Systems and methods for patient structure estimation during medical imaging

Abstract

This invention provides methods and systems for estimating patient structures prior to scanning with a medical imaging system. As an example, the method may include acquiring a depth image of a patient positioned on the stage of the medical imaging system, correcting the depth image based on histogram data from the depth image, and extracting the patient's three-dimensional structures based on the corrected depth image.

Description

Technical Field

[0001] The implementation schemes of the subject matter disclosed herein relate to medical imaging systems, and more specifically to accurate patient structure estimation prior to medical imaging. Background Technology

[0002] Non-invasive radiographic imaging techniques allow for the acquisition of images of the internal structures of a patient or object without the need for invasive procedures. Specifically, techniques such as computed tomography (CT) use various physical principles, such as differential transmission of X-rays through a target volume, to acquire image data and construct tomographic images (e.g., a three-dimensional representation of the interior of the human body or other imaging structures). In modern CT imaging systems, a gantry—a circular frame with an X-ray tube on one side and a detector on the other—rotates around a patient positioned on a stage, generating thousands of cross-sectional views of the patient in a single rotation. For these imaging techniques to be used effectively, the patient or object must be properly positioned and oriented within the imaging system. Summary of the Invention

[0003] In one embodiment, a method for a medical imaging system includes: acquiring a depth image of a patient positioned on the stage of the medical imaging system; correcting the depth image based on histogram data from the depth image; and extracting the patient's three-dimensional structure based on the corrected depth image. This eliminates the loss of depth information caused by uneven exposure and holes in the extracted three-dimensional patient structure / avatar / mesh / point cloud. Thus, accurate estimation of the patient's structure and orientation can be achieved prior to imaging.

[0004] It should be understood that the above brief description is provided to introduce selected concepts further described in the detailed embodiments in a simplified form. This is not intended to identify key or essential features of the claimed subject matter, the scope of which is uniquely defined by the claims following the detailed embodiments. Furthermore, the claimed subject matter is not limited to embodiments that address any shortcomings mentioned above or in any part of this disclosure. Attached Figure Description

[0005] The invention will be better understood by referring to the following description of non-limiting embodiments, in which:

[0006] Figure 1 A drawing view of an imaging system according to one embodiment is shown.

[0007] Figure 2 A block diagram of an exemplary imaging system according to one embodiment is shown.

[0008] Figure 3A block diagram of an exemplary algorithm for estimating patient structures prior to imaging, according to one embodiment, is shown.

[0009] Figures 4A to 4D A flowchart is shown illustrating a method for identifying and counteracting the effects of variable lighting and reflections on patient structure estimation.

[0010] Figure 5 Various conditions are shown, among which Figures 4A to 4D The method can generate filtered point clouds of patient structures without artifacts caused by variable lighting and reflections.

[0011] Figure 6 As shown in Figures 4A to 4D An example of point cloud segmentation described in the method.

[0012] Figure 7 It shows the use of Figures 4A to 4D An example of a method for estimating three-dimensional patient structures under dim lighting conditions.

[0013] Figures 8A to 8C It shows that it can be executed Figures 4A to 4D The method generates a series of images.

[0014] Figure 9 A flowchart is shown illustrating the method used to determine whether a patient is positioned to achieve the desired examination result. Detailed Implementation

[0015] The following description relates to various implementations of medical imaging systems. Specifically, systems and methods are provided for determining accurate three-dimensional (3D) depth estimates of patient structures prior to radiographic imaging, free from illumination, reflection, and exposure artifacts. Patient structure, position, and orientation all affect radiographic imaging results. Inappropriate patient position and / or orientation during or prior to scanning can significantly affect both image noise and patient surface dose. As an example, placing the patient in an off-center location can lead to imaging artifacts and unwanted radiation exposure to more sensitive areas of the body.

[0016] The required patient position and orientation (e.g., posture) for a radiological examination are based on the body part to be imaged, the suspected defect or disease, and the patient's condition, with the positioning protocol determined by the radiologist. The prescribed protocol is then executed by the technician operating the imaging system to obtain accurate diagnostic information and minimize the patient's X-ray exposure. Alternatively, the technician may manually adjust the height and lateral position of the scanning stage on which the patient is positioned for alignment during the radiological examination. However, the technician may introduce technical errors due to the high workload and inefficiency of manual positioning, for example. These technical errors can result in images acquired during the radiological examination that show overexposure, underexposure, or mispositioning of the patient. Therefore, the radiologist may decide to refuse and repeat the scan for an accurate diagnosis. In such examples, a second radiograph might be requested immediately when the patient is available. Alternatively, the patient may have to return for an additional appointment for a rescan. Both options increase patient discomfort, patient radiation exposure, cognitive stress on the scanning operator, and the amount of time until a diagnosis is made.

