Detecting anatomical anomalies in 2D medical images

Abstract

The invention relates to detecting anatomical anomalies in medical images. In order to detect anatomical anomalies, a computer-implemented method (100) and system is disclosed that detects 2D contours (130) of anatomical features in a medical image and compares these contours to 2D contours (140) predicted based on a 3D reference model in order to detect (150) anatomical anomalies. This approach can improve the accuracy of anatomical anomaly detection, thereby shortening the time to be in a medical facility and potentially improving the operator's experience and patient's prognosis.

Description

Technical Field

[0001] The present invention relates to a computer-implemented method, a computer program product and a system configured for detecting anatomical abnormalities in 2D medical images. Background Art

[0002] Medical images are often used for diagnostic purposes. For example, human observers observe medical images to check for anatomical abnormalities. Examples of such abnormalities include bone fractures and tumor tissue, which can be identified in X-ray images, ultrasound images, CT (computed tomography) images, or other medical images.

[0003] However, review of medical images by human observers is time consuming and prone to error, depending on the observer's expertise, experience, time pressure, and fatigue. This is particularly true in the hectic environment of a busy clinic and when anatomical abnormalities are not easily discernible.

[0004] US2022 / 0044041 discloses an algorithm for detecting fractures, classifying fractures, and providing guidance on how to treat fractures. Summary of the invention

[0005] In particular, it is an object of the invention to provide a computer implemented anatomical abnormality detection in medical imaging. The invention is defined by the independent claims. Advantageous embodiments are defined in the dependent claims.

[0006] One aspect of the present invention provides a computer-implemented method for detecting anatomical abnormalities in 2D medical images. The method comprises:

[0007] Obtaining the 2D medical image;

[0008] Detect 2D contours of anatomical features;

[0009] obtaining a 3D model representing the anatomical feature;

[0010] predicting a 2D contour of the anatomical feature based on the 3D model;

[0011] The anatomical abnormality is detected based on the detected 2D contour and the predicted 2D contour. In this way, the 2D contour detected in the X-ray image can be compared with the known reference contour obtained from the 3D model to detect and identify differences, which may be anatomical abnormalities. This approach can improve the accuracy of anatomical abnormality detection, thereby reducing the time spent in the medical institution and potentially improving the operator experience and patient outcomes.

[0012] The step of predicting the 2D contour of the anatomical feature based on the 3D model comprises:

[0013] estimating a pose of the anatomical feature based on the 2D medical image;

[0014] adjusting the 3D model based on the estimated pose;

[0015] generating a 2D projection based on the pose-adjusted 3D model; and

[0016] The 2D contour of the anatomical feature is predicted based on the 2D projection. In other words, the observation position and posture of the imaged target can be estimated based on the detected 2D contour. The 3D model can then be rotated and translated to match the position of the imaged anatomical feature relative to the virtual imaging system, and the 3D model can be adjusted by rotating the joints of the 3D model itself (e.g., changing the flexion state of the ankle joint) to match the posture of the imaged anatomical feature, thereby creating a healthy reference model with the same observation and posture parameters of the imaged target, which can be used to predict the 2D reference contour.

[0017] The step of detecting the 2D contour may also include segmenting the 2D medical image. For example, individual pixels of the acquired medical image may be labeled as "bone / no bone" so as to identify bones in the image and thereby detect the 2D contour. Alternatively, the step of detecting the 2D contour may include a machine learning algorithm that has been trained end-to-end. In addition, the step of detecting the 2D contour may include identifying anatomical features. For example, identifying that the 2D contour is a 2D contour of one or more bones of an ankle or wrist. The method may also be used to optimize the workflow of any further processing, or to provide guidance to a user or machine for further steps.

[0018] The method may further include classifying the anatomical abnormality. The step of classifying the anatomical abnormality may also be performed by a machine learning algorithm. In this manner, the output of the computer-implemented method is not binary (e.g., an anatomical abnormality is detected, or an anatomical abnormality is not detected), but rather provides the user with insight into the classification of the abnormality (e.g., whether the anatomical abnormality is severe, and / or whether the anatomical abnormality is a specific type of fracture and / or ligament rupture).

