Method and system for improving ultrasound plane acquisition

Abstract

This invention provides a method for determining a global confidence index for 2D ultrasound images extracted from a 3D ultrasound volume, wherein the global confidence index indicates the suitability of the 2D ultrasound images for medical measurements. The method includes obtaining a 3D ultrasound volume of an object and extracting a set of at least one 2D ultrasound image from the 3D ultrasound volume. Then, a set of geometric indicators is obtained using a first neural network, wherein each geometric indicator indicates a geometric feature of an anatomical structure of the object. The set of 2D ultrasound images is then processed using a second neural network, wherein the output of the second neural network is a set of anatomical indicators, and wherein the anatomical indicators at least indicate the presence of anatomical landmarks. A global confidence index is determined for each 2D ultrasound image in the set of 2D ultrasound images based on the geometric indicators and the anatomical indicators.

Description

Technical Field

[0001] This invention relates to a method and system for obtaining a global confidence index indicating the suitability of ultrasound images for biometric measurements. The invention also relates to enabling less skilled users to perform biometric measurements based on ultrasound volume when using a global confidence index. Background Technology

[0002] Ultrasound imaging technology has transformed ultrasound imaging into a powerful diagnostic tool because such technology provides a powerful visualization of the anatomical structure of the object being studied, at a cost that is only a fraction of other diagnostic tools, such as MRI.

[0003] Ultrasound imaging is routinely used during pregnancy to assess fetal development in the mother's uterus and measure fetal anatomy (this is known as fetal biometry). The traditional way clinicians acquire images of the fetus for each desired view involves manipulating the ultrasound probe while making acoustic contact with the mother's abdomen until the desired anatomical orientation is in the plane of the 2D imaging probe. If multiple views are generated using this procedure, there is a high risk of suboptimal measurements because acquiring and analyzing these views requires a high level of skill (in particular, fetal echocardiography is highly operator-dependent), and the fetus may move during the procedure, requiring the clinician to manually reorient the fetus each time it moves.

[0004] Of particular interest in analyzing fetal development are so-called biometric measurements, which are used to check whether the fetus is developing normally, such as whether it is within expected tolerances. Such biometric measurements typically rely on anatomical guidelines defined in international guidelines to facilitate them. However, providing such measurements is a time-consuming activity, as it requires detecting and labeling anatomical features such as bones and joints within the fetus to provide target landmarks. Labeling can be difficult because well-known artifacts, scattering, and shadowing effects caused by the fetus's position in the mother's uterus can impair ultrasound images. Therefore, fetal biometrics requires highly skilled personnel to acquire suitable planes in order to perform the biometric measurements.

[0005] Ultrasound imaging can also be used, for example, to assess a subject's liver. The optimal plane for the liver must be acquired by searching for hepatic veins within the imaging plane, requiring highly skilled clinicians to perform such measurements.

[0006] US 2009 / 0074280 discloses a system for determining planar locations within 3D ultrasound data for volume to obtain a standard view. Using a standard view improves consistency among users. 3D volume is analyzed using volumetric features, Haar wavelet features, and gradient features.

[0007] US 2016 / 081663 discloses a method for imaging an object. Candidate structures are identified in each of a plurality of 3D image frames. A subgroup of image frames containing the target structure is identified from the 3D frames. For this subgroup of 3D image frames, a plurality of 2D scan planes are determined, and these plurality of 2D scan planes are sorted using a determined sorting function to identify the desired scan plane.

[0008] There is a need to improve methods so that less skilled or non-skilled users can perform standard bioassays. Summary of the Invention

[0009] This invention is defined by the claims.

[0010] According to an example of one aspect of the invention, a method is provided for determining a global confidence index for a two-dimensional, 2D ultrasound image extracted from a three-dimensional, 3D ultrasound volume, wherein the global confidence index includes a measure of the suitability of the 2D ultrasound image for bioassay measurements, the method comprising:

[0011] Obtain the 3D ultrasonic volume of the object;

[0012] Extract a set of at least one 2D ultrasound image from the 3D ultrasound volume;

[0013] A set of geometric indicators is obtained using a first neural network, wherein each geometric indicator indicates the geometric features of the anatomical structure of the object;

[0014] The set of 2D ultrasound images is processed using a second neural network, wherein the output of the second neural network is a set of anatomical indicators, and wherein the anatomical indicators at least indicate the presence of anatomical landmarks; and

[0015] A global confidence index for each 2D ultrasound image in the set of 2D ultrasound images is determined based on the geometric and anatomical indicators.

[0016] This method provides clinicians with a global confidence index for 2D ultrasound images based on 3D ultrasound volume. First, a set of 2D ultrasound images is extracted from the 3D ultrasound volume. To obtain the global confidence index, the sets of geometric and anatomical indicators must first be determined.