[0017] Therefore, various technologies have been employed to expedite the radiology workflow. These include integrating time-of-flight (ToF) or depth cameras into the radiology lab. Prior to radiographic imaging, ToF or depth cameras can be used to generate 3D depth images of the patient. These 3D depth images can be used to determine patient structures, including anatomical key points, body contours, body volume / thickness, and the patient's position / orientation relative to the worktable. The patient structures can then be compared to positioning parameters proposed by the radiologist. If the determined patient structures do not align with the positioning parameters, the technician can reposition the patient before the scan, thereby reducing the incidence of rejection and repeat scans.

[0018] However, depth cameras can still be affected by several noise sources, such as variable lighting and / or reflective areas present in the inspection room. Such noise can cause areas in the 3D depth image to lack depth information or result in depth holes. These depth holes can lead to inaccurate patient acuity determination and ultimately, technicians may inaccurately determine whether the positioning parameters prescribed by the radiologist are being followed correctly. Therefore, despite using patient acuity estimation prior to the scan, rejections and repeat scans can still occur.

[0019] Therefore, according to the embodiments disclosed herein, a method and system are provided for identifying and counteracting the effects of variable illumination and reflections on depth camera-based patient structure estimation. In one embodiment, a depth image of a patient positioned on the stage of a medical imaging system can be captured, and bounding exposure depth imaging (BEDI) and / or illumination variation coefficient (CoIV)-based corrections can be applied to the captured depth image, enabling accurate 3D patient structure estimation.

[0020] exist Figure 1 and Figure 2 Examples of computed tomography (CT) imaging systems that can be used to acquire images according to the techniques of the present invention are provided. Figures 1 to 2 The CT imaging system includes a stage that can be positioned within the gantry of the CT imaging system. The gantry includes an X-ray projector and detector for imaging a subject positioned on the stage. The stage's position can be adjusted to place the subject within the gantry in the desired imaging location. Furthermore, the subject can be positioned on the stage in various postures and orientations (such as the exemplary posture) to obtain the desired radiographic images. Figure 3 The document provides a high-level overview of exemplary algorithms that can be used for patient body shape estimation prior to imaging. For example, Figure 3 The algorithm may include methods for identifying and counteracting the effects of variable lighting and reflection on patient structure estimation, such as Figures 4A to 4D The method shown is used to accurately generate a 3D model of the patient's structure for use in scan result prediction and patient pose interpretation, as can be used Figure 9 The exemplary method is executed. Figure 5 Examples of variable lighting and exposure conditions are provided, in which Figures 4A to 4D The method proposed can generate accurate 3D point clouds of patient structures. Figure 6 An example of intelligent segmentation of objects surrounding a patient is shown, which can... Figures 4A to 4D The method shown occurs during the process to achieve accurate 3D patient structure estimation. Figure 7 Provided Figures 4A to 4D The method shown can generate an example of accurate 3D patient structure estimation under dim lighting conditions. Figures 8A to 8C It is shown as a series of consecutive images Figures 4A to 4D The method presented in the text.

[0021] While computed tomography (CT) systems have been described by way of example, it should be understood that this technique can also be useful when applied to other medical imaging systems and / or medical imaging devices that utilize boreholes and worktables, such as x-ray imaging systems, magnetic resonance imaging (MRI) systems, positron emission tomography (PET) imaging systems, single-photon emission computed tomography (SPECT) imaging systems, and combinations thereof (e.g., multimodal imaging systems such as PET / CT, PET / MR, or SPECT / CT imaging systems). The present invention discussion of CT imaging modalities provides only one example as a suitable imaging modality.

[0022] Figure 1An exemplary CT imaging system 100 is illustrated. Specifically, the CT imaging system 100 is configured to image a subject 112 (such as a patient, inanimate object, one or more manufactured parts, industrial parts) and / or foreign objects (such as implants, stents, and / or contrast agents present in the body). Throughout this disclosure, the terms "subject" and "patient" are used interchangeably, and it should be understood that, at least in some examples, a patient is a type of subject that can be imaged by the CT imaging system, and a subject may include a patient. In one embodiment, the CT imaging system 100 includes a gantry 102, which may further include at least one x-ray radiation source 104 configured to project an x-ray radiation beam (or x-ray) 106 (see [link to documentation]). Figure 2 This is used for imaging patients. Specifically, the X-ray radiation source 104 is configured to project X-rays 106 toward a detector array 108 positioned on the opposite side of the gantry 102. Although Figure 1 Only a single X-ray radiation source 104 is depicted, but in some embodiments, multiple radiation sources may be used to project multiple X-rays 106 toward multiple detectors to obtain projection data corresponding to the patient at different energy levels.

