Ultrasound data segmentation

Abstract

A method for segmenting a target anatomical structure in ultrasound data. Scan converted ultrasound data is obtained in a scan converted space in a Cartesian coordinate system. The scan converted ultrasound data is transformed into descan ultrasound data in a descan space in a toric coordinate system. The descan ultrasound data is an estimate of the ultrasound data as obtained by the original acquisition procedure. Segmentation of the target anatomical structure can therefore be performed on the ultrasound data in the descan space. The resulting segmented data can then be rescanned back to the Cartesian coordinate system for display with the ultrasound data.

Description

Technical Field

[0001] This invention relates to methods for improving ultrasound imaging. In particular, it relates to the segmentation of anatomical structures in ultrasound data. Background Technology

[0002] Image segmentation in digital image processing is used to divide a digital image or volume into multiple segments (e.g., groups of pixels / voxels), each covering a region of the image or volume. Different segments are typically identified and distinguished based on similar shared graphical properties (e.g., color, intensity, or texture) of the pixels / voxels contained within the segmented region.

[0003] In the field of medical imaging, segmentation is valuable for identifying various anatomical structures or features, such as organs, air and fluid passages (e.g., blood vessels or the digestive tract), valves, or chambers. The resulting segmented contours allow clinicians to more quickly evaluate medical images or volumes (enabling faster diagnostic or treatment decisions). Segmentation can also be used to feed subsequent processing techniques.

[0004] Ultrasound imaging is an important field of medical imaging. It plays a crucial role in the assessment and diagnosis of large areas of the body. Diagnosis, treatment, and planning are often based on a clear depiction of relevant anatomical locations or features within ultrasound images, and segmentation can be used to automate this depiction process.

[0005] Automated segmentation and quantification of organs from 3D ultrasound volumes is an active research area, but it has not yet reached clinical routine. Several methods have been proposed, such as the use of classification techniques or marker-based methods. However, these methods are often imprecise and typically contain segmentation gaps.

[0006] Automatic segmentation of 3D organs from 3D ultrasound arrays suffers from major problems related to image acquisition. The acquired volume is stored in a rectangular volume, but only contains information within a so-called cone shape. This leads to inefficient representation of the target anatomical structure.

[0007] More precisely, ultrasound data acquisition methods have the following drawbacks, which affect the training of automatic segmentation algorithms, especially when they involve deep learning:

[0008] (i) Inefficient use of memory, because data outside the cone shape must also be stored. Memory space is a critical factor for success when using current AI techniques for segmentation. In some acquired ultrasound volumes, up to 60% of the voxels are black and do not represent information (they are outside the cone beam coverage).

[0009] (ii) There is a learning bias due to the location of the cone boundary being acquired.

[0010] (iii) Data augmentation requires complex transformations. Data augmentation is a crucial element for achieving high performance in deep learning. In fact, deep learning requires large, rich training datasets to enable its application to new images and avoid overfitting. In the absence of large datasets, an established technique is to apply geometric / grayscale transformations to the training dataset to generate more training images. To simulate the possible presence of organs in other locations within an image while maintaining the realism of images including cone shapes, complex transformations need to be carefully applied in the Cartesian domain. In conventional image processing methods, masks can be used to ignore extrapyramidal information, but this is not feasible when using novel segmentation techniques involving deep learning.

[0011] Therefore, a method is needed to overcome these acquisition shortcomings, with the ultimate goal of training an automatic segmentation algorithm in ultrasound imaging.

[0012] CN 107909585 discloses a method for segmenting the intima of blood vessels for intravascular ultrasound imaging.

[0013] Lo Vercio, Lucas, et al.'s paper, "Assessment of image features for vessel wallsegmentation in intravascular ultrasound images" (International Journal of Computer Assisted Radiology and Surgery, Vol. 11, No. 8, pp. 1397-1407 (XP036013989)), discloses a method for segmenting vessel walls from intravascular ultrasound images.