[0019] According to an embodiment of the present invention, the 3D model is obtained based on:

[0020] The 2D medical image,

[0021] The anatomical features, and / or

[0022] User input.

[0023] The 3D model may also be any of the following:

[0024] Computer-aided design models; and

[0025] A reference model is constructed based on a 3D medical image of the anatomical feature in at least one posture.

[0026] In some examples, the method further includes:

[0027] Generates the difference between:

[0028] The detected 2D contours, and

[0029] The predicted 2D contour.

[0030] Additionally, a difference map may be generated. The difference map contains information suitable for display on a display.

[0031] In some examples, the anatomical feature is any of the following:

[0032] Bone tissue, and

[0033] Ligament tissue.

[0034] In some examples, the anatomical abnormality is any of the following:

[0035] fractures, and

[0036] Ligament rupture.

[0037] In some examples, the step of classifying the anatomical abnormality classifies the anatomical abnormality based on any of:

[0038] Weber classification,

[0039] Pauwels classification,

[0040] Smith and Colles classification,

[0041] Salter-Harris classification, and

[0042] AO / OTA classification.

[0043] In some examples, the medical image is an X-ray image.

[0044] Another aspect of the present invention provides a computer program product comprising instructions for enabling a processor to execute an embodiment of the above method.

[0045] Another aspect of the present invention provides a system for detecting anatomical abnormalities in 2D medical images. The system includes a processor, wherein the processor is configured to:

[0046] Obtaining the 2D medical image;

[0047] Detect 2D contours of anatomical features;

[0048] obtaining a 3D model representing the anatomical feature;

[0049] Predicting a 2D contour of the anatomical feature based on the 3D model, wherein the step of predicting the 2D contour of the anatomical feature based on the 3D model comprises:

[0050] estimating a pose of the anatomical feature based on the 2D medical image;

[0051] adjusting the 3D model based on the estimated pose;

[0052] generating a 2D projection based on the pose-adjusted 3D model; and

[0053] predicting a 2D contour of the anatomical feature based on the 2D projection;

[0054] The anatomical abnormality is detected based on the detected 2D contour and the predicted 2D contour.

[0055] In one example, the system may also include an X-ray source and a detector configured to capture an X-ray image.

[0056] These and other aspects of the invention will be apparent from and elucidated with reference to the embodiments described hereinafter. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] Figure 1 A flow chart according to an embodiment of the present invention is shown;

[0058] Figure 2A shows a medical image according to an embodiment of the present invention;

[0059] Figure 2B shows the detected contours of anatomical features in a medical image according to an embodiment of the present invention;

[0060] Figure 2C shows predicted contours of anatomical features in a medical image according to an embodiment of the present invention;

[0061] Figure 3 shows the pose and observation parameters of an imaged target according to an embodiment of the present invention;

[0062] Figure 4 An articulated 3D model of an ankle according to an embodiment of the present invention is shown;

[0063] Figure 5A The classification of ankle fractures according to an embodiment of the present invention is shown;

[0064] Figure 5B The classification of femoral neck fractures according to an embodiment of the present invention is shown;

[0065] Figure 5C shows classification of fractures by epiphysis or bone growth plate according to an embodiment of the present invention; and

[0066] Figure 6 A processor circuit according to an embodiment of the invention is shown. DETAILED DESCRIPTION

[0067] The present invention will be described herein with reference to the accompanying drawings. It should be understood that the description and specific examples provided, while indicating exemplary embodiments, are for illustrative purposes only and are not intended to limit the scope of the present invention. It should also be understood that the drawings are merely schematic and are not drawn to scale. It should also be understood that the same reference numerals are used to indicate the same or similar parts throughout the drawings.