[0023] In some embodiments, an X-ray radiation source 104 projects an X-ray fan-shaped or conical beam 106, which is collimated to lie in the xy plane of a Cartesian coordinate system and is generally referred to as the “imaging plane” or “scanning plane.” The X-ray beam 106 passes through a subject 112. After being attenuated by the subject 112, the X-ray beam 106 is incident on a detector array 108. The intensity of the attenuated radiation beam received at the detector array 108 depends on the attenuation of the X-ray 106 by the subject 112. Each detector element of the detector array 108 generates a separate electrical signal, which is a measure of the beam intensity at the detector location. Intensity measurements from all detectors are acquired individually to produce a transmission distribution.

[0024] In a third-generation CT imaging system, a gantry 102 is used to rotate an X-ray radiation source 104 and a detector array 108 around a subject 112 within the imaging plane, causing the angle at which the X-ray beam 106 intersects the subject 112 to continuously change. A complete gantry rotation occurs when the gantry 102 completes one full 360-degree rotation. A set of X-ray attenuation measurements (e.g., projection data) from the detector array 108 at a given gantry angle is called a “view.” Therefore, a view is each incremental position of the gantry 102. A “scan” of the subject 112 comprises a set of views acquired at different gantry angles or viewing angles during one rotation of the X-ray radiation source 104 and the detector array 108.

[0025] In axial scanning, the projection data is processed to construct an image corresponding to a two-dimensional slice taken through the subject 112. One method for reconstructing an image from a set of projection data is known in the art as filtered backprojection. This method converts attenuation measurements from the scan into an integer called a “CT number” or “Henry’s unit” (HU), which is used to control the brightness of the corresponding pixel on, for example, a cathode ray tube display.

[0026] In some examples, the CT imaging system 100 may include a depth camera 114 positioned on or outside the gantry 102. As shown, the depth camera 114 is mounted on a ceiling 116 positioned above a subject 112 and oriented to image the subject when the subject 112 is at least partially outside the gantry 102. The depth camera 114 may include one or more light sensors, including one or more visible light sensors and / or one or more infrared (IR) light sensors. In some embodiments, the one or more IR sensors may include one or more sensors within both the near-IR and far-IR ranges to enable thermal imaging. In some embodiments, the depth camera 114 may also include an IR light source. The light sensor may be any 3D depth sensor, such as a time-of-flight (ToF) sensor, a stereo sensor, or a structured light depth sensor operable to generate a 3D depth image, while in other embodiments, the light sensor may be a two-dimensional (2D) sensor operable to generate a 2D image. In some such embodiments, the 2D light sensor may be used to infer depth based on an understanding of light reflection phenomena to estimate 3D depth. Regardless of whether the light sensor is a 3D depth sensor or a 2D sensor, the depth camera 114 can be configured to output a signal encoding an image to a suitable interface that can be configured to receive the signal encoding the image from the depth camera 114. In other examples, the depth camera 114 may also include other components, such as a microphone, to enable the reception and analysis of directional and / or non-directional sound from the observed subject and / or other sources.

[0027] In some embodiments, the CT imaging system 100 also includes an image processing unit 110 configured to reconstruct an image of the patient target volume using a suitable reconstruction method, such as an iterative or analytical image reconstruction method. For example, the image processing unit 110 may use an analytical image reconstruction method such as filtered backprojection (FBP) to reconstruct an image of the patient target volume. As another example, the image processing unit 110 may use iterative image reconstruction methods such as adaptive statistical iterative reconstruction (ASIR), conjugate gradient (CG), maximum likelihood expectation maximization (MLEM), model-based iterative reconstruction (MBIR), etc., to reconstruct an image of the patient target volume.

[0028] As used herein, the phrase "reconstructed image" is not intended to exclude embodiments of the invention in which data representing an image is generated rather than a visual image. Therefore, as used herein, the term "image" broadly refers to both a visual image and the data representing a visual image. However, many embodiments generate (or are configured to generate) at least one visual image.