[0017] In fetal biometry, geometric indicators can indicate head or abdominal circumference, biparietal diameter, and / or anteroposterior diameter, depending on the anatomy of the 2D ultrasound image (whether the image is of the head or abdomen). Abdominal circumference is typically used in conjunction with head circumference and femur length to determine fetal weight and age. If the 2D ultrasound image is of the femur, geometric indicators can indicate femur length. The set of geometric indicators may also include the head index and / or ratio between femur length and abdominal circumference during fetal biometry, as these are well-known measurements indicative of fetal health.

[0018] To identify anatomical landmarks, 2D ultrasound images are fed into a second neural network that has been trained to recognize a set of anatomical landmarks. For example, in an ultrasound image of the fetal abdomen, the second neural network can be trained to recognize the stomach, ribs, spine, umbilical vein, and left portal vein. Experts commonly use these anatomical landmarks in fetal biometry to identify suitable ultrasound image planes for measuring abdominal circumference.

[0019] The second neural network is configured to output a set of anatomical indicators. For example, it is well known that the optimal plane for measuring head circumference during fetal biometry is the transthalamic coronal section. In the transthalamic coronal section, the thalamus and cavum septum pellucidum are visible anatomical landmarks. The second neural network can be trained to recognize these anatomical landmarks, wherein the anatomical indicators at least indicate the presence of anatomical landmarks in the 2D ultrasound image.

[0020] In fetal biometry, accurate and reliable medical measurements of head circumference, abdominal circumference, and femur length are crucial for determining fetal age, weight, size, and health status. A global confidence index (GCI) indicates whether a 2D ultrasound image is a suitable plane for performing medical measurements such as fetal head circumference, abdominal circumference, and femur length. For example, the GCI allows selection of the most appropriate image from a set of 2D ultrasound images. The GCI also allows users to determine or estimate how close a 2D ultrasound image is to the optimal plane for the biometric measurement. Both geometric and anatomical indicators are used to determine the GCI. The GCI assists clinicians in determining which 2D ultrasound image to use when measuring head circumference in fetal biometry, where only 3D ultrasound volume is required, thus eliminating the need for specialists in fetal biometry.

[0021] The first neural network may be a 2D neural network, and obtaining the set of geometric indicators includes processing the set of 2D ultrasound images using the 2D neural network, wherein the output of the 2D neural network is the set of geometric indicators.

[0022] In this method for determining a set of geometric indicators, 2D ultrasound images are input into a first 2D neural network, which has been trained to segment and indicate geometric features, such as the head, abdomen, and / or femur. The output may be, for example, the perimeter or area of ​​the segmented anatomical structure.

[0023] Alternatively, the first neural network may be a 3D neural network, and wherein obtaining the set of geometric indicators includes processing the 3D ultrasound volume using the 3D neural network, wherein the 3D neural network is trained to identify 3D anatomical structures within the 3D ultrasound volume, and the output of the 3D neural network is the set of geometric indicators.

[0024] In this method, a 3D ultrasound volume is input into a first 3D neural network. The first 3D neural network is trained to recognize anatomical structures (e.g., ribs, liver, stomach, etc.) within the 3D volume. Geometric indicators can then be based on various 2D planes that intersect with the anatomical structures and measure, for example, the size, length, or width of the anatomical structures.

[0025] The first 3D neural network can be used for anatomical segmentation in 3D (e.g., stomach, ribs, etc.). This allows for obtaining a set of 2D planes that truncate the 3D anatomical segmentation and thus obtaining 2D ultrasound images based on the 2D planes, and is able to ignore any 2D planes that are not truncated in the 3D anatomical segmentation.

[0026] A geometric indicator can depend on one or more other geometric indicators.

[0027] The geometric indicators obtained from the 3D neural network can be calculated at the truncation point between the set of 2D ultrasound images and the 3D anatomical structure.

[0028] The method may further include displaying a subset of 2D ultrasound images and displaying a global confidence index corresponding to each 2D ultrasound image in the subset of 2D ultrasound images, wherein the subset of 2D ultrasound images is determined by selecting the following 2D ultrasound images based on comparing the corresponding global confidence indices of the 2D ultrasound images:

[0029] The 2D ultrasound image has the highest value of the global confidence index for the 2D ultrasound images in the set; or

[0030] The 2D ultrasound image has the lowest value of the global confidence index for the 2D ultrasound images in the set; or

[0031] The 2D ultrasound image satisfies the predetermined value of the global confidence index; or

[0032] The 2D ultrasound image satisfies the user-determined value of the global confidence index.

[0033] Therefore, a high global confidence index value can indicate a good fit, or a low global confidence index value can indicate a good fit. The relative value of the global confidence index between 2D ultrasound images can be used to select a subset, or the absolute value (which can be user-defined) can be used as a threshold for comparing the 2D ultrasound image with it.

[0034] Displaying a subset of 2D ultrasound images corresponding to a global confidence index for a certain range and / or value can help clinicians learn which plane is best suited for 2D ultrasound images. For example, three 2D ultrasound images can be displayed, where one image lies on the plane determined to be the best fit, while the others lie on planes that have been shifted to either side of the best fit plane. This helps clinicians learn when they are approaching the optimal plane for biometric measurements.