[0068] The present invention provides a computer-implemented method of analyzing and evaluating 2D (two-dimensional) medical images for the purpose of identifying anatomical abnormalities.

[0069] Figure 1 An exemplary flow chart of an embodiment of a method 100 according to the present invention is shown, the method 100 comprises the following steps:

[0070] In step 110, a 2D medical image 210 is obtained. The 2D medical image may be any type of medical image, including but not limited to, for example, X-ray images, CT (computed tomography) images, ultrasound images, MR (magnetic resonance) images, or any other type of medical image.

[0071] An exemplary medical image 210 contemplated in the present invention is Figure 2A A 2D X-ray image of ankle joint 211 is shown.

[0072] In step 120, 2D contours of anatomical features are detected. This may, for example, involve segmentation of 2D medical images or an end-to-end machine learning approach.

[0073] Broadly speaking, segmentation can be understood as the task of classifying an image into labels on a pixel-by-pixel basis. Alternatively, the classification is based on a larger image region as a whole (e.g., a contour or surface). The label represents the semantics of the corresponding pixel or image region. In this way, for example, bone pixels (i.e., pixels with the label "bone") can be identified in a classification operation. The pixel is then considered to represent bone tissue, and the same is true for other tissue types. Thus, segmentation can be used to identify the coverage of organs in an image. In an alternative embodiment, segmentation can be performed using a neural network or a convolutional neural network (e.g., as disclosed in WO2022 / 084074). The present invention also contemplates algorithms disclosed herein that include prior shape knowledge.

[0074] From image segmentation, a segmented image is usually obtained, in which each pixel and / or region has a specific label assigned to it. This label can be a binary "bone / no bone" or a multi-level "bone / ligament tissue / no bone / ...". According to an embodiment of the present invention, a part of the medical image corresponding to the anatomical feature of interest (e.g., "bone"), or more specifically, "tibia / fibula / calcaneus / talus", can be selected to generate a label such as Figure 2B 2D profile 212 is shown.

[0075] In step 130, a 3D (three-dimensional) model 330 is obtained that represents the imaged object 211 or the identified anatomical feature 213. The model may be selected from the list of 3D models based on at least one of the following:

[0076] The medical image 210 obtained in step 110,

[0077] the anatomical features identified in step 120, and

[0078] User input.

[0079] For example, the image 210 obtained in step 110 may be annotated to indicate that the image is an image of an ankle joint. In another example, the identified anatomical features or a combination of identified anatomical features may indicate that the image contains at least an ankle joint. For example, a DICOM tag may be associated with the anatomical features being imaged (which may be used to identify a 3D model). In yet another example, a user of a computing system executing method 100 may select a 3D model from a list of models. The user may select from a whole of all 3D models, or from a reduced set of articulated 3D models (which may be pre-selected based on 2D medical images and / or anatomical features identified in the 2D medical images). In another example, a 3D model may be selected based on background information of the acquisition location (e.g., a radiology department of a hospital), based on, for example, user diagnostic request data and / or a radiology standard operating procedure for the requested examination process. For example, prior to image acquisition, a user or operator may indicate that an ankle joint X-ray image is desired, which may determine a relevant 3D model, or create a pre-selection of a relevant 3D model. The 3D model may be articulated so that different poses of the imaged object can be taken into account, for example, the 3D model can reproduce a specific flexion state of the ankle joint.

[0080] The representative 3D model can be a computer-aided design (CAD) model created using any CAD software, or a reference model extracted from 3D imaging using known techniques. For example, a sequence of MR images of the ankle joint of a healthy subject can be obtained at different flexion angles. An image can then be segmented manually or automatically according to known techniques. The segmented image can then be used to create a surface model using, for example, a triangular mesh. For each triangle in the mesh, features for model adaptation can then be trained using, for example, image intensity, and the resulting model can be used to depict the corresponding bone surface in other MR images. This process can generate a 3D model for each MR image. The posture in each image and / or corresponding 3D model can also be determined by, for example, measuring the angle between the main axis of the tibia and the main axis of the calcaneus in the flexed state. Smooth interpolation of the surface model between different postures recorded in the MR images can make it possible to estimate the grid constellation in any posture therebetween.

