Method and system for selecting region of interest in image

Abstract

To solve the problem in which the efficacy of a treatment regimen may be compromised if selection of regions of interest (ROIs) of the same patient in different visits is not precise.SOLUTION: A computer-implemented method for selecting one or more regions of interest (ROIs) in an image. The method comprises the following steps: identifying one or more objects that have been segmented from the image; identifying predefined landmarks of the objects; determining reference morphometrics pertaining to the objects by performing morphometry on the objects by reference to the landmarks; selecting one or more ROIs from the objects according to the reference morphometrics, where the selecting step comprises identifying the location of the ROIs relative to the reference morphometrics; and outputting the selected one or more ROIs.SELECTED DRAWING: Figure 1

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. This paper concerns systems, particularly those that apply artificial intelligence to medical image analysis, such as morphometry. The region of interest may be located within such a medical image, but is not limited to these applications. Possible medical imaging applications include computer-aided histology. These include computed tomography (CT), magnetic resonance (MR), ultrasound, and lesion scanner imaging.

[0002] Related Applications This application is a joint venture with U.S. Patent Application No. 16 / 448,285 (filed June 21, 2019). No. 6,313,635, filed on Dec. 13, 2003, and which claims priority thereto, the entire contents of which are incorporated herein by reference at the time of filing. Be taken in. [Background technology]

[0003] A region of interest (ROI) is an area of ​​an object or part of an object that is under consideration. The definition of the contour is that of the object (or part of the object) that typically appears in the image. The question is how the concept is commonly used in medical imaging applications. Select or identify an ROI in a medical image for evaluation, e.g., a tumor on a mammogram. The contour of the region of interest can be defined as an ROI, which can be used, for example, to measure the size of a tumor. During coronary calcium scanning, for example, The endocardial border can be identified as an ROI at the end of the diastolic and end-diastolic phases, and this allows the evaluation of cardiac function. For this purpose, the femoral neck can be identified as the ROI for lumbar DXA scan. It should be performed to assess bone mineral density; the distal radius ROI should be placed on the wrist CT scan. This may be an option and may be performed to assess bone microarchitecture.

[0004] To ensure that clinically relevant regions are identified for clinical analysis Therefore, the selection of ROIs is important in medical image analysis. For example, in the evaluation of bone microstructure, Although the distal region of the radius can be selected for this purpose, inappropriate selection of the region may result in invalid bone microarchitectural evaluation. If the ROI is selected too close, the region will contain insufficient trabecular tissue. If you select an ROI that is too distal, the region will contain too little cortex. This will be the case.

[0005] In addition, in some diagnostic analyses, the quantitative analysis items of the selected ROI are analyzed on a large scale. The ROI should be compared with the reference value calculated from the data sample. This makes such comparisons invalid. If the ROI selection is not accurate and reproducible, Different ROIs at different visits for the same patient are analyzed and compared, and treatment planning The effectiveness of the method may be impaired.

[0006] The ROI can be selected manually or semi-manually. In the manual method, the operator selects a predetermined It requires drawing the outline of the ROI based on rules. For example, see U.S. Patent No. 7,609,810. No. 7 discloses a method for selecting the femoral neck from a lumbar image in bone densitometry. The literature describes a femoral neck box measuring 1.5 cm by 5.0 cm in width and length. The box was then re-arranged so that its width (1.5 cm) was parallel to the x-axis (or the midline of the femur). The user must center the femoral neck box on the midline of the femur. The femoral neck box corners can be moved along and perpendicular to the femoral neck box corners of the greater trochanter bone. In addition, if the femoral neck is very short, You will be encouraged to reduce the width of your cervical box.

[0007] Some known systems involve semi-manual ROI selection. These methods usually require some initial input from the operator, based on which For example, XtremeCT® (Brussels, Switzerland) is used to identify and select ROIs. Scanco Medical AG's High Resolution Peripheral Quantitative CT (HR-pQCT) system in Tyseren This requires the operator to select the ROI to be scanned. First, a 2D anterior-posterior projection is obtained of the proximal limb. On the 2D projection, the operator The anatomical landmarks are intuitively identified and horizontal reference lines are manually drawn. From the reference line, a standard line is drawn for the anatomical volume to be scanned. Add a distance offset.

