Identifying anatomical objects in medical images

Abstract

A method and system for performing anatomical object recognition on a medical image. Position information that provides information about an anatomical object measurement position in the medical image, and receives (410) and processes (420) the medical image to produce anatomical object information. The anatomical object information identifies an anatomical object in a form that can be used to train a machine learning model.

Description

Technical Field

[0001] The present invention relates to the field of medical image processing and, in particular, to identifying anatomical objects in medical images. Background Art

[0002] There has been a long-standing desire to perform accurate automatic recognition of anatomical objects in medical images, for example, to automatically identify clinically relevant regions within medical images. Recently, machine learning models have been used to perform this automatic recognition. Such algorithms offer a distinct advantage over traditional forms of computer programming techniques because they can be generalized and trained to recognize image features without the need to write custom computer code.

[0003] However, a well-known drawback of machine learning models is that they typically require large amounts of training data to perform tasks accurately and robustly. For object recognition tasks, this places a significant burden on clinicians to accurately and repeatedly generate diverse examples of training data.

[0004] Therefore, it is desirable to generate training data to train machine learning models to perform object recognition tasks. More specifically, it is increasingly desirable to reduce the burden on clinicians to manually perform object recognition on medical images to generate such training data. Summary of the Invention

[0005] The invention is defined by the independent claims. The dependent claims define advantageous embodiments.

[0006] According to an example based on one aspect of the present invention, a computer-implemented method for performing anatomical object recognition on a medical image is provided, the computer-implemented method comprising: receiving a medical image; receiving position information, the position information identifying one or more measurement positions of an anatomical object with respect to the medical image, wherein each measurement position represents a position defined by a clinician during a geometric measurement process using the medical image; and processing the position information to generate anatomical object information, the anatomical object information identifying a region containing and / or bounding an anatomical object, wherein processing the position information comprises performing segmentation or object recognition on the medical image using the position information.

[0007] The present disclosure recognizes that during the course of a standard or routine medical analysis of a patient / object, clinicians often perform a geometric measurement procedure in which one or more geometric measurements of an anatomical object are captured. A geometric measurement, also known as a geometric measurement, is any measurement of a dimension or distance-based parameter, such as distance, length, width, diameter, radius, perimeter, side length, area, or volume. Other examples are well known to those skilled in the art.

[0008] The geometric measurement process typically involves a clinician identifying a measurement location (e.g., at least one or more starting and ending locations) for a measurement on a medical image and performing one or more measurements using this point or points to determine the measurement. This is particularly common in fetal monitoring (e.g., to identify fetal head circumference, fetal bladder size, or fetal abdominal size), but is also performed in other medical monitoring processes, such as identifying the size of a tumor or growth, the length of the spine, the length of a bone, and the like.

[0009] It has been recognized herein that these measurement locations can effectively represent locations of identified anatomical objects, for example, locations on the boundaries of the (investigated) anatomical object (or other specific locations, such as the center or intersection). It is proposed to exploit this understanding to automatically segment medical images or perform object recognition using the measurement locations. Specifically, a bounding box or contour can be positioned to intersect or contain the measurement location.

[0010] The proposed techniques can be used to generate training data suitable for training machine learning models to perform image segmentation tasks or object recognition in medical images. These processes have technical purposes because they are related to the recognition of objects represented by images. The proposed techniques improve the available training dataset or training data by providing additional examples, making any subsequently trained machine learning models more robust, insensitive to variation, and / or accurate.

[0011] In some embodiments, the one or more measurement locations include two or more measurement endpoints; and each measurement endpoint represents a starting position or an ending position of a measurement of the anatomical object by a clinician using a medical image during a geometric measurement procedure.

[0012] The endpoints may be assumed to identify the boundaries of the anatomical object, thereby facilitating increased ease and accuracy in correctly identifying or segmenting the boundaries of the anatomical object.

[0013] In some examples, the one or more measurement locations include: a first set of measurement locations representing a first starting location and a first ending location of geometric measurements acquired across a first dimension of the anatomical object; and a second set of measurement locations representing a second starting location and a second ending location of geometric measurements taken across a different second dimension of the anatomical object.