[0081] Figure 4 The 3D model 430 contemplated by the present invention is shown. The 3D model can be articulated and can be modified to fit any pose encountered in a 2D medical image. For example, by Figure 3 The image sequence in shows the calcaneus rotated from 80 degrees to 135 degrees with respect to the tibia.

[0082] In step 140, the 2D contour of the anatomical feature in the medical image is predicted based on the 3D model. The contour is predicted according to the following steps:

[0083] estimating 142 observation parameters and / or pose of an imaged object from the 2D medical image,

[0084] modifying or adjusting 144 the 3D model according to the estimated viewing parameters and / or pose,

[0085] creating 146 a 2D projection from the observed and / or pose-adjusted 3D model, wherein the projection represents the 2D medical image 210, and

[0086] A 2D outline 214 of the anatomical feature shown in the 2D projection of the 3D model is generated 148 .

[0087] The observation parameters and postures referred to in step 142 may be (eg Figure 3 For the ankle joint) such as any of the following:

[0088] flexion angle δ between the tibia and calcaneus;

[0089] viewing direction, which includes azimuth and polar rotation; and

[0090] In the case of X-ray images, detector position, translation, rotation and scaling.

[0091] This list should not be construed as exhaustive, and other observation and / or posture parameters are contemplated by the present invention.

[0092] According to S. et al., “CNN-based pose estimation for assessing quality of ankle-joint X-ray images” (2022), estimate pose and / or observation parameters in step 142.

[0093] For example, a mesh 3D model of an ankle joint can be sampled at random flexion angles and viewing directions, and a projection can be generated by identifying triangles in the mesh in the 3D model that are approximately tangentially passed by the X-ray beam in the selected viewing direction, projecting the edges of the triangle onto a virtual detector, and semantically grouping and discretizing the triangle on the detector pixel matrix. The resulting set of discrete contours in the projection forms the input to a posture estimation network. Such a posture estimation network can be a standard regression network involving a ReLU (rectified linear unit) activation function with alternating step-1 and step-2 convolution blocks and a fully connected layer. Then, a network trained according to the above principles can be used to predict the posture of the detected 2D contour. For example, the 2D contour of the imaged ankle joint is fed into a neural network, and since the neural network has been trained, the neural network can provide the flexion angle and viewing direction of the imaged ankle joint.

[0094] The pose and / or observation parameters estimated in step 142 may then be used as input to step 144 to modify or adjust the 3D model to reproduce the corresponding pose and / or observation parameters. For example, the flexion angle of the 3D ankle model 430 may be as follows: Figure 4 The variation shown, in which the angle between the tibia and the calcaneus changes from 80 degrees to 135 degrees.

[0095] After adjusting the 3D model to the estimated pose, the 2D projection is determined in step 146 by, for example, the following operations: identifying triangles in the mesh in the 3D model that are approximately tangentially passed by the X-ray beam in the selected viewing direction, projecting the edges of the triangles onto the virtual detector, and semantically grouping and discretizing the triangles on the detector pixel matrix; thereby also directly providing the predicted 2D contour of step 148. Figure 2C An exemplary 2D contour 214 predicted in steps 140 , 142 - 148 from the 3D model of the ankle joint is shown.

[0096] In an optional step, a difference between a 2D contour generated based on the 2D medical image and a 2D contour generated based on the articulated 3D model can be determined. The difference can be calculated based on, for example, point-by-point subtraction or pixel-by-pixel subtraction. In some embodiments, a difference map can be created, wherein the difference map can also be visualized on a user interface (not shown).