[0008] However, whether manual or semi-manual, both are labor-intensive and tedious, and operators ROI selection was inconsistent between operators and within individual operators. Variability in ROI selection is an important concern in cross-sectional and multicenter observational studies, and post hoc For the analysis, cross-sectional data (where multiple operators may have been involved in data collection) This is also an important concern when pooling data sets. Selecting an ROI based on should include consideration of clinical significance but be based on heuristic rules Selection may result in poor clinical evaluation. For example, as mentioned above, the femoral neck is fixed. The volume of the wrist to be scanned is fixed. Based on the distance measured, either the femoral neck or distal wrist volume selections are used to determine the individual Characterize the skeletal condition of each patient. Since human femurs vary in size, the femoral neck also varies in size. The fixed size box is suitable for patients with small femurs and other musculoskeletal The material can include muscle and fat tissue, and in patients with large femurs, The wrist volume to be scanned is within the anatomical limits. The marks are selected at a fixed distance from each other, although bone morphology varies significantly along the bone axis. Therefore, these techniques provide a method to obtain microstructure information for patients with different bone lengths. Comparisons of ratings will generally not lead to clinically useful conclusions.

[0009] Another existing method for ROI selection involves coregistration. For example, one voxel-based morphometry approach for brain MRI data first This involves spatially normalizing all of a subject's scans to the same stereotactic space. This involves performing a registration for each image of the template image, which is This is done by minimizing the sum of squares of the residuals. Then, the gray matter (corresponding to the ROI) is positive. From the normalized images, a smoothing process is applied, and finally a statistical analysis is performed to determine the groups. This method allows for selection of ROIs taking into account brain morphology. However, this method is primarily used in group comparison studies and not in single subject studies. This is rarely used as a method to align all brain scans to the same template. This is because matching would eliminate differences in brain anatomical structure between subjects.

[0010] Registration is also used to improve ROI selection prediction for patient monitoring. When monitoring microstructural changes in the same ROI in repeated CT scans, In longitudinal studies, 3D registration is required to select the same ROIs. follow-up scans (e.g., second and third scans) This involves rotating and translating the image to the coordinates of the first scan (or reference image). The registration procedure involves comparing baseline and follow-up scans of the same subject. OIs can be precisely selected, although in most diagnostic assays, the The measurements are compared to reference data calculated from a large population. However, this does not overcome the problem of suboptimal ROI selection across subjects. do not have. [Prior art documents] [Patent documents]

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

[0012] According to a first aspect of the present invention, one or more regions of interest (ROIs) are identified in an image. The present invention provides a computer-implemented method for selecting a number of interests, the method comprising: : identifying one or more segmented objects from the image; identifying predefined landmarks on the object; Morphometric measurements are performed on the object by referencing the landmarks, thereby determining a reference morphometric associated with selecting one or more ROIs from the object according to the reference morphometrics. and identifying a location of the ROI relative to the reference morphometrics. P and and outputting the selected one or more ROIs.

[0013] In an embodiment, the step of determining the reference morphometrics comprises determining a detected landmark Measuring basic morphometrics on an object by referencing the measured determining a reference morphometrics based on the base morphometrics; For example, determining the reference morphometrics may involve the use of one or more trained deep learning references. The method may include using a reference morphometric model.

[0014] In an embodiment, one or more ROIs are selected from the object according to a reference morphometric. The steps further include determining a shape and a size of the ROI.

[0015] In an embodiment, the method further comprises segmenting the object from the image. Specifically, the step of segmenting the objects includes generating a mask for each segmented object. The step of segmenting the object includes generating a segmentation map using one or more trained machine learning segmenters. By way of example, the method may include using a segmentation model. The method includes a step of performing a stationary configuration.