[0035] The present invention also provides a method for selecting 2D ultrasound images during fetal bioassay, the method comprising:

[0036] The global confidence index is determined using the method defined above, wherein the 3D ultrasound volume is the fetal biometric 3D ultrasound volume;

[0037] Displaying a 2D rendering of the 3D ultrasound volume;

[0038] Displaying a virtual ultrasound probe, wherein the virtual ultrasound probe is configured to perform virtual three-dimensional navigation around the 3D ultrasound volume according to a fetal biometrics workflow;

[0039] The 2D ultrasound images are selected from the set of 2D ultrasound images based on the position of the virtual ultrasound probe relative to the 3D ultrasound volume and also using a global confidence index corresponding to the highest level of fitness.

[0040] The method for selecting the 2D ultrasound image may further include: for the selected 2D ultrasound image having a selected plane:

[0041] The virtual ultrasound probe is rotated about an axis perpendicular to the normal of the selected plane and passing through the center of the selected plane, wherein the selection of the 2D ultrasound image is also based on the rotation of the virtual ultrasound probe about the vertical axis; and

[0042] The virtual ultrasound probe is translated along an axis parallel to the normal of the selected plane, and the selection of the 2D ultrasound image is also based on the translation of the virtual ultrasound probe along the parallel axis.

[0043] A method for simulating planar extraction during fetal bioassays helps inexperienced clinicians navigate a real ultrasound probe to the appropriate location to obtain ultrasound images. The 3D volume of the fetus and a virtual ultrasound probe can be displayed on a monitor. The virtual ultrasound probe can then automatically navigate around the 3D volume of the fetus and show the clinician which 2D ultrasound images will be obtained based on the placement of the real ultrasound probe in certain locations. A global confidence index for each 2D ultrasound image can also be displayed to indicate which ultrasound images are best suited for bioassay measurements.

[0044] Methods for selecting 2D ultrasound images may also include:

[0045] Displays the selected 2D ultrasound image;

[0046] Display the corresponding global confidence index; and

[0047] The virtual ultrasound probe is displayed at the following position on the 3D ultrasound volume: the position indicates where the real ultrasound probe is placed on the real 3D volume in order to obtain the selected 2D ultrasound image.

[0048] The present invention also provides a computer program including code units, which, when the program is run on a processing system, are used to implement the methods mentioned above.

[0049] The present invention also provides a system for determining a global confidence index for a two-dimensional, 2D ultrasound image extracted from a three-dimensional, 3D ultrasound volume, wherein the global confidence index comprises a measure of the suitability of the 2D ultrasound image for medical measurements, the system comprising:

[0050] An ultrasonic probe, used to obtain the 3D ultrasonic volume of an object;

[0051] The processor is configured as follows:

[0052] Extract a set of at least one 2D ultrasound image from the 3D ultrasound volume;

[0053] A set of geometric indicators is computed using a first neural network, wherein each geometric indicator indicates the geometric features of the anatomical structure of the object;

[0054] The set of 2D ultrasound images is processed using a second neural network, wherein the output of the second neural network is a set of anatomical indicators, and wherein the anatomical indicators at least indicate the presence of anatomical landmarks; and

[0055] A global confidence index for each 2D ultrasound image in the set of 2D ultrasound images is determined based on the geometric and anatomical indicators.

[0056] The system may also include a display for displaying one or more of the following:

[0057] Global confidence index; and

[0058] At least one 2D ultrasound image.

[0059] The present invention also provides a system for selecting 2D ultrasound images during fetal bioassay, comprising:

[0060] The system for determining a global confidence index as defined above, wherein the 3D ultrasound volume is a fetal 3D ultrasound volume, and wherein the processor is further configured to:

[0061] 2D ultrasound images are selected from the set of 2D ultrasound images based on the position of the virtual ultrasound probe relative to the 3D ultrasound volume and also using a global confidence index (114) corresponding to the highest level of fitness.

[0062] Furthermore, the display is configured as follows:

[0063] Displaying a 2D rendering of the 3D ultrasound volume;

[0064] A virtual ultrasound probe is displayed, wherein the virtual ultrasound probe is configured to perform virtual three-dimensional navigation around the 3D ultrasound volume according to a fetal biometrics workflow; and

[0065] The selected 2D ultrasound image and its corresponding global confidence index are displayed.

[0066] The processor can also be configured to: target a selected 2D ultrasound image with a selected plane:

[0067] The virtual ultrasound probe is rotated about an axis perpendicular to the normal of the selected plane and passing through the center of the selected plane; the virtual ultrasound probe is translated along an axis parallel to the normal of the selected plane; and

[0068] The 2D ultrasound image is also selected based on the rotation of the virtual ultrasound probe about the vertical axis and the translation of the virtual ultrasound probe along the parallel axis.

