation network for placenta segmentation to segment first type of placenta volume data from the first type of volume data; segmenting other types of volume data based on the spatial position of the first type of placenta volume data in the first type of volume data and the spatial position correspondence between the different types of volume data to obtain other types of placenta volume data, where the other types of volume data refer to volume data that is not the first type among the at least two different types of volume data; and using each type of placenta volume data obtained as the segmentation result;

[0035] Calculate relevant parameters of the placenta based on the segmentation results.

[0036] In one embodiment, the processor is further configured to:

[0037] inputting the second type of volume data into a hybrid segmentation network for umbilical cord segmentation, wherein the hybrid segmentation network for umbilical cord segmentation is used to process the input volume data to segment umbilical cord volume data from the input volume data;

[0038] A skeleton line is extracted from the segmented umbilical cord body data, and a center line is extracted based on the extracted skeleton line; and relevant parameters of the umbilical cord are calculated based on the extracted center line, wherein the relevant parameters of the umbilical cord include the umbilical cord length.

[0039] In one embodiment, the hybrid segmented network includes a primary network, an intermediate layer, and a secondary network;

[0040] The primary network is used to perform a coarse segmentation on a target area in the input volume data and output a probability map with a low resolution relative to the original volume data, wherein the target area is a placenta area or an umbilical cord area; wherein the primary network includes a first encoder, a first decoder and a first skip connection; the first encoder is used to gradually extract features of the volume data and reduce the spatial resolution of the volume data; the first encoder includes a plurality of convolutional layers and pooling layers, the convolutional layers are used to capture features of the volume data in three-dimensional space, and the pooling layers are used to downsample so that the spatial resolution of the volume data gradually becomes smaller; the first decoder is used to gradually restore the spatial resolution of the volume data and perform feature fusion, the first decoder includes a plurality of deconvolutional layers and upsampling layers, the deconvolutional layers perform deconvolution operations, and the upsampling layers are used to perform spatial upsampling; the first skip connection exists between the first encoder and the first decoder, and is used to fuse the outputs of different levels in the first encoder with the inputs of corresponding levels in the first decoder;

[0041] The intermediate layer is used to generate candidate spatial regions using an adaptive threshold algorithm for the probability map, and extract corresponding high-resolution sub-volume data from the original volume data based on the candidate spatial regions;

[0042] The secondary network is used to refine the edge of the target area in the sub-volume data and output the target volume data, the target area and the target volume data are the placenta area and the placenta volume data, respectively, or the umbilical cord area and the umbilical cord volume data, respectively; the secondary network includes a second encoder, a second decoder and a second jump connection, the second encoder includes multiple residual dense blocks, the second decoder includes multiple upper layers, the second jump connection is a jump connection based on the convolutional attention feature fusion module, connecting the second encoder and the second decoder; wherein, the output of the residual dense block of the previous level is used as the input of the residual dense block of the next level, and as the input of the jump connection based on the convolutional attention feature fusion module of this layer, the jump connection based on the convolutional attention feature fusion module is used to learn the importance weight of each channel, and adaptively adjust the weight of the feature channel, and then dynamically adjust the importance of the different spatial positions of the features in the second encoder, and then perform channel-weighted splicing with the features of the second decoder.

[0043] In one embodiment, the convolution layer uses depthwise separable convolution, including depthwise convolution and pointwise convolution. The depthwise convolution performs convolution on each channel separately, and the pointwise convolution is used to combine the output channels of the depthwise convolution.

[0044] In one embodiment, the processor is further configured to: after obtaining the segmentation result, perform post-processing on the segmentation result; the post-processing includes at least one of the following:

[0045] Mark the connected regions in the segmentation results to obtain multiple different connected regions; sort the spatial volumes of the connected regions, and select the connected region with the largest spatial volume as the processed segmentation result;

[0046] The segmentation result is filled using the region growing algorithm, the area outside the placenta is filled, and the placenta and the holes inside the placenta are retained to obtain the processed segmentation result.

[0047] In one embodiment, the first type of volume data is tissue type volume data;

[0048] The second type of volume data is velocity type volume data, and the third type of volume data is energy type volume data; or the second type of volume data is energy type volume data, and the third type of volume data is velocity type volume data.

[0049] In one embodiment, the processor calculates relevant parameters of the placenta based on the segmentation results, including: calculating the volume of the placenta and / or the blood supply parameters of the placenta; the blood supply parameters of the placenta include one or more of the vascularization index, blood flow index and vascularization blood flow index.

[0050] In one embodiment, the processor is further configured with a decision network; the decision network obtains auxiliary decision results and risk probabilities for characterizing the risks of fetal hypoxia, intrauterine growth restriction and preeclampsia based on input evaluation through a random forest algorithm.

[0051] According to a third aspect, an embodiment provides an ultrasound imaging method for fetal measurement, comprising:

[0052] Acquiring at least two different types of volume data regarding a region containing the placenta, the volume data being three-dimensional data, the different types of volume data being of the same size and corresponding in spatial position; the at least two different types of volume data including first type of volume data and at least one of second type of volume data and third type of volume data; the first type of volume data being tissue type volume data, the second type of volume data being velocity type volume data, and / or the third type of volume data being energy type volume data;

[0053] Inputting the first type of volume data into a hybrid segmentation network for placenta segmentation to segment first type of placenta volume data from the first type of volume data; segmenting other types of volume data based on the spatial position of the first type of placenta volume data in the first type of volume data and the spatial position correspondence between the different types of volume data to obtain other types of placenta volume data, where the other types of volume data refer to volume data that is not the first type among the at least two different types of volume data; and using each type of placenta volume data obtained as the segmentation result;

[0054] Calculating placenta-related parameters based on the segmentation result; the placenta-related parameters include placenta volume and / or placenta blood supply parameters; the placenta blood supply parameters include one or more of a vascularization index, a blood flow index, and a vascularized blood flow index, wherein the vascularization index is the percentage of the amount of blood flow data in the placenta to the segmentation result, the blood flow index is the ratio of the average blood flow intensity of the blood flow data in the placenta to the average energy intensity of the energy volume data, and the vascularized blood flow index is the product of the vascularization index and the blood flow index;

[0055] inputting the second type of volume data into a hybrid segmentation network for umbilical cord segmentation, wherein the hybrid segmentation network for umbilical cord segmentation is used to process the input volume data to segment umbilical cord volume data from the input volume data;

[0056] A skeleton line is extracted from the segmented umbilical cord body data, and a center line is extracted based on the extracted skeleton line; and relevant parameters of the umbilical cord are calculated based on the extracted center line, wherein the relevant parameters of the umbilical cord include the umbilical cord length.

[0057] In one embodiment, the hybrid segmented network includes a primary network, an intermediate layer, and a secondary network;

[0058] The primary network is used to perform a coarse segmentation on a target area in the input volume data and output a probability map with a low resolution relative to the original volume data, wherein the target area is a placenta area or an umbilical cord area; wherein the primary network includes a first encoder, a first decoder and a first skip connection; the first encoder is used to gradually extract features of the volume data and reduce the spatial resolution of the volume data; the first encoder includes a plurality of convolutional layers and pooling layers, the convolutional layers are used to capture features of the volume data in three-dimensional space, and the pooling layers are used to downsample so that the spatial resolution of the volume data gradually becomes smaller; the first decoder is used to gradually restore the spatial resolution of the volume data and perform feature fusion, the first decoder includes a plurality of deconvolutional layers and upsampling layers, the deconvolutional layers perform deconvolution operations, and the upsampling layers are used to perform spatial upsampling; the first skip connection exists between the first encoder and the first decoder, and is used to fuse the outputs of different levels in the first encoder with the inputs of corresponding levels in the first decoder;

[0059] The intermediate layer is used to generate candidate spatial regions using an adaptive threshold algorithm for the probability map, and extract corresponding high-resolution sub-volume data from the original volume data based on the candidate spatial regions;

[0060] The secondary network is used to refine the edge of the target area in the sub-volume data and output the target volume data, the target area and the target volume data are the placenta area and the placenta volume data, respectively, or the umbilical cord area and the umbilical cord volume data, respectively; the secondary network includes a second encoder, a second decoder and a second jump connection, the second encoder includes multiple residual dense blocks, the second decoder includes multiple upper layers, the second jump connection is a jump connection based on the convolutional attention feature fusion module, connecting the second encoder and the second decoder; wherein, the output of the residual dense block of the previous level is used as the input of the residual dense block of the next level, and as the input of the jump connection based on the convolutional attention feature fusion module of this layer, the jump connection based on the convolutional attention feature fusion module is used to learn the importance weight of each channel, and adaptively adjust the weight of the feature channel, and then dynamically adjust the importance of the different spatial positions of the features in the second encoder, and then perform channel-weighted splicing with the features of the second decoder.