[0016] In an embodiment, the method comprises preparing the image for segmentation by pre-processing the image. For example, the pre-processing of the image may further include a step of preparing the image for processing by applying a Gaussian blur function to the image. Applying histogram equalization to reduce noise and / or histogram equalization to reduce image contrast This may include improving trust.

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

[0018] In an embodiment, the method further comprises: computing the density of one or more selected ROIs based on the attenuation of the indicated material; The method may further include determining the concentration of the protein using adjacent tissue such as muscle or fat or a calibration filter. (may include phantoms).

[0019] According to a second aspect of the invention, one or more regions of interest (ROIs) are identified in an image. A system for selecting interests of interest is provided, the system comprising: Identifying one or more objects segmented from the image and also determining a predefined range of objects an object and landmark classifier configured to identify landmarks; Morphometric measurements are performed on the object by referencing the landmarks, thereby a morphometer configured to determine a reference morphometrics associated with the configured to select one or more ROIs from the object according to a reference morphometrics; a region selector for identifying the location of the ROI and comparing it with the reference morphometrics; a region selector, which includes determining the shape and size of the ROI in relation to the region; and a result output configured to output the selected ROI.

[0020] In an embodiment, the morphometer detects the object by referencing detected landmarks. Determine the reference morphometrics by measuring the baseline morphometrics on the body. and determining a reference morphometrics based on the measured basic morphometrics. For example, the morphometer may be configured to use one or more pre-trained deep learning reference To determine the reference morphometrics by using the morphometric model The method may further comprise:

[0021] In an embodiment, the region selector selects the shape of the ROI relative to the reference morphometrics. and further configured to determine a size.

[0022] In an embodiment, the system includes a segmenter configured to segment an object from an image. For example, the segmenter may further include a segmentation processor. The segmenter is further configured to generate a segmentation result based on one or more trained machine learning A segmentation model can be used to segment the object.

[0023] In an embodiment, the system performs image segmentation by pre-processing the image. The image processing device may further include an image preprocessing device configured to prepare the image for processing. The processor may be configured to reduce noise by applying a Gaussian blur function to the image. and / or enhancing the contrast of the image by applying histogram equalization to the image. It is configured to make

[0024] In an embodiment, the system is adapted to normalize the object before performing morphometric measurements on the object. The image processing apparatus further includes a normalizer configured to: This normalizes the object.

[0025] In an embodiment, the system may further include a step of: determining whether or not the density of one or more selected ROIs is based on the attenuation of the indicated material; The density determiner further comprises a density determiner configured to determine using:

[0026] According to a third aspect of the present invention, a method for performing a method for detecting a plurality of stimuli, the method comprising the steps of: We have implemented a method for selecting a region of interest (ROI) in an image. In accordance with this aspect, there is provided a computer program comprising a program code configured to: Also provided in the present application is a computer readable medium comprising such a computer program. can be.

[0027] Any of the various individual features of each of the above-described aspects of the invention, as well as the claims Any of the various individual features of the embodiments described herein, including those described in the claims, can be combined in any suitable and desired manner. [Brief description of the drawings]

[0028] In order that the present invention may be more clearly defined, reference will now be made, by way of example only, to the accompanying drawings, in which: The form will be explained below. [Figure 1] FIG. 1 is a schematic diagram of a segmentation system according to an embodiment of the present invention. [Diagram 2]FIG. 2 is a diagram of a schematic workflow for the operation of the segmentation system of FIG. 1 according to an embodiment of the present invention. [Diagram 3] FIG. 2 shows morphometric measurements taken on a femur by the segmentation system of FIG. 1. [Figure 4] 4A-4F are diagrams illustrating intermediate results produced by the segmentation system of FIG. 1 at various stages of the workflow of FIG. [Diagram 5] FIG. 2 shows morphometric measurements taken on the radius by the segmentation system of FIG. 1. [Figure 6] 6A-6N are diagrams illustrating intermediate results produced by the segmentation system of FIG. 1 at various stages of the workflow of FIG. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

