Medical Image Processing

Abstract

An apparatus for determining the location of a midsagittal axis for a medical image depicting the heart. The image is processed to identify a first portion representing a portion of the spine and a bounding box that encloses a representation of the subject's internal thorax. The internal thorax is the portion of the thorax that is included between the thorax and the spine or skin, but does not include any calcified tissue such as bone. The midsagittal axis is determined to intersect with a portion of the spine and extend perpendicular to the side of the bounding box that is closest to the portion of the spine.

Description

[Technical Field]

[0001] The present invention relates to the field of medical imaging, and in particular to medical imaging of the heart. [Background technology]

[0002] The use of medical imaging to analyze and evaluate the condition of a subject's heart, particularly the heart of a fetus, is increasing. For example, ultrasound diagnosis of congenital heart disease is a widely adopted technique and is increasingly recommended during prenatal surveillance of women worldwide.

[0003] One important biometry measurement for use in assessing fetal cardiac status is the cardiac angle (CA), which is the angle between the midsagittal line of the heart and the cardiac axis. Traditionally, the cardiac angle is calculated manually by a clinician who reviews one or more appropriate medical images based on their expertise and experience. Summary of the Invention [Problem to be solved by the invention]

[0004] It is desirable to improve the accuracy and repeatability (eg, clinician independence) of determining cardiac angles. [Means for solving the problem]

[0005] The invention is defined by the claims.

[0006] According to an example according to one aspect of the present invention, a computer-implemented method is provided for determining the location of the midsagittal axis on a medical image providing a four-chamber view plane of the heart.

[0007] The computer implementation includes acquiring a medical image, the medical image including a representation of a subject's thorax within a four-chamber view plane, and processing the medical image to identify a bounding box having edges that intersect with boundaries of a first portion of the image representing a portion of the subject's spine and a second portion of the image representing an interior of the subject's thorax, the interior thorax including only the subject's soft tissue contained within the subject's thorax.

[0008] The computer-implemented method also includes determining a position of the midsagittal axis such that the midsagittal axis intersects the first portion of the image and extends perpendicular to an edge of the bounding box that is closest to the first portion of the image.

[0009] Accurate identification of the location of the midsagittal axis is useful for deriving several different biometric measurements, such as cardiac angle, but also provides information useful for guiding clinicians to understand the anatomical structures represented in medical images.

[0010] The proposed approach utilizes a novel concept of the internal thorax to identify the midsagittal axis. The internal thorax refers to the portion of the chest contained within the thoracic cage and does not include any bones such as the ribs or spine. The use of the internal thorax has been found to allow for more accurate identification of the midsagittal axis compared to the use of the entire chest or the ribs themselves. This is particularly advantageous in a fetal monitoring environment.

[0011] In the rest of the contests herein, the bounding boxes are oval (i.e., elliptical or square).

[0012] In some examples, determining the position of the midsagittal axis is configured such that the midsagittal axis originates from the first portion of the image and extends away from the edge of the bounding box that is closest to the first portion of the image.

[0013] In some examples, determining the position of the midsagittal axis is further configured such that the midsagittal axis intersects a centroid of the first portion of the image.

[0014] In some examples, processing the medical image includes using one or more machine learning methods to process the medical image. In particularly advantageous embodiments, this step includes processing the medical image using at least one machine learning method having a U-net architecture. This form of architecture has been identified as being particularly accurate and effective at identifying representative portions (e.g., contours or bounding boxes) of particular features within an image, i.e., performing image segmentation tasks. It will be appreciated that other forms of machine learning algorithms may be used in other embodiments of the present invention.

[0015] The interior of the thorax can exclude calcified tissue such as bone. In particular, the internal thorax excludes the posterior portion of the spine and ribs and runs anteriorly to the skin.

[0016] The medical images are preferably ultrasound medical images, however it is possible to apply the proposed technique using other forms of medical images such as CT images or magnetic resonance images.

[0017] The step of acquiring the medical image may include acquiring a plurality of potential medical images of the subject, and processing the plurality of potential medical images to select one of the potential medical images as the medical image.

[0018] In some examples, processing the plurality of potential images includes processing each potential medical image to determine whether the potential medical image represents a four-chamber view plane, and removing each potential medical image that is determined not to represent a four-chamber view plane.

[0019] It has been recognized that medical images preferably provide a four-chamber view plane of the heart to facilitate accurate determination of the midsagittal axis.

[0020] Processing the plurality of potential images may include processing each potential medical image to determine whether the potential medical image includes a representation of the subject's entire chest, and removing each potential medical image that is determined not to include a representation of the subject's entire chest. Accuracy of identifying the internal chest is significantly reduced if the medical image does not include the subject's entire chest. Accuracy can be improved by removing medical images that do not include a representation of the entire chest.

[0021] Also provided is a computer-implemented method for determining a cardiac angle measurement in a subject, the computer-implemented method including: determining a position of a midsagittal axis on a medical image in a four-chamber view plane by performing any method disclosed herein; processing the medical image using one or more second machine learning methods to determine the position of the cardiac axis; and determining an angle between the midsagittal axis and the cardiac axis as the cardiac angle measurement.

[0022] This embodiment provides a technique for determining cardiac angle measurements. The proposed approach is accurate and repeatable, allowing for ease of comparison between different cardiac angle measurements (e.g., at different points in a single patient's medical timeline or between different patients). More specifically, this embodiment provides an automated solution for measuring CA, which is useful for rapid analysis and medical reporting.

[0023] Processing the medical image using one or more second machine learning methods may include processing the medical image using at least one second machine learning method having a U-net architecture.

[0024] In some examples, processing the medical image to determine the location of the cardiac axis includes processing the medical image using one or more second machine learning methods to identify a third portion of the image representing the subject's ventricular septum, and fitting a line to the third portion of the image that intersects with the midsagittal axis to determine the location of the cardiac axis.

[0025] The computer-implemented method may further include controlling an output interface to provide a visual representation of the determined cardiac angle measurements, which provides a user with additional information about the subject for clinician decision making or clinical analysis.

[0026] Also proposed is a computer program product comprising computer program code means which, when executed on a computing device having a processing system, causes the processing system to perform all of the steps of any of the methods disclosed herein.

[0027] A processing system for determining the location of the midsagittal axis on a medical image providing a four-chamber view plane of the heart is also proposed.