[0061] According to the ultrasound imaging system and method for fetal measurement of the above-mentioned embodiment, various types of volume data including the placenta are obtained, and the placenta is segmented using corresponding segmentation methods based on the respective characteristics of different types of volume data. This solution is not only simple and feasible, but also saves computing resources, and the segmentation results are reliable and accurate. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] Figure 1 is a schematic structural diagram of an ultrasound imaging system according to an embodiment;

[0063] Figure 2 is a schematic structural diagram of an ultrasound imaging system according to an embodiment;

[0064] Figure 3 is a flow chart of an ultrasound imaging method for fetal measurement according to an embodiment;

[0065] Figure 4 is a flow chart for segmenting volume data in one embodiment;

[0066] Figure 5 is a flow chart for segmenting volume data in one embodiment;

[0067] Figure 6 is a flow chart of an ultrasound imaging method for fetal measurement according to an embodiment;

[0068] Figure 7 is a schematic diagram of a hybrid segmentation network according to an embodiment;

[0069] Figure 8 is a schematic diagram of a primary network according to an embodiment;

[0070] Figure 9 is a schematic diagram of a primary network according to an embodiment;

[0071] Figure 10 is a schematic diagram of a secondary network according to an embodiment;

[0072] Figure 11 A schematic diagram of a secondary network according to an embodiment. DETAILED DESCRIPTION

[0073] The present invention will be further described in detail below by means of specific embodiments in conjunction with the accompanying drawings. Similar elements in different embodiments are numbered with associated similar elements. In the following embodiments, many detailed descriptions are provided to enable the present application to be better understood. However, those skilled in the art will readily appreciate that some of the features may be omitted in different circumstances, or may be replaced by other elements, materials, or methods. In some cases, some operations related to the present application are not shown or described in the specification. This is to avoid the core portion of the present application being overwhelmed by excessive descriptions, and for those skilled in the art, it is not necessary to describe these related operations in detail. They will fully understand the related operations based on the description in the specification and the general technical knowledge in the art.

[0074] In addition, the features, operations, or characteristics described in the specification may be combined in any appropriate manner to form various embodiments. Furthermore, the steps or actions in the method description may be reordered or adjusted in a manner readily apparent to those skilled in the art. Therefore, the various sequences in the specification and drawings are provided solely for the purpose of clearly describing a particular embodiment and are not intended to be mandatory, unless otherwise specified.

[0075] Component numbers used herein, such as "first" and "second," are used solely to distinguish the components being described and do not convey any sequential or technical meaning. References to "connection" and "coupling" in this application, unless otherwise specified, include both direct and indirect connections (couplings).

[0076] Please refer to Figure 1 In some embodiments, an ultrasound imaging system 100 is provided, which includes an ultrasound probe 10, a transceiver control circuit 20, and a processor 30; please refer to Figure 2 Some embodiments may also include a display 40, and each component is described below.

[0077] The ultrasound probe 10 is used to transmit ultrasound waves toward a region of interest, such as an area containing the placenta, and receive returned ultrasound echoes to obtain ultrasound echo signals. In some specific embodiments, the ultrasound probe 10 includes multiple array elements for converting electrical pulse signals into and from ultrasound waves, thereby transmitting ultrasound waves toward the region of interest and receiving returned ultrasound echoes. The array elements can transmit ultrasound waves in response to an excitation electrical signal or convert received ultrasound waves into electrical signals. Therefore, each array element can be used to transmit ultrasound waves toward tissue in the region of interest or to receive ultrasound echoes (referred to as ultrasound echoes) returned from the tissue. During ultrasound testing, transmit and receive sequences can be used to control which array elements are used for transmitting and which are used for receiving ultrasound waves, or to control the time slots used for transmitting and receiving ultrasound echoes. All array elements participating in ultrasound transmission can be simultaneously excited by an electrical signal, thereby transmitting ultrasound waves simultaneously; alternatively, array elements participating in ultrasound transmission can be excited by multiple electrical signals separated by a predetermined time interval, thereby continuously transmitting ultrasound waves separated by a predetermined time interval.

[0078] In some embodiments, the ultrasound probe 10 may be a one-dimensional ultrasound probe provided with a robotic arm, and the robotic arm is used to drive the one-dimensional ultrasound probe to move in space.

[0079] In some embodiments, the ultrasound probe 10 may be a volume probe.

[0080] In some embodiments, the ultrasound probe 10 may be an area array probe.

[0081] The transceiver control circuit 20 is used to control the ultrasound probe 10 to transmit ultrasound waves and receive ultrasound echoes. For example, the transceiver control circuit 20 is used to control the ultrasound probe 10 to transmit ultrasound waves toward the region of interest and to control the ultrasound probe 10 to receive ultrasound echoes reflected from the region of interest. In some specific embodiments, the transceiver control circuit 20 is used to generate a transmit sequence and a receive sequence and output them to the ultrasound probe 10. The transmit sequence is used to control some or all of the multiple array elements in the ultrasound probe 10 to transmit ultrasound waves toward the region of interest. The parameters of the transmit sequence include the number of array elements used for transmission and ultrasound transmission parameters (such as amplitude, frequency, number of transmissions, transmission interval, transmission angle, waveform, and / or focus position). The receive sequence is used to control some or all of the multiple array elements to receive ultrasound echoes after they pass through tissue. The parameters of the receive sequence include the number of array elements used for reception and echo reception parameters (such as reception angle and depth). Depending on the application of the ultrasound echoes or the image generated based on the ultrasound echoes, the ultrasound parameters in the transmit sequence and the echo parameters in the receive sequence may vary.

[0082] In some embodiments, the processor 30 is used to process the ultrasonic echo signal received by the ultrasonic probe 10; in some embodiments, the processor 30 processes the ultrasonic echo signal and can perform one or more data processing links, such as signal demodulation, amplification, filtering, downsampling, beam synthesis, 3D reconstruction, etc. of the ultrasonic echo signal; it should be noted that these processing links are not necessary in all scenarios. In some scenarios, some of the processing links may be required, and in other scenarios, another part of the processing links may be required.

[0083] In some embodiments, the processor 30 includes, but is not limited to, a central processing unit (CPU), a microcontroller unit (MCU), a field-programmable gate array (FPGA), a digital signal processing unit (DSP), and other devices for interpreting computer instructions and processing data in computer software. In some embodiments, the processor 30 is configured to execute various computer applications stored in a non-transitory computer-readable storage medium, thereby enabling the ultrasound imaging system 100 to perform corresponding functions.

[0084] In some embodiments, the display 40 can be used to display information, such as parameters and images calculated by the processor 30. Those skilled in the art will appreciate that, in some embodiments, the ultrasound imaging system 100 may not have an integrated display 40, but may instead be connected to a computer device (e.g., a computer) to display information via a display module (e.g., a display screen) of the computer device.

[0085] The above is some description of the ultrasound imaging system 100 .

[0086] The ultrasound imaging system 100 of the present application can be used for fetal measurement. In such an example, the ultrasound imaging system 100 can perform ultrasound scanning of a region of a pregnant woman containing the placenta to obtain two-dimensional or three-dimensional ultrasound data. In some embodiments, the ultrasound probe 10 is used to transmit ultrasound waves to the region containing the placenta and receive returned ultrasound echoes to obtain ultrasound echo signals. The processor 30 generates one or more different types of volume data based on the ultrasound echo signals. The volume data is three-dimensional data and can also be referred to as three-dimensional volume data, three-dimensional ultrasound data, three-dimensional volume data, etc. The different types of volume data have the same size and one-to-one correspondence in spatial location.