[0097] In step 150, anatomical abnormalities are detected based on the 2D contours detected in step 120 and the 2D contours predicted in step 140, respectively. The anatomical abnormalities can be detected by visual inspection, for example, a difference map shown on a user interface (not shown). In another exemplary embodiment, the anatomical abnormalities can be detected automatically. For example, based on whether the difference or deviation between (i) a 2D contour generated based on a 2D medical image and (ii) a 2D contour generated based on an articulated 3D model exceeds a specific threshold. In yet another exemplary embodiment, a machine learning algorithm can be used to detect anatomical abnormalities. For example, a user or an expert can annotate a 2D medical image to identify anatomical abnormalities. These annotated images can also be processed according to steps 120 and 140 to detect 2D contours and predict 2D contours, wherein the detected 2D contours, the predicted 2D contours, and / or the differences between the detected 2D contours and the predicted 2D contours can be added to the annotated images to be used in the training process.

[0098] The identification of the anatomical abnormality may be binary (either the abnormality is present or not) or non-binary (in which case a probability of the abnormality being present may be provided). In an advantageous embodiment, the anatomical abnormality may also be classified based on size, location, or any other form of classification to indicate severity in optional step 160. In other words, the anatomical abnormality may be classified into a plurality of categories. Figure 5A (discussed in more detail below) shows, for example, the classification of ankle fractures, while Figure 5B (also discussed in more detail below) shows, for example, the classification of femoral neck fractures. According to an embodiment of the present disclosure, the classification of anatomical abnormalities is performed using a trained machine learning algorithm. For example, one may use, for example, Figures 5A-5C The classification network is trained using the annotated data in and the detected 2D contours 212 and the predicted 2D contours 214 as input.

[0099] For classification training of machine learning models, annotated medical images are provided. For example, an X-ray image of an ankle joint with the following annotations indicating whether there is a fracture and whether there is a fracture according to e.g. Figure 5A The annotation may be performed by an expert (e.g., a clinician, radiologist, researcher, or any other qualified person). For example, an annotated image may be an ankle X-ray image 210 with an associated label of no fracture. Another example may be an ankle X-ray image with a fracture and a corresponding label identifying the classification of the fracture. For example, whether the fracture is a type A fracture, a type B fracture, or a type C fracture in the Weber classification.

[0100] In the same manner as the images to be analyzed by the method 100, the annotated images used for training are also subjected to steps 120 and 140 of the method 100 in order to perform the following operations:

[0101] detecting 2D contours of anatomical features; and

[0102] The 2D contour of the anatomical feature is predicted from a healthy reference (eg, from the 3D model obtained in step 130 ).

[0103] The detected 2D contours and the predicted 2D contours can then form the input for the machine learning model, and the annotations of the training images can form the output. Typically, training is performed using a subset of the training images and corresponding annotations, and the remainder of the training images and corresponding annotations are used to evaluate the trained model. When a predetermined criterion is reached, training is stopped to ensure that the model has been trained well enough without being overtrained. In some embodiments, noise can also be added to the training data manually or automatically to avoid overfitting.

[0104] Those skilled in the art will appreciate that although the method 100 is presented in a specific order, the order of the steps may be changed, additional steps may be added between the steps, and / or some steps may be deleted.

[0105] It should also be understood that the individual steps in method 100 can be performed in real time, for example, during the examination process, or after the examination process. For example, a 2D medical image can be obtained in real time during image acquisition and analyzed according to method 100, or a 2D medical image can be first obtained during the acquisition step and analyzed only upon request by an operator and / or expert. Thus, method 100 can be implemented within an image acquisition system or in a separate computing unit external to the image acquisition system. In another embodiment, method 100 can also be performed in the cloud.

[0106] Figures 5A-5C The predefined categories contemplated by the present invention are disclosed. The categories in these figures are merely illustrative and in no way limit the concept of the present invention.