[0028] The processing system is configured to acquire the medical image, the medical image including a representation of the subject's chest in the four-chamber view plane, and process the medical image to identify a bounding box having an edge that intersects with a boundary of a first portion of the image representing a portion of the subject's spine and a second portion of the image representing the interior of the subject's thorax, the interior thorax including only the subject's soft tissue contained within the subject's thorax, and to determine a position of the midsagittal axis such that the midsagittal axis intersects the first portion of the image and extends perpendicular to the edge of the bounding box that is closest to the first portion of the image.

[0029] Also provided is an ultrasound system comprising any of the processing systems disclosed herein, an ultrasound probe configured to generate medical images for acquisition by the processing system, and an output interface configured to provide a visual representation of the midsagittal axis and / or values ​​of parameters derived from the midsagittal axis.

[0030] An example of a parameter value derived from the midsagittal axis is a cardiac angle measurement. Other examples will be apparent to those skilled in the art.

[0031] These and other aspects of the invention will be apparent from and elucidated with reference to the embodiments described hereinafter.

[0032] For a better understanding of the present invention and to show more clearly how it may be carried into effect, reference will now be made, by way of example only, to the accompanying drawings in which: [Brief explanation of the drawings]

[0033] [Figure 1] 1 is a flowchart illustrating the proposed method. [Figure 2] The proposed technology is shown. [Figure 3] 1 is a flow chart showing suggested steps for use in the method. [Figure 4] 1 is a flowchart illustrating another proposed method. [Figure 5] The proposed technology is shown. [Figure 6] The proposed system is shown. DETAILED DESCRIPTION OF THE INVENTION

[0034] The present invention will now be described with reference to the drawings.

[0035] It should be understood that the detailed description and specific examples, while indicating exemplary embodiments of the devices, systems, and methods, are for purposes of illustration only and are not intended to limit the scope of the invention. These and other features, aspects, and advantages of the devices, systems, and methods of the present invention will become better understood from the following description, appended claims, and accompanying drawings. It should be understood that the figures are schematic only and are not drawn to scale. It should also be understood that the same reference numerals are used throughout the figures to indicate the same or similar parts.

[0036] The present invention provides a mechanism for determining the location of the midsagittal axis for a medical image depicting the heart. The image is processed to identify a first portion representing the spine and a portion of a bounding box surrounding a representation of the subject's internal thorax. The internal thorax is the portion of the thorax that is included between the thorax and the spine or skin, but does not include any calcified tissue such as bone. The midsagittal axis is determined to intersect with the portion of the spine and extend perpendicular to the side of the bounding box that is closest to the portion of the spine.

[0037] The embodiments are based on the recognition that the use of the internal chest facilitates accurate automatic determination of the position of the midsagittal axis within a four-chamber view of the heart.

[0038] While the proposed approach can be used in any form of cardiac / heart monitoring, it finds particular advantages in fetal cardiac monitoring, and even more specifically in ultrasound-based fetal cardiac monitoring.

[0039] For the avoidance of doubt, throughout this specification, the adjectives "closest" or "farthest" in the phrase "closest [ / farthest] edge of the bounding box" relate to the noun "edge" and not the noun "bounding box." Thus, the phrase "edge [ / farthest] of the bounding box closest to the first portion of the image" identifies a particular edge of the bounding box, rather than a particular bounding box.

[0040] 1 is a flow chart illustrating a computer-implemented method 100 according to one embodiment, which is designed to identify or determine the location of the midsagittal axis for a medical image providing a four-chamber view plane of the heart, preferably a fetal heart.

[0041] Method 100 includes step 110 of acquiring a medical image. The medical image may be acquired (e.g., retrieved) from a memory or storage device, or may be provided by an image generating device, for example. Detailed examples of step 110 are provided later in this disclosure.

[0042] The medical images are preferably non-invasively acquired medical images such as ultrasound images, X-ray or CT images, or magnetic resonance images, etc. Ultrasound images are particularly advantageous because such medical images are readily available and because the generation of ultrasound images is commonly used in prenatal monitoring of the fetus, so that they provide adequate detail for implementing the proposed approach.

[0043] The method 100 also includes a step 120 of processing the medical image to identify a first portion of the image representing a portion of the subject's spine, and a bounding box having edges that intersect with the boundary of a second portion of the image representing the subject's internal thorax.

[0044] The internal thorax includes only the subject's soft tissue contained within the subject's thorax. The internal thorax thereby excludes calcified tissue such as bone (e.g., of the spine or thorax). More specifically, the internal thorax excludes the posterior portion of the spine and ribs and extends beneath the skin.

[0045] As shown, step 120 may be performed in separate substeps.

[0046] Thus, step 120 may include a first sub-step 121 of processing the medical image to identify a first portion of the image that represents a portion of the subject's spine.

[0047] Similarly, step 120 may include a second sub-step 122 of processing the medical image to identify a bounding box having edges that intersect with the boundary of a second portion of the image representing the subject's internal thorax, the internal thorax including only the subject's soft tissue contained within the thorax.

[0048] More specifically, the bounding box may be sized to completely contain the second portion of the image and intersect its edges, i.e., so that none of the second portion of the image lies outside the bounding box. The bounding box is preferably the smallest sized bounding box that satisfies these constraints.

[0049] Sub-steps 121 and 122 can be performed simultaneously or separately.

[0050] In some examples, one or both of sub-steps 121 and 122 are performed using one or more (suitably trained) machine learning methods.

[0051] One approach for performing substep 122 is to use one or more machine learning methods to identify a contour that bounds the second portion of the image. A bounding box can then be fit to this contour, e.g., a bounding box that intersects with the edges of the second portion of the image. In particular, the bounding box can be a box (i.e., a cube) of minimal size that completely contains the contour (representing the inner chest) that bounds the second portion of the image, i.e., none of the contour extends inside the bounding box. One exemplary technique for determining such a bounding box is the "minAreaRect" function used in OpenCV. Other exemplary techniques will be apparent to those skilled in the art.

[0052] Another approach for performing sub-step 122 is to use one or more machine learning methods that are directly trained to determine the bounding boxes, which can be achieved by appropriate training of the machine learning methods.

[0053] Alternatively, the two functions of sub-steps 121 and 122 may be performed using the same machine learning method or methods, for example using a single dual-purpose machine learning method.

[0054] When used, the one or more machine learning methods preferably include at least one machine learning method having a U-net architecture, such machine learning methods being particularly suited to the task of performing image segmentation, thus providing improved identification of the first portion and / or bounding box.

[0055] Alternative approaches (e.g., not utilizing any machine learning process) for performing sub-step 121 and / or sub-step 122 will be apparent to those skilled in the art. By way of example, one or more shape models or automated finite element modeling techniques may be used to identify regions or contours representative of the first portion and / or associated internal breast. In particular, any suitable segmentation technique may be used to perform sub-step 121 and / or sub-step 122.