[0087] In some specific embodiments, the processor 30 generates any one, any two, or all three of the first type of volume data, the second type of volume data, and the third type of volume data based on the ultrasound echo signal. In some embodiments, the volume data generated by the processor 30 based on the ultrasound echo signal includes at least the first type of volume data; in some embodiments, the volume data generated by the processor 30 based on the ultrasound echo signal also includes the second type of volume data and / or the third type of volume data.

[0088] In some embodiments, the first type of volume data is tissue volume data (referred to as tissue volume data for short). Tissue volume data mainly represents the structural characteristics of tissues through the grayscale values ​​of pixels.

[0089] In some embodiments, the second type of volume data is velocity type volume data (which may be simply referred to as velocity volume data).

[0090] In some embodiments, the third type of volume data is energy-type volume data (may be referred to as energy volume data for short).

[0091] An ultrasound imaging process may look like this:

[0092] In B mode, a set of delayed focused pulses is sent to the ultrasound probe 10 via the transceiver control circuit 20. The ultrasound probe 10 transmits ultrasound toward the tissue under test, i.e., the area of ​​the pregnant woman's body containing the placenta, and receives the ultrasound reflected from the tissue under test after a certain delay, i.e., the ultrasound echo. The ultrasound echo signal undergoes beamforming processing to complete focusing delay, weighting, and channel summation. After further signal processing, the ultrasound echo signal of a complete scanning cycle of the ultrasound probe 10 is processed as described above to obtain a volume of pre-reconstructed volume data (generally in polar coordinates). Then, through the 3D reconstruction link, the polar coordinate volume data is converted into rectangular coordinate volume data, i.e., tissue type volume data (also referred to as tissue volume data).

[0093] Similarly, 3D scanning in color mode can generate velocity volume data (also referred to as velocity volume data). Furthermore, 3D scanning in color mode generally generates both velocity volume data and tissue volume data. Typically, the tissue volume data and velocity volume data are of the same size and have a one-to-one correspondence in spatial location.

[0094] Similarly, 3D scanning in Power mode generates energy-type volume data (also referred to as energy volume data). Furthermore, 3D scanning in Power mode generally generates both energy volume data and tissue volume data. Typically, the tissue volume data and energy volume data are of the same size and have a one-to-one correspondence in spatial location.

[0095] In addition, after obtaining the relevant volume data, the ultrasound imaging system 100 can also accept user interaction operations in the subsequent image rendering process, and obtain visual cross-sectional images and 3D rendering images through calculation.

[0096] The ultrasound imaging system 100 of the present application can execute some or all of the steps in the ultrasound imaging method 101 for fetal measurement. For example, the ultrasound imaging system 100 can execute some or all of the steps in the ultrasound imaging method 101 for fetal measurement through its processor 30. The ultrasound imaging method 101 for fetal measurement is described in detail below.

[0097] Please refer to Figure 3 In some embodiments, the ultrasound imaging method 101 for fetal measurement may include the following steps:

[0098] Step 110: Acquire one or more different types of volume data containing the placenta. The volume data is volume data containing the placenta, that is, acquire one or more types of volume data containing the placenta.

[0099] In some embodiments, step 110 acquires at least two different types of volume data including the placenta.

[0100] In some embodiments, step 110 acquires any two or three of the first type of volume data, the second type of volume data, and the third type of volume data. In some examples, step 110 acquires at least the first type of volume data; in some embodiments, processor 30 also acquires the second type of volume data and / or the third type of volume data.

[0101] In some embodiments, the first type of volume data is tissue-type volume data (may be referred to as tissue volume data for short).

[0102] In some embodiments, the second type of volume data is velocity type volume data (which may be simply referred to as velocity volume data).

[0103] In some embodiments, the third type of volume data is energy-type volume data (may be referred to as energy volume data for short).

[0104] Step 120: Segmenting the placenta volume data from the volume data containing the placenta.

[0105] In order to calculate the relevant parameters of the placenta, the volume data needs to be segmented to obtain the placenta volume data.

[0106] Please refer to Figure 4 In some embodiments, step 120 of segmenting the volume data includes the following steps:

[0107] Step 121: Segment the first type of volume data to obtain the first type of placenta volume data. That is, step 121 segments the first type of volume data containing the placenta to obtain the first type of placenta volume data (hereinafter referred to as the first type of placenta volume data).

[0108] In some specific embodiments, step 121 inputs the first type of volume data (i.e., the first type of volume data including the placenta) into a hybrid segmentation network for placenta segmentation (which may be referred to as hybrid segmentation network HSN1) to segment the first type of placenta volume data from the first type of volume data, wherein the hybrid segmentation network HSN1 for placenta segmentation is constructed based on deep learning or neural networks, etc., which will be described in detail below.

[0109] Step 122: Based on the spatial position of the first type of placental volume data, continue to segment the placental volume data from other types of volume data.

[0110] Because different types of volume data have the same size and a one-to-one correspondence in spatial positions, after segmenting the first type of placental volume data from the first type of volume data in step 121, based on the spatial position of the first type of placental volume data within the first type of volume data and the one-to-one correspondence between the spatial positions of the different types of volume data, corresponding types of placental volume data can be segmented from other types of volume data, for example, segmenting the second type of placental volume data from the second type of volume data, and segmenting the third type of placental volume data from the third type of volume data. The resulting placental volume data for each type serves as the overall segmentation result, which is then used to calculate placental-related parameters in the subsequent step 130.

[0111] Let's take the first type of volume data as tissue volume data and the second type of volume data as velocity volume data as an example. The tissue volume data containing the placenta is input into the hybrid segmentation network HSN1 for placenta segmentation to segment the tissue volume data of the placenta itself from the tissue volume data containing the placenta. Since the tissue volume data containing the placenta and the velocity volume data containing the placenta are of equal size and correspond one to one in spatial position, the velocity volume data of the placenta itself is segmented from the velocity volume data containing the placenta based on the spatial position of the tissue volume data of the placenta itself in the tissue volume data containing the placenta.

[0112] Similarly, taking the first type of volume data as tissue volume data and the third type of volume data as energy volume data as an example, the tissue volume data containing the placenta is input into the hybrid segmentation network HSN1 for placenta segmentation to segment the tissue volume data of the placenta itself from the tissue volume data containing the placenta. Since the tissue volume data containing the placenta and the energy volume data containing the placenta are of equal size and correspond one to one in spatial position, based on the spatial position of the tissue volume data of the placenta itself in the tissue volume data containing the placenta, the velocity volume data of the placenta itself is correspondingly segmented from the energy volume data containing the placenta.

[0113] Similarly, taking the first type of volume data as tissue volume data, the second type of volume data as velocity volume data, and the third type of volume data as energy volume data as an example, the tissue volume data containing the placenta is input into the hybrid segmentation network HSN1 for placenta segmentation to segment the tissue volume data of the placenta itself from the tissue volume data containing the placenta; since the tissue volume data containing the placenta and the velocity volume data containing the placenta are of equal size and correspond one-to-one in spatial position, the velocity volume data of the placenta itself is correspondingly segmented from the velocity volume data containing the placenta based on the spatial position of the tissue volume data of the placenta itself in the tissue volume data containing the placenta; since the tissue volume data containing the placenta and the energy volume data containing the placenta are of equal size and correspond one-to-one in spatial position, the velocity volume data of the placenta itself is correspondingly segmented from the energy volume data containing the placenta based on the spatial position of the tissue volume data of the placenta itself in the tissue volume data containing the placenta.

[0114] Compared with other volume data such as velocity volume data and energy volume data, tissue volume data has richer and clearer structural information. Therefore, by performing placental segmentation on tissue volume data, the spatial position of the segmented placental tissue volume data is anchored and benchmarked, and then the corresponding placental volume data is segmented from the type of volume data according to the corresponding spatial position relationship. This solution is not only simple and feasible, but also saves computing resources. In addition, the segmentation results of other types of volume data are more reliable and accurate than those directly segmented through methods such as neural networks.

[0115] Considering that in some scenarios, the scanned volume data may or may not include tissue volume data, it is necessary to provide different segmentation strategies.

[0116] Please refer to Figure 5 In some embodiments, step 120 of segmenting the volume data includes the following steps:

[0117] Step 123: Based on the different types of volume data, a segmentation strategy is selected from the first segmentation strategy and the second segmentation strategy for execution, wherein each segmentation strategy still requires the use of the hybrid segmentation network HSN1 for placenta segmentation.