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

[0030] Referring to FIG. 1, the system 10 includes a region of interest (ROI) selection unit. The user interface 14 includes a selection controller 12 and a GUI 16. The user interface 14 typically includes one or more displays, one or more of which may be GUs. I16), a keyboard and mouse, and optionally a printer. The OI selection controller 12 includes at least one processor 18 and a memory 20 . The system 10 may 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 The system 10 may be implemented as a distributed ROI selection system. For example, all or some components of memory 20 may be The user interface 14 may be located at a location remote from the processor 18; may be located remotely from the memory 20 and / or the processor 18. .

[0031] The memory 20 is in data communication with the processor 18 and is typically volatile. and non-volatile memory (and one or more of each memory type). (which may include a RAM (random access memory), a ROM, and one or more This includes mass storage devices.

[0032] As will be described in more detail below, the processor 18 includes an image processor 22 and a segmenter 23. 4, an object and anatomical landmark classifier 26, a normalizer 28, and (reference morphometer a morphometer 30 (including a morphometer 32), a ROI (region of interest) selector 34, Optional Index Determiner 35, Optional Density Determiner 36, and I / O Interface 37 and result output 38. The memory 20 includes a program code 40 and an image storage section 4 2, a trained segmentation model,44 and an object and anatomical landmark identification model. 46, morphometry settings 48, reference morphometry model 50, and ROI selection settings 52. The ROI selection controller 12 is programmed, at least in part, from a memory 20. It is implemented by the processor 18 executing the code 40 .

[0033] In general terms, the I / O interface 37 transfers medical images to memory 2 for processing. 0 into image storage 42. Then, the I / O interface 37 outputs the results, for example to a result output 38 and / or a GUI 1. For example, the I / O interface 37 outputs a CT scan in DICOM format to the The scan may then be read into image store 42. After processor 18 processes the scan, the selected The ROIs can be, for example, presented to the user, stored, or used for further analysis. (or any combination thereof) and output.

[0034] The image processor 22 of the processor 18 is adapted to process the image before segmentation. For example, the processor 18 may be configured to remove noise in the input image or to The contrast of the force image can be improved. The image may also be processed after scanning. For example, the processor 18 may Processing can be carried out to facilitate subsequent morphometric measurements.

[0035] The segmenter 24 performs segmentation using a trained segmentation model 44. In this embodiment, this is done using artificial intelligence, i.e., deep learning neural networks. Use machine learning algorithms such as ML. Different models are trained for different applications. For example, to segment the brain images for one or more MRI scans, We used MRI scans as training data for the mentation model 44 and deep convolutional neural network model 44. It is possible that the model was trained using neural networks. The study involves the home visitor annotating the scan. To segment the radius, annotate the segmentation model44. A deep convolutional neural network was trained on the CT scans of a wrist that had been scanned. It is possible that they may have been training using

[0036] The morphometer 30 measures the object according to the morphometric settings 48. For different objects, different morphometrics are measured, and in each case The appropriate settings are stored as morphometric settings 48. For example, for the radius: The morphometric measurements required are different from those required for femoral measurements. For the same object, different ROI selections will result in different morphometric measurements. For example, femoral troch region from a lumbar DXA scan For femoral neck selection, it is sufficient to measure the neck axis and shaft axis. In some cases, it may be necessary to also measure the femoral head center and femoral head fracture plane.

[0037] The ROI selector 34 selects the region of interest according to the morphometric measurements and ROI selection settings 52. Select an ROI from a segmented object, for example the femoral neck from a lumbar DXA scan. When selecting, the femoral neck box height should be adjusted to the lumbar shaft length for various analytical purposes. It may be defined as various percentage ratios.

[0038] System Workflow FIG. 2 outlines a general workflow 54 of the system 10. For reference, in S56, the original image (e.g., DICOM, TIFF or PNG format) is The data is read into memory 20 of system 10. Memory 20 is preferably a high speed data access For example, the system 10 may be configured to allow a computer (e.g., When implemented as a combination of software and hardware on a PC, may be loaded into the RAM of memory 20.