[0014] Nicolas Toussaint et al.'s paper, "Weakly supervised localization for fetal ultrasound images" (Arxiv.org, Cornell University Library (XP080901171)), discloses a method for detecting and locating fetal anatomical regions in 2D ultrasound images. Summary of the Invention

[0015] This invention is defined by the claims.

[0016] According to an example of one aspect of the present invention, a method for segmenting a target anatomical structure in a 3D ultrasound volume is provided, the method comprising:

[0017] Obtain the scan-transformed ultrasound volume in the scan transformation space of the Cartesian coordinate system;

[0018] The scanned ultrasound volume is transformed into a descanned ultrasound volume in a toroidal coordinate system, wherein the two centers of the toroidal coordinate system are respectively equivalent to the optical and mechanical centers of the ultrasound volume obtained through the original acquisition process, thereby significantly reducing the proportion of empty data; and

[0019] Segmentation of the target anatomical structure on the ultrasound volume in the descanning space is performed to generate segmentation data.

[0020] Ultrasound volumes typically contain “empty” data with meaningless information. When obtained using phased arrays or curved arrays, the ultrasound images used to generate ultrasound volumes are usually represented in a cone shape; however, they are displayed on a rectangular (or square) screen. Therefore, a large proportion of the pixels (or voxels for 3D volumes) are empty and contain meaningless information.

[0021] Descan space refers to the amount of data required to store the scan-converted ultrasound volume in the initial coordinate system. For example, in Cartesian coordinates, image data can be stored in a matrix where each matrix element represents the color of a pixel in RGB. However, for scan-converted ultrasound images, due to the acquisition of the ultrasound volume, meaningful data is in a cone shape, so a proportion of the data (for scan-converted ultrasound images) stored in the matrix is ​​empty data.

[0022] The ultrasound volume is obtained in a scan transformation space (e.g., a cube containing a cone of ultrasound volume) in a Cartesian coordinate space with empty data. It is obtained in a Cartesian coordinate system because it represents how the image would look in real life and is therefore more intuitive for clinicians.

[0023] However, due to the lack of data, the segmentation algorithm gives more weight to the shape of cones and regions with empty data, rather than segmenting the target anatomy during training. Therefore, the scan-converted ultrasound volume is "de-scanned," which involves performing an inverse scan transformation on the scan-converted ultrasound volume. This is done by transforming the scan-converted ultrasound volume to a toroidal coordinate system that represents how the ultrasound volume was acquired (and thus can be considered a de-scanning coordinate system). By transforming the scan-converted ultrasound volume to a coordinate system that represents how it was acquired, the proportion of empty data can be significantly reduced.

[0024] Then, the segmentation algorithm can be trained and applied to the descanned ultrasound volume in the descanned space, with greater accuracy and precision than in the acquisition space.

[0025] Segmentation methods, for example, are based on convolutional neural networks. Descanning transformation methods improve upon convolutional neural network-based segmentation methods by making efficient use of memory.

[0026] The method may further include transforming the descan ultrasound volume in the descan space to the scan conversion space after the segmentation has been performed.

[0027] Once the descanned ultrasound volume has been segmented for the target anatomical structure (e.g., kidney, fetus, etc.), the segmented data (in descanned space) can be transformed back to Cartesian coordinates, allowing it to be superimposed on the initial image. Therefore, the transformation of the segmentation results provides visualization of the segmented portions with correct Cartesian geometry. In this way, clinicians can analyze ultrasound volumes without distortion, with the added benefit that the data has now been segmented (for example, to calculate the area / volume of the segmented target anatomical structure).

[0028] The method may also include displaying one or more of the following:

[0029] The scanning conversion ultrasound volume in the scanning conversion space;

[0030] The descanned ultrasound volume in the descanned space;

[0031] The segmented ultrasound volume in the descanning space; or

[0032] The segmented ultrasound volume in the scanning conversion space.

[0033] The method may further include determining the volume of the target anatomical structure based on the segmentation results of the target anatomical structure.