[0118] Step 124 : In the case of volume data of the first type, a first segmentation strategy is executed.

[0119] Specifically, in step 124, when it is determined that the volume data acquired in step 110 includes the first type of volume data, the first segmentation strategy is executed and the volume data is segmented by the hybrid segmentation network HSN1 for placenta segmentation:

[0120] Inputting the first type of volume data into a hybrid segmentation network HSN1 for placenta segmentation to segment the first type of placenta volume data from the first type of volume data;

[0121] Segmenting the other types of volume data based on the spatial position of the first type of placental volume data in the first type of volume data and the spatial position correspondence between different types of volume data to obtain other types of placental volume data, wherein the other types of volume data refer to volume data other than the first type among the at least two different types of volume data acquired in step 110;

[0122] The obtained placenta body data of each type are collectively used as the segmentation result in step 124 .

[0123] Let's take the first type of volume data as tissue volume data and the second type of volume data as velocity volume data as an example. The tissue volume data containing the placenta is input into the hybrid segmentation network HSN1 for placenta segmentation to segment the tissue volume data of the placenta itself from the tissue volume data containing the placenta. Since the tissue volume data containing the placenta and the velocity volume data containing the placenta are of equal size and correspond one to one in spatial position, the velocity volume data of the placenta itself is segmented from the velocity volume data containing the placenta based on the spatial position of the tissue volume data of the placenta itself in the tissue volume data containing the placenta. Similarly, taking the first type of volume data as tissue volume data and the third type of volume data as energy volume data as an example, the tissue volume data containing the placenta is input into the hybrid segmentation network HSN1 for placenta segmentation to segment the tissue volume data of the placenta itself from the tissue volume data containing the placenta. Since the tissue volume data containing the placenta and the energy volume data containing the placenta are of equal size and correspond one to one in spatial position, based on the spatial position of the tissue volume data of the placenta itself in the tissue volume data containing the placenta, the velocity volume data of the placenta itself is correspondingly segmented from the energy volume data containing the placenta. Similarly, taking the first type of volume data as tissue volume data, the second type of volume data as velocity volume data, and the third type of volume data as energy volume data as an example, the tissue volume data containing the placenta is input into the hybrid segmentation network HSN1 for placenta segmentation to segment the tissue volume data of the placenta itself from the tissue volume data containing the placenta; since the tissue volume data containing the placenta and the velocity volume data containing the placenta are of equal size and correspond one-to-one in spatial position, the velocity volume data of the placenta itself is correspondingly segmented from the velocity volume data containing the placenta based on the spatial position of the tissue volume data of the placenta itself in the tissue volume data containing the placenta; since the tissue volume data containing the placenta and the energy volume data containing the placenta are of equal size and correspond one-to-one in spatial position, the velocity volume data of the placenta itself is correspondingly segmented from the energy volume data containing the placenta based on the spatial position of the tissue volume data of the placenta itself in the tissue volume data containing the placenta.

[0124] Step 125 : In the absence of volume data of the first type, executing a second segmentation strategy.

[0125] Specifically, in step 125, when it is determined that the volume data acquired in step 110 does not include the first type of volume data, the second segmentation strategy is executed and the volume data is segmented by the hybrid segmentation network HSN1 for placenta segmentation:

[0126] Inputting each type of volume data obtained in step 110 into the hybrid segmentation network HSN1 for placenta segmentation, so as to segment the corresponding type of placenta volume data from each type of volume data;

[0127] Based on the spatial positions of the obtained various types of placental volume data in the respective types of volume data, the spatial positions of the placental volume data obtained based on the weighted calculation are obtained. Based on the spatial positions of the weighted calculation of the placental volume data, the corresponding types of placental volume data are segmented from each type of volume data and used as the segmentation results in step 125.

[0128] Taking the second type of volume data as velocity volume data and the third type of volume data as energy volume data as an example, the velocity volume data containing the placenta is input into the hybrid segmentation network HSN1 for placenta segmentation to segment the velocity volume data of the placenta itself from the velocity volume data containing the placenta, thereby obtaining the spatial position of the placenta volume data in the velocity volume data (which may be recorded as spatial position 1); similarly, the energy volume data containing the placenta is input into the hybrid segmentation network HSN1 for placenta segmentation to segment the energy volume data of the placenta itself from the energy volume data containing the placenta. The placenta volume data is measured, so that the spatial position of the placenta volume data in the energy volume data can be obtained (which may be recorded as spatial position 2), and then a weighted calculation is performed based on the placenta volume data at spatial position 1 and spatial position 2 (for example, the weighting coefficients can both be 0.5) to obtain the spatial position of the placenta volume data after weighted calculation (which may be recorded as spatial position 3), and then based on the placenta volume data at spatial position 3, the final velocity volume data of the placenta itself and the energy volume data of the placenta itself are segmented from the velocity volume data and the energy volume data, thereby obtaining the final segmentation result for subsequent parameter calculation, etc.

[0129] In steps 121 and 124, the first type of volume data is input into the hybrid segmentation network HSN1 for placenta segmentation for processing and segmentation, thereby obtaining the first type of placenta volume data. When the first type of placenta volume data is obtained, it can be used as preliminary first type of placenta volume data, and then used as final first type of placenta volume data after post-processing. In some embodiments, the post-processing may include at least one of the following:

[0130] Marking the connected regions in the segmentation result (i.e., the preliminary first type of placental body data) to obtain multiple different connected regions; sorting the spatial volumes of the connected regions, and selecting the connected region with the largest spatial volume as the processed segmentation result;

[0131] The segmentation result is filled using the region growing algorithm, the area outside the placenta is filled, and the placenta and the holes inside the placenta are retained to obtain the processed segmentation result.

[0132] In addition, in this article, before any type of volume data is input into the hybrid segmentation network HSN1, a data budget can be performed first. For example, anisotropic diffusion filtering is used to perform three-dimensional noise reduction and contrast enhancement on the volume data, retaining the placental edge structure and highlighting the interface between the placenta and the amniotic membrane.

[0133] Step 130: Calculate relevant parameters of the placenta based on the segmentation results.

[0134] In some embodiments, placenta-related parameters may include placental volume and / or placental blood supply parameters.

[0135] The volume of the placenta, also known as the placental capacity, can be calculated based on the histological data of the placenta.

[0136] Placental blood supply parameters can be calculated using velocity volume data and / or energy volume data. Both velocity volume data and energy volume data can reflect information related to blood flow, and therefore, both can also be referred to as blood flow data. In some examples, placental blood supply parameters include one or more of a vascularization index, a blood flow index, and a vascularized blood flow index.

[0137] The vascularization index (VI) is the percentage of blood flow data within the placenta in the segmentation result. For example, it can be calculated as the percentage of the number of pixels with blood flow signals in the placental volume data relative to the total number of pixels in the placental volume data. The number of pixels with blood flow signals in the placental volume data can be determined using velocity volume data and / or energy volume data. The vascularization index (VI) can indicate the abundance or sparseness of blood vessels within the placenta.

[0138] The blood flow index (FI) is the average intensity of the placental blood flow data. Specifically, it calculates the average blood flow signal intensity for pixels in the placental volume data that have a blood flow signal. Pixels without a blood flow signal are not included in the calculation. For velocity volume data, for example, the blood flow index (FI) can be the average velocity intensity of the placental blood flow data, where the result can be normalized to a range of 0–100. For energy volume data, for example, the blood flow index (FI) can be the average energy intensity of the placental blood flow data, where the result can be normalized to a range of 0–100.

[0139] The vascularization flow index (VFI) is the average blood flow signal intensity of all pixels in the placental data (including pixels without blood flow signals). This means that even pixels without blood flow signals in the placental data are included in the calculation. For example, it can be calculated using the formula VFI = VI * FI / 100.

[0140] Another important parameter for fetal measurement is the length of the umbilical cord. Although tissue volume data provides richer and more specific structural information than other volume data, such as velocity volume data and energy volume data, considering the rich blood flow signals in the umbilical cord, this application uses blood volume data, such as velocity volume data or energy volume data, to segment the umbilical cord volume data and further calculate relevant umbilical cord parameters.

[0141] Please refer to Figure 6 In some embodiments, the ultrasound imaging method 101 for fetal measurement includes the following steps:

[0142] Step 140: Segment the umbilical cord volume data from the second type of volume data.