[0014] In some examples, the first dimension is substantially perpendicular to the second dimension.

[0015] In some examples, one of the width or the height of the region is defined by the distance between the first starting position and the first ending position; and the other of the width or the height of the region is defined by the distance between the second starting position and the second ending position.

[0016] In some examples, processing the location information includes identifying as the region an outline or bounding box that intersects the location of each measured location identified in the location information.

[0017] This technique provides a simple yet effective mechanism for performing image segmentation or object recognition.

[0018] In an alternative example, the step of processing the medical image and the position information may include identifying as the region an outline or a bounding box containing the location of each measurement location identified in the position information.

[0019] In some examples, each measurement location represents a position of the jaws of a virtual caliper used by a clinician to determine measurements of an anatomical object during a geometric measurement procedure.

[0020] In some examples, each measurement location is manually defined by a clinician during a geometric measurement procedure.

[0021] This approach ensures that the clinician defines the locations of the measurement locations so that the accuracy of object identification will be similar to that of the clinician performing the full object identification themselves.

[0022] In some examples, receiving the location information includes receiving a medical report associated with the medical image, the medical report including the location information.

[0023] Medical reports are more likely to contain accurate location information. Therefore, by using the data contained in medical reports for medical images, more accurate object recognition can be performed.

[0024] In some examples, the medical report comprises a structured medical report, and preferably a DICOM encoded report.

[0025] Using structured medical reports increases the ease and reliability of identifying the location information contained therein (because it will be in a predictable location or formatted in a predictable manner). Structured medical reports are also less likely to contain erroneous data than other forms of medical reports because such structured medical reports require data to be entered in a predefined manner. Therefore, using structured medical reports increases the accuracy and reliability of object recognition using location information.

[0026] In some examples, the medical report also includes measurement information identifying measurements generated during the geometric measurement procedure.

[0027] A computer-implemented method for performing anatomical object recognition on medical images is also proposed, comprising performing the above method for each medical image.

[0028] Preferably, the position information of each medical image is previously obtained, e.g., has been previously generated and stored in a memory or storage unit. This suggests that the proposed method can be performed on old, historical, and recently available data to improve the amount of data available for training machine learning models.

[0029] In some examples, for each medical image, receiving the position information includes receiving a medical report associated with the medical image, the medical report including the position information and measurement information identifying measurements generated during the geometric measurement procedure.

[0030] This embodiment exploits the recognition that medical reports containing measurement information are likely to have a high or sufficiently accurate identification of the measurement locations. In other words, the location information contained in such medical reports is considered reliable because it corresponds to the precise measurements used in the reports. That is, these images have been annotated by clinicians.

[0031] This method ensures that the reliability of anatomical object identification is similar or identical to that in which the anatomical object identification is performed directly by a clinician.

[0032] In some examples, the method further includes: adding each medical image and anatomical object information to a training dataset to train a machine learning model to perform anatomical object recognition to produce an updated training dataset; and then using the updated training dataset to train or retrain the machine learning model.

[0033] According to another aspect of the present invention, there is provided a computer program product comprising computer program code, which, when executed by a processor, causes the processor to perform the steps of the above method.

[0034] According to yet another aspect of the present invention, there is provided a processing system for performing anatomical object recognition on a medical image, the processing system being configured to perform the steps of any method described or claimed herein.

[0035] 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

[0036] For a better understanding of the invention, and to show more clearly how it may be put into practice, reference will now be made, by way of example only, to the accompanying drawings, in which:

[0037] Figure 1 illustrates a system for performing anatomical object recognition on medical images according to an embodiment;

[0038] Figure 2illustrates a fetal ultrasound image with caliper positions for fetal head circumference measurement;

[0039] Figure 3 Pictured Figure 2 The same fetal ultrasound image in , where the bounding box defines the fetal head;

[0040] Figure 4 illustrates a computer-implemented method of performing anatomical object recognition on a medical image according to an embodiment; and

[0041] Figure 5 A computer-implemented method of performing anatomical object recognition on a medical image according to an embodiment is illustrated. DETAILED DESCRIPTION

[0042] The present invention will be described with reference to the accompanying drawings.

