Method and system for selecting an area of ​​interest within an image

Abstract

To provide a computer-implemented method for selecting one or more regions of interest (ROIs) in an image.SOLUTION: The method comprises the following steps: a step 60 of identifying one or more objects segmented from the image; a step 62 of identifying predefined landmarks of the objects; a step of determining reference morphometrics pertaining to the objects by performing morphometrics on the objects by reference to the landmarks; a step 68 of selecting one or more ROIs from the objects according to the reference morphometrics; a step 70 comprising identifying the location of the ROIs relative to the reference morphometrics; and a step 74 of outputting the selected one or more ROIs.SELECTED DRAWING: Figure 2

Description

Technical Field

[0001] The present invention relates to a method and system for selecting a two-dimensional or three-dimensional region of interest in an image, and more particularly to applying artificial intelligence to medical image analysis such as morphometry, but is not limited to these applications. The region of interest can be a 2D region or 3D volume in such a medical image. Possible medical imaging applications include computed tomography (CT), magnetic resonance (MR), ultrasound, lesion scanner imaging, and the like.

[0002] Related Applications This application is based on and claims the priority of U.S. Patent Application No. 16 / 448,285 (filing date: June 21, 2019), and the entire content thereof at the time of filing is incorporated by reference.

Background Art

[0003] The region of interest (ROI) relates to the definition of the contour of an object or a part of an object to be considered. Typically, the manner in which the object (or part of the object) appears in the image is questioned, and this concept is commonly used in medical imaging applications. For analysis or evaluation, the ROI in a medical image is selected or identified. For example, the contour of a tumor on a mammogram can be defined as the ROI, which can be used, for example, to measure the size of the tumor; during a coronary calcium scan, the endocardial boundary can be identified as the ROI at various stages of the cardiac cycle (e.g., end-systole and end-diastole), which can be done for the evaluation of cardiac function; during a lumbar DXA scan, the femoral neck can be identified as the ROI, which can be done for the calculation of bone mineral density; during a wrist CT scan, the distal radius ROI can be selected, which can be done for the evaluation of bone microstructure.

[0004] The selection of a region of interest (ROI) is crucial in medical image analysis to ensure that clinically relevant regions are identified for clinical analysis. For example, the distal radius region can be selected for bone microstructure assessment, but inappropriate region selection can render the assessment ineffective. Selecting an ROI that is too proximal will result in a region containing insufficient trabecular tissue; selecting an ROI that is too distal will result in a region containing only thin cortex.

[0005] Furthermore, some diagnostic analyses require comparing the quantitative analysis of selected ROIs with reference values ​​calculated from large data samples. Inaccurate ROI selection renders such comparisons invalid. Accurate selection of ROIs with respect to medical imaging is essential for disease monitoring or treatment. If ROI selection is not accurate and reproducible, different ROIs from different visits to the same patient will be analyzed and compared, potentially compromising the effectiveness of treatment planning.

[0006] The ROI can be selected manually or semi-manually. Manual methods require an operator to outline the ROI according to predetermined rules. For example, U.S. Patent No. 7,609,817 discloses a method for selecting the femoral neck from a lumbar image in bone densitometry. This patent discloses a femoral neck box with dimensions of 1.5 cm × 5.0 cm, where the box is centered in a new coordinate system such that its width (1.5 cm) is parallel to the x-axis (or the midline of the femur). The user can move the femoral neck box along and perpendicular to the midline of the femur, and is prompted to position the corners of the femoral neck box to align with the bone edge of the greater trochanter. Furthermore, if the femoral neck is very short, the user is prompted to reduce the width of the femoral neck box.

[0007] Some known systems include semi-manual ROI selection. In such semi-manual methods, the operator typically provides some initial input, which is used to identify and select the ROI. For example, XtremeCT® (a high-resolution quantitative CT (HR-pQCT) system for peripheral bones from Scanco Medical AG in Blützelen, Switzerland) requires the operator to select the ROI to be scanned. The operator first acquires a 2D anterior-posterior projection of the proximal limb. On the 2D projection, the operator visually identifies anatomical landmarks and manually intersects them with horizontal reference lines. From the reference lines, a standard distance offset is applied to the anatomical volume to be scanned.