[0056] Thus, step 120 may include processing the medical image using one or more segmentation techniques and / or algorithms to identify a bounding box having edges that intersect with the boundaries of a first portion of the image that represents a portion of the subject's spine and a second portion of the image that represents the subject's internal thorax.

[0057] Examples of suitable segmentation techniques are, for example, Pal, Nikhil R., and Sankar K. Pal. "A review on image segmentation techniques." (Pattern recognition 26.9 (1993): 1277-1294; Sharma, Neeraj, and Lalit M. Aggarwal.), "Automated medical image segmentation techniques." (Journal of medical physics 35.1 (2010): 3;), or Ramesh, KKD, et al. "A review of medical image segmentation algorithms." (EAI Endorsed Transactions on Pervasive Health and Technology 7.27 (2021): e6-e6).

[0058] The method 100 also includes determining 130 the position of the midsagittal axis such that the midsagittal axis intersects the first portion of the image and extends perpendicular to the edge of the bounding box that is closest to the first portion of the image.

[0059] In this manner, the midsagittal axis is positioned to extend parallel to two parallel edges of the bounding box that are perpendicular to the edge of the bounding box that is closest to the first portion of the image. In this manner, the midsagittal axis may be perpendicular to the edge of the bounding box that is closest to the first portion and / or that is farthest from the first portion.

[0060] In some examples, step 130 includes determining the position of the mid-sagittal axis to start at the first portion of the image and extend away from (and perpendicular to) the edge (of the bounding box) closest to the first portion of the image, hi some examples, step 130 includes ending the mid-sagittal axis at the edge (of the bounding box) farthest from the first portion of the image.

[0061] In this way, the midsagittal axis can start at the first portion of the image and terminate at the edge (of the bounding box) that is furthest from the first portion of the image. The midsagittal axis is perpendicular to the closest edge of the bounding box (with respect to the first portion of the image) and is therefore perpendicular to the furthest edge of the bounding box.

[0062] In some examples, determining the position of the midsagittal axis is configured such that the midsagittal axis intersects with a center point or center of gravity of the first portion of the image. This provides a simple technique for defining the intersection with the first portion of the image. For example, the midsagittal axis may originate from the center point or center of gravity of the first portion of the image and extend away from (and perpendicular to) the edge of the bounding box that is closest to the first portion of the image.

[0063] The embodiment is based on the recognition that the axis that intersects the spine (in a 2D plane) and extends in the general direction of the internal thorax away from the spine represents the sagittal medial axis. More specifically, the use of the internal thorax to perform the calculation avoids any discontinuities or errors caused by calcified tissue.

[0064] According to the medical literature, the midsagittal axis or line is the line that divides the representation of the chest region in an image into two equal halves, i.e., bisecting the chest region. This is a complex task to accomplish using purely automated means. The techniques proposed above have been identified as providing excellent accuracy in determining the location of the midsagittal line.

[0065] More specifically, the proposed approach utilizes a subanatomical structure of the type defined herein, namely, the internal thorax. This is effectively the portion of the thorax that remains after removing any bony structures (e.g., ribs and vertebrae) and their extraneous tissue. The identification or segmentation of the spinal anatomy is used as an anchor or intersection point that identifies the point where the midsagittal axis should intersect, e.g., the start of the midsagittal axis. The line that is perpendicular to the edge of the rectangle closest to the spine contour and that passes through the (centroid of) the spine contour is used as the sagittal medial axis.

[0066] Method 100 may further include controlling an output interface (not shown) to provide a visual representation of the medical image and / or the mid-sagittal axis. Specifically, this may include providing a representation of the mid-sagittal axis overlaying the medical image at an associated position in the visual representation of the acquired medical image, as determined using the preceding steps of method 100. This provides a clinician with useful information for conducting a clinical evaluation or analysis of the subject.

[0067] Figure 2 shows an example image 200 with features identified or determined by the proposed approach, conceptually illustrated, and is provided for improved conceptual understanding of the proposed technique.

[0068] It is understood that any image that provides a four-chamber view of the heart (of a subject) should include a view of a portion of the subject's spine and a view of the subject's internal thorax. Thus, image 200 can be assumed to include a representation of the spine and a representation of the internal thorax.

[0069] The first contour 210 identifies or bounds a representation of a portion of the spine in the example image. Thus, the first contour 210 comprises a first portion of the image that represents the portion of the spine in the example image.

[0070] The second contour 220 identifies or bounds the representation of the internal thorax within the exemplary image 200. Thus, the second contour includes a second portion of the image that represents the internal thorax of the subject.

[0071] Bounding box 230 is a bounding box having edges that intersect with the boundary of a second portion of the image representing the subject's internal thorax. Bounding box 230 is formed in the shape of a square or rectangle (i.e., a cuboid).

[0072] The first dashed line 240 represents the subject's mid-sagittal axis. As shown in Figure 2, the first dashed line 240 (and thus the mid-sagittal axis) is positioned so that it intersects the first contour (i.e., the first portion of the image) and is perpendicular to the nearest and / or farthest edge of the first contour 210 or boundary from the first portion of the image.

[0073] It is recognized that not all medical images (e.g., of a fetal heart) are suitable for processing to generate an accurate midsagittal axis and / or to generate measurements derived using the midsagittal axis. In particular, it is recognized that the accuracy of identifying the midsagittal axis (and / or measurements derived therefrom) is improved when the medical image provides a suitably accurate four-chamber view of the heart, a representation of the entire chest, and / or a representation of relevant or desired anatomical structures for performing the proposed method. For purposes of determining the midsagittal axis, relevant anatomical structures include portions of the spine (of a subject) and the internal thorax (of a subject).

[0074] By way of explanation, when only a portion of the thorax is included in a medical image, identifying the location of the midsagittal axis is prone to error. The orientation of the midsagittal axis depends on a bounding box placed on the thorax region. When the thorax is only partially represented, there is insufficient data to process to identify the location of the bounding box, which can lead to incorrect positioning of the bounding box.

[0075] In the example where the medical image is an ultrasound image, the image may contain only a partial representation of the rib cage if it was acquired in a zoomed view, or if there is, for example, some shadow along its border that blocks the view of the rib cage.

[0076] 1, for improved performance, step 110 can be adapted to suppress or limit the acquired images to meet one or more of these criteria, i.e., one medical image may be selected from multiple possible medical images.

[0077] FIG. 3 is a flow chart illustrating one embodiment of a process for selecting one of a plurality of medical images as the medical image for purposes of performing step 110. As shown in FIG.