[0069] The display can also be configured as follows:

[0070] The virtual ultrasound probe is displayed at the following position on the 3D ultrasound volume: the position indicates where the real ultrasound probe is placed on the real 3D volume in order to obtain the selected 2D ultrasound image.

[0071] These and other aspects of the invention will become apparent and elucidated with reference to one or more embodiments described below. Attached Figure Description

[0072] To better understand the invention and to more clearly illustrate how it can be practiced, reference will now be made to the accompanying drawings by way of example only, in which:

[0073] Figure 1 The method for obtaining the global confidence index is shown;

[0074] Figure 2 The method for using a first 3D neural network is shown;

[0075] Figure 3 The method for using a first 2D neural network is shown;

[0076] Figure 4 This illustrates 3D segmentation performed via a first 3D neural network;

[0077] Figure 5 This illustrates 2D segmentation performed via a first 2D neural network;

[0078] Figure 6 An example of a set of anatomical indicators is shown;

[0079] Figure 7 An example of an optimal ultrasound plane is shown;

[0080] Figure 8 A first example of waist circumference under different virtual probe translations is shown;

[0081] Figure 9A and Figure 9B Examples of waist circumference measurements under different virtual probe translations and rotations are shown;

[0082] Figure 10 An illustration shows a geometric indicator for a set of angle combinations; and

[0083] Figure 11 Three examples of 2D ultrasound images are shown. Detailed Implementation

[0084] The invention will be described with reference to the accompanying drawings.

[0085] It should be understood that while the detailed description and specific examples indicate exemplary embodiments of the apparatus, system, and method, these descriptions and specific examples are intended for illustrative purposes only and are not intended to limit the scope of the invention. These and other features, aspects, and advantages of the apparatus, system, and method of the invention will be better understood from the following description, claims, and drawings. It should be understood that these drawings are merely schematic and not drawn to scale. It should also be understood that the same reference numerals are used throughout all the drawings to indicate the same or similar parts.

[0086] This invention provides a method for determining a global confidence index for 2D ultrasound images extracted from a 3D ultrasound volume, wherein the global confidence index indicates the suitability of the 2D ultrasound images for medical measurements. The method includes obtaining a 3D ultrasound volume of an object and extracting a set of at least one 2D ultrasound image from the 3D ultrasound volume. A set of geometric indicators is then obtained using a first neural network, wherein each geometric indicator indicates a geometric feature of an anatomical structure of the object. The set of 2D ultrasound images is then processed using a second neural network, wherein the output of the second neural network is a set of anatomical indicators, and wherein each anatomical indicator indicates at least the presence of anatomical landmarks. A global confidence index is then determined for each 2D ultrasound image in the set of 2D ultrasound images based on the geometric and anatomical indicators.

[0087] Figure 1 A method for obtaining a global confidence index 114 is illustrated. The principle involves having a user acquire a 3D ultrasound volume 102 containing the target anatomical structure and extracting a 2D ultrasound image 104 from the 3D volume 102. For a given anatomical structure, specific alignment of the central slice of the volume on the corresponding 2D standard plane is not required. Using the acquired volume 102, the first step is to measure the geometry (geometric indicators) of the target anatomical structure. Ultrasound data is input into a first neural network 106, which is trained to measure the geometry of the target anatomical structure (e.g., the circumference of the liver, the length of the femur, etc.) and output a set of geometric indicators 108.

[0088] In parallel, a second neural network 110 is run to evaluate the effectiveness of each extraction plane and output a set of anatomical indicators 112. A global confidence index 114 is derived by combining the measurements of geometric indicators 108 and anatomical indicators 112. A global confidence index 114 is provided for each two-dimensional (2D) ultrasound image.

[0089] exist Figure 1 In the example, the geometric indicator is derived from the 3D volume 102, while the anatomical indicator is derived from the 2D image 104. This is just one example, and this will become clearer with the other examples below.

[0090] The global confidence index of 114 can be used to suggest the most suitable target plane to the user. The target plane can be displayed together with its corresponding global confidence index of 114. Additional display content can allow for evaluation of the corresponding global confidence index of 114 for planes near the target plane.

[0091] Therefore, the global confidence index 114 can be used to assist inexperienced clinicians in acquiring a suitable plane that most closely resembles the optimal plane (e.g., the transthalamic coronal section used for head circumference biometry) for 2D ultrasound measurements. For example, the global confidence index 114 can be used in fetal biometry to acquire a plane for measuring the head, abdomen, and femur. In another example, when imaging the liver of a subject, the global confidence index 114 can be used to acquire the 2D ultrasound plane best suited for the hepatic veins.

[0092] Finding the global confidence index 114 involves: performing 3D organ segmentation or 2D biometric segmentation using (multiple) neural networks, and scoring 2D anatomical frames and deriving a score for each frame. The global confidence index 114 can then be used for liver applications to guide clinicians (e.g., inexperienced clinicians) in obtaining suitable ultrasound planes / images for biometric measurements.