[0043] It should be understood that the detailed description and specific examples, while indicating exemplary embodiments of the apparatus, system, and method, are intended for illustrative purposes only and are not intended to limit the scope of the invention. These and other features, aspects, and advantages of the apparatus, system, and method of the present invention will be better understood from the following description, the appended claims, and the accompanying drawings. It should be understood that the drawings are schematic only and are not drawn to scale. It should also be understood that the same reference numerals are used throughout the drawings to represent the same or similar parts.

[0044] The present invention provides a method and system for performing anatomical object recognition on medical images. The method includes receiving and processing the medical images to generate anatomical object information, which provides information about the measured locations of anatomical objects in the medical images. The anatomical object information is used to identify anatomical objects in a form that can be used to train a machine learning model.

[0045] Embodiments are based, at least in part, on the recognition that geometric measurements taken during a clinician's work provide medical images that have been annotated by clinical experts. By converting the measurement data into data that can be used to train a machine learning model to perform object recognition, an accurate training dataset can be generated without requiring additional clinician time.

[0046] For example, the illustrative embodiments may be employed with medical imaging systems that measure anatomical objects for monitoring or diagnosis, such as ultrasound systems, MRI systems, CT systems, and X-ray systems.

[0047] Figure 1 Illustrated is a system 100 for performing anatomical object recognition on medical images according to an embodiment.

[0048] A medical image can be any image that allows a clinician to make (geometric) measurements of an anatomical object, e.g., for monitoring or diagnostic purposes. For example, a medical image can be an ultrasound image (e.g., a fetal ultrasound image, an echocardiogram, or a vascular ultrasound image), a magnetic resonance (MR) image, a computed tomography (CT) image, or an X-ray image.

[0049] The medical image may be a 2D medical image or a 3D medical image.

[0050] The system 100 includes a processing system or processor 110 and a memory unit 120. The processing system or processor itself is an embodiment of the present invention. In this specification, the terms processing system and processor may be used interchangeably.

[0051] The processing system 110 is configured to receive a medical image and position information 130 identifying the location of one or more measurement locations with respect to the medical image. Each measurement location represents a location defined or identified by a clinician during a geometric measurement procedure.

[0052] For a 2D medical image, the measured position may identify the position of a pixel of the 2D medical image. Similarly, for a 3D medical image, the measured position may identify the position of a voxel of the 3D medical image.

[0053] Of course, during the geometric measurement process, each measurement location may be identified by multiple clinicians (e.g., a team of clinicians or a supervisor-trainee clinician pair). Alternatively, each measurement location may be identified by a single clinician with appropriate experience or training.

[0054] An example of a measurement location is a measurement endpoint, a basis representing the start or end location of a measurement taken by a clinician during a geometric measurement procedure.

[0055] In an advantageous embodiment, the one or more measurement locations include two or more measurement endpoints, each measurement endpoint representing a starting or ending location of a measurement of the anatomical object. In other words, the location information may include locations in at least one set of locations, each set comprising a starting location and an ending location, such that the distances between the locations in each set correspond to measurements taken during the geometric measurement procedure.

[0056] In an example, each measurement location may represent a position of the jaws of a virtual caliper used by a clinician to determine measurements of an anatomical object during a geometric measurement procedure.

[0057] The use of virtual calipers to determine measurements of anatomical objects in medical images is well known and widely used in medical imaging systems. Virtual calipers can be manually positioned by the clinician or automatically positioned. Applications that automatically position calipers on medical images during geometric measurement procedures typically allow the user to manually correct the automated position.