[0143] Specifically, step 140 inputs the second type of volume data into a hybrid segmentation network for umbilical cord segmentation (which may be referred to as a hybrid segmentation network HSN2) to segment the second type of umbilical cord volume data from the second type of volume data, wherein the hybrid segmentation network HSN2 for umbilical cord segmentation is constructed based on deep learning or neural networks, etc., which will be explained in detail below.

[0144] In some embodiments, the second type of volume data may be velocity volume data or energy volume data.

[0145] Step 141: Calculate relevant parameters of the umbilical cord based on the segmented umbilical cord volume data, wherein the relevant parameters of the umbilical cord include the umbilical cord length.

[0146] Specifically, step 141 can extract skeleton lines from the umbilical cord volume data (umbilical cord velocity volume data or energy volume data) segmented in step 140, extract center lines based on the extracted skeleton lines, and calculate relevant parameters of the umbilical cord, such as umbilical cord length, based on the extracted center lines.

[0147] Skeleton lines can be extracted from the umbilical cord volume data using a skeleton extraction algorithm to extract a single-pixel wide three-dimensional skeleton from the umbilical cord volume data. Skeleton extraction algorithms can use topological refinement methods, distance transform methods, Laplace shrinkage methods, etc. The topological refinement method is based on the principle of medial axis transformation and iteratively erodes the model surface until a single-pixel skeleton is retained. The distance transform method calculates the distance from the internal points of the model to the boundary and extracts local maxima as candidate skeleton points. The Laplace shrinkage method first converts the segmentation results into a surface mesh model, constructs a cotangent Laplace matrix, and iteratively shrinks the mesh surface until it converges to a skeleton line.

[0148] The aforementioned 3D skeleton often contains numerous branches, and it is generally impossible to directly extract a complete centerline. Therefore, centerline extraction must be performed based on the extracted skeleton line. Therefore, the 3D skeleton is combined with the Dijkstra shortest path algorithm (also known as the Dijkstra algorithm) to search for a minimum-cost path, generating a single-pixel-wide, branch-free centerline. This process ensures topological continuity through global optimization, avoiding the breakage problems associated with traditional methods caused by local noise. Since the extracted centerline is a single-pixel-wide, branch-free curve, the umbilical cord length can be calculated based on the 3D centerline. For example, the curve length can be calculated by accumulating the distance between two adjacent points.

[0149] It can be understood that in some embodiments of the present invention, the ultrasound imaging method 101 for fetal measurement may include step 110 , step 140 , and step 141 , but not step 120 and step 130 .

[0150] In this paper, a hybrid segmentation network HSN1 for placenta segmentation is used to process input volume data to segment placenta volume data from the input volume data; a hybrid segmentation network HSN2 for umbilical cord segmentation is used to process input volume data to segment umbilical cord volume data from the input volume data. In some examples, the network structures of the hybrid segmentation network HSN1 for placenta segmentation and the hybrid segmentation network HSN2 for umbilical cord segmentation can be the same or similar, and the difference lies in the different training data.

[0151] Please refer to Figure 7 In some embodiments, the hybrid segmented network HSN1 and the hybrid segmented network HSN2 may each include a primary network 1100, an intermediate layer 1200, and a secondary network 1300, which will be described in detail below.

[0152] In some embodiments, the primary network 1100 is configured to coarsely segment the target region in the input volume data and output a probability map having a lower resolution than the original volume data. It is understood that the pixel values ​​of the pixels in the probability map are used to indicate whether the corresponding pixels belong to the target region. In the hybrid segmentation network HSN1 for placenta segmentation, the target region is the placenta region; in the hybrid segmentation network HSN2 for umbilical cord segmentation, the target region is the umbilical cord region.

[0153] Please refer to Figure 8 In some embodiments, the primary network 1100 includes a first encoder 1110, a first decoder 1120, and a first jump connection 1130, which are described in detail below.

[0154] In some embodiments, the first encoder 1110 is used to gradually extract features from the volume data and reduce its spatial resolution. The first encoder 1110 may include multiple convolutional layers (ConvL) and multiple pooling layers (PL). Generally, the number of convolutional layers (ConvL) and pooling layers (PL) is the same, for example, three layers each. The convolutional layers (ConvL) are used to capture the features of the volume data in three-dimensional space, while the pooling layers (PL) are used to downsample the volume data to gradually reduce its spatial resolution. In this paper, the convolutional layers (ConvL) employ depthwise separable convolution, including depthwise convolution and pointwise convolution. Depthwise convolution performs convolution on each channel separately, while pointwise convolution combines the output channels of the depthwise convolution. Depthwise separable convolution significantly reduces computational complexity without sacrificing accuracy. The volume data input to the primary network 1100 can be sized 128*128*64 to reduce computational overhead. The first encoder 1110 contains only three layers, and fast feature extraction is achieved by compressing the number of channels to [16, 32, 64].

[0155] In some embodiments, the first decoder 1120 is used to gradually restore the spatial resolution of the volume data and perform feature fusion. The first decoder 1120 includes multiple deconvolution layers (DeConvL) and upsampling layers (UpS). The deconvolution layers (DeConvL) perform deconvolution operations, also known as transposed convolutions. The upsampling layers (UpS) perform spatial upsampling.

[0156] In some embodiments, a first skip connection 1130 exists between the first encoder 1110 and the first decoder 1120. The first skip connection 1130 is used to fuse the outputs of different levels in the first encoder 1110 with the inputs of corresponding levels in the first decoder 1120. Introducing the first skip connection 1130 between the first encoder 1110 and the first decoder 1120 allows the feature maps of different levels in the first encoder 1110 to be fused with the corresponding levels in the first decoder 1120, allowing the first decoder 1120 to obtain richer feature information from the first encoder 1110, thereby helping to improve the accuracy of the segmentation results.

[0157] Figure 9 This is an example of a primary network 1100. In the figure, Conv(+BN)+ReLu, Conv represents a convolution operation, BN stands for batch normalization (optional), and ReLu is the activation function. The numbers in the figure represent the dimensions; Conv 1*1*1 indicates a 1*1*1 convolution operation.

[0158] In some embodiments, the middle layer 1200 is used to generate candidate spatial regions using an adaptive threshold algorithm on the probability map, and extract corresponding high-resolution sub-volume data from the original volume data based on the candidate spatial regions.

[0159] In some embodiments, the secondary network 1300 is used to refine the edges of the target region in the sub-volume data output by the intermediate layer 1200 and output the target volume data. It is understood that in the hybrid segmentation network HSN1 for placenta segmentation, the target region is the placenta region and the target volume data is the placenta volume data; in the hybrid segmentation network HSN2 for umbilical cord segmentation, the target region is the umbilical cord region and the target volume data is the umbilical cord volume data.

[0160] Please refer to Figure 10 In some embodiments, the secondary network 1300 includes a second encoder 1310, a second decoder 1320, and a second hop connection 1330, which are described in detail below.

[0161] In some embodiments, the second encoder 1310 includes multiple residual dense blocks RDB; the second decoder 1320 includes multiple up-adopting layers UpS; the second jump connection 1330 is a jump connection based on a convolutional attention feature fusion module, and the second jump connection 1330 connects the second encoder 1310 and the second decoder 1320; wherein, the output of the residual dense block RDB of the previous level is used as the input of the residual dense block RDB of the next level, and as the input of the second jump connection 1330 based on the convolutional attention feature fusion module of this layer, and the second jump connection 1330 based on the convolutional attention feature fusion module is used to learn the importance weight of each channel, and adaptively adjust the weight of the feature channel, and then dynamically adjust the importance of different spatial positions of the features in the second encoder 1310, and then perform channel-weighted splicing with the features of the second decoder 1320.

[0162] The input of the secondary network 1300 can be sub-volume data of size 96*96*48. The second encoder 1310 introduces a residual dense block (RDB). Each downsampling stage can include three densely connected convolution groups, which enhances the placenta edge representation capability through feature reuse. The introduction of the residual dense block (RDB) can achieve efficient feature extraction, alleviate the gradient vanishing problem, and reduce the number of parameters. A convolutional attention feature fusion module (CBAM) is designed at the jump connection. Specifically, the features of the second encoder 1310 are first dynamically adjusted in importance for each channel through the channel attention module. The importance of the different spatial positions of the features of the second encoder 1310 is then dynamically adjusted. The features are then channel-weighted and spliced ​​with the second decoder 1320. The segmentation result output by the second decoder 1320 is the target volume data. As mentioned above, in the hybrid segmentation network HSN1 for placenta segmentation, the target volume data is the placenta volume data; in the hybrid segmentation network HSN2 for umbilical cord segmentation, the target volume data is the umbilical cord volume data.