[0078] In this embodiment, step 110 may include step 310 of acquiring a plurality of potential medical images of the subject. The potential medical images may be acquired from a storage unit, memory, or a medical imaging device (such as an ultrasound machine). In particular, the potential medical images may include a series of images captured during a medical imaging procedure.

[0079] The possible medical images may represent, for example, different 2D slices of a 3D image of the thorax, in a preferred example each 2D slice being perpendicular to the spine or spinal axis of the subject.

[0080] Step 110 can then include processing (320, 330) the plurality of potential medical images to select one of the potential medical images as the medical image. Thus, one of the potential medical images can be selected as the medical image.

[0081] One approach for performing such a processing procedure comprises an iterative process 320 that operates to eliminate any medical images (out of multiple possible images) that do not meet the criteria.

[0082] For example, the iterative process 320 may operate to process each potential medical image to determine whether the potential medical image represents a four-chamber view plane, and to eliminate each potential medical image determined not to represent a four-chamber view plane.

[0083] In some examples, process 320 operates to process each potential medical image to determine whether the potential medical image includes a representation of the subject's entire chest, and removes each potential medical image that is determined not to include a representation of the subject's entire chest.

[0084] In some examples, process 320 may operate to process each potential medical image to determine whether the potential medical image includes a representation of a desired anatomical structure of the subject, and to remove each potential medical image determined to not include a representation of the desired anatomical structure. For purposes of determining the midsagittal axis, relevant anatomical structures include a portion of the spine (of the subject) and the internal thorax (of the subject).

[0085] One exemplary technique for performing process 320 is described below.

[0086] The process includes a step 321 of selecting a next possible image from a plurality of possible images, for example, a sequence of possible images or a sequence of possible images.

[0087] The process then moves to step 322, where it is determined whether the selected potential image provides a four-chamber view of the (fetal) heart. Methods for performing such a step are described, for example, in International Patent Application Publication No. WO 2022 / 243178 A1; Sundaresan, Vaanathi, et al., "Automated characterization of the fetal heart in ultrasound images using fully convolutional neural networks" (2017 IEEE 14th International Symposium on Biomedical Imaging (ISBI 2017). IEEE, 2017); or Park, Jin Hyeong, et al., "Automated cardiac view classification of echocardiogram" (2007 IEEE 11th International Conference on Computer Vision. IEEE, 2007).

[0088] In response to determining in step 322 that the potential medical image does not provide a four-chamber view of the heart, the potential medical image is removed in step 329 and the next potential medical image is selected in step 321. Otherwise, the method continues.

[0089] The process then proceeds to step 323, where it is determined whether the selected potential image includes a representation of the subject's entire chest.

[0090] In a simple example, step 323 may be performed using a suitably trained machine learning method.

[0091] In a more detailed example, the following three-step algorithm, based on simple image processing techniques, can be used to detect the presence or absence of a whole breast in a potential medical image. The algorithm described below is specifically designed for use with ultrasound images, which are an example of a medical image that may be used in some embodiments, but is equally applicable to other forms of medical imaging.

[0092] First, a fan beam contour is detected. The fan beam represents the boundary of the entire anatomical region represented in the medical image. It will be apparent that a portion of a conventional ultrasound image includes a sector that contains a representation of the imaged anatomical region. The area outside this sector may contain other information (e.g., anatomical information, subject data, and / or annotations).

[0093] One approach to finding the fan beam contours is to first find all closed contours in the ultrasound image. A closed contour is a contour that encloses only non-zero or intensity values ​​in the ultrasound image, a term well known in the art. An example of a suitable function is "findContours" used in OpenCV. A convex hull can then be fitted over the largest identified contours. The convex hull represents the fan beam boundary, i.e., acts as the fan beam contour.

[0094] An internal breast contour is determined. The internal breast contour is a contour that surrounds or bounds (or is predicted to surround or bound) a representation of the internal breast in the medical image. The internal breast contour may be determined by processing the potential medical image with an appropriately trained machine learning algorithm, such as a trained U-net neural network. Accordingly, image segmentation may be performed on the potential medical image to identify the neural network.

[0095] It is then determined whether there is an intersection between the fan beam contour and the internal breast contour. The occurrence of an intersection indicates that the potential medical image does not contain a representation of the entire breast. Thus, the absence of an intersection indicates that the potential image does contain a representation of the entire breast.

[0096] To increase robustness against false positives (i.e., erroneous identification that a potential medical image contains a representation of the entire chest), the fan beam contour may be reduced or scaled down in size (for example), or the inner chest contour may be expanded or scaled up in size (e.g., dilated), or both, before determining any intersections between the fan beam contour and the inner chest contour. By way of example only, the inner chest contour may be dilated with a kernel size of 10. This effectively means that the size of the inner chest contour is increased by going around a contour with a thickness of 5 pixels. This smooths the inner chest contour boundary and ensures that the entire inner chest is considered for analysis (i.e., increases the likelihood that the inner chest contour includes the entire inner chest). As another example, the fan beam contour may be reduced by a first predetermined percentage, or the inner chest contour may be reduced by a second predetermined percentage, or both. The first and / or second predetermined percentages may each be between 5% and 20%, for example between 5% and 10%.

[0097] If it is determined in step 323 that the potentially selected image does not include a representation of the subject's entire chest, the potentially selected image is removed in step 329 and a new potential image is selected in step 321. Otherwise, process 320 continues.

[0098] The process then proceeds to step 324, where it is determined whether the selected potential image contains a representation of the relevant anatomical structures on which subsequent processing will be relied upon. For example, in this embodiment, in which the midsagittal axis is identified, step 324 may include determining whether the potential image contains a representation of a portion of the spine and / or internal thorax.

[0099] In a simple example, step 324 may be performed using one or more appropriately trained machine learning methods. For example, one or more classification algorithms may process a medical image to predict whether the medical image contains a representation of a particular object. Exemplary classification algorithms are well known in the art and include any appropriately trained machine learning network.

[0100] If it is determined in step 324 that the selected potential image does not contain a representation of each of the relevant anatomical structures, the selected potential image is removed in step 329 and a new potential image is selected in step 321. Otherwise, process 320 continues.

[0101] Of course, any of steps 322, 323, and 324 may be omitted in some embodiments, for example, if it is known in advance that a potential medical image meets the requirements of said steps. It will also be understood that steps 323, 323, and 324 may be performed in any order or even combined.