[0093] Figure 2 A method for using a first 3D neural network 106a is illustrated. Given a 3D ultrasound volume, in step 202, the anatomical structure is segmented directly in 3D, and in step 204, the anatomical structure is segmented using the first 3D neural network 106a. In step 206, additional processing may be performed to extract the major axis of the structure (e.g., the direction parallel to the spine in abdominal imaging), and in step 208, 2D ultrasound planes are then extracted from the planes intersecting the 3D volume. Geometric indicators 108 are then determined based on the intersection between the 3D segmentation result and each 2D ultrasound plane. The geometric indicator 108 for each 2D plane may depend on the geometric indicators 108 of the other 2D ultrasound planes.

[0094] The display can be used to show the extracted contents of 2D ultrasound images. For example, a virtual 2D probe can be displayed to simulate multiple locations of an equivalent 2D probe that will image a portion of the volume. For each 2D image, measurements of the target anatomical structure are then calculated using 3D segmentation results previously obtained by taking the intersection of the plane and the volume.

[0095] Figure 3A method for using an alternative first 2D neural network 106b is illustrated. If a 3D segmentation network 106a is not available, the first 2D neural network 106b can be used, for example, as a 2D segmentation network. After obtaining the 3D ultrasound volume in step 302 and orienting the 3D ultrasound volume in step 304, a 2D ultrasound image is extracted from the volume in step 306. The 2D ultrasound image is then input into the first 2D neural network 106b, and a set of geometric indicators 108 is given to each 2D ultrasound image in the corresponding 2D ultrasound image.

[0096] When using the first 2D neural network 106b, the method for orienting the 3D ultrasound volume can be, for example, based on the segmentation of the 2D ultrasound image or the 3D ultrasound volume.

[0097] In this case, a virtual probe can be used directly to simulate multiple locations and obtain portions of the ultrasound volume. Then, for example, a 2D segmentation network 106b is used on each 2D ultrasound image in the 2D ultrasound images corresponding to the location of the virtual probe to segment the corresponding anatomical structures in each image and perform measurements on the anatomical structures, thereby determining a set of geometric indicators 108.

[0098] Figure 4 The diagram illustrates 3D segmentation 402 performed by a first 3D neural network 106a. The first 3D neural network can be trained using a learning database consisting of 3D volumes 102 of the target anatomical structure associated with the segmentation results in 3D. For example, the learning database could be the 3D volume 102 of the stomach with corresponding 3D segments 402. The 3D neural network is trained to infer the 3D segments 402 of the corresponding anatomical structures. The output of the first 3D neural network can be a 3D mesh or mask represented in volumetric coordinates.

[0099] Figure 5 The diagram illustrates 2D segmentation 502 performed by a first 2D neural network 106b. The first 2D neural network 106b can be trained using a database of 2D images 104 containing given anatomical structures (abdomen, head, etc.). The first 2D neural network 106b is trained to segment the corresponding anatomical structure and output, for example, the segmented perimeter 502. Then, following clinical routines, geometric objects 504 (ellipses, circles, etc.) can be fitted to the network's output to provide object measurements.

[0100] Figure 6An example of a set of anatomical landmarks 112 is shown. The anatomical landmarks are shown in an ultrasound image 104 of the fetal head: head 112a, thalamus 112b, choroid plexus 112c, falx cerebri 112d, and cavum septum pellucidum 112e. A second neural network 110 can be trained using a database containing 2D images 104 of anatomical structures required according to clinical guidelines. Each plane can then be associated with a score corresponding to its anatomical content. The output of the second neural network 110 can be, for example, a square surrounding each anatomical landmark and a percentage of the neural network's confidence in the presence of the anatomical landmark.

[0101] Figure 7 An example of an optimal ultrasound plane 702 is shown. Inspired by clinical workflows, the virtual probe can be virtually navigated within a volume using an organized approach:

[0102] First, a translation is performed along the detected principal axis, and planes are extracted at different translation positions. Then, each plane is evaluated using a first neural network 106 and a second neural network 110 based on geometric indicators 108 and anatomical indicators 112.

[0103] Secondly, select the plane corresponding to the optimal global confidence index, and denote this plane as P. OptTrans 702. Point C (refer to 704) is the distance between the probe footprint and plane P. OptTrans The intersection of 702.

[0104] Then the possible rotations were defined:

[0105] (1) From C704 and the center O706 of the considered 2D frame, vector It is the first axis of rotation.

[0106] (2) to P OptTrans The normal 702 and The cross product between them is defined as denoted as The second axis of rotation.

[0107] From the perspective of discrete range And the value of θ, first perform the angle. (around axis) The rotation is then performed at an angle θ (around the axis). Rotate the plane and extract the corresponding plane.