[0034] The scanning coordinate system is the toroidal coordinate system.

[0035] The toroidal coordinate system is used because it best represents the acquisition of the ultrasonic volume. For example, on phased arrays and curved arrays, when capturing a 3D volume, the ultrasonic probe rotates around the center of a rotor (i.e., the motor that rotates the ultrasonic probe). Therefore, the toroidal coordinate system provides a lower proportion of empty data.

[0036] The method may further include estimating the acquisition geometry of the ultrasound volume to derive a transformation for converting the scan-to-scan ultrasound volume to a de-scan ultrasound volume, wherein estimating the acquisition geometry includes one or more of the following operations:

[0037] Estimate the optical center of the sample;

[0038] The estimated mechanical center was collected;

[0039] Estimate the minimum depth for data acquisition;

[0040] Estimate the maximum depth for data acquisition;

[0041] Estimated perspective; and

[0042] Estimate the sampling sweep angle.

[0043] In typical imaging acquisitions, the imaging lines of the scan-converted ultrasound volume (SCE) are likely not saved along with the SCE. Therefore, due to the lack of the acquisition geometry required to perform the transformation, performing descan may not be directly possible. In such cases, it may be necessary to estimate the acquisition geometry based on the SCE before transforming the SCE to the descan space.

[0044] If there are separate optical and mechanical centers, a toroidal transformation is produced.

[0045] An estimate can be obtained by analyzing images of the scanned transformed ultrasound volume in a Cartesian coordinate system.

[0046] The present invention also provides a computer program product including computer program code units, which, when run on a computing device having a processing system, cause the processing system to perform all the steps of the aforementioned method.

[0047] The present invention also provides a system for segmenting target anatomical structures in a descan ultrasound volume, the system comprising:

[0048] The processor is configured as follows:

[0049] Receive the scan-transformed ultrasound volume within the scan transformation space in the Cartesian coordinate system;

[0050] The scanned ultrasound volume is transformed into a descanned ultrasound volume in a toroidal coordinate system, wherein the two centers of the toroidal coordinate system are respectively equivalent to the optical and mechanical centers of the ultrasound volume obtained through the original acquisition process, thereby significantly reducing the proportion of empty data; and

[0051] Segmentation of the target anatomical structure on the ultrasound volume in the descanning space is performed to generate segmentation data.

[0052] The processor can also be configured to transform the segmented data (in the descan space) back into the scan transformation space after the segmentation has been performed.

[0053] The system may also include a display for displaying one or more of the following:

[0054] The ultrasonic volume in the scanning conversion space;

[0055] The ultrasonic volume in the descanning space;

[0056] The segmented ultrasound volume in the descanning space; or

[0057] The segmented ultrasound volume in the scanning conversion space.

[0058] The processor can also be configured to determine the volume of the target anatomical structure based on the segmentation results of the target anatomical structure.

[0059] The scanning coordinate system is the toroidal coordinate system.

[0060] The processor can be configured to transform the ultrasound volume to a descanned space within the toroidal coordinate system by: estimating the acquisition geometry of the ultrasound volume to derive a transformation for transforming the scan-converted ultrasound volume to a descanned ultrasound volume, wherein the processor is configured to estimate the acquisition geometry based on one or more of the following operations:

[0061] Estimate the optical center of the sample;

[0062] The estimated mechanical center was collected;

[0063] Estimate the minimum depth for data acquisition;

[0064] Estimate the maximum depth for data acquisition;

[0065] Estimated perspective; and

[0066] Estimate the sampling sweep angle.