[0102] In some examples, the method then performs a check step 325 to check whether all medical images have been processed. In response to all medical images having been processed, the method may then proceed to step 330 to select one of the remaining potential medical images as the medical image. If not, the method may return to step 321.

[0103] Alternatively, the method may move directly to step 330 without performing step 325, i.e., the first medical image identified as meeting the relevant criteria is selected as the medical image to be processed.

[0104] If step 325 is performed, step 330 may include performing a random or pseudo-random selection of one of the potential medical images as the medical image to be processed. Approaches for performing such random or pseudo-random selection of one or more options are well established in the art. Alternatively, step 330 may select one or more of the medical images based on one or more characteristics of the medical images, such as average contrast or average brightness.

[0105] In some examples, if step 325 is performed, step 330 may include performing a quality analysis of all remaining potential medical images and then selecting the potential medical image with the best quality assessment value for further processing. Approaches for determining image quality values ​​are well known in the art, such as those disclosed by U.S. Patent Application Publication No. 2023 / 068399 A1, International Patent Application No. WO 2022 / 096471 A1, or European Patent Application Publication No. 4,080,449 A1.

[0106] Of course, if no potential images remain after performing process 320, the method can end. Alternatively, the entire process 320 with the first potential image acquired may be repeated without performing one of steps 322, 323, and 324 (if present). For example, step 323 may be omitted.

[0107] In some examples, if no possible images remain after performing process 320, a user-perceptible warning may be generated, for example, in a user interface.

[0108] In some examples, if no possible images remain after performing process 320, one of the possible images may nevertheless be selected and further processed (i.e., referring to FIG. 1, step 120 may be performed on the possible image).

[0109] In such an example, rather than determining the location of the midsagittal axis (i.e., performing step 130), a user-adjustable representation of the midsagittal axis may be provided in a user interface overlaying the display of a selected one of the potential medical images. The user can then manipulate the relative position of the midsagittal line by interacting with the representation to define the location of the midsagittal axis. The initial positioning of the representation may be such that it is centered on the internal chest to reduce the user's burden when moving or manipulating the representation.

[0110] FIG. 4 illustrates a computer-implemented method 400 for determining cardiac angle measurements in a subject that utilizes the above-described method 100 for determining the location of the midsagittal axis.

[0111] The computer-implemented method includes performing process 410 of processing the medical image (acquired in step 110 of method 100) using one or more second machine learning methods to determine the location of the cardiac axis. The cardiac axis is an axis extending in the direction of a representation of the subject's ventricular (IV) septum in the medical image. An alternative label for the cardiac axis is the cardiac line.

[0112] The computer-implemented method then performs step 420 of determining the angle between the midsagittal axis and the cardiac axis as the cardiac angle measurement.

[0113] In a simple example, process 410 involves processing medical images to directly predict the cardiac axis using a suitable trained machine learning algorithm, an example of a suitable machine learning algorithm is a machine learning algorithm with a U-net architecture.

[0114] In a preferred example, process 410 includes step 411 of processing the medical image using one or more second machine learning methods to identify a third portion of the image representing the subject's ventricular (IV) septum.

[0115] Thus, step 411 may include performing a segmentation process on the medical image to identify, as a segmentation result, such as a contour or bounding box, a portion that contains or bounds a representation of the subject's IV septum. The third portion is thus defined or bounded by the segmentation result. This process may be performed by a suitably trained second machine learning method or algorithm.

[0116] Next, process 410 performs step 412 of fitting a line to a third portion of the image that intersects with the midsagittal axis to determine the location of the cardiac axis. Procedures for fitting lines to segmentation results (i.e., contours or bounding boxes) are established in the art and generally operate to identify the location of a line that intersects the segmentation result (bounding the third portion of the image) while maintaining the maximum average distance (for all points along the line) from the boundary of the segmentation result. One suitable function for performing step 412 is the "fit line" function used in OpenCV. Generally, such a function can act as a "best fit line" and is well established in the field of image processing.

[0117] Step 420 can be trivially performed by determining or measuring the smallest angle between the midsagittal axis and the cardiac axis. In other words, after calculating both axes (or lines), the deviation or angle (in degrees or radians) between the two axes is the cardiac angle measurement.

[0118] The method 400 may further include controlling 430 an output interface to provide a visual representation of the determined cardiac angle measurements, which provides a clinician with useful information for performing a clinical evaluation or analysis of the subject.

[0119] 5 shows an example image 500 with features identified or determined by the proposed technique for identifying the location of the cardiac axis. FIG. 2 is provided for improved conceptual understanding of the proposed technique.

[0120] It is understood that any image that provides a four-chamber view of the heart (of a subject) should include a view of the subject's IV septum. Thus, image 500 can be assumed to include a representation of the subject's IV septum.

[0121] A first dashed line 240 representing the subject's midsagittal axis is again shown for improved contextual understanding.

[0122] The third contour 510 identifies or bounds a representation of the IV septum in the example image. Thus, the third contour 510 includes a third portion of the image that represents the IV septum in the example image.

[0123] A second dashed line 520 represents the subject's cardiac axis and represents a line fit to the third portion of the image.

[0124] The angle θ between the first dashed line 240 and the second dashed line is used as a cardiac measurement.

[0125] A first dashed line 240 represents the subject's mid-sagittal axis. As shown in Figure 2, the first dashed line 240 (and thus the mid-sagittal axis) is positioned so that it intersects the first contour (i.e., the first portion of the image) and is perpendicular to the edge of the bounding box that is closest to the first contour 210 or first portion of the image.

[0126] 3, it was previously explained how, for improved performance, step 110 can be adapted to suppress or limit the acquired images to meet one or more criteria to achieve improved accuracy in identifying the midsagittal axis. The same principles apply to improving accuracy in identifying the cardiac axis. It is therefore advantageous if the acquired images also meet the cardiac axis criteria.

[0127] In particular, it has been recognized that the accuracy of identifying the cardiac axis (and / or measurements derived therefrom) is improved when the medical image provides an appropriately accurate four-chamber view of the heart, a representation of the entire chest, and / or a representation of the relevant or desired anatomical structures for performing the proposed method. For purposes of determining the cardiac axis, the relevant anatomical structures include the subject's IV septum.

[0128] Thus, step 324 may be suitably adapted to check whether the potential medical image includes an adequate representation of the subject's IV septum.

[0129] More specifically, when segmentation of the IV septum is used to determine the cardiac axis, the accuracy of calculating the position of the cardiac axis depends on the accuracy of the segmentation result that bounds the representation of the IV septum (i.e., the identification of the third portion of the medical image) (i.e., referring to FIG. 4, the accuracy of step 411). If the image quality is poor, the visibility of the IV septum in the image will be poor, resulting in inaccurate or imprecise identification of the third portion of the image (i.e., inaccurate segmentation). This will lead to an inaccurate calculation of the position of the cardiac axis.