[0108] For each of the extracted planes, a global confidence index 114 is then determined. The global confidence index 114 for each plane is then compared, and the plane with the global confidence index 114 indicating the maximum fit for bioassay measurements is considered the target (optimally available) plane for obtaining 2D ultrasound images. This is because, although plane P... OptTrans 702 could be the most suitable plane during translation, but it can exist at different angles. and / or θ) reach P OptTrans 702 provides a better plane for performing the required bioassay measurements.

[0109] The global confidence index, which indicates best fit, can be determined based on either the maximum or minimum value, depending on the format of the global confidence index.

[0110] A subset of 2D ultrasound images can be selected from the extracted plane for display along with the corresponding global confidence index. For example, the subset of 2D ultrasound images can be determined based on a comparison of the corresponding global confidence index values ​​of the 2D ultrasound images. A subset of 2D images with a desired number of global confidence index values ​​indicating maximum fit (which could be the highest or lowest value of the global confidence index mentioned above) can be selected. Alternatively, the images used for display can be based on a predetermined threshold satisfying the global confidence index or a user-defined threshold satisfying the global confidence index.

[0111] Figure 8 A first example of waist circumference under different virtual probe translations is shown. In the example, P OptTrans 702 corresponds to the position where curve 802 reaches its maximum value. Clearly, curve 802 can be fitted to the measurement point of the waist circumference using any known curve fitting technique. Alternatively, P... OptTrans 702 can correspond to point 804, which has the highest value.

[0112] Figure 9A and Figure 9B Examples of abdominal circumference measurements under different virtual probe translations and rotations are shown. The abdominal circumference measurement can form part of the geometric indicator 108 in fetal biometry. As described above, the virtual probe translates along the main axis of the abdomen and performs measurements at different (discrete) locations. The plane with the optimal geometric indicator 108 corresponds to the plane 902 with the largest circumference in this case. Figure 9A The diagram shows how to select P. OptTrans 702 is the cross-section of the translated cross-section tested. The circumference 902 can be used to determine the geometric indicator (in this case, the waist circumference).

[0113] Once the most suitable plane is found, different possible rotations are performed. Figure 9B It is shown that at a fixed P OptTrans The indication at position 702 and The cross-section is a combination of variations. Therefore, it is possible to derive the optimal geometric indicator based on different circumferences under different rotations. However, the optimal geometric indicator may not be related to the optimal global confidence index, because a plane with a high geometric indicator may not contain the anatomical landmarks required for bioassay measurements.

[0114] exist Figure 9B In this context, circumference 904 is the maximum circumference measured after rotation. Therefore, instead of using circumference 902 to determine the geometric indicator, circumference 904 can provide a more accurate geometric indicator for the waist circumference.

[0115] Figure 10 A diagram illustrating the geometric indicators for a set of angle combinations is shown. The x-axis and y-axis represent the angles around the axis. The different rotations of θ and z-axis represent the geometric indicators indicating waist circumference. The global confidence index 114 can then be calculated using the optimal geometric indicator value 1002 in combination with other geometric indicators 108 and anatomical indicators 112.

[0116] Figure 11 Three examples of 2D ultrasound images are shown. A visualization tool allows the user to understand the quality of the ultrasound plane given a global confidence index (derived from a combination of geometric and anatomical indicators). For any given plane (image), the associated view can be colored according to this global confidence index 114. The underlying color map ranges from green (good anatomical view, suitable for measurement) to red (incorrect view). For example, Figure 11 Image a) shows an incorrect view and will be colored red. Image b) shows a slightly incorrect view of the anatomy, but it does show some anatomical structures of interest, so it will be colored yellow. Image c) shows a good anatomical view (suitable for measurement) and will therefore be colored green. Note that this function can be coupled with instructions regarding the positioning of views in the 3D volume 102 at hand to provide the user with general context.

[0117] As explained above, the detection of anatomical and geometric indicators is based on user-defined machine learning algorithms.

[0118] A machine learning algorithm is any self-trained algorithm that processes input data to produce or predict output data. Here, the input data includes 2D ultrasound images and / or 3D ultrasound volumes, and the output data includes a set of geometric indicators and a set of anatomical indicators.

[0119] Suitable machine learning algorithms for use in this invention will be apparent to those skilled in the art. Examples of suitable machine learning algorithms include decision tree algorithms and artificial neural networks. Other machine learning algorithms (e.g., logistic regression, support vector machines, or Naive Bayes models) are suitable alternatives.

[0120] Artificial neural networks (or simply neural networks) are inspired by the human brain. A neural network consists of multiple layers, each containing multiple neurons. Each neuron performs mathematical operations. Specifically, each neuron can include different weighted combinations of a single type of transformation (e.g., the same type of transformation, sigmoid, etc., but weighted differently). In processing input data, the mathematical operations of each neuron are performed on the input data to produce a numerical output, and the outputs of each layer in the neural network are sequentially fed into the next layer. The final layer provides the output.