[0107] Figure 5A shows the Weber classification of the ankle joint. Figure 5A , three different types of fractures are shown, type A 511, type B 512, and type C 513. According to Wikipedia (https: / / en.wikipedia.org / wiki / Danis%E2%80%93Weber_classification), these three different types of fractures can be described as:

[0108] Type A 511 - Fracture of the distal fibula at the syndesmosis (the connection between the distal ends of the tibia and fibula). Typical features are:

[0109] Below ankle level

[0110] Complete tibiofibular syndesmosis

[0111] Deltoid ligament intact

[0112] Occasionally fractures of the medial malleolus

[0113] Usually stable: Nevertheless, open reduction and internal fixation (ORIF) is occasionally necessary, particularly in cases of medial malleolus fractures.

[0114] Type B 512 - Fibula fracture at the level of the ligamentous attachment. Typical features are:

[0115] At the level of the ankle, extending superiorly and laterally toward the fibula

[0116] Complete or partial tear of the syndesmosis, but no widening of the tibiofibular joint

[0117] The medial malleolus may be fractured or the deltoid ligament may be torn

[0118] Variable stability

[0119] Type C 513 - ligamentous proximal fibula fracture. Typical features are:

[0120] Above ankle level

[0121] Syndesmotic rupture, widening of the distal tibiofibular joint

[0122] Presence of medial malleolus fracture or deltoid ligament injury

[0123] Unstable: ORIF required

[0124] Type B 512 fractures and Type C 513 fractures imply some degree of damage to the ligamentous attachments themselves (which may not be directly visible on the radiograph). They are inherently unstable and are more likely to require surgical repair for a good prognosis. Type A fractures are generally stable and can be managed with simple measures (e.g., plaster casts). Therefore, correct identification of ankle fractures according to the Weber classification may improve the prognosis and / or recovery of the imaged subject.

[0125] Figure 5BThe Pauwels classification of femoral neck fractures is shown. These fractures are characterized by the fracture angle. The greater the angle, the more unstable the fracture and the greater the shear stress at the fracture site. This shearing can lead to higher rates of nonunion. According to the description of the Pauwels classification in Wikipedia (https: / / en.wikipedia.org / wiki / Pauwel%27s_angle), the fractures are divided into type I 521 fractures, type II 522 fractures, and type III 523 fractures as follows:

[0126] Type I 521 - fracture angle less than 30 degrees

[0127] Type II 522 – fracture angle between 30 and 50 degrees

[0128] Type III 523 - fracture angle greater than 50 degrees.

[0129] Figure 5C The Salter-Harris classification of fractures involving the epiphysis or growth plate (particularly the temporary calcification zone) is shown. It is therefore a form of fracture in children, occurring in 15% of long bone fractures in children. There are nine types of Salter-Harris fractures; type I531 to type V535 described by Robert B. Salter and W. Robert Harris in 1963 (e.g. Figure 5C The first five types are described as follows:

[0130] Type I 531 – Transverse fracture across the growth plate (also called “physis”);

[0131] Type II 532 - fracture through the growth plate and metaphysis, with preservation of the epiphysis: The spine takes approximately 12-90 weeks or more to heal;

[0132] Type III 533—fracture through the growth plate and epiphysis, with preservation of the metaphysis;

[0133] Type IV 534 – fractures through all three parts of the bone (growth plate, metaphysis, and epiphysis);

[0134] Type V 535 – Compression fracture of the growth plate (causing a decrease in the perceived space between the epiphysis and metaphysis on radiographs).

[0135] FIG. 5A to FIG. 5CThe predefined categories in are provided as a reference to demonstrate possible classifications according to embodiments of the present invention. In particular, by using predefined categories, the operator is allowed to see not only that a fracture may have occurred, but also what type of fracture may have occurred, so that subsequent steps can be improved to improve the prognosis and / or recovery of the imaged subject.

[0136] Another predefined classification may be the AO / OTA (Arbeitsgemeinschaft für Osteosynthesefragen / Association of Orthopaedic Traumatology) classification, which defines fractures according to anatomical location, fracture type and fracture subgroups.