[0039] At S58, the image processor 22 may, if required or desired, Preprocessing the image. This prepares the image for segmentation. This can improve segmentation results (note that this requires that the For example, preprocessing can involve applying a Gaussian blur function to the image. You can also apply histogram equalization to reduce noise and histogram equalization to improve image contrast. This may involve improving the quality of the product or doing both.

[0040] At S60, segmenter 24 segments an object of interest from the image, preferably If included, this involves generating a mask for the segmented object. In an embodiment, the segmentation model 44 is generally For different objects, different learning models are trained. For example, MRI scans Segmenting the brain from a can and the radius from a wrist CT scan For each of these, a different model is trained. Although other embodiments may use contour detection or blob detection. Alternative segmentation of the object of interest is performed using more conventional image processing algorithms such as This involves a quantization step, which may provide acceptable accuracy for some applications.

[0041] Additionally, this step may include one or more segmentations (e.g., perimeter Segment bone material from soft tissue and one or more specific bones from the bone material. In such cases, different segmentation methods are used for each segmentation. For example, the first segmentation may be trained in advance. Using the pre-trained segmentation model 44, we also performed a second segmentation can use more conventional image processing methods (or vice versa), or One can use two pre-trained segmentation models (one for each segmentation step). (One model can be used for each group.)

[0042] At S62, the object of interest is obtained by segmentation, and the object and The landmark classifier 26 recognizes landmarks related to the object and For example, segmenter 24 may use this step to distinguish the radius and ulna from the surrounding material in the image. When the bones are segmented, the radius is recognized and identified, and the radial head and styloid are identified. The ulnar notch and ulnar ridge are recognized and identified. The identified landmarks are used for morphometric analysis. This will be used later for measuring the

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

[0044] In S64, a normalizer 28 normalizes the segmented objects to improve the accuracy of subsequent morphometric analysis. Normalization usually involves coordinate transformation, especially for 3D scans. This is true: objects are segmented from images presented in the image coordinate system. On the other hand, morphometrics are performed on the site. For the example of a body part, we transform the object in the image coordinate system to the body part coordinate system: Please note that morphometrics are simplified.

[0045] Second, the morphometrics itself is performed, and the automatic ROI selection is performed based on one or more clinical See also: Functionally significant morphometrics (herein referred to as reference morphometrics) In some other cases, this is accomplished by directly measuring the reference morphometrics. Although it can be determined, typically this is measured after segmentation and landmark detection.

[0046] For example, for DXA waist scans, waist length is one of the reference morphometrics. This is calculated from several basic morphometrics. The risk itself is determined by referencing detected landmarks. The calculation itself is arbitrary. This can be done by any suitable (including conventional) image processing algorithm.

[0047] As another example, radius length can be used as a ROI when selecting bone volumes for bone microarchitecture analysis. The radial length cannot be measured directly. Because scanning the entire forearm is generally not feasible (because (This is because such scans involve excessive radiation exposure.) Thus, for example, Basic morphometrics such as mid-diaphyseal diameter, maximum distal diameter, and minimum distal diameter were first measured. In this embodiment, the radius length is predicted based on the basic morphometrics described above. A pre-trained deep learning model is used to determine radial length.

[0048] Thus, in S66, the morphometer 30 generates one or more segmented and identified The basic morphometrics are determined for the processed object, which is called morphometrics. For different objects and different requirements regarding ROI selection, The morphometer 30 determines different underlying morphometric measurements. The morphometer 30 is configured to perform image processing such as the tangent of the femoral head and The desired base morphometrics, such as the neck axis and lumbar extremity, are measured, and these measurements are Refer to the part landmarks identified by the object and part landmark identifier 26 in 62. For example, femoral neck ROI selection from lumbar DXA images is done using morphology. The Ometer 30 uses landmarks such as the femoral head tangent, femoral head center, neck axis, and shaft axis. There will be.