[0008] However, whether manual or semi-manual, both methods are labor-intensive and cumbersome, and ROI selection is inconsistent among operators, as is the ROI selection of individual operators. Fluctuations in ROI selection are a significant concern in cross-sectional and multicenter observational studies, and also when pooling cross-sectional datasets for post-hoc analysis (where multiple operators may have been involved in data collection). Furthermore, selecting ROIs based on heuristic rules presents another problem: ROI selection for evaluation should include consideration of clinical significance, but selection based on heuristic rules can lead to inferior clinical evaluation. For example, as mentioned above, the femoral neck is defined as a box of fixed size, and the wrist volume to be scanned is selected based on a fixed distance. Either the femoral neck or distal wrist volume is used to characterize the skeletal status of individual patients. Because human femurs vary in size, the femoral neck also varies in size. A fixed-size box may include other musculoskeletal materials, such as muscle and adipose tissue, in patients with small femurs, while in patients with large femurs, the box may encompass only a portion of the femoral neck. The volume of the wrist to be scanned is selected at a fixed distance from anatomical landmarks. However, bone morphology can vary significantly along the bone axis. Therefore, comparing microstructural assessments of patients with different bone lengths using these methods will usually not lead to clinically useful conclusions.

[0009] Another existing method for ROI selection involves co-registration. For example, one voxel-based morphometry technique for brain MRI data involves spatially normalizing all subject scans to the same stereotactic space. This involves aligning each image to the same template image, which is done by minimizing the sum of squared residuals between them. Then, gray matter (corresponding to the ROI) is selected from the normalized images, smoothing is applied, and finally, statistical analysis is performed to localize the differences between groups. This method allows for ROI selection that takes brain morphology into account. However, this method is mainly used in group comparison studies and rarely for single subjects because aligning all brain scans to the same template eliminates the differences in anatomical structure of the brain between subjects.

[0010] Registration is also used to improve ROI selection prediction during patient monitoring. When monitoring changes in bone microstructure, analysis of the same ROI across repeated CT scans is crucial. In longitudinal studies, 3D registration is used to select the same ROI; this involves rotating and translating follow-up scans (e.g., second or third scans) to the coordinates of the first scan (or reference image). The registration procedure allows for precise selection of baseline and follow-up scan ROIs for the same subject. However, in many diagnostic analyses, measurements obtained for a patient are compared to reference data derived from a larger population. Even with registration techniques, the problem of suboptimal ROI selection across subjects remains. [Prior art documents] [Patent Documents]

[0011] [Patent Document 1] U.S. Patent No. 7,609,817 [Overview of the project]

[0012] According to a first aspect of the present invention, a computer implementation method for selecting one or more regions of interest (ROIs) within an image is provided, the method being: A step of identifying one or more objects segmented from an image, Steps include identifying predefined landmarks of an object, The process involves performing morphometric measurements on an object by referencing landmarks to determine the reference morphometrics associated with the object, and A step of selecting one or more ROIs from an object according to reference morphometrics, comprising identifying the location of the ROIs in relation to the reference morphometrics, This includes the step of outputting one or more selected ROIs.

[0013] In embodiments, the step of determining the reference morphology further includes measuring the underlying morphology on an object by referencing detected landmarks and determining the reference morphology based on the measured underlying morphology. For example, determining the reference morphology may include using one or more trained deep learning reference morphological models.

[0014] In the embodiment, the step of selecting one or more ROIs from an object according to reference morphometrics further includes determining the shape and size of the ROIs.

[0015] In embodiments, the method further includes the step of segmenting objects from an image. For example, the step of segmenting objects includes generating a mask for each segmented object. The step of segmenting objects may include using one or more trained machine learning segmentation models. For example, the method includes the step of performing multiple segments.

[0016] In embodiments, the method further includes the step of preparing the image for segmentation by preprocessing the image. For example, the image preprocessing may include applying a Gaussian blur function to the image to reduce noise and / or applying histogram equalization to improve the contrast of the image.