[0067] These and other aspects of the invention will become apparent and will be explained with reference to one or more embodiments described below. Attached Figure Description

[0068] 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:

[0069] Figure 1 The ultrasound image in the rectangular scan conversion space is shown;

[0070] Figure 2 A histogram showing the percentage of meaningful data in a large database of volumetric kidney samples;

[0071] Figure 3 A flowchart of the method for performing the segmentation is shown;

[0072] Figure 4 The ultrasonic volume transformed to polar coordinates is shown;

[0073] Figure 5 The ultrasound volume in the scan conversion space is shown;

[0074] Figure 6 The torus is shown in Cartesian and torus coordinate systems;

[0075] Figure 7 The ultrasonic volume is shown in spherical coordinates;

[0076] Figure 8 This illustrates organ segmentation in the descanning coordinate system and the acquisition coordinate system;

[0077] Figure 9 The results of organ segmentation in the toroidal coordinate system and the acquisition coordinate system are shown. Detailed Implementation

[0078] The invention will be described with reference to the accompanying drawings.

[0079] It should be understood that while the detailed description and specific examples indicate exemplary embodiments of the apparatus, system, and method, they are 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 present invention will become better understood from the following description, claims, and drawings. It should be understood that the drawings are merely schematic and not drawn to scale. It should also be understood that in all the drawings, the same reference numerals are used to indicate the same or similar parts.

[0080] This invention provides a method for segmenting target anatomical structures in ultrasound data. Scan-transformed ultrasound data is obtained in the scan transformation space of a Cartesian coordinate system. The scan-transformed ultrasound data is then transformed into descanned ultrasound data in the descan space of a toroidal coordinate system. The descanned ultrasound data is an estimate of the ultrasound data obtained through the original acquisition process. Therefore, segmentation of the target anatomical structure can be performed on the ultrasound data in the descan space. The resulting segmented data can then be rescanned back into the Cartesian coordinate system for display along with the ultrasound data.

[0081] Figure 1The image illustrates scan-transformed ultrasound data 102 within a rectangular scan transformation space. Scan lines are typically used to create viewable ultrasound data 102 (e.g., images or volumes) from the raw data based on how the ultrasound data 102 was acquired (based on the acquisition process), allowing visualization in the Cartesian domain where distances and angles are measurable. Therefore, with the aid of scan lines, the raw data captured by the ultrasound probe is scan-transformed to Cartesian coordinates to provide a “true” view of the ultrasound data. Furthermore, the scan-transformed ultrasound data 102 is typically stored in a scan transformation space (in Cartesian coordinates), and any post-acquisition analysis or data processing must generally be performed within this scan transformation space.

[0082] The scan-converted ultrasound data 102 in Cartesian coordinates typically contains meaningful data 104 in a cone shape surrounded by “empty” (or meaningless) data 106. Even if a significant proportion of the scan-converted ultrasound data 102 does not contain meaningful data 104, due to the nature of image data storage in Cartesian coordinates (typically, bitmap images), the entire scan-converted space needs to be stored.

[0083] 3D scan-converted ultrasound volumes typically contain a significant proportion of empty data 106 (melanosomes), which are not clinically necessary but are required for the scan-converted ultrasound volume to be stored in Cartesian coordinates.

[0084] Figure 2 A histogram showing the percentage of meaningful data 104 in the large database of volumetric kidney acquisitions is presented. The large database of volumetric kidney acquisitions contains 621 volumes. For each volumetric kidney acquisition, the ratio (in percentage %) of the volume (meaningful data 104) within the imaging cone across the entire Cartesian volume is calculated.

[0085] The x-axis shows the ratio of the meaningful data 104 to the total data stored in the scan transformation space (in percentage %) and in steps of 0.4%. The y-axis shows the number of cases for each ratio range.

[0086] As can be seen, the ratios range from 35.6% to 44.8%, with the average ratio of meaningful data 104 around 38%. Therefore, the proportion of empty data 106 in volumetric kidney acquisitions (only required for storing data in Cartesian coordinates) is around 62%.

[0087] This means that, on average, 62% of all stored data (used for volumetric kidney acquisition) is essentially wasted storage space. Data compression could be used to make data storage more efficient, but voxels would require memory storage for use with deep learning algorithms. Furthermore, due to the high proportion of empty data 106 in the converted ultrasound data 102, any machine learning algorithm trained using this data (e.g., a segmentation algorithm) would place an undue high importance on the empty data 106.