[0121] Methods for training machine learning algorithms are well-known. Typically, such methods involve obtaining a training dataset that includes training input data entries and corresponding training output data entries. An initialized machine learning algorithm is applied to each input data entry to generate a predicted output data entry. The machine learning algorithm is then modified using the error between the predicted output data entry and its corresponding training output data entry. This process can be repeated until the error converges and the predicted output data entry is sufficiently similar to the training output data entry (e.g., ±1%). This is often referred to as supervised learning.

[0122] For example, in machine learning algorithms formed by neural networks, the weighted mathematical operations of each neuron can be modified until the error converges. Known methods for modifying neural networks include gradient descent, backpropagation, and others.

[0123] The training input data entries correspond to example 2D ultrasound images and / or 3D ultrasound volumes annotated with target anatomical structures and anatomical landmarks. Technicians will readily develop processors to perform any of the methods described herein. Therefore, each step of the flowchart can represent a different action performed by the processor and can be executed by the corresponding module of the processor.

[0124] A processor can be implemented in a variety of ways using software and / or hardware to perform a range of required functions. A processor typically uses one or more microprocessors, which can be programmed using software (e.g., microcode) to perform the required functions. A processor can be implemented as a combination of dedicated hardware for performing some functions and one or more programmed microprocessors and associated circuitry for performing other functions.

[0125] Examples of circuits that may be used in various embodiments of this disclosure include, but are not limited to, conventional microprocessors, application-specific integrated circuits (ASICs), and field-programmable gate arrays (FPGAs).

[0126] In various implementations, the processor may be associated with one or more storage media, such as volatile and non-volatile computer memories, e.g., RAM, PROM, EPROM, and EEPROM. The storage media may be encoded with one or more programs that, when run on one or more processors and / or controllers, perform the required functions. The various storage media may be fixed within the processor or controller or may be portable, allowing one or more programs stored thereon to be loaded into the processor.

[0127] Those skilled in the art, through studying the accompanying drawings, the disclosure, and the claims, will be able to understand and implement variations of the disclosed embodiments when practicing the claimed invention. In the claims, the word "comprising" does not exclude other elements or steps, and the words "a" or "an" do not exclude a plurality.

[0128] Although certain measures are described in different dependent claims, this does not mean that combinations of these measures cannot be used advantageously.

[0129] Computer programs can be stored / distributed on suitable media, such as optical storage media or solid-state media supplied together with or as part of other hardware, but can also be distributed in other forms, such as via the Internet or other wired or wireless telecommunications systems.

[0130] If the term “suitable” is used in the claims or description, it should be noted that the term “suitable” is intended to be equivalent to the term “configured as”.

[0131] No reference numerals in the claims should be construed as limiting the scope.

Claims

1. A method for determining a global confidence index (114) for each image in a set of two-dimensional, 2D ultrasound images extracted from a three-dimensional, 3D, ultrasound volume, wherein, The global confidence index (114) comprises a measure of the suitability of 2D ultrasound images (104) for bioassay measurements, and the method includes: Obtain the 3D ultrasonic volume of the object (102); Extract a set of at least one 2D ultrasound image (104) from the 3D ultrasound volume (102); The 3D ultrasound volume (102) is processed using a first neural network (106) to obtain a set of geometric indicators (108), wherein each geometric indicator (108) indicates the geometric features of the anatomical features of the object; The set of 2D ultrasound images (104) is processed using a second neural network (110), wherein the output of the second neural network (110) is a set of anatomical indicators (112), and wherein the anatomical indicators (112) at least indicate the presence of anatomical landmarks; and The global confidence index (114) for each 2D ultrasound image (104) in the set is determined based on the geometric indicator (108) and the anatomical indicator (112).

2. The method according to claim 1, wherein, The first neural network (106) is a 2D neural network (106b), and wherein obtaining the set of geometric indicators (108) includes processing the set of 2D ultrasound images (104) using the 2D neural network (106b), wherein the output of the 2D neural network (106b) is the set of geometric indicators (108).

3. The method according to claim 1, wherein, The first neural network (106) is a 3D neural network (106a), and wherein obtaining the set of geometric indicators (108) includes processing the 3D ultrasound volume (102) using the 3D neural network (106a), wherein the 3D neural network (106a) is trained to identify 3D anatomical structures within the 3D ultrasound volume (102), and the output of the 3D neural network (106a) is the set of geometric indicators (108).

4. The method according to claim 3, wherein, The geometric indicator (108) obtained from the 3D neural network (106a) is calculated at the truncation between the set of 2D ultrasound images (104) and the 3D anatomical structure.

5. The method according to any one of claims 1-4, wherein, The geometric indicator (108) depends on one or more other geometric indicators.