[0130] One approach for checking whether a potential medical image contains an adequate representation of a subject's IV septum is described below.

[0131] First, the medical image is processed (e.g., using one or more segmentation techniques, such as machine learning methods) to identify a cardiac contour, which is a contour or segmentation result that bounds a portion of the medical image that represents the subject's heart. The cardiac contour is configured to be elliptical in shape.

[0132] It is then checked whether the heart contour is sufficiently eccentric, for example if a measure of the eccentricity of the heart contour violates some predetermined value.

[0133] In one example, the measure of eccentricity is determined by dividing the length of the minor axis of the heart contour by the length of the major axis, such that a lower measure of eccentricity indicates a more eccentric shape. Thus, determining whether the measure of eccentricity of the heart contour violates some predetermined value may include determining whether the measure of eccentricity is less than the predetermined value, for example, less than 0.7.

[0134] Other approaches to determining efficiency, in contrast, may indicate increasing eccentricity to increase the eccentricity measurement, and the method may be modified accordingly (as would be known to one skilled in the art).

[0135] If the cardiac contour is determined to be sufficiently eccentric (e.g., the eccentricity measure violates a certain predetermined value), then it is determined whether the angle between the major axis of the ellipse (which acts as the cardiac axis) and the determined cardiac axis (e.g., calculated using step 410 above) is less than a predetermined angle, e.g., less than 40 degrees or less than 30 degrees. The example values ​​of angles given in this context are merely exemplary, and one skilled in the art would be able to easily modify any such values ​​for use in various use case scenarios.

[0136] In response to the determined angle being less than the predetermined angle, the potential medical image is deemed to include an adequate representation of the subject's IV septum. Of course, this means that step 410 need not be repeated in subsequent processing of the medical image to avoid redundant processing.

[0137] In response to the determined angle being greater than a predetermined angle, the potential image is deemed not to contain an adequate representation of the subject's IV septum. Thus, the potential medical image can be rejected. Alternatively, the potential medical image can be subjected to further processing, such as another approach to checking for suitability, as outlined below.

[0138] If the cardiac contour is not sufficiently elliptical (e.g., the eccentricity measure does not exceed a predetermined value), the cardiac axis should be determined in another way. To do this, segmentation is performed on the medical image to identify the locations of the right atrium (RA), left atrium (LA), left ventricle (LV), and right ventricle (RV) chambers.

[0139] Procedures for performing such segmentation are Arafati, Arghavan, et al. "Multi-label 4-chamber segmentation of echocardiograms using fully convolutional networks." (2018) or Cao, Yu, Patrick McNeillie, and Tanveer Syeda-Mahmood. "Segmentation of anatomical structures in four-chamber view echocardiogram images." (2014 22nd International conference on pattern recognition. IEEE, 2014).

[0140] It is then possible to define multiple possible cardiac axes, each of which is parallel to a line intersecting the LA and LV or the RA and RV, or is parallel to an axis or line that is perpendicular to the line intersecting the LA and RA or the LV and RV.

[0141] It is then determined whether the angle between any potential cardiac axis and the determined cardiac axis (e.g., calculated using step 410 above) is less than a predetermined angle, e.g., less than 40 degrees or less than 30 degrees. The example values ​​of angles given in this context are merely exemplary, and one skilled in the art would be able to easily modify any such values ​​for use in various use case scenarios.

[0142] In response to either of these angles (between the potential and determined cardiac axes) being less than a predetermined angle, the potential medical image is deemed to contain an adequate representation of the subject's IV septum. Of course, this means that step 410 need not be repeated in subsequent processing of the medical image to avoid redundant processing.

[0143] In response to any of these angles (between the potential cardiac axis and the determined cardiac axis) being greater than a predetermined angle, the potential image is deemed not to contain an adequate representation of the subject's IV septum.

[0144] Therefore, potential medical images may be eliminated in some instances. Alternatively, potential medical images may be subjected to further processing, e.g., other approaches to checking suitability, as outlined below.

[0145] Another approach for checking whether a potential medical image contains an adequate representation of a subject's IV septum is described below, which can be performed independently of the previous approach or, as mentioned above, if the previous approach fails.

[0146] It is recognized that for accurate segmentation, the mitral valve (MV) and tricuspid valve (TV) should be aligned or nearly aligned with one another. Each of these valves defines a valve axis along which the valve lies. Blood flow is generally perpendicular to the valve axis. The cardiac axis should be perpendicular or nearly perpendicular to the valve axes of both of these valves. It is determined whether these criteria are met.

[0147] Thus, it can be determined whether the angle between the MV and the TV is less than a first predetermined angle (e.g., less than 45 degrees or less than 30 degrees), whether the angle between the MV and the cardiac axis is greater than a second predetermined angle (e.g., between 60 degrees and 135 degrees or between 70 degrees and 90 degrees), and whether the angle between the TV and the cardiac axis is greater than a second predetermined angle (e.g., between 60 degrees and 135 degrees or between 70 degrees and 90 degrees). The example values ​​of angles given in this context are merely exemplary, and one skilled in the art could easily modify any such values ​​for use in various use case scenarios.

[0148] Depending on whether these criteria are met, it may be determined that the potential medical image is suitable for use in identifying the location of the cardiac axis, i.e., should not be rejected for this reason alone.

[0149] In an example utilizing the method of acquiring medical images as shown in FIG. 3 , if a potential medical image is not identified but continues to be processed anyway, rather than determining the location of the midsagittal axis (i.e., performing step 130) and / or determining the cardiac axis (i.e., performing step 410), a user-adjustable representation of the midsagittal axis and / or cardiac axis may be provided in a user interface overlaying the display of the medical image. The user can then manipulate the relative positions of the midsagittal line and / or cardiac axis by interacting with the representation to define the location of the midsagittal axis and / or cardiac axis. The initial positioning of each representation may be centered within the internal chest to reduce user effort when moving or manipulating the representation.

[0150] 3, process 320 is performed on each of a plurality of possible medical images to identify the medical image to be processed. However, the principles of process 320 may instead be applied to a single image, e.g., a user-identified image, to accept or reject the single image for subsequent processing.

[0151] In other words, a method of acquiring medical images is proposed that includes receiving a single potential medical image and determining whether to accept or reject the single potential image by performing steps 322, 323 and / or 324. This single potential image may be a medical image identified by a user or clinician.