[0163] Figure 11 FIG13 is an example of a secondary network 1300 , and the numbers in the figure represent the dimensions.

[0164] In response to the challenges of complex morphology and blurred boundaries of the three-dimensional ultrasound placenta, this application designs the above-mentioned hybrid segmentation network structure which is mainly a cascade of the primary network 1100 and the secondary network 1300. Progressive optimization is achieved through the dual-stage collaboration of coarse segmentation and fine segmentation. The primary network coarsely segments the placenta area, and the secondary network refines the edges.

[0165] After calculating relevant parameters from the segmentation results, such as placenta-related parameters and / or umbilical cord-related parameters, auxiliary decision-making and diagnosis can be further performed based on the constructed decision network.

[0166] In some embodiments, the ultrasound imaging system 100 (or its processor 30 ) can be configured with a decision network. The decision network uses a random forest algorithm to evaluate inputs and generate auxiliary decision results and risk probabilities for characterizing the risk of one or more of fetal hypoxia, intrauterine growth restriction, and preeclampsia. This is described in detail below.

[0167] The decision network can be constructed as follows:

[0168] (1) Data is collected using the ultrasound imaging method 101 for fetal measurement described herein or other methods to obtain volume data, placental parameters (e.g., placental volume, placental blood supply parameters), and / or umbilical cord parameters (e.g., umbilical cord length). Obstetricians then annotate the case outcomes, including intrauterine growth restriction (IUGR), preeclampsia, and / or placental abruption, and establish a gestational age-specific reference curve, such as normal placental volume at 24 weeks: 200 ± 30 cm 3 , 40 weeks 500±50cm 3 wait.

[0169] (2) Perform feature engineering and association analysis.

[0170] For example, the umbilical cord length is categorized into short, normal, and long. Short can be less than 30 cm, normal can be between 30 and 100 cm, and long can be greater than 100 cm. It should be noted that these values ​​are for example purposes only and are not intended to limit actual use.

[0171] Another example is analyzing the placental volume deviation: calculating the percentage deviation between the actual volume and the expected value for the gestational age;

[0172] For another example, analyzing the abnormal blood flow index: a risk score is generated when the comprehensive VI is less than 0.4, the VFI is less than 0.5, and the FI continues to decline (for example, the weekly change rate is greater than 15%). It should be noted that these values ​​are only used for examples and are not limited to these values ​​in actual use.

[0173] (3) Training of decision network.

[0174] In the selection of the decision network model or network structure, the XGBoost and random forest fusion architecture can be used. The input features include the classification of umbilical cord length, the deviation of placental volume, the vascularization index VI, the blood flow index FI, the dynamic change rate of the vascularization blood flow index VFI and the gestational age parameters.

[0175] Dynamic weight allocation is performed during training, followed by validation and optimization. During dynamic weight allocation, an attention mechanism can be introduced to assign higher decision weights to high-risk gestational ages (e.g., 28 to 32 weeks) and abnormal parameters (e.g., VFI less than 0.3). During validation and optimization, the decision boundary is adjusted for false-positive cases (e.g., normal placental volume but abnormal VFI) through 5-fold cross-validation (validated on an external test set).

[0176] After training, a risk decision model based on placental volume, blood supply parameters (VI, FI, VFI) and umbilical cord length was obtained.

[0177] The decision network is constructed and trained through the above process.

[0178] During actual use, the three-dimensional volume data of the pregnant woman (including the volume data of the placenta) is obtained through the ultrasound imaging system 100, and the placental volume data is obtained through the above-mentioned ultrasound imaging method 101 such as that for fetal measurement, as well as parameters such as the volume of the placenta and the blood supply parameters of the placenta (vascularization index VI, blood flow index FI, vascularization blood flow index VFI, etc.); the relevant data are input into the decision-making network to obtain auxiliary decision-making results and risk probabilities.

[0179] For example, the auxiliary decision results and risk probabilities output by the decision network can be three-level risk results:

[0180] (1) Low risk (e.g., probability less than 10%): Placental volume and blood flow parameters are within the normal range, and routine prenatal checkups are recommended;

[0181] (2) Medium risk (e.g., probability between 10% and 30%): a single parameter is abnormal (e.g., the umbilical cord is too long but blood flow is normal), prompting enhanced fetal monitoring;

[0182] (3) High risk (e.g., probability between 30% and 70%): Multiple parameters are abnormal (e.g., placental volume less than 10% and VFI less than 0.4), and hospitalization for monitoring and initiation of glucocorticoids to promote lung maturation are recommended.

[0183] Furthermore, decision-making explanations can be made: visual reports can be generated, core risk factors can be marked (for example, "an excessively long umbilical cord causes a 20% decrease in placental perfusion"), and historical data trends can be compared (for example, VFI decreases by 15% for three consecutive weeks).

[0184] This document is described with reference to various exemplary embodiments. However, those skilled in the art will recognize that changes and modifications may be made to the exemplary embodiments without departing from the scope of this document. For example, the various operational steps and components used to perform the operational steps may be implemented in different ways (e.g., one or more steps may be deleted, modified, or incorporated into other steps) depending on the specific application or any number of cost functions associated with the operation of the system.

[0185] In the above-described embodiments, all or part of the embodiments may be implemented using software, hardware, firmware, or any combination thereof. Furthermore, as will be appreciated by those skilled in the art, the principles herein may be embodied in a computer program product on a computer-readable storage medium pre-installed with computer-readable program code. Any tangible, non-transitory computer-readable storage medium may be used, including magnetic storage devices (hard disks, floppy disks, etc.), optical storage devices (CD-ROMs, DVDs, Blue Ray disks, etc.), flash memory, and / or the like. These computer program instructions may be loaded onto a general-purpose computer, a special-purpose computer, or other programmable data processing device to form a machine, such that the instructions executed on the computer or other programmable data processing device generate a device that implements a specified function. These computer program instructions may also be stored in a computer-readable memory, which may instruct the computer or other programmable data processing device to operate in a specific manner, such that the instructions stored in the computer-readable memory may form an article of manufacture including a device that implements a specified function. The computer program instructions may also be loaded onto a computer or other programmable data processing device to execute a series of operational steps on the computer or other programmable device to generate a computer-implemented process, such that the instructions executed on the computer or other programmable device provide the steps for implementing the specified function.

[0186] Although the principles of this invention have been shown in various embodiments, many modifications of structure, arrangement, proportion, elements, materials and components that are particularly suitable for specific environments and operational requirements can be used without departing from the principles and scope of this invention. The above modifications and other changes or amendments are intended to be included within the scope of this invention.

[0187] The foregoing detailed description has been described with reference to various embodiments. However, those skilled in the art will recognize that various modifications and changes can be made without departing from the scope of this disclosure. Therefore, the present disclosure will be considered in an illustrative rather than a restrictive sense, and all such modifications will be included within its scope. Similarly, the advantages, other advantages and solutions to the problems of the various embodiments have been described above. However, the benefits, advantages, solutions to the problems and any elements that can produce these, or make them more specific, should not be interpreted as critical, required or necessary. The term "comprising" and any other variants used in this article are all non-exclusive inclusions, so that a process, method, article or device that includes a list of elements includes not only these elements, but also other elements that are not explicitly listed or do not belong to the process, method, system, article or device. In addition, the term "coupled" and any other variants used in this article refer to physical connections, electrical connections, magnetic connections, optical connections, communication connections, functional connections and / or any other connections.

[0188] Those skilled in the art will recognize that many changes can be made to the details of the above embodiments without departing from the basic principles of the invention. Therefore, the scope of the present invention should be determined solely by the claims.