[0088] Figure 3 A flowchart of a method for performing segmentation on a 3D ultrasound volume is shown. Once the 3D ultrasound volume is obtained in step 302, the acquisition geometry can be estimated in step 304 (if unknown). The acquisition geometry may include estimating the scan lines of the ultrasound data and finding the optical and mechanical centers of the acquisition process.

[0089] In step 304, the acquisition geometry is estimated based on how the ultrasound data is acquired. For example, if an ultrasound volume is obtained using a phased array ultrasound probe rotating around the skin of the object, the ultrasound volume will have a minimum and maximum depth (relative to the ultrasound transducer array), an optical center (where the ultrasound transducer resides), a mechanical center (where the rotor of the ultrasound probe resides), a viewing angle (based on the angle of the ultrasound image obtained by the ultrasound probe), and a sweep angle (based on the rotation of the ultrasound probe around the skin by the rotor of the ultrasound probe). These quantities of the acquisition geometry can be estimated based on the ultrasound volume. In step 306, using the acquisition geometry, the volume can be "de-scanned".

[0090] Descanning is the process of "cancelling" the scan transformation to remove as much empty data as possible. This can be done by choosing a coordinate system to simulate the acquisition process. The toroidal coordinate system is a good approximation because it requires two centers (equivalent to the optical and mechanical centers of acquisition in ultrasound imaging).

[0091] Once the volume has been descanned, a segmentation algorithm can be applied to the volume in step 308 to segment the target anatomical structure. Working in the toroidal coordinate system for segmentation is more powerful than other coordinate systems and provides better automatic segmentation performance. The target anatomical structure may include one or more of the following: organs, blood vessels, heart chambers, bones, muscles, etc.

[0092] The segmentation algorithm can then be trained using the descanned volume. Then, in step 310, the segmented data can be returned to Cartesian coordinates (for rescanning), allowing clinicians (e.g., on a monitor) to view the volume more intuitively. The original volume is shown, for example, with the rescanned segmented data. Optionally, the image data of the original volume can be rescanned along with the segmented data, but preferably the original image data is used (without descanning and rescanning).

[0093] For the purpose of illustrating the general steps, Figure 4 An example of how to descan a scan-converted ultrasound image 102a is shown. A 2D ultrasound image is used instead of an ultrasound volume in this example to simplify the illustration while demonstrating the descan steps.

[0094] Figure 4 a) A scan-transformed ultrasound image 102a in Cartesian coordinates is shown. Empty data 106 is required to store a meaningful image 104a in a rectangular format (e.g., bitmap). Target anatomical structures 402 are shown to aid in explaining how the descan process works.

[0095] Figure 4 b) shows a scan-converted ultrasound image 102a interpolated using imaging lines 404. Imaging lines 404 can be stored together with the scan-converted ultrasound image 102a. However, in most cases, the stored scan-converted ultrasound data 102 does not contain data about imaging lines 404. In this case, imaging lines 404 can be estimated from the scan-converted ultrasound image 102a. Two imaging lines 404 can be estimated based on the sides of the scan-converted ultrasound image 102a. Based on these imaging lines 404, an optical center 406 can be estimated, and additional imaging lines 404 can be constructed. The viewing angle can also be estimated based on the imaging lines 404 as the angle between the two furthest imaging lines 404 determined at the edge.

[0096] Figure 4 c) shows the descanned ultrasound image 408 in polar coordinates (again, a simplified 2D image is used as an example). Therefore, the radial component is based on... Figure 4 The distance of each data point found in b) from the optical center 406. The angular component of each data point can also be found by measuring the angle between the imaging line containing the data point and the edge imaging line of the scan-converted ultrasound image 102a in Cartesian coordinates. As can be seen, there is no empty data 106 in the descanned ultrasound image 408, while it can still be stored in rectangular form (e.g., an RGB matrix bitmap). Furthermore, the target anatomical structure 402 represents a higher proportion of the total data (compared to the scan-converted ultrasound image 102a).