[0017] In embodiments, the method further includes the step of normalizing an object before performing morphometric measurements on the object. For example, the normalization of the object may include a coordinate transformation.

[0018] In embodiments, the method may further include the step of determining the density of one or more selected ROIs using the attenuation of the material to be indicated (for example, adjacent tissues such as muscle or fat, or a calibration phantom).

[0019] A second aspect of the present invention provides a system for selecting one or more regions of interest (ROIs) within an image, wherein the system: An object and landmark classifier configured to identify one or more objects segmented from an image, and to identify predefined landmarks of the objects, A morphometer configured to perform morphometric measurements on an object by referencing landmarks and thereby determine the reference morphometrics associated with the object, A region selector configured to select one or more ROIs from an object according to reference morphometrics, comprising identifying the location of the ROI and determining the shape and size of the ROI in relation to the reference morphometrics, It includes a result output configured to output the selected ROI.

[0020] In an embodiment, the morphometer is configured to determine a reference morphometry by measuring a basic morphometry on an object by referring to the detected landmarks, and to determine the reference morphometry based on the measured basic morphometry. For example, the morphometer can be further configured to determine the reference morphometry by using one or more trained deep learning reference morphometry models.

[0021] In an embodiment, the region selector is further configured to determine the shape and size of the ROI in relation to the reference morphometry.

[0022] In an embodiment, the system further includes a segmenter configured to segment an object from an image. By way of example, the segmenter is further configured to generate a mask for each segmented object. The segmenter can segment the object using one or more trained machine learning segmentation models.

[0023] In an embodiment, the system further includes an image preprocessor configured to prepare the image for segmentation by preprocessing the image. By way of example, the preprocessor is configured to reduce noise by applying a Gaussian blur function to the image and / or to improve the contrast of the image by applying histogram equalization to the image.

[0024] In an embodiment, the system further includes a normalizer configured to normalize the object before performing morphometry on the object. By way of example, the normalizer normalizes the object by coordinate transformation.

[0025] In an embodiment, the system further includes a density determiner configured to determine the density of one or more selected ROIs using the attenuation of the indicated material.

[0026] According to a third aspect of the present invention, a computer program is provided comprising program code configured to implement the method for selecting a region of interest (ROI) in an image according to the first aspect, when executed by one or more processors. In this aspect, a computer-readable medium comprising such a computer program is also provided.

[0027] Any of the various individual features of each of the above-described aspects of the present invention, as well as any of the various individual features of the embodiments described herein, including the claims, can be combined in an appropriate and desired manner. [Brief explanation of the drawing]

[0028] To more clearly define the present invention, exemplary embodiments will be described below with reference to the accompanying drawings. [Figure 1] This is a schematic diagram of a segmentation system according to an embodiment of the present invention. [Figure 2] This figure shows a schematic workflow of the operation of the segmentation system shown in Figure 1, according to an embodiment of the present invention. [Figure 3] This figure shows the morphological measurements performed on the femur using the segmentation system shown in Figure 1. [Figure 4] Figures 4A to 4F illustrate the intermediate results generated by the segmentation system in Figure 1 at various stages of the workflow in Figure 2. [Figure 5] This figure shows the morphological measurements performed on the radius using the segmentation system shown in Figure 1. [Figure 6] Figures 6A to 6N show the intermediate results generated by the segmentation system in Figure 1 at various stages of the workflow in Figure 2. [Modes for carrying out the invention]

[0029] system Figure 1 is a schematic diagram of a segmentation system according to an embodiment of the present invention.

[0030] Referring to Figure 1, System 10 comprises an ROI (region of interest) selection controller 12 and a user interface 14 (including a GUI 16). The user interface 14 typically includes one or more displays (one or more of which can display the GUI 16), a keyboard and mouse, and optionally a printer. The ROI selection controller 12 includes at least one processor 18 and memory 20. System 10 can be implemented, for example, as a combination of software and hardware on a computer (e.g., as a personal computer or mobile computing device) or as a dedicated ROI selection system. System 10 can optionally be distributed; for example, all or some components of memory 20 can be located remotely from the processor 18; and the user interface 14 can be located remotely from memory 20 and / or the processor 18.