[0058] When a virtual caliper is used to determine measurements from a medical image, the position of the virtual caliper's jaws is typically saved along with the medical imaging data. For example, a DICOM-encoded report for a medical imaging examination typically contains the measurements taken, the calculations and caliper positions used to make the measurements, and the medical image(s) on which the measurements were taken. Thus, processing system 110 can perform anatomical object recognition on historical images (i.e., images for which a geometric measurement process has concluded) as well as during a geometric measurement process.

[0059] An alternative example of a measurement location is to identify the location of an elongated structure (e.g., a spine or bone), for example, the centerline of the elongated structure. In some geometric measurement procedures, a clinician can identify multiple locations that identify the path taken by such an elongated structure to measure the elongated structure. These locations are suitable as measurement locations.

[0060] The processing system 110 may receive the location information 130 from the storage unit 120. For example, the storage unit may form part of a picture archiving and communication system (PACS), an electronic medical record (EMR) system, a clinical information system (CIS), a radiology information system (RIS), a cardiovascular information system (CVIS), or any other system that stores clinical data related to medical imaging.

[0061] Alternatively, the position information may be obtained by the processing system via user input (eg, the processing system may receive measurements for each measurement location as measurements are taken during a geometric measurement procedure).

[0062] In some examples, the processing system may receive location information 130 by receiving a medical report associated with the medical image that includes the location information. The medical report may be anonymized using any suitable anonymization / de-identification method to protect the patient's anonymity. This technique allows for processing historical information that was previously unavailable for training a machine learning model for object recognition.

[0063] The medical report may comprise a structured medical report, i.e., the location information is stored in a predictable location and / or formatted in a predictable manner. This approach increases the ease and reliability of correctly identifying the location information in the medical report. Preferably, the medical report comprises a DICOM (Digital Imaging and Communications in Medicine) report. These have a fixed and constrained format, from which information or data can be easily and reliably extracted.

[0064] In some examples, the medical report also includes measurement information identifying the or any (one or more) measurements generated during the geometric measurement procedure. In other words, the position information is associated with the measurement information, and the measurement information identifies to which anatomical object the measurement corresponds and preferably identifies the measurement type of the anatomical object (e.g., fetal head circumference, biparietal diameter, occipitofrontal diameter, abdominal circumference, femur length, etc.).

[0065] Upon receiving the position information 130 , the processing system 110 processes the position information to generate anatomical object information that identifies a region containing and / or defining an anatomical object.

[0066] Various methods can be used to generate anatomical object information that identifies a region containing and / or defining an anatomical object. It should be appreciated that the precise mechanism used to identify the region may depend, at least in part, on the type or landmark of the anatomical object to be identified (e.g., the anatomical object being measured). The processing system 110 can be configured to determine the type or landmark of the anatomical object and generate the anatomical object information based on the determined type or landmark of the anatomical object. The type or landmark of the anatomical object can be determined in a variety of ways. For example, it can be received along with the measurement location.

[0067] As a first example, a measurement location can represent the start / end location of a measurement of a first anatomical object. A measurement location may represent the location of a second anatomical object, which can be used to define anatomical object information. For example, a geometric measurement of the tibia may begin at the knee and end at the ankle. In this case, the starting location of the measurement identifies the location of the knee. Therefore, a bounding box containing the starting location may represent a bounding box that identifies the location of the knee. As another example, a geometric measurement of the spine may begin at the base of the spine and extend to the head. In this case, the starting location of the measurement identifies the location of the base of the spine.

[0068] In these examples, the bounding box can, for example, have a predetermined size and / or shape. In some examples, the size of the bounding box is defined by the length of a geometric measurement, for example, the bounding box is scaled to match the proportions of the geometric measurement.

[0069] This first example provides an example scenario where a single measurement location can be used to generate anatomical object information identifying the anatomical object location.Other approaches utilize multiple measurement locations in order to generate anatomical object information.

[0070] As another example, the region bounding the anatomical object can be identified by defining the region as a contour or bounding box that intersects the location of each measurement location identified in the location information 130. This approach is particularly useful if the geometric measurement process is performed by identifying locations on the object boundary, such as is typically the case when performing geometric measurements of the heart, lungs, fetal head, fetal bladder, etc.