[0029] The CT imaging system 100 also includes a stage 115 on which the patient 112 is positioned for imaging. The stage 115 may be electrically powered, allowing adjustment of its vertical and / or lateral positions. Therefore, the stage 115 may include a motor and a motor controller, as will be described below with respect to… Figure 2 The table motor controller moves the table 115 by adjusting the motor to properly position the subject within the frame 102 to obtain projection data corresponding to the target volume of the subject. The table motor controller can adjust the height of the table 115 (e.g., its vertical position relative to the ground on which the table is located) and the lateral position of the table 115 (e.g., its horizontal position along an axis parallel to the axis of rotation of the frame 102).

[0030] Figure 2 It shows something similar to Figure 1 An exemplary imaging system 200 of a CT imaging system 100. In one embodiment, imaging system 200 includes detector array 108 (see...). Figure 1 The detector array 108 also includes a plurality of detector elements 202 that together collect the X-ray beam 106 passing through the subject 112 (see [link]). Figure 1 To acquire the corresponding projection data, in one embodiment, the detector array 108 is fabricated in a multi-slice configuration comprising multiple rows of cells or detector elements 202. In such a configuration, one or more additional rows of detector elements 202 are arranged in parallel for acquiring projection data. In some examples, individual detectors or detector elements 202 of the detector array 108 may include photon counting detectors that register interactions of individual photons into one or more energy bins. It should be understood that the methods described herein can also be implemented using energy integration detectors.

[0031] In some embodiments, the imaging system 200 is configured to traverse different angular positions around the subject 112 to acquire the desired projection data. Therefore, the gantry 102 and the components mounted thereon can be configured to rotate about a center of rotation 206 to acquire projection data, for example, at different energy levels. Alternatively, in embodiments where the projection angle relative to the subject 112 varies over time, the mounted components can be configured to move along a generally arcuate path rather than along a circumference.

[0032] In one embodiment, the imaging system 200 includes a control mechanism 208 to control the movement of components, such as the rotation of the gantry 102 and the operation of the x-ray radiation source 104. In some embodiments, the control mechanism 208 further includes an x-ray controller 210 configured to provide power and timing signals to the x-ray radiation source 104. Additionally, the control mechanism 208 includes a gantry motor controller 212 configured to control the rotational speed and / or position of the gantry 102 based on imaging requirements.

[0033] In some embodiments, control unit 208 further includes a data acquisition system (DAS) 214 configured to sample analog data received from detector element 202 and convert the analog data into digital signals for subsequent processing. The data sampled and digitized by DAS 214 is transferred to a computer or computing device 216. In one example, computing device 216 stores the data in storage device 218. For example, storage device 218 may include hard disk drives, floppy disk drives, optical disc read / write (CD-R / W) drives, digital versatile optical disc (DVD) drives, flash memory drives, and / or solid-state storage drives.

[0034] Additionally, computing device 216 provides commands and parameters to one or more of the DAS 214, x-ray controller 210, and rack motor controller 212 to control system operations, such as data acquisition and / or processing. In some embodiments, computing device 216 controls system operations based on operator input. Computing device 216 receives operator input, such as commands and / or scan parameters, via an operator console 220 operably coupled to computing device 216. Operator console 220 may include a keyboard (not shown) or a touchscreen to allow the operator to specify commands and / or scan parameters.

[0035] Although Figure 2 Only one operator console 220 is shown, but more than one operator console may be coupled to the imaging system 200, for example, to input or output system parameters, request inspections, and / or view images. Furthermore, in some embodiments, the imaging system 200 may be coupled via one or more configurable wired and / or wireless networks (such as the Internet and / or VPNs) to multiple monitors, printers, workstations, and / or similar devices, for example, located locally or remotely within an institution or hospital, or in completely different locations.

[0036] In one implementation, for example, the imaging system 200 includes or is coupled to a Picture Archiving and Communication System (PACS) 224. In an exemplary implementation, the PACS 224 is further coupled to a remote system (such as a radiology information system, a hospital information system) and / or coupled to an internal or external network (not shown) to allow operators in different locations to provide commands and parameters and / or obtain access to image data.

[0037] The computing device 216 uses operator-provided and / or system-defined commands and parameters to operate the table motor controller 226, which in turn controls the table motor 228, which is adjustable. Figure 1 The position of the worktable 115 is shown. Specifically, the worktable motor controller 226 moves the worktable 115 via the worktable motor 228 to properly position the subject 112 in the frame 102 to obtain projection data corresponding to the target volume of the subject 112. For example, the computing device 216 can send commands to the worktable motor controller 226 to instruct the worktable motor controller 226 to adjust the vertical and / or lateral position of the worktable 115 via the motor 228.