[0031] The memory 20 can communicate data with the processor 18 and typically includes both volatile and non-volatile memory (and may include one or more of each memory type), and includes RAM (Random Access Memory), ROM, and one or more mass storage devices.

[0032] As will be described in more detail later, the processor 18 includes an image processor 22, a segmenter 24, an object and anatomical landmark classifier 26, a normalizer 28, a morphometer 30 (including a reference morphometry determinator 32), an ROI (region of interest) selector 34, an optional index determinator 35, an optional density determinator 36, an I / O interface 37, and a result output 38. The memory 20 includes program code 40, an image storage unit 42, a trained segmentation model 44, an object and anatomical landmark classifier model 46, a morphometry setting 48, a reference morphometry model 50, and an ROI selection setting 52. The ROI selection controller 12 is implemented, at least in part, by the processor 18 executing the program code 40 from the memory 20.

[0033] Broadly speaking, the I / O interface 37 is configured to read medical images into the image storage unit 42 of the memory 20 for processing. Once the system 10 has completed processing, the I / O interface 37 outputs the results, for example, to the result output 38 and / or GUI 16. For example, the I / O interface 37 can read a CT scan in DICOM format into the image storage unit 42. After the processor 18 has processed the scan, the selected ROI can be output, for example, presented to the user, stored, used for further analysis (or any combination thereof).

[0034] The image processor 22 of processor 18 is configured to process the image before segmentation. For example, processor 18 can remove noise from the input image or improve the contrast of the input image. Processor 18 can also process the image after segmentation. For example, processor 18 can process the segmented object to facilitate subsequent morphometry.

[0035] The segmenter 24 performs segmentation using a trained segmentation model 44, which in this embodiment utilizes artificial intelligence, i.e., machine learning algorithms such as deep learning neural networks. Different models are trained for different applications. For example, to segment brain images from one or more MRI scans, the segmentation model 44 may have been trained using a deep convolutional neural network with the MRI scans as training data. Such training involves experts annotating the scans. Alternatively, to segment the radius from a wrist CT scan, the segmentation model 44 may have been trained using a deep convolutional neural network with annotated wrist CT scans as training data.

[0036] The morphometer 30 measures objects according to the morphometric settings 48. Different objects require different morphometric measurements, and the appropriate settings for each case are stored as morphometric settings 48. For example, the morphometric measurements required for the radius are different from those required for the femur. Even for the same object, different morphometric measurements are required for different ROI selections. For example, for femoral trochanter region selection from a lumbar DXA scan, it is sufficient to measure only the cervical axis and shaft axis. For femoral neck selection, it may be necessary to measure the femoral head center and femoral head fracture plane as well.

[0037] The ROI selector 34 selects an ROI from a segmented object according to the morphometric measurement and ROI selection settings 52. For example, when selecting the femoral neck from a lumbar DXA scan, the height of the femoral neck box can be defined as various percentage ratios of the lumbar axis length to suit various analytical purposes.

[0038] System workflow Figure 2 schematically illustrates a typical workflow 54 of system 10. Referring to Figure 2, in S56, the source image (e.g., in DICOM, TIFF, or PNG format) is loaded into the memory 20 of system 10. Preferably, the memory 20 allows system 10 high-speed data access. For example, if system 10 is implemented as a combination of software and hardware on a computer (e.g., a PC), the image may be loaded into the RAM of memory 20.

[0039] In S58, the image processor 22 preprocesses the loaded image if required or desired. This prepares the image for segmentation, which may improve the segmentation results (including ensuring that the results reach a desired quality threshold). For example, preprocessing may include applying a Gaussian blur function to the image to reduce noise, or applying histogram equalization to improve the image contrast, or doing both.