[0071] The region containing the anatomical object can be identified by defining an outline or bounding box encompassing the location of each measurement location identified in the position information. This approach is particularly useful, for example, if a geometric measurement process is used to measure the length of an object along its centerline. This technique is also useful if geometric measurements are known to be unreliable, such as those taken during early fetal growth or without the aid of contrast agents.

[0072] These methods provide techniques for generating anatomical object information that identifies the location of a measured anatomical object.

[0073] The size of the region that encompasses but does not intersect each measurement location can be defined by adding a predetermined boundary around the region intersecting each measurement location. The predetermined boundary can, for example, include an absolute value (e.g., a predetermined number of pixels) or can be defined as a percentage of the distance between the measurement locations. In some examples, the size of the predetermined boundary can vary for regions of different sizes.

[0074] In some examples, the one or more measurement locations may include a first set of measurement locations representing a first starting location and a first ending location of measurements taken across a first dimension of the anatomical object, and a second set of measurement locations representing a second starting location and a second ending location of measurements taken across a second, different dimension of the anatomical object. The first dimension may be substantially perpendicular to the second dimension.

[0075] When the one or more measurement locations include a first set of measurement locations and a second set of measurement locations, one of the width or height of the region may be defined by the distance between a first starting location and a first ending location in the first set of locations, and the other of the width or height of the region may be defined by the distance between a second starting location and a second ending location in the second set of locations.

[0076] In other words, the distance between the first starting position and the first ending position and the distance between the second starting position and the second ending position can collectively define the width and height of the region containing and / or bounding the anatomical object. In some examples, a predetermined margin can be added to the distance between each set of positions to define the width and / or height of the region.

[0077] Figure 2 and Figure 3 The diagram illustrates this concept. Figure 2 A fetal ultrasound image 200 is shown with caliper positions for fetal head circumference measurement. X1 and X2 mark the positions of a first starting position and a first ending position, respectively, forming a first set of measurement positions. Y1 and Y2 mark the positions of a second starting position and a second ending position, respectively, forming a second set of measurement positions.

[0078] Figure 3 The same fetal ultrasound image 300 is shown with a bounding box 300 defining the fetal head. The height of the bounding box is defined as the distance between positions X1 and X2, while the width of the bounding box is defined as the distance between positions Y1 and Y2. In this way, the bounding box intersects the measured positions X1, X2, Y1, and Y2.

[0079] Although Figure 3 A rectangular bounding box is shown, but one skilled in the art will appreciate that the region containing and / or bounding the anatomical object may be defined by an outline having a different shape, such as an ellipse, a circle, or any other suitable shape.

[0080] return Figure 1 The system 100 may also include a user interface 140 in communication with the processing system 110. In some cases, such as when the processing system performs anatomical object recognition during a geometric measurement procedure, the user interface may include a user interface of a medical imaging system performing the geometric measurement procedure.

[0081] The user interface 140 can receive user input that allows the user of the system 100 to adjust the settings of the anatomical object identification to meet their requirements, and the processing system 110 can control these settings based on the received user input. For example, the user input can define the size of the outline or bounding box (e.g., whether the area contains or defines the anatomical object), or the size of the border around the measurement location. The output format of the anatomical object information generated by the system (e.g., json, txt, xml, etc.) can also be defined by the user input so that the output data meets the requirements of various AI vendors.

[0082] In some examples, the processing system 110 can perform anatomical object recognition as described above for each of the plurality of medical images. In this way, anatomical object information can be generated for the plurality of medical images.

[0083] For the avoidance of doubt, it should be noted that the position information used for each of the plurality of medical images need not be generated by the same clinician during the same geometric measurement procedure. Rather, the position information and each instance can represent position information generated by a different clinician and / or a different geometric measurement procedure, reflecting the instance when the associated medical image underwent the geometric measurement procedure.

[0084] The processing system can then add each medical image and the anatomical object information generated for each image to a training dataset to train a machine learning model to perform anatomical object recognition (e.g., a segmentation model or an automated measurement model). This generates an updated training dataset without requiring additional clinician time to annotate the images added to the training dataset. The machine learning model can then be trained or retrained using the updated training dataset.