[0038] As previously described, the DAS 214 samples and digitizes the projection data acquired by detector element 202. Subsequently, the image reconstructor 230 uses the sampled and digitized X-ray data to perform high-speed reconstruction. Although Figure 2 Image reconstructor 230 is shown as a separate entity; however, in some embodiments, image reconstructor 230 may be part of computing device 216. Alternatively, image reconstructor 230 may not be present in imaging system 200, and alternatively, computing device 216 may perform one or more functions of image reconstructor 230. Furthermore, image reconstructor 230 may be located locally or remotely and may be operatively connected to imaging system 200 using wired or wireless networks. Specifically, one exemplary embodiment may use computing resources in a "cloud" network cluster for image reconstructor 230.

[0039] In one embodiment, image reconstructor 230 stores the reconstructed image in storage device 218. Alternatively, image reconstructor 230 transmits the reconstructed image to computing device 216 to generate usable patient information for diagnosis and evaluation. In some embodiments, computing device 216 transmits the reconstructed image and / or patient information to display 232, which is communicatively coupled to computing device 216 and / or image reconstructor 230. In one embodiment, display 232 allows an operator to evaluate the anatomical structures of the image. Display 232 may also allow an operator, for example via a graphical user interface (GUI), to select the volume of interest (VOI) and / or request patient information for subsequent scanning or processing.

[0040] As further described herein, computing device 216 may include computer-readable instructions executable to send commands and / or control parameters to one or more of the DAS 214, X-ray controller 210, gantry motor controller 212, and stage motor controller 226 according to an examination imaging protocol that includes a clinical task / intention, also referred to herein as a Clinical Intent Identifier (CID) for the examination. For example, the CID may inform the objective of the procedure based on clinical indications (e.g., general scan or lesion detection, anatomical structure of interest, quality parameters, or other objectives) and may further define the required patient position and orientation (e.g., posture) during the scan (e.g., supine and foot-first). The operator of system 200 can then position the patient on the stage according to the patient position and orientation specified by the imaging protocol. Furthermore, computing device 216 can set and / or adjust various scan parameters (e.g., dose, gantry rotation angle, kV, mA, attenuation filter) according to the imaging protocol. For example, the imaging scheme can be selected by the operator from a number of imaging schemes stored in the memory of the computing device 216 and / or a remote computing device, or the imaging scheme can be automatically selected by the computing device 216 based on the received patient information.

[0041] During the examination / scanning phase, it may be desirable to expose the subject to the lowest possible radiation dose while still maintaining the required image quality. Additionally, reproducible and consistent image quality may be required between examinations, between subjects, and between different imaging system operators. Therefore, the imaging system operator may perform manual adjustments to the stage position and / or subject position to, for example, center the desired patient anatomy within the gantry aperture. However, such manual adjustments are prone to error. Therefore, the CID associated with the selected imaging protocol can be mapped to various subject positioning parameters, including subject posture and orientation, stage height, anatomical reference used for scanning, and start and / or end scan positions.

[0042] Therefore, the depth camera 114 can be operatively and / or communicatively coupled to the computing device 216 to provide image data for determining the subject's structure, including posture and orientation. Furthermore, various methods and procedures described herein for determining patient structure based on image data generated by the depth camera 114 can be stored as executable instructions in the non-transitory memory of the computing device 216.

[0043] Additionally, in some examples, computing device 216 may include camera image data processor 215, which includes instructions for processing information received from depth camera 114. The information received from depth camera 114 (which may include depth information and / or visible light information) can be processed to determine various subject parameters, such as subject identity, subject physique (e.g., height, weight, patient thickness), and current subject position relative to the worktable and depth camera 114. For example, prior to imaging, an image reconstructed from point cloud data generated by camera image data processor 215 based on images received from depth camera 114 can be used to estimate the body contour or structure of subject 112. Computing device 216 can use these subject parameters to perform, for example, patient-scanner contact prediction, scan range overlay, and scan keypoint calibration, as will be described in more detail herein. Furthermore, data from depth camera 114 may be displayed via display 232.

[0044] In one implementation, information from depth camera 114 can be used by camera image data processor 215 to perform tracking of one or more subjects within the field of view of depth camera 114. In some examples, image information (e.g., depth information) can be used to perform skeletal tracking, where multiple joints of the subject are identified and analyzed to determine the subject's movement, posture, position, etc. The position of the joints during skeletal tracking can be used to determine the aforementioned subject parameters. In other examples, image information can be used directly to determine the aforementioned subject parameters without skeletal tracking.