[0040] In S60, the segmenter 24 segments the object of interest from the image, and, if desired, includes generating a mask for the segmented object. In this embodiment, the segmentation model 44 is generally used for segmentation. Different learning models are trained for different objects. For example, different models are trained for segmenting the brain from an MRI scan and for segmenting the radius from a wrist CT scan. Machine learning, such as a deep learning model, is used to achieve accurate segmentation. However, in other embodiments, alternative segmentation of the object of interest is performed using more conventional image processing algorithms such as contour detection or blob detection, which may provide acceptable accuracy for some applications.

[0041] Furthermore, this step may include one or more segmentations (for example, segmenting bone material from surrounding soft tissue and then segmenting one or more specific bones from the bone material). In such cases, different segmentation methods may be used for each segmentation; for example, a pre-trained segmentation model 44 may be used for the first segmentation and a more conventional image processing method for the second segmentation (or vice versa), or two pre-trained segmentation models may be used (one model may be used for each segmentation step).

[0042] In S62, the object of interest is acquired by segmentation, and the object and part landmark classifier 26 recognizes and identifies part landmarks associated with the object. For example, if the segmenter 24 segments the radius and ulna from the surrounding material of the image in this step, the radius is recognized and identified, and then, for example, the radial head, styloid process, and ulnar notch are recognized and identified. The identified part landmarks are later used for morphometric measurements.

[0043] Object and part landmarks are identified by a pre-trained deep learning model, which is stored within the object and part landmark identification model 46.

[0044] In S64, the normalizer 28 normalizes the segmented object to facilitate subsequent morphometric processing. Normalization typically involves coordinate transformation, which is particularly relevant for 3D scans. An object is segmented from an image presented in an image coordinate system, while morphometrics are performed, for example, on parts. Therefore, regarding the example of parts, please note that morphometrics are simplified by transforming the object in the image coordinate system to a part coordinate system.

[0045] Next, the morphometrics themselves are performed, and automated ROI selection is carried out by referencing one or more clinically significant morphometrics (referred to here as reference morphometrics). While in some other cases the reference morphometrics can be measured directly, this is generally measured after segmentation and landmark detection.

[0046] For example, in a DXA lumbar scan, lumbar length is one of the reference morphometrics, which is calculated from several base morphometrics, the base morphometrics themselves being determined by referencing detected landmarks. The calculation itself can be performed by any appropriate (including conventional) image processing algorithm.

[0047] To give another example, radial length is used as a reference morphological parameter by the ROI selector when selecting bone volume for bone microstructure analysis. Radial length cannot be measured directly because scanning the entire forearm is generally not feasible (because such a scan would involve excessive radiation exposure). Therefore, basic morphological parameters such as distal width, median diaphysis diameter, maximum distal diameter, and minimum distal diameter are measured first. In this embodiment, a deep learning model, pre-trained to predict radial length based on the aforementioned basic morphological parameters, determines the radial length.

[0048] Therefore, in S66, the morphometer 30 determines the basic morphometrics for one or more segmented and identified objects, which is done in accordance with the morphometrics setting 48. Different basic morphometrics are determined for different objects and different requirements for ROI selection. The morphometer 30 is configured to perform image processing to make such measurements. The morphometer 30 measures desired basic morphometrics such as the tangent to the femoral head, the cervical axis, and the lumbar end, and this measurement is made by referring to the site landmarks identified by the object and site landmark classifier 26 in S62. For example, for the selection of a femoral neck ROI from a lumbar DXA image, the morphometer 30 will use site landmarks such as the tangent to the femoral head, the center of the femoral head, the cervical axis, and the diaphysis axis.

[0049] Some reference morphometrics can be measured directly in certain applications. For example, if distal radius width is selected as the reference morphometrics, it can be measured directly by referencing a detected landmark. In such situations, the morphometer 30 measures one or more required reference morphometrics without requiring the use of a reference morphometrics determiner. That is, the reference morphometrics are the base morphometrics.

[0050] However, this is generally not valid; in many cases, the reference morphometrics are calculated by the reference morphometrics determinator 32 based on the base morphometrics. This is done in S68. For example, for a lumbar DXA scan, the reference morphometrics determinator 32 can determine the length of the line between the apex of the femoral head and the center point of the final slice between the bones (both points have been identified in S66) when calculating the lumbar length. The derivation of the line can be done using any appropriate image processing algorithm.