[0085] Machine learning models are trained using self-training algorithms that process input data to produce or predict output data.

[0086] Here, the input data may include one or more medical images, and the output data may include anatomical object information (eg, segmentation information and / or anatomical measurements) of the one or more medical images.

[0087] Suitable machine learning models for use with the present invention will be readily apparent to those skilled in the art. Examples of suitable machine learning models include decision tree algorithms and artificial neural networks. Other machine learning models, such as logistic regression, support vector machines, or naive Bayesian models are suitable alternatives.

[0088] The structure of an artificial neural network (or neural network for short) is inspired by the human brain. Neural networks consist of layers, each containing multiple neurons. Each neuron performs a mathematical operation. Specifically, each neuron can include a different weighted combination of a single type of transformation (for example, the same type of transformation, such as a sigmoid, but with different weights). As it processes input data, each neuron performs a mathematical operation on the input data, producing a numerical output. The output of each layer in the neural network is sequentially fed into the next layer. The final layer provides the output.

[0089] Methods for training machine learning models are well known. Typically, such methods include obtaining a training dataset comprising training input data entries and corresponding training output data entries. An initialized machine learning model 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 model. This process can be repeated until the error converges and the predicted output data entry is sufficiently similar to the training output data entry (e.g., ±1%, ±5%, ±10%). This is generally referred to as a supervised learning technique.

[0090] For example, in the case where the machine learning model is formed by a neural network, the mathematical operation (weight) of each neuron can be modified until the error converges. Known methods for modifying neural networks include gradient descent, back propagation algorithms, etc.

[0091] In one example, the training input data entries correspond to medical images added to the training dataset, and the training output data entries correspond to anatomical object information generated for each image. This provides a technique for performing object recognition in a single image.

[0092] In another example, the training input data entries correspond to sets of medical images added to the training dataset, and the training output data entries correspond to sets of corresponding anatomical object information generated for each set of medical images. Each set of medical images can be a chronological sequence of medical images of the same object (e.g., medical images examined by the same clinician during a single geometric measurement procedure). For example, this approach can allow for training a machine learning model to recognize or track an object across multiple images (e.g., a sequence of images).

[0093] In some examples, all and corresponding anatomical object information in a plurality of medical images (eg, all medical images in a particular database) may be added to a training dataset.

[0094] In other examples, only a subset of the plurality of medical images and corresponding anatomical object information may be subjected to anatomical object recognition and / or added to the training dataset. In other words, the plurality of medical images and corresponding anatomical object information may be filtered before anatomical object recognition is performed and / or before being added to the training dataset, for example, to provide a specialized training dataset for a particular purpose or to ensure that the training dataset meets one or more predetermined requirements.

[0095] For example, the subset of the plurality of medical images may include medical images of a particular anatomical object (e.g., a fetal head) or medical images in which a particular measurement (e.g., a fetal head circumference measurement) has been taken. In some examples, the desired anatomical object or measurement result for selecting the subset of the plurality of medical images may be determined via user input.

[0096] In another example, the subset of the plurality of medical images may include medical images for which patient consent has been obtained for use as training data. In another example, the subset of the plurality of medical images may include medical images of automatically placed calipers that have been manually corrected by a clinician, as this increases the likelihood that the caliper position is accurate.

[0097] In some examples, all available medical images can be used to initially train a machine learning model, and a subset of medical images generated after training can be used to retrain the model. For example, if the machine learning model is a model for automatically placing virtual calipers, or a model for automatically measuring anatomical objects in medical images, the output of the trained model can be corrected by a clinician. By performing anatomical object recognition based on the corrected measurement positions, the model can then be retrained and improved using images for which the model output has been corrected.

[0098] After generating the updated training set, processing system 110 may store the updated training set in storage unit 120 and / or export the updated training set (e.g., to a cloud-based storage system). In some examples, processing system 110 may automatically store the updated training set. In other examples, user input may define when and where the updated training set is stored.