[0049] Some reference morphometrics may be measured directly in some applications. For example, If the distal radius width is selected as the reference morphometric, the detected landmarks are referenced. In this situation, it can be measured directly by specifies one or more reference morphometrics required by the reference morphometrics determiner The measurement is performed without the need for the use of the reference morphometrics. It is.

[0050] However, this is not generally true; in many cases, the reference morphometric The morphometrics are calculated by a reference morphometrics determiner 32 based on the basic morphometrics. This is done in S68. For example, for a lumbar DXA scan, the reference morph The hip length is calculated by the hip metric determiner 32 based on the apex of the femoral head and the final interosseous slap. Find the line between the center points of the chairs (both points were identified in S66) and calculate the length of that line. The derivation of the lines can be done using any suitable image processing algorithm.

[0051] However, the reference morphometrics determiner 32 in this embodiment may be implemented using one or more pre-trained Using the deep learning reference morphometric model 50, As mentioned above, the deep learning reference morphometric model The diameters of the distal bone, the diaphysis, the maximum distal diameter, and the minimum distal diameter (which were measured using the S66 standard) were used. Trained to predict radius length based on measurements of morphometric parameters (the cornerstone of morphometrics) Therefore, the reference morphometrics determiner 32 in this embodiment is ,Based solely on the analysis of the distal part of the subject’s forearm, this deep learning ,reference morphometrics Model 50 can be used to determine radial length (which is the reference morphometric for this example).

[0052] Next, an ROI selector 34 selects one or more ROIs from the object for each purpose. This is done according to the ROI selection setting 52, and the reference morphometrics are S6 8. The ROI selector 34 performs the ROI selection in two steps: An ROI selector 34 identifies the location of each respective ROI; In S72, the ROI selector 34 selects the shape and size of each ROI (whether two-dimensional or three-dimensional). For example, for the selection of the distal radius region, the ROI selector 34 determines the size of the region. First, the location of the ROI was identified as a percentage of the radius length from the distal end of the radius, and the ROI was calculated. The I selector 34 allows the thickness of the selected volume to be expressed as a different percentage of the radius length. Define.

[0053] In S74, the results including the selected ROI are output. to pass it on to the user for output (e.g., to the user interface) Alternatively, the action may involve presenting the content via a monitor on the face 14. This involves passing these results to another system (e.g., via the I / O interface 37). In a further alternative, the action may be to add the results to, for example, an optional index. This can be used as input for further analysis by a density determiner 35 or a density determiner 36, etc. (or both). Any combination of these is also contemplated.

[0054] The index determiner 35 uses a segmenter to segment the selected ROI into different structures. The system is configured to calculate the characterization indexes that characterize the For example, after selecting a bone volume ROI from a CT wrist scan, the index determiner 35 controls the segmenter 24 to further segment the selected ROI into cortical, transitional, and trabecular regions. Segment the bone and then use indices to characterize the segments (e.g., trabecular structure and porosity). The segmentation and / or index determination may be performed, for example, by calculating the "selected No. 9,064,320, entitled "Method and System for Image Analysis of Tissue Structures" This can be done using the techniques described in.

[0055] The density determiner 36 may be configured to determine the density of a second material (e.g., muscle tissue or other adjacent material, or a material having a known density). The attenuation of the calibration phantom (of the phantom) is used as the indicator to determine the density of the selected ROI. The method described in U.S. Pat. No. 9,064,320 is also used here. A second material is scanned with the same scanner and the attenuation values ​​are converted to density values. It is possible to establish a formula for converting

[0056] And the process ends.

[0057] Femoral neck ROI selection from lumbar DXA images FIG. 3 and FIG. 4A-F show the femoral neck CT scan of the system 10 from a lumbar DXA image. An example of the application to ROI selection is shown in Figure 3, where morphometry measurements are performed on the femur. 4A-4F show the workflow 54 steps generated by the system 10. The interim results are presented.