[0051] However, the reference morphometrics determinator 32 of this embodiment has the function of determining the reference morphology using one or more pre-trained deep learning reference morphometrics models 50. As described above, the deep learning reference morphometrics model can be trained to predict the radial length based on measurements of distal width, diaphysis diameter, maximum distal diameter, minimum distal diameter, etc. (these are the basic morphometrics measured in S66). Therefore, the reference morphometrics determinator 32 of this embodiment can determine the radial length (which is the reference morphology in this example) using the deep learning reference morphometrics model 50 based solely on the analysis of the distal portion of the subject's forearm.

[0052] Next, the ROI selector 34 selects one or more ROIs from the object according to each purpose, which is done in accordance with the ROI selection settings 52, and the reference morphometrics are calculated in S68. The ROI selector 34 performs ROI selection in the following two steps: in S70, the ROI selector 34 identifies the position of each respective ROI, and in S72, the ROI selector 34 determines the shape and size of each ROI (whether two-dimensional or three-dimensional). For example, with regard to the selection of the distal radius region, the ROI selector 34 first identifies the position of the ROI as a percentage of the radial length from the distal end of the radius, and then the ROI selector 34 defines the thickness of the selected volume as another percentage of the radial length.

[0053] Results including the ROIs selected in S74 are output. This action may involve passing the results to result output 38 and outputting them to the user, for example (e.g., presenting them to the user via the monitor of the user interface 14). Alternatively, this action may involve passing these results to another system (e.g., via the I / O interface 37). In further alternatives, this action may involve using the results as input for further analysis by, for example, an arbitrary index decisioner 35 or density decisioner 36 (or both). Any combination of these is also conceivable.

[0054] The index determination unit 35 is configured to segment the selected ROI into different structures using a segmenter and to calculate characterization indices that characterize these segments. For example, after selecting a bone volume ROI from a CT wrist scan, the index determination unit 35 controls the segmenter 24 to further segment the selected ROI into cortical, transitional, and trabecular regions, and then calculates indices that characterize the segments (e.g., trabecular structure and porosity). Segmentation and / or index determination can be performed using, for example, the method described in U.S. Patent No. 9,064,320, entitled "Method and System for Image Analysis of Selected Tissue Structures".

[0055] The density determination unit 36 ​​is configured to determine the density of a selected ROI using the attenuation of a second material (e.g., muscle tissue or other adjacent material, or a calibration phantom with a known density) as the indicated element, again employing the method described in U.S. Patent No. 9,064,320. The second material is scanned by the same scanner, and a formula can be established to convert the attenuation value into a density value.

[0056] And the process is finished.

[0057] Selection of ROI for the femoral neck from lumbar DXA images Figures 3 and 4A to 4F show an example of the application of system 10 to ROI selection of the femoral neck from lumbar DXA images. Figure 3 shows morphometric measurements performed on the femur. Figures 4A to 4F show the intermediate results generated by system 10 at various steps in workflow 54.

[0058] As generally shown in Figure 3, 76: (After the femoral head and the apex of the femoral head are recognized and identified in S62,) the tangent line (L1) of the femoral head 78 is identified; the cervical axis (L2) is identified as the centerline passing through the femoral neck; the diaphysis axis (L3) is identified as the centerline passing through the diaphysis; the minimum transverse line (L4) is identified as the shortest line segment parallel to L1 and crossing the femoral neck; the lumbar end line (L5) is identified as a line passing through the intersection of L2 and the diaphysis and parallel to L1; the axial intersection (P1) is identified as the intersection of the cervical axis (L2) and the diaphysis axis (L3); and the center point of the femoral head (P2) is identified.

[0059] Based on the identified reference lines and points, the following morphometrics are measured by the reference morphometrics determiner 32: the lumbar axis length (D1) is calculated as the distance between L1 and L5; the distance between the axis intersection (P1) and the femoral head center point (P2) is calculated as D2; and the distance between the femoral head center point (P2) and the minimum transverse line (L4) is calculated as D3.