[0097] The descan coordinate system is used instead of the toroidal coordinate system. Figure 5 A portion of the torus 502 in both Cartesian and toroidal coordinate systems is shown. This transformation is inspired by the acquisition process, as it performs a descan transformation, which can be viewed as a dedicated preprocessing step for the segmentation algorithm. Using acquisition geometry parameters, the Cartesian-to-torus transformation can be applied to scan-transformed ultrasound data 102 to simulate ultrasound lines of similar volumes that have been generated. The toroidal coordinate system can be used to represent meaningful 3D ultrasound data 104b.

[0098] 3D ultrasound volume is typically determined based on a number of 2D ultrasound images. An optical center 406 is obtained from at least two 2D ultrasound images. The optical center creates a circular portion, and the center of the circle is the mechanical center 504.

[0099] Once the mechanical center 504 and the optical center 406 are obtained, the distance between them, R0, can be calculated. Therefore, Cartesian coordinates (x, y, z) can be converted to toroidal coordinates.

[0100]

[0101] R = R0 + rcosθ (2)

[0102] Equations (1) and (2) can be used to transform the volume from Cartesian coordinates to torus coordinates. In the acquired volume, points are named (x, y, z). To find its points in the transformation domain... The corresponding intensity at that point, the equation will Compute as a function of (x, y, z), where,

[0103] [R,Θ,Φ] is the range of values ​​for the parameters chosen for descanning. This defines the size of the descan volume computed as input to the segmentation training and testing algorithms.

[0104] The true values ​​of these parameters are often unknown because no pre-scanned transformation data is stored. Therefore, there is freedom to set these values. This is an advantage of learning methods. For example, for coarse and fast algorithms, small values ​​can be chosen, such as (64, 64, 64). In practice, it is preferable to choose a wider range of values, which still allows training deep learning algorithms in memory. It is also possible to define unequal values ​​for these parameters, such as (128, 128, 64). This can reflect the proportional distance in the original acquisition volume.

[0105] Therefore, a variety of choices of these parameters can be used to apply the scan. This means that when learning a network for a given organ, the method can be scaled to the available memory on a computer. At the end of this step, a 3D dataset of dimension (R, Θ, Φ) is obtained, where each voxel carries important information.

[0106] It can calculate the minimum and maximum values ​​of r based on the minimum and maximum acquisition depths, respectively; it can calculate the value of θ based on the viewing angle; and it can calculate the value of r based on the sweep angle. The value of .

[0107] Once the descan ultrasound data 408 has been segmented in the descan space, it can be returned to the scan transformation space in Cartesian coordinates using equations (3) and (4).

[0108]

[0109]

[0110] Figure 6 The scan-conversion ultrasound volume 102b in the scan-conversion space is shown. The white cone represents meaningful data 104b, while the dashed space represents zero values ​​106 that do not encode any useful information. The larger dashed circle outside the scan-conversion space represents the mechanical center 504, and the smaller black circle represents the optical center 406 in the A-plane.

[0111] Using meaningful data 104b, parameters of the acquisition geometry were estimated. These represent: the center of the imaging beam in plane A (optical center 406), the minimum and maximum imaging depths, the viewing angle in plane A, the sweep center (mechanical center 504) used for imaging other planes, and the corresponding sweep angles.

[0112] Figure 7 The ultrasonic volume 104b is shown in spherical coordinates. The spherical coordinate system is an extreme case of the toroidal coordinate system, for which the optical center 406 and the mechanical center 504 are both located at the same position (or can be approximated as being in the same position). In such a case, the ultrasonic volume 104b can be fully defined in the spherical coordinate system without requiring null data. However, since it is assumed that the optical center 406 and the mechanical center 504 are located at the same position (which is sometimes impractical), using the spherical coordinate system will be less accurate than using the toroidal coordinate system. Therefore, using the toroidal coordinate system allows for scan data with different optical and mechanical centers, and thus enables efficient operation for a wider variety of initial scan conditions.