[0137] The anatomical location is according to the AO / OTA classification, which is described by a two-digit number, where the first digit indicates the major anatomical part (e.g., upper arm (humerus), lower arm, thigh, lower leg (tibia / fibula), etc.), and the second digit indicates the specific part of the bone involved (e.g., proximal, middle, distal).

[0138] The fracture type is coded as type A, B, or C, where:

[0139] Type A indicates a simple fracture;

[0140] Type B indicates a wedge-shaped fracture with more than two fragments and some contact between the major fragments; and

[0141] Type C indicates a complex or comminuted fracture, possibly with multiple separate fragments.

[0142] Fracture subgroups indicated by a single number provide additional details, such as fracture pattern or configuration regarding the nature of the fracture (eg, whether the fracture is oblique, transverse, spiral, etc.).

[0143] Although predefined classifications for bone and / or ligament fractures have been discussed above, the present invention also contemplates other types of classifications, including but not limited to, for example, cancer type and / or stage. For example, a segmented tumor may be automatically classified as a stage I tumor, a stage II tumor, a stage III tumor, or a stage IV tumor based on its size and / or location.

[0144] The method 100 may also include an optional step 170 of providing treatment recommendations based on the identified anatomical abnormalities and / or the classification results of the identified anatomical abnormalities. For example, the treatment recommendation may be that a C-type ankle fracture classified by Weber requires immediate surgery, while for a hairline fracture and / or a stress fracture, rest and / or a cast may be recommended. The recommendation may be based on annotated data, for example, the classified image may also be annotated to provide treatment recommendations based on the classification as described above.

[0145] Figure 6 600 according to various aspects of the present disclosure. As shown, the processor circuit 600 may include a processor 606, a memory 603, and a communication module 608. These elements may communicate with each other directly or indirectly (eg, via one or more buses).

[0146] As contemplated by the present disclosure, the processor 606 may include a central processing unit (CPU), a graphics processing unit (GPU), a digital signal processor (DSP), an application specific integrated circuit (ASIC), a controller, a field programmable gate array (FPGA) device, another hardware device, a firmware device, or any combination thereof, which are configured to perform the operations described herein. The processor 606 may also be implemented as a combination of computing devices, for example, a combination of a DSP and a microprocessor, a plurality of microprocessors, a combination of one or more microprocessors and a DSP core, or any other such configuration. The processor 606 may also implement various deep learning networks, which may include hardware implementations or software implementations. The processor 606 may also include a preprocessor implemented in hardware or implemented in software.

[0147] As contemplated by the present disclosure, memory 603 may be any suitable storage device, such as a cache memory (e.g., a cache memory of processor 606), a random access memory (RAM), a magnetoresistive RAM (MRAM), a read-only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), flash memory, a solid-state storage device, a hard drive, other forms of volatile and non-volatile memory, or a combination of different types of memory. The memory may be distributed among multiple memory devices and / or located remotely relative to the processor circuitry. In one embodiment, memory 603 may store instructions 605. Instructions 605 may include instructions that, when executed by processor 606, cause processor 606 to perform the operations described herein.

[0148] Instructions 605 may also be referred to as code. The terms "instructions" and "code" should be broadly interpreted as including (one or more) any type of computer-readable statements. For example, the terms "instructions" and "code" may refer to one or more programs, routines, subroutines, functions, processes, etc. "Instructions" and "codes" may include a single computer-readable statement or multiple computer-readable statements. Instructions 205 may be in the form of an executable computer program or script. For example, routines, subroutines, and / or functions may be defined in a programming language, including but not limited to C, C++, C#, Pascal, BASIC, API calls, HTML, XHTML, XML, ASP scripts, JavaScript, FORTRAN, COBOL, Perl, Java, ADA, .NET, etc. Instructions may also include instructions for machine learning and / or deep learning.