[0060] Figure 4A is the original image in lumbar DXA scan format, which is input to system 10 (see S56); Figure 4B is an inverted version 80' of image 80 (provided to assist in naked-eye examination and interpretation of Figure 4A). Figure 4C shows the result of applying a pre-trained segmentation model 44 (preferably deep learning type) by the segmenter 24 (see S60), which is done to segment the femur from the image and to generate a segmented mask 82 for the segmented object (femur in this example).

[0061] Figure 4D shows the results of morphometric measurements performed on the segmented femur (see S66). Figure 4E shows the selection of the femoral neck ROI by ROI selector 34, indicated by reference numeral 86 (see S70 and S72). The minimum transverse line (L4 in Figure 3) is the centerline of the region, and the region width is selected as a percentage of the lumbar axis length (D1 in Figure 3) (10% in this example). (This percentage is configurable and stored within the ROI selection setting 52). ROI selection can be changed for different purposes or studies.

[0062] Figure 4F shows, indicated by reference numeral 88, that the width of the selected ROI is 10% of D2 (see Figure 3).

[0063] Selection of distal radial ROI from wrist CT images Figures 5 and 6A to 6N illustrate an application example of system 10 when selecting the distal radial region as the ROI from wrist CT images. Figure 3 shows morphometric measurements performed on the radius. Figures 6A to 6N show intermediate results generated in the steps of the ROI selection workflow 50. (Figures 6B, 6D, 6F, 6H, 6J, 6L, and 6N are inverted versions of Figures 6A, 6C, 6E, 6G, 6I, 6K, and 6M, respectively, and are provided here to support naked-eye examination and interpretation of Figures 6A, 6C, 6E, 6G, 6I, 6K, and 6M.)

[0064] As shown in Figure 5, 90, the distal radius breadth (RDB) is measured as the distance from the innermost point of the ulnar notch to the outermost point of the styloid process; the radius length (RL) is measured as the distance from the nearest end of the radial head to the tip of the styloid process.

[0065] Figure 6A is the original image in the form of a wrist CT scan. Figure 6C shows the bone segmented by the segmenter 24 from the surrounding tissue (using a pre-trained deep learning segmentation model 44) (see S60). Figure 6E shows the radius segmented by the segmenter 24 from other bones (in this case, using a conventional connection component detection algorithm) (see S60). Figure 6G shows the segmented radius, which has been transformed from the image coordinate system to the radial region coordinate system by the normalizer 28 (see S64). Figure 6I shows how the distal radius width (arrow-shaped lines at both ends) is measured. Figure 6K shows how the region of interest (dashed box) has been selected. The selected ROI is at a distance R from the tip of the distal radius. DB It starts from this point. The height of the region is R DB It has been selected as half of the distal radius. Figure 6M is an image of the selected distal radius.

[0066] Those skilled in the art will realize that many modifications can be made without departing from the scope of the present invention, and it should be noted in particular that further embodiments can be brought about by utilizing certain features of the embodiments of the present invention.

[0067] Even if there are references to prior art in this specification, it should be noted that such references do not constitute an admission that the prior art is part of well-known technology in any country.

[0068] In the attached claims and the aforementioned detailed description of the invention, unless the context explicitly or implicitly requires a different interpretation, the terms “equipped with,” “equipped” (third-person singular present tense), and “equipped with” are used in a comprehensive sense (i.e., to specify the presence of the declared features), but this does not preclude the existence or addition of further features in the various embodiments of the invention. [Explanation of Symbols]

[0069] Segmentation System 12 ROI Selection Controller 14. User Interface 16 GUI 18 processors 20 memory 22 Image Processors 24 Segmenter 26 Object and anatomical landmark identifyrs 28 Normalizer 30 Morphometers 32 Reference Morphological Measurement Determinant 34 ROI Selector 35 Index Determinant 36 Density determiner 37 I / O Interfaces 38 Result Output 40 Program Code 42 Image storage unit 44 Trained Segmentation Models 46 Object and Anatomical Landmark Identification Models 48 Morphological Measurement Settings 50 Reference Morphological Measurement Model 52 ROI Selection Settings