[0152] Of course, in a further example, if it is determined that a single potential image should be eliminated, it is still possible to perform the proposed approach using that potential image, but this preferably with a user-perceivable indicator (e.g., a visual warning or other form of indicator) that any results may be inaccurate. The user-perceivable indicator may indicate the reason for the failure of the single potential image, for example, why it does not provide a four-chamber view.

[0153] In such an example, rather than determining the location of the midsagittal axis (i.e., performing step 130) and / or determining the location of the cardiac axis (i.e., performing step 410), a user-adjustable representation of the midsagittal axis and / or cardiac axis may be provided in a user interface overlaying the display of the medical image. The user can then manipulate the relative positions of the midsagittal line and / or cardiac axis by interacting with the representation to define the location of the midsagittal axis and / or cardiac axis. The initial positioning of each representation may be centered in the internal chest to reduce user effort when moving or manipulating the representation.

[0154] In some examples, if it is determined that a single potential image should be removed, a user-perceptible warning may be generated, for example, in a user interface.

[0155] In all of the above methods, a step of acquiring a medical image is performed. As mentioned above, the medical image provides a four-chamber view plane of the heart. Thus, the medical image may be a 2D medical image or a 2D slice of a 3D medical image (to provide the four-chamber view plane).

[0156] In one example, acquiring the medical image includes extracting 2D planes from a 3D image of the thorax, which may be performed using, for example, the following procedure: identifying the spinal axis using spinal triangulation / body detection techniques, slicing the 3D image into multiple 2D planes perpendicular to the spinal axis, and processing the multiple 2D planes to select one of the 2D planes that provides a four-chamber view as the medical image.

[0157] This approach can be viewed as a variation of the method 110 shown in Figure 3. In particular, step 310 in the method 110 shown by Figure 3 can include acquiring a 3D image of the thorax, identifying the location of the spinal axis within the 3D image, and extracting multiple 2D slices of the 3D image (which then act as potential medical images). Process 320 can then be performed to identify 2D images that meet desired requirements.

[0158] The proposed embodiments may utilize one or more machine learning algorithms to perform tasks on the images, such as segmentation tasks, classification tasks, or axis localization tasks.

[0159] A machine learning algorithm is any self-training algorithm that processes input data to generate or predict output data. In the context of this disclosure, input data includes medical images, and output data includes results of tasks performed on the medical images (e.g., segmentation results, classification results, etc.).

[0160] Machine learning algorithms suitable for use in the present 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, such as logistic regression, support vector machines, or naive Bayesian models, are suitable alternatives.

[0161] The structure of an artificial neural network (or simply a neural network) is inspired by the human brain. A neural network is composed of layers, each layer containing multiple neurons. Each neuron contains a mathematical operation. In particular, each neuron may comprise a different weighted combination of a single type of transformation (e.g., the same type of transformation, such as sigmoid, but with different weightings). In the process of processing input data, the mathematical operation of each neuron is performed on the input data to generate a numerical output, and the output of each layer in the neural network is sequentially fed to the next layer. The final layer provides the output.

[0162] Methods for training machine learning algorithms are well known. Typically, such methods involve obtaining a training dataset including 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 error between the predicted output data entry and the corresponding training output data entry is used to modify the machine learning algorithm. This process can be repeated until the error converges and the predicted output data entries are sufficiently similar to the training output data entries (e.g., ±1%). This is commonly known as a supervised learning technique.

[0163] For example, if a machine learning algorithm is formed from a neural network, the mathematical operations (weights) of each neuron can be modified until the error converges. Known methods for modifying neural networks include gradient descent, backpropagation algorithms, etc.

[0164] A particularly advantageous form of machine learning method for use in performing segmentation tasks is a neural network with a U-net architecture. An overview of the U-net architecture and its applicability to medical image segmentation is provided by Siddique, Nahian, et al. "U-net and its variants for medical image segmentation: A review of theory and applications" (IEEE Access 9 (2021): 82031-82057). Those skilled in the art will be able to readily use this document or documents cited therein to implement machine learning algorithms capable of performing segmentation tasks.

[0165] However, the performance of segmentation tasks is not limited to machine learning methods with U-net architectures. Another suitable example of a usable segmentation technique is Seo, Hyunseok, et al. "Machine learning techniques for biomedical image segmentation: an overview of technical aspects and introduction to state-of-the-art applications." (Medical physics 47.5 (2020): e148-e167).

[0166] Tchito Tchapga, Christian, et al. "Biomedical image classification in a big data architecture using machine learning algorithms." (Journal of Healthcare Engineering 2021 (2021): 1-11) describes various approaches for performing classification tasks using various machine learning techniques. Those skilled in the art will be able to use any such mechanism in an appropriate embodiment.

[0167] 6 illustrates a system 600 according to one embodiment. The system includes a processing system 610, which is itself an embodiment of the present invention.

[0168] The processing system 610 is configured to perform any of the methods described herein. Accordingly, each step of the flowchart may represent a different action performed by the processing system and may be performed by a respective module of the processing system.

[0169] The system 600 may also include a memory 620 and / or a medical imaging system 630 configured to generate medical images acquired by the processing system 610. Thus, the processing system may acquire medical images, or potential medical images if relevant, from the memory and / or the medical imaging system.

[0170] System 600 also includes an output interface 640, such as a display or screen. Processing system 610 may be configured to control the operation of output interface 640 to control the visual representation of information provided to a clinician or other user. The visual representation may include a visual representation of a medical image, a mid-sagittal axis, a cardiac axis, and / or a cardiac angle measurement, where relevant, depending on the embodiment. Techniques for controlling the visual representations provided on a display or screen are well established in the art.

[0171] In one example, system 600 is an ultrasound system comprising at least a processing system 610, a medical imaging system 630 embodied as an ultrasound probe, and an output interface 640. The ultrasound probe generates medical images in the form of ultrasound images for processing by the processing system. Thus, the images obtained by processing system 610 are ultrasound images generated by the ultrasound probe.

[0172] Examples of suitable ultrasound probes for producing medical ultrasound images are well established in the art, and techniques for obtaining four-chamber views of the heart using ultrasound probes are similarly well established and known.

[0173] The processing system 610 can be implemented in numerous ways using software and / or hardware to perform the various functions required. The processor is one example of a processing system that uses one or more microprocessors that can be programmed using software (e.g., microcode) to perform the necessary functions. However, the processing system can be implemented with or without a processor, and can also be implemented as a combination of dedicated hardware to perform some functions and processors (e.g., one or more programmed microprocessors and associated circuitry) to perform other functions.