[0149] The communication module 608 can include any electronic circuit and / or logic circuit to facilitate direct or indirect data communication between the processor circuit 600 and, for example, an external display (not shown) and / or an imaging device or system (e.g., an X-ray imaging system). In this regard, the communication module 608 can be an input / output (I / O) device. The communication can be performed via any suitable means. For example, the communication means can be a wired link, such as a universal serial bus (USB) link or an Ethernet link. Alternatively, the communication means can be a wireless link, such as an ultra-wideband (UWB) link, an Institute of Electrical and Electronics Engineers (IEEE) 802.11 WiFi link, or a Bluetooth link.

[0150] It should be noted that the above embodiments illustrate rather than limit the present invention, and that those skilled in the art will be able to design many alternative embodiments without departing from the scope of the appended claims. In the claims, any reference numerals in brackets should not be interpreted as limiting the claims. The word "comprising" does not exclude the presence of elements or steps not listed in the claims. The word "one" or "an" in front of an element does not exclude the presence of multiple such elements. The present invention can be implemented by means of hardware comprising several different elements and / or by means of a suitably programmed processor. In a device-type claim that lists several devices, several of these devices can be implemented by the same hardware. The measures listed in different dependent claims can be advantageously used in combination.

Claims

1. A computer-implemented method (100) for detecting anatomical abnormalities in a 2D medical image (210), the method comprising: Obtaining (110) the 2D medical image (210); detecting (120) a 2D contour (212) of an anatomical feature; obtaining (130) a 3D model (430) representing the anatomical feature; A 2D contour (214) of the anatomical feature is predicted (140) based on the 3D model (430), wherein the step (140) of predicting the 2D contour of the anatomical feature based on the 3D model comprises: estimating (142) a pose of the anatomical feature based on the 2D medical image (210); adjusting (144) the 3D model (430) based on the estimated pose; generating (146) a 2D projection based on the pose adjusted 3D model, and predicting (148) a 2D contour (214) of the anatomical feature (213) based on the 2D projection; The anatomical abnormality is detected (150) based on the detected 2D contour (212) and the predicted 2D contour (214).

2. The method (100) according to claim 1, wherein: The step of detecting the 2D contour (212) includes segmenting the 2D medical image (210).

3. The method (100) according to any one of the preceding claims, further comprising: The anatomical abnormality is classified (160).

4. The method (100) according to claim 3, wherein: The step of classifying the anatomical abnormality (160) is performed by a machine learning algorithm.

5. The method (100) according to any one of the preceding claims, wherein: The 3D model (430) is obtained based on: the 2D medical image (210), The anatomical feature (213), and / or User input.

6. The method (100) according to any one of the preceding claims, wherein: The 3D model (430) is any of the following: Computer-aided design models; and A reference model is constructed based on a 3D medical image of the anatomical feature in at least one posture.

7. The method (100) according to any one of the preceding claims, further comprising: Generates the difference between: the detected 2D contour (212), and The predicted 2D contour (214).

8. The method (100) according to any one of the preceding claims, wherein: The anatomical feature is any of the following: Bone tissue, and Ligament tissue.

9. The method (100) according to any one of the preceding claims, wherein: The anatomical abnormality is any of the following: fractures, and Ligament rupture.

10. The method (100) according to claim 3 or 4, wherein: The step of classifying the anatomical abnormality classifies the anatomical abnormality based on any of the following: Weber classification, Pauwels classification, Smith and Colles classification, Salter-Harris classification, and -AO / OTA classification.

11. The method (100) according to any of the preceding claims, further comprising: A treatment recommendation is generated ( 170 ) based on the detected anatomical abnormality.

12. The method (100) according to any one of the preceding claims, wherein: The medical image is an X-ray image.

13. A computer program product comprising instructions for enabling a processor to perform the method according to any one of the preceding claims.

14. A system for detecting anatomical abnormalities in a 2D medical image, the system comprising a processor (206) configured to perform the method according to any one of claims 1-13.

15. The system of claim 14, further comprising an X-ray source and an X-ray detector.

Images