Claims

1. A computer implementation method for selecting one or more regions of interest (ROIs) within an image, wherein the method is: The steps include identifying one or more objects of interest segmented from the aforementioned image, A step of identifying a predefined anatomical landmark of the object, wherein the anatomical landmark is selected from a group of lines, the shortest line segment parallel to the tangent to the femoral head and crossing the femoral neck, or a group of lumbar end lines passing through the intersection of the midline passing through the femoral neck and the shaft and parallel to the tangent to the femoral head. A step of determining reference morphometrics associated with the object by performing morphometric measurements on the object by referring to the aforementioned anatomical landmarks, wherein the morphometric measurements include calculating, as a reference distance, the distance between the tangent to the femoral head and the lumbar end line, or the distance between the center point of the femoral head and the shortest line segment. A step of selecting one or more ROIs from the object according to the reference morphological metrics, comprising identifying the position of the ROI in relation to the reference morphological metrics, wherein the center line of the ROI is the shortest line segment, and the width of the ROI is determined as a predetermined percentage of the reference distance. A method comprising the step of outputting one or more selected ROIs.

2. A method according to claim 1, wherein the step of selecting one or more ROIs from the object according to the reference morphological features further includes determining the shape and size of the ROIs.

3. A method according to claim 1 or 2, further comprising the step of segmenting the object from the image.

4. In the method according to claim 3, the step of segmenting the object is: (a) including generating a mask for each segmented object, and / or (b) Using one or more trained machine learning segmentation models, method.

5. A method according to claim 3 or 4, comprising the step of performing a plurality of segmentations.

6. A method according to any one of claims 1 to 5, further comprising the step of preparing the image for segmentation by preprocessing the image.

7. A method according to claim 6, wherein the image preprocessing includes applying a Gaussian blur function to the image to reduce noise and / or applying histogram equalization to improve the contrast of the image.

8. A system for selecting one or more regions of interest (ROIs) within an image, wherein the system: An object and landmark classifier configured to identify one or more objects of interest segmented from the aforementioned image and to identify predefined anatomical landmarks of the objects, wherein the anatomical landmarks are selected from a group of lines, the shortest line segment parallel to the tangent to the femoral head and crossing the femoral neck, or a group of lumbar end lines passing through the intersection of the midline passing through the femoral neck and the shaft and parallel to the tangent to the femoral head. A morphometer configured to perform morphometric measurements on an object by referring to the aforementioned anatomical landmarks and thereby determine reference morphometrics associated with the object, wherein the morphometric measurements include calculating, as a reference distance, the distance between the tangent to the femoral head and the lumbar end line, or the distance between the center point of the femoral head and the shortest line segment, A region selector configured to select one or more ROIs from the object according to the reference morphological metrics, comprising identifying the position of the ROI in relation to the reference morphological metrics, wherein the center line of the ROI is the shortest line segment, and the width of the ROI is determined as a preset percentage of the reference distance, A system comprising a result output configured to output the selected ROI.

9. The system according to claim 8, wherein the region selector is further configured to determine the shape and size of the ROI in relation to the reference morphometrics.

10. A system according to claim 8 or 9, further comprising a segmenter configured to segment the object from the image.

11. In the system according to claim 10, the segmenter is (a) Generate a mask for each segmented object, and / or (b) A system further configured to segment the object using one or more trained machine learning segmentation models.

12. The system according to any one of claims 8 to 11 further includes an image preprocessor, the image preprocessor being (a) By preprocessing the image, (b) By preprocessing the image and applying a Gaussian blur function to the image to reduce noise, and / or (c) By preprocessing the image and applying histogram equalization to the image to improve the contrast of the image, A system for preparing the aforementioned images for segmentation.

13. A computer program comprising program code configured to implement the method for selecting a region of interest (ROI) in an image according to any one of claims 1 to 7 when executed by one or more processors.

14. A computer-readable medium comprising the computer program described in claim 13.

Images