6. The method according to any one of claims 1-4, further comprising displaying a subset of 2D ultrasound images and displaying a global confidence index (114) corresponding to each 2D ultrasound image in the subset of 2D ultrasound images, wherein, The subset of 2D ultrasound images was determined by selecting the following 2D ultrasound images based on a comparison of the corresponding global confidence index (114) of the 2D ultrasound images: The 2D ultrasound image has the highest value of the global confidence index for the 2D ultrasound images in the set; or The 2D ultrasound image has the lowest value of the global confidence index for the 2D ultrasound images in the set; or The 2D ultrasound image satisfies the predetermined value of the global confidence index; or The 2D ultrasound image satisfies the user-determined value of the global confidence index.

7. A method for selecting 2D ultrasound images during fetal bioassay, the method comprising: The global confidence index (114) is determined using the method according to any one of the preceding claims, wherein the 3D ultrasound volume (102) is the 3D fetal ultrasound volume; Displaying a 2D ultrasound image (104) extracted from the 3D ultrasound volume (102). Displaying a virtual ultrasound probe, wherein the virtual ultrasound probe is configured to perform virtual three-dimensional navigation around the 3D ultrasound volume (102) according to a fetal biometrics workflow; 2D ultrasound images are selected from the set of 2D ultrasound images based on the position of the virtual ultrasound probe relative to the 3D ultrasound volume (102) and also using a global confidence index (114) corresponding to the highest level of fitness.

8. The method according to claim 7, further comprising: For the selected 2D ultrasound image with the selected plane: The virtual ultrasound probe is rotated about an axis passing through the center of the selected plane and perpendicular to the normal of the selected plane, wherein the selection of the 2D ultrasound image is also based on the rotation of the virtual ultrasound probe about the axis perpendicular to the normal of the selected plane; and The virtual ultrasound probe is translated along an axis parallel to the normal of the selected plane, and the selection of the 2D ultrasound image is also based on the translation of the virtual ultrasound probe along the axis parallel to the normal of the selected plane.

9. The method according to claim 8, further comprising: Displays the selected 2D ultrasound image; Display the corresponding global confidence index; and The virtual ultrasound probe is displayed at the following position on the 3D ultrasound volume (102): the position indicates the location on the real volume where the real ultrasound probe is placed in order to obtain the selected 2D ultrasound image.

10. A computer program product comprising a computer program that, when run on a processing system, is used to implement the method according to any one of claims 1 to 9.

11. A system for determining a global confidence index (114) for each image in a set of two-dimensional, 2D ultrasound images extracted from a three-dimensional, 3D, ultrasound volume, wherein, The global confidence index (114) comprises a measure of the suitability of 2D ultrasound images (104) for bioassay measurements, and the system includes: An ultrasonic probe used to obtain the 3D ultrasonic volume of an object (102). The processor is configured as follows: Extract a set of at least one 2D ultrasound image (104) from the 3D ultrasound volume (102); The 3D ultrasound volume (102) is processed using a first neural network (106) to calculate a set of geometric indicators (108), wherein each geometric indicator (108) indicates the geometric features of the anatomical features of the object; The set of 2D ultrasound images (104) is processed using a second neural network (110), wherein the output of the second neural network (110) is a set of anatomical indicators (112), and wherein the anatomical indicators (112) at least indicate the presence of anatomical landmarks; and The global confidence index (114) for each 2D ultrasound image (104) in the set is determined based on the geometric indicator (108) and the anatomical indicator (112).

12. The system of claim 11, further comprising a display for displaying one or more of the following: Global confidence index (114); and At least one 2D ultrasound image (104).

13. A system for selecting 2D ultrasound images during fetal biometry, comprising: The system according to claim 12, wherein the 3D ultrasound volume (102) is a 3D fetal ultrasound volume, and wherein the processor is further configured to: 2D ultrasound images are selected from the 3D ultrasound volume based on the position of the virtual ultrasound probe relative to the 3D ultrasound volume (102) and also using a global confidence index (114) corresponding to the highest level of fitness. Furthermore, the display is configured as follows: Displaying a 2D ultrasound image (104) extracted from the 3D ultrasound volume (102). A virtual ultrasound probe is displayed, wherein the virtual ultrasound probe is configured to perform virtual three-dimensional navigation around the 3D ultrasound volume (102) according to a fetal biometrics workflow; and Display the selected 2D ultrasound image and the corresponding global confidence index (114).

14. The system according to claim 13, wherein, The processor is also configured to: for the selected 2D ultrasound image having a selected plane: The virtual ultrasound probe is rotated about an axis passing through the center of the selected plane and perpendicular to the normal of the selected plane; The virtual ultrasound probe is translated along an axis parallel to the normal to the selected plane; and The 2D ultrasound image is also selected based on the rotation of the virtual ultrasound probe about the axis perpendicular to the normal of the selected plane and the translation of the virtual ultrasound probe along the axis parallel to the normal of the selected plane.

15. The system according to any one of claim 13 or 14, wherein, The display is also configured to: The virtual ultrasound probe is displayed at the following position on the 3D ultrasound volume (102): the position indicates the location on the real volume where the real ultrasound probe is placed in order to obtain the selected 2D ultrasound image.

Images