[0099] Figure 4 Illustrated is a computer-implemented method 400 for performing anatomical object recognition on medical images according to an embodiment.

[0100] Method 400 begins at step 410 by receiving location information identifying one or more measurement locations with respect to a medical image. Each measurement location represents a location defined by a clinician during a geometric measurement procedure using the medical image.

[0101] In some examples, each measurement location can represent a jaw position of a virtual caliper used by a clinician to determine measurements of an anatomical object during a geometric measurement procedure.The position of each measurement location can be manually defined by a clinician during the geometric measurement procedure.

[0102] As described above, the one or more measurement locations may include a first set of measurement locations representing a first starting location and a first ending location for measurements taken along a first dimension of the anatomical object, and a second set of measurement locations representing a second starting location and a second ending location for measurements taken along a second, different dimension of the anatomical object. The first dimension may be substantially perpendicular to the second dimension.

[0103] In some examples, step 410 includes receiving a medical report associated with the medical image, the medical report including the position information. The medical report may include a structured medical report, and preferably a DICOM-encoded report. The medical report may also include measurement information identifying measurement values ​​generated during the geometric measurement procedure.

[0104] In step 420, the position information is processed to generate anatomical object information that identifies a region containing and / or bounding the anatomical object. This step may include identifying a contour or bounding box that intersects or contains the location of each measurement location identified in the position information as the region.

[0105] As described above, one of the width or height of the region may be defined by the distance between a first starting position and a first ending position, and the other of the width or height of the region may be defined by the distance between a second starting position and a second ending position.

[0106] Figure 5 Illustrated is a computer-implemented method 500 for performing anatomical object recognition on a medical image, according to an embodiment.

[0107] Method 500 begins at step 510, where anatomical object recognition is performed on each medical image using method 400 described above. In some examples, receiving the position information can include receiving a medical report associated with the medical image, the medical report including the position information and measurement information identifying measurements generated during the geometric measurement procedure.

[0108] In some examples, the method may also include step 520, in which each medical image and anatomical object information is added to a training dataset to train a machine learning model to perform anatomical object recognition, thereby generating an updated training dataset; and step 530, in which the machine learning model is subsequently trained or retrained using the updated training dataset.

[0109] It should be understood that the disclosed methods are computer-implemented methods. Thus, the concept of a computer program is also proposed, comprising a code for implementing any of the described methods when the program is run on a processing system.

[0110] Those skilled in the art can easily develop a processing system including an input interface, a data processor, and an output interface for performing any method described herein. In particular, each step of the flow chart can represent a different action performed by the processing system and can be performed by the relevant modules of the processing system.

[0111] Embodiments utilize data processors. Data processors can be implemented in a variety of ways using software and / or hardware to perform the various functions required. A processor is an example of a data processor that employs one or more microprocessors that can be programmed using software (e.g., microcode) to perform the required functions. However, a data processor can be implemented with or without a processor, and can also be implemented as a combination of dedicated hardware for performing some functions and a processor (e.g., one or more programmed microprocessors and associated circuitry) for performing other functions.

[0112] Examples of data processor 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).

[0113] In various implementations, a processor or data processor 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 data processors, perform the desired functions. The various storage media may be fixed within the processor or data processor, or may be transportable so that one or more programs stored thereon can be loaded into the processor or data processor.

[0114] It should be understood that the disclosed methods are preferably computer-implemented methods. Thus, the concept of a computer program is also provided, which includes code for implementing any of the described methods when the program is run on a processing system (e.g., a computer). Thus, different parts, lines, or code blocks of a computer program according to an embodiment may be executed by a processing system or computer to perform any of the methods described herein.

[0115] Also proposed is a non-transitory storage medium storing or carrying a computer program or computer code which, when executed by a processing system (eg having a data processor), causes the processing system to perform any of the methods described herein.

[0116] In some alternative embodiments, the functions recorded in the block diagram(s) or flow diagram(s) may not occur in the order shown 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 on the functionality involved.