Claims

1. An ultrasound imaging system for fetal measurement, characterized in that: include: An ultrasound probe, configured to transmit ultrasound waves to an area containing the placenta and receive returned ultrasound echoes to obtain ultrasound echo signals; a transceiver control circuit, configured to control the ultrasonic probe to transmit ultrasonic waves and receive ultrasonic echoes; and Processor for: generating at least two different types of volume data based on the ultrasound echo signal, the volume data being three-dimensional data, the different types of volume data having the same size and corresponding spatial positions; the at least two different types of volume data comprising at least two of a first type of volume data, a second type of volume data, and a third type of volume data; Based on different types of volume data, a segmentation strategy is selected from a first segmentation strategy and a second segmentation strategy, and the volume data is segmented using a hybrid segmentation network for placenta segmentation to obtain a segmentation result, wherein the hybrid segmentation network for placenta segmentation is used to process the input volume data to segment placenta volume data from the input volume data; wherein: determining whether the at least two different types of volume data include the first type of volume data, and if so, selecting a first segmentation strategy and segmenting the volume data using the hybrid segmentation network for placenta segmentation; and if not, selecting a second segmentation strategy and segmenting the volume data using the hybrid segmentation network for placenta segmentation; The selecting a first segmentation strategy and segmenting the volume data using the hybrid segmentation network for placenta segmentation includes: inputting the first type of volume data into the hybrid segmentation network for placenta segmentation to segment first type of placenta volume data from the first type of volume data; segmenting other types of volume data based on the spatial position of the first type of placenta volume data in the first type of volume data and the spatial position correspondence between the different types of volume data to obtain other types of placenta volume data, wherein the other types of volume data refer to volume data that is not the first type among the at least two different types of volume data; and using each type of placenta volume data obtained as the segmentation result; The selecting of the second segmentation strategy and segmenting the volume data using the hybrid segmentation network for placenta segmentation comprises: inputting each type of volume data of the at least two different types of volume data into the hybrid segmentation network for placenta segmentation to segment the corresponding type of placenta volume data from each type of volume data; obtaining the spatial position of the placenta volume data obtained based on weighted calculation based on the obtained spatial positions of all types of placenta volume data in the respective types of volume data; and segmenting the corresponding type of placenta volume data from each type of volume data based on the weighted calculation spatial position of the placenta volume data, and using the segmentation result as the segmentation result; Calculate relevant parameters of the placenta based on the segmentation results.

2. The ultrasound imaging system for fetal measurement according to claim 1, wherein: The at least two different types of volume data include the second type of volume data; the processor is further configured to: inputting the second type of volume data into a hybrid segmentation network for umbilical cord segmentation, wherein the hybrid segmentation network for umbilical cord segmentation is used to process the input volume data to segment umbilical cord volume data from the input volume data; A skeleton line is extracted from the segmented umbilical cord body data, and a center line is extracted based on the extracted skeleton line; and relevant parameters of the umbilical cord are calculated based on the extracted center line, wherein the relevant parameters of the umbilical cord include the umbilical cord length.

3. The ultrasound imaging system for fetal measurement according to claim 1 or 2, characterized in that: The hybrid segmentation network includes a primary network, an intermediate layer, and a secondary network; The primary network is used to perform a coarse segmentation on a target area in the input volume data and output a probability map with a low resolution relative to the original volume data, wherein the target area is a placenta area or an umbilical cord area; wherein the primary network includes a first encoder, a first decoder and a first skip connection; the first encoder is used to gradually extract features of the volume data and reduce the spatial resolution of the volume data; the first encoder includes a plurality of convolutional layers and pooling layers, the convolutional layers are used to capture features of the volume data in three-dimensional space, and the pooling layers are used to downsample so that the spatial resolution of the volume data gradually becomes smaller; the first decoder is used to gradually restore the spatial resolution of the volume data and perform feature fusion, the first decoder includes a plurality of deconvolutional layers and upsampling layers, the deconvolutional layers perform deconvolution operations, and the upsampling layers are used to perform spatial upsampling; the first skip connection exists between the first encoder and the first decoder, and is used to fuse the outputs of different levels in the first encoder with the inputs of corresponding levels in the first decoder; The intermediate layer is used to generate candidate spatial regions using an adaptive threshold algorithm for the probability map, and extract corresponding high-resolution sub-volume data from the original volume data based on the candidate spatial regions; The secondary network is used to refine the edge of the target area in the sub-volume data and output the target volume data, the target area and the target volume data are the placenta area and the placenta volume data, respectively, or the umbilical cord area and the umbilical cord volume data, respectively; the secondary network includes a second encoder, a second decoder and a second jump connection, the second encoder includes multiple residual dense blocks, the second decoder includes multiple upper layers, the second jump connection is a jump connection based on the convolutional attention feature fusion module, connecting the second encoder and the second decoder; wherein, the output of the residual dense block of the previous level is used as the input of the residual dense block of the next level, and as the input of the jump connection based on the convolutional attention feature fusion module of this layer, the jump connection based on the convolutional attention feature fusion module is used to learn the importance weight of each channel, and adaptively adjust the weight of the feature channel, and then dynamically adjust the importance of the different spatial positions of the features in the second encoder, and then perform channel-weighted splicing with the features of the second decoder.

4. The ultrasound imaging system for fetal measurement according to claim 3, wherein: The convolution layer adopts depth-wise separable convolution, including depth-wise convolution and point-wise convolution. The depth-wise convolution performs convolution on each channel separately, and the point-wise convolution is used to combine the output channels of the depth-wise convolution.

5. The ultrasound imaging system for fetal measurement according to claim 1, wherein: The processor is further configured to: after obtaining the segmentation result, perform post-processing on the segmentation result; the post-processing includes at least one of the following: Mark the connected regions in the segmentation results to obtain multiple different connected regions; sort the spatial volumes of the connected regions, and select the connected region with the largest spatial volume as the processed segmentation result; The segmentation result is filled using the region growing algorithm, the area outside the placenta is filled, and the placenta and the holes inside the placenta are retained to obtain the processed segmentation result.

6. The ultrasound imaging system for fetal measurement according to claim 1 or 2, characterized in that: The first type of volume data is tissue type volume data; The second type of volume data is velocity type volume data, and the third type of volume data is energy type volume data; or the second type of volume data is energy type volume data, and the third type of volume data is velocity type volume data.

7. The ultrasound imaging system for fetal measurement according to claim 6, wherein: The processor calculates relevant parameters of the placenta based on the segmentation results, including: calculating the volume of the placenta and / or the blood supply parameters of the placenta; the blood supply parameters of the placenta include one or more of a vascularization index, a blood flow index, and a vascularized blood flow index; the vascularization index is the percentage of the amount of blood flow data in the placenta to the segmentation results, the blood flow index is the ratio of the average blood flow intensity of the blood flow data in the placenta to the average energy intensity of the energy volume data, and the vascularized blood flow index is the product of the vascularization index and the blood flow index.

8. An ultrasound imaging system for fetal measurement, characterized in that: include: An ultrasound probe, configured to transmit ultrasound waves to an area containing the placenta and receive returned ultrasound echoes to obtain ultrasound echo signals; a transceiver control circuit, configured to control the ultrasonic probe to transmit ultrasonic waves and receive ultrasonic echoes; and Processor for: generating at least two different types of volume data based on the ultrasound echo signal, the volume data being three-dimensional data, the different types of volume data having the same size and corresponding spatial positions; the at least two different types of volume data including first type of volume data, and the at least two different types of volume data further including at least one of second type of volume data and third type of volume data; Inputting the first type of volume data into a hybrid segmentation network for placenta segmentation to segment first type of placenta volume data from the first type of volume data; segmenting other types of volume data based on the spatial position of the first type of placenta volume data in the first type of volume data and the spatial position correspondence between the different types of volume data to obtain other types of placenta volume data, where the other types of volume data refer to volume data that is not the first type among the at least two different types of volume data; and using each type of placenta volume data obtained as a segmentation result; Calculating relevant parameters of the placenta based on the segmentation results; The processor is further configured to: inputting the second type of volume data into a hybrid segmentation network for umbilical cord segmentation, wherein the hybrid segmentation network for umbilical cord segmentation is used to process the input volume data to segment umbilical cord volume data from the input volume data; A skeleton line is extracted from the segmented umbilical cord body data, and a center line is extracted based on the extracted skeleton line; and relevant parameters of the umbilical cord are calculated based on the extracted center line, wherein the relevant parameters of the umbilical cord include the umbilical cord length.