[0058] As generally shown at 76 in Figure 3: (S62 identifies the femoral head and the apex of the femoral head) After the femoral head 78 is identified, the tangent (L1) of the femoral head 78 is identified; the neck axis (L2) is identified; The femoral neck is identified as the center line through the femoral neck; the diaphyseal axis (L3) is identified as the center line through the diaphysis. The line of least transverse (L4) is identified as the shortest line that is parallel to L1 and intersects the femoral neck. The lumbar epiphysis (L5) is identified as the 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 diaphyseal axis (L3); the femoral head A central point (P2) is identified.

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

[0060] FIG. 4A shows the original image in the form of a lumbar DXA scan that is input to the system 10. FIG. 4B is a reversed version 80' of image 80 (see FIG. 4A for naked eye inspection). (Provided to aid in inspection and interpretation.) FIG. 4C shows a (preferably deep learning) pre-trained The results of applying the refined segmentation model 44 to the segmenter 24 are shown in Fig. (see S60), which is done to segment the femur from the image, and , generating a segmented mask 82 for the segmented object (in this example, the femur) This is done to.

[0061] FIG. 4D shows the results of morphometric measurements performed on the segmented femurs (see S66). FIG. 4E shows the result of the selection of the femoral neck ROI by the ROI selector 34, reference numeral 86. (See S70 and S72). The minimum transverse line (L4 in Figure 3) is the center line of the area, The width of the region is selected as a percentage (10% in this example) of the lumbar axial length (D1 in Figure 3). (This percentage is configurable and can be set in the ROI selection settings 52. The ROI selection can be varied for different purposes and studies.

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

[0063] Distal radius ROI selection from wrist CT images 5 and 6A-6N show the system 10 for extracting an ROI from a wrist CT image at the distal radius. An example of application of the method to select bone regions is shown in Fig. 3. 6A-6N show the ROIs generated by the steps of the ROI selection workflow 50. The intermediate results are shown in Figs. 6B, 6D, 6F, 6H, 6J, 6L, and 6N, respectively. 6A, 6C, 6E, 6G, 6I, 6K, and 6M, where FIG. To aid in the naked eye inspection and interpretation of 6C, 6E, 6G, 6I, 6K, and 6M (Provided for the purpose of

[0064] As shown in Figure 5, 90, the radius distal breadth (RDB) is the width of the ulnar bone. Radial length (RL) is measured as the distance from the medial most point of the scar to the lateral most point of the styloid process. The radius length is measured from the most proximal end of the radial head to the tip of the styloid process. do.

[0065] Figure 6A shows the original image in the form of a wrist CT scan. Figure 6C shows the original image (pre-trained deep learning). The segmenter 24 segments the tissue from the surrounding tissue (using a trained segmentation model 44). Figure 6E shows a bone that has been repaired from another bone (in this case a conventional repair). The segmenter 24 uses a contiguous component detection algorithm to segment the FIG. 6G shows a segmented radius with the The image coordinate system is transformed into the radial coordinate system by the image transformer 28 (S6 4). Figure 6I shows how the distal radial width (double-ended arrow-shaped line) is measured. K shows the region of interest (dashed box) selected. The selected ROI is Distance R from the tip of the distal radius DB The height of the region is R DB As half of FIG. 6M is an image of the selected distal radius.

[0066] Those skilled in the art will recognize that many modifications can be made without departing from the scope of the invention. In particular, certain features of the embodiments of the present invention may be used to provide further embodiments. Note that it is possible.

[0067] If any references to prior art are made in this specification, such references are not to be construed as limiting the scope of the invention. It should be noted that a prior art application does not constitute an admission that the prior art forms part of the public knowledge in any country. I want to.