[0117] Those skilled in the art will be able to understand and implement variations to the disclosed embodiments when practicing the claimed invention by studying the drawings, the disclosure and the claims. In the claims, the word "comprising" does not exclude other elements or steps, and the word "one" or "an" does not exclude a plurality. A single processor or other unit may implement the functions of several items recited in the claims. The measures recited in mutually different dependent claims may be advantageously combined. If it is described above that a computer program can be stored / distributed on a suitable medium such as an optical storage medium or a solid-state medium provided together with other hardware or as part of other hardware, it may also be distributed in other forms such as via the Internet or other wired or wireless telecommunications systems. If the term "arrangement" is used in the claims or description, it should be noted that the term "arrangement" is intended to be equivalent to the term "system" and vice versa. Any figure marks in the claims should not be interpreted as limiting the scope.

Claims

1. A computer-implemented method (400) for performing anatomical object recognition on a medical image, the computer-implemented method comprising: receiving (410) the medical image and position information identifying, with respect to the medical image, one or more measurement locations about an anatomical object, wherein each measurement location represents a location defined by a clinician during a geometric measurement procedure using the medical image; and The position information is processed to generate (420) anatomical object information, the anatomical object information identifying a region of the medical image containing and / or defining the anatomical object, processing the position information including using the position information to perform segmentation or object recognition on the medical image.

2. The computer-implemented method of claim 1 , wherein: The one or more measurement locations include two or more measurement endpoints; and Each measurement endpoint represents a starting position or an ending position of a measurement of an anatomical object performed by a clinician during a geometric measurement procedure using the medical image.

3. The computer-implemented method of claim 1 or 2, wherein the one or more measurement locations include: a first set of measurement locations (X1, X2) representing first starting locations and first ending locations for geometric measurements taken across a first dimension of the anatomical object; and A second set of measurement locations (Y1, Y2) representing second starting locations and second ending locations for geometric measurements taken across a second, different dimension of the anatomical object.

4. The computer-implemented method of claim 3, wherein: The first dimension is substantially perpendicular to the second dimension.

5. The computer-implemented method of claim 3 or 4, wherein: One of a width or a height of the region is defined by a distance between the first starting position and the first ending position; and The other of the width or the height of the region is defined by a distance between the second starting position and the second ending position.

6. The computer-implemented method according to any one of claims 1 to 5, wherein: The step of processing the position information includes identifying as the region an outline or bounding box that intersects the position of each measured position identified in the position information.

7. The computer-implemented method of any one of claims 1 to 6, wherein: Each measurement location represents a position of the jaws of a virtual caliper used by the clinician to determine geometric measurements of the anatomical object during the geometric measurement procedure.

8. The computer-implemented method according to any one of claims 1 to 7, further comprising the step of determining a type or identification of the anatomical object, and the anatomical object information is generated based on the determined type or identification of the anatomical object.

9. The computer-implemented method of any one of claims 1 to 8, wherein: The step of receiving the position information includes receiving a medical report associated with the medical image, the medical report including the position information.

10. The computer-implemented method of claim 8, wherein: The medical report also includes measurement information identifying measurements generated during the geometric measurement procedure.

11. A computer-implemented method (500) of performing (510) anatomical object recognition on a medical image, comprising: For each medical image, anatomical object recognition is performed on the medical image according to the method according to any one of claims 1 to 10.

12. The computer-implemented method of claim 11 , further comprising: generating a training dataset including medical images and anatomical object information; and A machine learning model is trained using the training data to perform anatomical object recognition.

13. The computer-implemented method of claim 12, further comprising: adding (520) each medical image and anatomical object information to the training dataset to generate an updated training dataset; and The machine learning model is then trained or retrained (530) using the updated training dataset.

14. A computer program product comprising computer program code which, when executed by a processor (110), causes the processor to perform the steps of the method according to any one of claims 1 to 13.

15. A processing system (110) for performing anatomical object recognition on a medical image, the processing system being configured to perform the steps of the method according to any one of claims 1 to 13.

Images