[0113] Figure 8The organ segmentation results 806 in the descanning coordinate system and the acquisition coordinate system are shown. The left-handed image 802 shows an ultrasound image from an ultrasound volume with a kidney segmentation result in Cartesian coordinates. The right-handed image 804 shows a descanned ultrasound image from the ultrasound volume after the ultrasound volume has been transformed to a toroidal coordinate system with the corresponding descanning segmentation result. In the right-handed image 804, the horizontal axis represents the radius r, and the vertical axis represents the polar angle θ of the toroidal coordinate system. Descanning segmentation results are generated based on the ground truth segmentation results and can be used for network training.

[0114] Large amounts of data are needed to train segmentation algorithms. Data augmentation is a method used to create large amounts of data from an initial small data size. For example, it is possible to create multiple training images by applying geometric transformations (e.g., rotation and translation) and color transformations (e.g., grayscale transformation) to a single ultrasound image / single ultrasound volume.

[0115] Descanned ultrasound data in toroidal coordinates allows for more intuitive and error-free geometric transformations to create large databases for training.

[0116] Furthermore, when training segmentation algorithms, data bias (e.g., data shape) must be avoided where possible. In scan-converted ultrasound data, the cone shape containing meaningful data is a source of data bias for segmentation algorithms. Using empty (meaningless) data to train segmentation algorithms can also lead to the algorithm ignoring certain parts of the data or searching for patterns in empty data during future segmentations.

[0117] It can fully describe (e.g., in toroidal coordinates) the ultrasound volume without requiring empty data for storage, thus eliminating data bias caused by empty data.

[0118] This algorithm can be any type of standard deep learning method, such as U-Net segmentation. At the output of the segmentation, the segmented ultrasound data can be transformed back to Cartesian coordinates. The acquisition geometry parameters are reused in the inverse transformation to transform the segmentation mask from, for example, toroidal coordinates back to the image Cartesian coordinates.

[0119] Figure 9The results of organ segmentation 908 in toroidal coordinates 904 and Cartesian coordinates 902 are shown. The same case is shown, with segmentation performed on the left-hand side image 902 without coordinate transformation, and on the right-hand side image 904 with coordinate transformation to the descan domain. Using the descan ultrasound image 904 provides more accurate results. The approximate reality 906 is shown with a black circle, and the segmentation algorithm result 908 is shown by a white boundary. The striped arrow 910 shows segmentation gaps in the result without descan transformation in the scan-transformed ultrasound image 902. The corresponding arrow 912 is shown in the image with descan transformation, demonstrating a significant reduction in segmentation gaps in the descan ultrasound image 904.

[0120] This method has also been successfully applied to fetal abdominal and adult kidney segmentation. The results obtained using this method are significantly superior to methods that do not utilize transformations within the toroidal domain. The results are shown below, demonstrating a significant improvement over conventional segmentation (without descanning).

[0121]

[0122] As discussed above, this system utilizes a processor to perform data processing. Processors can be implemented in various ways, using software and / or hardware, to perform a variety of required functions. A processor typically employs one or more microprocessors, which can be programmed using software (e.g., microcode) to perform the desired functions. A processor can be implemented as a combination of dedicated hardware performing some functions and one or more programmed microprocessors and associated circuitry performing other functions.

[0123] 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).

[0124] In various implementations, the processor may be associated with one or more storage media, such as volatile and non-volatile computer memories, such as 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 desired functions. The various storage media may be fixed within the processor or controller, or may be transportable, allowing one or more programs stored thereon to be loaded into the processor.

[0125] By studying the accompanying drawings, the disclosure, and the claims, those skilled in the art will be able to understand and implement variations of the disclosed embodiments in 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.

[0126] A single element or other unit may perform the functions of several items recorded in the claims.

[0127] Although specific measures are described in different dependent claims, this does not imply that combinations of these measures cannot be used advantageously.