[0068] In the appended claims and the foregoing detailed description of the invention, Therefore, unless the context otherwise requires, the terms "have" and "have" (third person singular Terms such as "has" and "has" are used in a comprehensive sense (i.e., the presence of the declared feature). Although the term "presence" is used to indicate the presence of additional features in various embodiments of the invention, This does not preclude the inclusion or addition of other elements. [Explanation of symbols]

[0069] Segmentation System 12 ROI Selection Controller 14 User Interface 16 GUI 18 Processors 20 Memory 22 Image Processor 24 Segmenta 26 Object and Anatomical Landmark Identifier 28 Normalizer 30 Morphometer 32 Reference Morphometric Determinator 34 ROI Selector 35 Index determiner 36 Density determiner 37 I / O Interface 38 Result output 40 Program Code 42 Image storage section 44 Pre-trained segmentation models 46 Object and Anatomical Landmark Identification Model 48 Morphometric Setup 50 Reference Morphometric Models 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, The steps include identifying a predefined landmark of the object, The steps include: performing morphometric measurements on the object by referring to the aforementioned landmark to determine the reference morphometrics associated with the object; A step of selecting one or more ROIs from the object according to the reference morphological metrics, comprising identifying the position of the ROIs in relation to the reference morphological metrics, The step includes outputting one or more selected ROIs, The step of determining the aforementioned reference morphometrics is: Measuring the underlying morphometrics on the object by referring to the detected landmark, and A method comprising determining the reference morphometrics based on the measured baseline morphometrics.

2. A method according to Claim 1, comprising determining the reference morphological features based on the measured underlying morphological features using one or more trained deep learning reference morphological features models.

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

4. A method according to any one of claims 1 to 3, further comprising the step of segmenting the object from the image.

5. In the method according to claim 4, 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.

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

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

8. The method according to claim 7, 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.

9. A method according to any one of claims 1 to 8, further comprising the step of normalizing the image of the object by coordinate transformation or other means before performing morphometry on the object.

10. A method according to any one of claims 1 to 9, further comprising the step of determining the density of one or more selected ROIs using the attenuation of the material to be indicated.

11. A method according to any one of claims 1 to 10, The aforementioned reference morphometrics include radius length, The aforementioned basic morphometrics include one or more of the distal radius width, radial shaft diameter, maximum distal radius diameter, and minimum distal radius diameter, and A method wherein one or more ROIs include bone volume for bone microstructure analysis.

12. A system for selecting one or more regions of interest (ROI) 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 landmarks of the said objects, A morphometer configured to perform morphometric measurements on an object by referring to the aforementioned landmark and thereby determine the reference morphometrics associated with the object, A region selector configured to select one or more ROIs from the object according to the reference morphometrics, the region selector includes identifying the location of the ROI and determining the shape and size of the ROI in relation to the reference morphometrics, The system includes a result output configured to output the selected ROI, The morphometer described above is Measuring the underlying morphometrics on the object by referring to the detected landmark, and Determining the reference morphometrics based on the measured base morphometrics, A system configured to determine reference morphometrics.

13. The system according to claim 12, wherein the morphometer is configured to determine the reference morphometrics based on the measured underlying morphometrics using one or more trained deep learning reference morphometrics models.

14. The system according to claim 12 or 13, wherein the region selector is further configured to determine the shape and size of the ROI in relation to the reference morphometrics.

15. A system according to any one of claims 12 to 14, further comprising a segmenter configured to segment the object from the image.

16. In the system according to claim 15, 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.

17. The system according to any one of claims 12 to 16, further comprising an image preprocessing device, wherein the image preprocessing device is (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.

18. A system according to any one of claims 12 to 17, further comprising a normalizer configured to normalize the image of the object by coordinate transformation or other means before morphometric measurement is performed on the object.

19. The system according to any one of claims 12 to 18, further comprising a density determiner configured to determine the density of one or more selected ROIs using the attenuation of the material to be specified.

20. In the system according to any one of claims 12 to 19, The aforementioned reference morphometrics include radius length, The aforementioned basic morphometrics include one or more of the distal radius width, radial shaft diameter, maximum distal radius diameter, and minimum distal radius diameter, and The system is configured such that the region selector selects one or more ROIs, including bone volume for bone microstructure analysis.

21. 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 11 when executed by one or more processors.

22. A computer-readable medium comprising the computer program described in Claim 21.

Images