9. The ultrasound imaging system for fetal measurement according to claim 8, wherein: The hybrid segmentation network includes a primary network, an intermediate layer, and a secondary network; The primary network is used to perform a coarse segmentation on a target area in the input volume data and output a probability map with a low resolution relative to the original volume data, wherein the target area is a placenta area or an umbilical cord area; wherein the primary network includes a first encoder, a first decoder and a first skip connection; the first encoder is used to gradually extract features of the volume data and reduce the spatial resolution of the volume data; the first encoder includes a plurality of convolutional layers and pooling layers, the convolutional layers are used to capture features of the volume data in three-dimensional space, and the pooling layers are used to downsample so that the spatial resolution of the volume data gradually becomes smaller; the first decoder is used to gradually restore the spatial resolution of the volume data and perform feature fusion, the first decoder includes a plurality of deconvolutional layers and upsampling layers, the deconvolutional layers perform deconvolution operations, and the upsampling layers are used to perform spatial upsampling; the first skip connection exists between the first encoder and the first decoder, and is used to fuse the outputs of different levels in the first encoder with the inputs of corresponding levels in the first decoder; The intermediate layer is used to generate candidate spatial regions using an adaptive threshold algorithm for the probability map, and extract corresponding high-resolution sub-volume data from the original volume data based on the candidate spatial regions; The secondary network is used to refine the edge of the target area in the sub-volume data and output the target volume data, the target area and the target volume data are the placenta area and the placenta volume data, respectively, or the umbilical cord area and the umbilical cord volume data, respectively; the secondary network includes a second encoder, a second decoder and a second jump connection, the second encoder includes multiple residual dense blocks, the second decoder includes multiple upper layers, the second jump connection is a jump connection based on the convolutional attention feature fusion module, connecting the second encoder and the second decoder; wherein, the output of the residual dense block of the previous level is used as the input of the residual dense block of the next level, and as the input of the jump connection based on the convolutional attention feature fusion module of this layer, the jump connection based on the convolutional attention feature fusion module is used to learn the importance weight of each channel, and adaptively adjust the weight of the feature channel, and then dynamically adjust the importance of the different spatial positions of the features in the second encoder, and then perform channel-weighted splicing with the features of the second decoder.

10. The ultrasound imaging system for fetal measurement according to claim 9, wherein: The convolution layer adopts depth-wise separable convolution, including depth-wise convolution and point-wise convolution. The depth-wise convolution performs convolution on each channel separately, and the point-wise convolution is used to combine the output channels of the depth-wise convolution.

11. The ultrasound imaging system for fetal measurement according to claim 8, wherein: The processor is further configured to: after obtaining the segmentation result, perform post-processing on the segmentation result; the post-processing includes at least one of the following: Mark the connected regions in the segmentation results to obtain multiple different connected regions; sort the spatial volumes of the connected regions, and select the connected region with the largest spatial volume as the processed segmentation result; The segmentation result is filled using the region growing algorithm, the area outside the placenta is filled, and the placenta and the holes inside the placenta are retained to obtain the processed segmentation result.

12. The ultrasound imaging system for fetal measurement according to claim 8, wherein: The first type of volume data is tissue type volume data; The second type of volume data is velocity type volume data, and the third type of volume data is energy type volume data; or the second type of volume data is energy type volume data, and the third type of volume data is velocity type volume data.

13. The ultrasound imaging system for fetal measurement according to claim 12, wherein: The processor calculates relevant parameters of the placenta based on the segmentation result, including: calculating the volume of the placenta and / or the blood supply parameters of the placenta; the blood supply parameters of the placenta include one or more of a vascularization index, a blood flow index and a vascularized blood flow index, the vascularization index is the percentage of the amount of blood flow data in the placenta to the segmentation result, the blood flow index is the ratio of the average blood flow intensity of the blood flow data in the placenta to the average energy intensity of the energy volume data, and the vascularized blood flow index is the product of the vascularization index and the blood flow index.

14. The ultrasound imaging system for fetal measurement according to any one of claims 8 or 13, wherein: The processor is also configured with a decision network; the decision network obtains auxiliary decision results and risk probabilities for characterizing the risks of fetal hypoxia, fetal intrauterine growth restriction and preeclampsia based on input evaluation through a random forest algorithm.

15. An ultrasound imaging method for fetal measurement, characterized in that: include: Acquiring at least two different types of volume data regarding a region containing the placenta, the volume data being three-dimensional data, the different types of volume data being of the same size and corresponding in spatial position; the at least two different types of volume data including first type of volume data and at least one of second type of volume data and third type of volume data; the first type of volume data being tissue type volume data, the second type of volume data being velocity type volume data, and / or the third type of volume data being energy type volume data; Inputting the first type of volume data into a hybrid segmentation network for placenta segmentation to segment first type of placenta volume data from the first type of volume data; segmenting other types of volume data based on the spatial position of the first type of placenta volume data in the first type of volume data and the spatial position correspondence between the different types of volume data to obtain other types of placenta volume data, where the other types of volume data refer to volume data that is not the first type among the at least two different types of volume data; and using each type of placenta volume data obtained as a segmentation result; Calculating placenta-related parameters based on the segmentation result; the placenta-related parameters include placenta volume and / or placenta blood supply parameters; the placenta blood supply parameters include one or more of a vascularization index, a blood flow index, and a vascularized blood flow index, wherein the vascularization index is the percentage of the amount of blood flow data in the placenta to the segmentation result, the blood flow index is the ratio of the average blood flow intensity of the blood flow data in the placenta to the average energy intensity of the energy volume data, and the vascularized blood flow index is the product of the vascularization index and the blood flow index; inputting the second type of volume data into a hybrid segmentation network for umbilical cord segmentation, wherein the hybrid segmentation network for umbilical cord segmentation is used to process the input volume data to segment umbilical cord volume data from the input volume data; A skeleton line is extracted from the segmented umbilical cord body data, and a center line is extracted based on the extracted skeleton line; and relevant parameters of the umbilical cord are calculated based on the extracted center line, wherein the relevant parameters of the umbilical cord include the umbilical cord length.

16. The ultrasound imaging method for fetal measurement according to claim 15, wherein: The hybrid segmentation network includes a primary network, an intermediate layer, and a secondary network; The primary network is used to perform a coarse segmentation on a target area in the input volume data and output a probability map with a low resolution relative to the original volume data, wherein the target area is a placenta area or an umbilical cord area; wherein the primary network includes a first encoder, a first decoder and a first skip connection; the first encoder is used to gradually extract features of the volume data and reduce the spatial resolution of the volume data; the first encoder includes a plurality of convolutional layers and pooling layers, the convolutional layers are used to capture features of the volume data in three-dimensional space, and the pooling layers are used to downsample so that the spatial resolution of the volume data gradually becomes smaller; the first decoder is used to gradually restore the spatial resolution of the volume data and perform feature fusion, the first decoder includes a plurality of deconvolutional layers and upsampling layers, the deconvolutional layers perform deconvolution operations, and the upsampling layers are used to perform spatial upsampling; the first skip connection exists between the first encoder and the first decoder, and is used to fuse the outputs of different levels in the first encoder with the inputs of corresponding levels in the first decoder; The intermediate layer is used to generate candidate spatial regions using an adaptive threshold algorithm for the probability map, and extract corresponding high-resolution sub-volume data from the original volume data based on the candidate spatial regions; The secondary network is used to refine the edge of the target area in the sub-volume data and output the target volume data, the target area and the target volume data are the placenta area and the placenta volume data, respectively, or the umbilical cord area and the umbilical cord volume data, respectively; the secondary network includes a second encoder, a second decoder and a second jump connection, the second encoder includes multiple residual dense blocks, the second decoder includes multiple upper layers, the second jump connection is a jump connection based on the convolutional attention feature fusion module, connecting the second encoder and the second decoder; wherein, the output of the residual dense block of the previous level is used as the input of the residual dense block of the next level, and as the input of the jump connection based on the convolutional attention feature fusion module of this layer, the jump connection based on the convolutional attention feature fusion module is used to learn the importance weight of each channel, and adaptively adjust the weight of the feature channel, and then dynamically adjust the importance of the different spatial positions of the features in the second encoder, and then perform channel-weighted splicing with the features of the second decoder.

Images