[0174] Examples of processing system components that may be used in various embodiments of the present disclosure include, but are not limited to, conventional microprocessors, application specific integrated circuits (ASICs), and field programmable gate arrays (FPGAs).

[0175] In various implementations, a processor or processing system may be associated with one or more storage media, such as volatile and non-volatile computer memory, such as RAM, PROM, EPROM, and EEPROM. The storage media may be encoded with one or more programs that, when executed on the one or more processors and / or processing systems, perform the necessary functions. The various storage media may be fixed within the processor or processing system or may be portable, such that the one or more programs stored thereon can be loaded into the processor or processing system.

[0176] It will be understood that the disclosed methods are preferably computer-implemented methods.

[0177] Thus, the concept of a computer program is also proposed, comprising code means for implementing any described methods when said program is executed on a processing system such as a computer. Thus, different parts, lines or blocks of code of a computer program according to an embodiment may be executed by a processing system or a computer in order to perform the methods described herein.

[0178] Also proposed is a non-transitory storage medium that stores or carries a computer program or computer code that, when executed by a processing system, causes the processing system to perform any of the methods described herein.

[0179] In some alternative implementations, the functions noted in the block diagrams or flowcharts may occur out of the order noted in the figures. For example, two blocks shown in succession may in fact be executed substantially concurrently or the blocks may sometimes be executed in the reverse order, depending upon the functionality involved.

[0180] Ordinal numbers (e.g., "first," "second," etc.) are used purely to distinguish between distinct elements for clarity, and reference to a non-"first" (e.g., "second" or "third") element does not require that a "first" element be present. One skilled in the art can re-label any such element appropriately (e.g., re-label a "second" element as a "first" element if only a second element is present).

[0181] Variations to the disclosed embodiments can be understood and effected by those skilled in the art in practicing the claimed invention, from a study of the drawings, the disclosure and the appended claims. In the claims, the word "comprising" does not exclude other elements or steps, and the indefinite articles "a" or "an" do not exclude a plurality. A single processor or other unit may fulfill the functions of several items recited in the claims. The mere fact that certain measures are recited in mutually different dependent claims does not indicate that a combination of these measures cannot be used to advantage.

[0182] Where a computer program is described above, it may be stored / distributed on a suitable medium such as an optical storage medium or a solid-state medium supplied together with or as part of other hardware, but may also be distributed in other forms, such as via the Internet or other wired or wireless telecommunications systems.

[0183] It should be noted that when the term "adapted for" is used in the claims or description, it is intended to be equivalent to the term "configured for." When the term "apparatus" is used in the claims or description, it is intended to be equivalent to the term "system," and vice versa. Any reference signs in the claims should not be construed as limiting the scope.

Claims

1. 1. A computer-implemented method for determining the location of a midsagittal axis on a medical image providing a four-chamber view plane of the heart, the computer-implemented method comprising: acquiring the medical image, the medical image including a representation of the subject's thorax in the four-chamber view plane; Processing the medical image a first portion of the image representing a portion of the subject's spine; a bounding box having an edge that intersects with a boundary of a second portion of the image representing the subject's internal thorax, the internal thorax including only the subject's soft tissue contained within the subject's thorax; and identifying determining a position of the midsagittal axis such that the midsagittal axis intersects a first portion of the image and extends perpendicular to an edge of a bounding box that is closest to the first portion of the image; A method comprising:

2. 2. The computer-implemented method of claim 1, wherein determining the position of the midsagittal axis is configured such that the midsagittal axis originates from a first portion of the image and extends away from an edge of the bounding box that is closest to the first portion of the image.

3. 3. The computer-implemented method of claim 1, wherein determining the position of the mid-sagittal axis is further configured such that the mid-sagittal axis intersects with a center of gravity of a first portion of the image.

4. 4. The computer-implemented method of claim 1, wherein processing the medical images comprises processing the medical images using one or more machine learning methods.

5. 5. The computer-implemented method of claim 1, wherein the internal thorax excludes any calcified tissue such as bone.

6. The computer-implemented method of claim 1 , wherein the medical image is an ultrasound medical image.

7. The step of acquiring a medical image includes: acquiring a plurality of possible medical images of the subject; processing the plurality of potential medical images to select one of the potential medical images as the medical image; 7. The computer-implemented method of claim 1, comprising:

8. The step of processing the plurality of possible images comprises: processing each potential medical image to determine whether the potential medical image represents a four-chamber view plane; removing each potential medical image determined not to represent a four-chamber view plane; 8. The computer-implemented method of claim 7, comprising:

9. The step of processing the plurality of possible images comprises: processing each potential medical image to determine whether the potential medical image includes a representation of the subject's entire chest; removing each possible medical image determined not to contain a representation of the subject's entire chest; 9. The computer-implemented method of claim 7 or 8, comprising:

10. 10. The computer-implemented method of claim 1, wherein the medical image provides a four-chamber view plane of a fetal heart and includes a representation of the rib cage of the fetal subject within the four-chamber view plane.

11. 1. A computer-implemented method for determining a cardiac angle measurement in a subject, the computer-implemented method comprising: determining the position of the midsagittal axis on a medical image in a four-chamber view plane by carrying out the method according to any one of claims 1 to 10; processing the medical image using one or more second machine learning methods to determine the location of the cardiac axis; determining the angle between the midsagittal axis and the cardiac axis as the cardiac angle measurement; 10. A computer-implemented method comprising:

12. The computer-implemented method of claim 1 , wherein the bounding box is rectangular.

13. 13. A computer program product comprising computer program code means which, when executed on a computing device having a processing system, causes said processing system to perform all of the steps of the method of any one of claims 1 to 12.

14. 1. A processing system for determining the location of a midsagittal axis on a medical image providing a four-chamber view plane of a heart, the processing system comprising: acquiring the medical image, the medical image including a representation of the subject's thorax in the four-chamber view plane; Processing the medical image a first portion of the image representing a portion of the subject's spine; a bounding box having an edge that intersects with a boundary of a second portion of the image representing the subject's internal thorax, the internal thorax including only the subject's soft tissue contained within the subject's thorax; and identifying determining a position of the midsagittal axis such that the midsagittal axis intersects a first portion of the image and extends perpendicular to an edge of a bounding box that is closest to the first portion of the image; a processing system configured to execute the

15. A processing system according to claim 14; an ultrasound probe configured to generate medical images for acquisition by the processing system; an output interface configured to provide a visual representation of the midsagittal axis and / or values ​​of parameters derived from the midsagittal axis; An ultrasound system comprising:

Images