[0128] Computer programs can be stored / distributed on suitable media, such as optical storage media or solid-state media provided with or as part of other hardware, but computer programs can also be distributed in other forms, such as via the Internet or other wired or wireless telecommunications systems.

[0129] If the term “suitable” is used in the claims or specification, it should be noted that the term “suitable” is intended to be equivalent to the term “configured as”.

[0130] Any reference numerals in the claims should not be construed as limiting the scope.

Claims

1. A method for segmenting a target anatomical structure in a 3D ultrasound volume, the method comprising: (302) Obtain the scan-transformed ultrasound volume in the scan transformation space in the Cartesian coordinate system; (306) The scanned ultrasound volume is transformed into a descanned ultrasound volume in the descanned space of the toroidal coordinate system, wherein the two centers of the toroidal coordinate system are respectively equivalent to the optical center and mechanical center of the ultrasound volume obtained through the original acquisition process, thereby significantly reducing the proportion of empty data; and (308) Perform segmentation of the target anatomical structure on the ultrasound volume in the descan space to generate segmentation data.

2. The method according to claim 1 further includes (310) transforming the segmentation result back to the scan conversion space after the segmentation has been performed.

3. The method of claim 2, further comprising displaying one or more of the following: The ultrasonic volume in the scanning conversion space; The ultrasonic volume in the descanning space; The segmented ultrasound volume in the descanning space; or The segmented ultrasound volume in the scanning conversion space.

4. The method according to any one of claims 1 to 3, wherein, The method further includes determining the volume of the target anatomical structure based on the segmentation results of the target anatomical structure.

5. The method according to any one of claims 1 to 3, further comprising estimating the acquisition geometry of the ultrasound volume to derive a transformation for converting the scan-to-scan ultrasound volume to a de-scan ultrasound volume, wherein, Estimating the acquisition geometry includes one or more of the following operations: Estimate the optical center of the sample; The estimated mechanical center was collected; Estimate the minimum depth for data acquisition; Estimate the maximum depth for data acquisition; Estimated perspective; as well as Estimate the sampling sweep angle.

6. A computer program product comprising computer program code units, which, when run on a computing device having a processing system, cause the processing system to perform all the steps of the method according to any one of claims 1 to 5.

7. A system for segmenting a target anatomical structure in a 3D ultrasound volume, the system comprising: The processor is configured as follows: (302) Receive the scan-transformed ultrasound volume in the scan-transformation space in the Cartesian coordinate system; (306) The scanned ultrasound volume is transformed into a descanned ultrasound volume in the descanned space of the toroidal coordinate system, wherein the two centers of the toroidal coordinate system are respectively equivalent to the optical center and mechanical center of the ultrasound volume obtained through the original acquisition process, thereby significantly reducing the proportion of empty data; and (308) Perform segmentation of the target anatomical structure on the ultrasound volume in the descan space to generate segmentation data.

8. The system according to claim 7, wherein, The processor is also configured (310) to transform the segmented data back into the scan transformation space after the segmentation has been performed.

9. The system of claim 8, further comprising a display for displaying one or more of the following: The ultrasonic volume in the scanning conversion space; The ultrasonic volume in the descanning space; The segmented ultrasound volume in the descanning space; or The segmented ultrasound volume in the scanning conversion space.

10. The system according to any one of claims 7 to 9, wherein, The processor is also configured to determine the volume of the target anatomical structure based on the segmentation results of the target anatomical structure.

11. The system according to any one of claims 7 to 9, wherein, The processor is configured to transform the ultrasound volume to a descanned space within the toroidal coordinate system by: estimating the acquisition geometry of the ultrasound volume to derive a transformation for transforming the scan-converted ultrasound volume to a descanned ultrasound volume, wherein the processor is configured to estimate the acquisition geometry based on one or more of the following operations: Estimate the optical center of the sample; The estimated mechanical center was collected; Estimate the minimum depth for data acquisition; Estimate the maximum depth for data acquisition; Estimated perspective; and Estimate the sampling sweep angle.

Images