[0045] Based on these patient positioning parameters, the computing device 216 can output one or more alerts to the operator related to patient posture / orientation and predicted examination (e.g., scan) results, thereby reducing the likelihood that the patient will be exposed to a higher-than-expected radiation dose and improving the quality and reproducibility of the images generated by the scan. As an example, estimated body structure can be used to determine whether the patient is in the imaging position prescribed by the radiologist, thereby reducing the incidence of repeated scans due to improper positioning. Furthermore, the amount of time spent by the imaging system operator positioning the patient can be reduced, allowing for more scans to be performed throughout the day and / or allowing for additional patient interaction.

[0046] Based on depth cameras (such as...) Figure 1 and Figure 2 The depth camera 114 described herein receives data to determine multiple exemplary patient orientations. For example, a controller (e.g., Figure 2 The computing device 216 can extract patient structure and pose estimates based on images received from a depth camera, thereby enabling different patient orientations to be distinguished from each other.

[0047] A first exemplary patient orientation may include a pediatric patient, and a second exemplary patient orientation may include an adult patient. Both the first and second exemplary patient orientations may include a patient lying supine in a supine position, but with different arm positions. For example, the first exemplary patient orientation may include two arms crossed and positioned above the head of a pediatric patient, while the second exemplary patient orientation may include two arms crossed and positioned above the abdomen of an adult patient. The supine position, patient physique, and arm positions can all be distinguished based on data received from a depth camera and by methods and algorithms (such as relative to...) described further herein. Figure 3 and Figures 4A to 4D (Description) Analysis.

[0048] As other examples, a third exemplary patient orientation may include a patient covered with a blanket, and a fourth exemplary patient orientation may include a patient wearing a medical gown. Additionally, a fifth exemplary patient orientation may include operator occlusion. As will be explained herein, the inclusion of a blanket in the third exemplary patient orientation, a medical gown in the fourth exemplary patient orientation, and operator occlusion in the fifth exemplary patient orientation does not affect the patient structure and pose estimation determined based on data received from a depth camera.

[0049] As a further example, the patient to be imaged can be placed in a series of positions. For example, a sixth exemplary patient orientation may include a patient in a supine position, a seventh exemplary patient orientation may include a patient in a prone position where he / she lies face down, and an eighth exemplary patient orientation may include a patient in a lateral decubitus position where the patient lies on his / her side. Supine, prone, and lateral decubitus positions (including sideways) can all be distinguished from each other based on data received from the depth camera. Therefore, there are multiple poses, orientations, patient body shape / physique, and potential occlusions (e.g., blankets, medical gowns, operators) that can be used to determine a 3D patient structure estimate prior to imaging.

[0050] Figure 3 An exemplary algorithm 400 is shown, which can be controlled by a controller (such as...) Figure 2 The computing device 216) is used to implement this so that, prior to radiographic imaging, data is collected from a depth camera (e.g., Figure 1 and Figure 2The depth camera 114 receives data to estimate patient structures (including pose and position). In the illustrated embodiment, algorithm 400 is utilized prior to CT imaging; however, it will be understood that the embodiments described herein can be implemented using other types of medical imaging modalities (e.g., MRI, X-ray, PET, interventional angiography systems). Furthermore, in some embodiments, the patient can be continuously monitored via algorithm 400 and data received from the depth camera before and during medical imaging.

[0051] In the illustrated implementation, at 402, algorithm 400 includes positioning the patient on the worktable of the CT imaging system (such as...). Figure 1 On workbench 115). The technician can locate the patient according to the prescribed protocol proposed by the radiologist (such as a Clinical Intent Identifier (CID) based on the examination selected by the radiologist). As an example, algorithm 400 can be displayed (e.g., on the workbench 115). Figure 2 The system can be displayed on monitor 232 and / or otherwise transmit a prescribed protocol based on the selected CID to guide the technician during patient positioning. For example, the patient can be positioned supine, prone, or lateral, head-first or feet-first relative to the gantry of the CT imaging system. Patient positioning can also be refined as outlined in the prescribed protocol, such as by adjusting limb positions to achieve the desired posture. Various straps and / or pillows can be used to help the patient maintain the correct position and remain still.

[0052] Furthermore, proper patient positioning within the CT imaging system means that the patient's midline (an imaginary line drawn between the patient's eye and pubic symphysis) is centered on the stage, and the stage height is adjusted so that the centroid of the area to be scanned coincides with the gantry's center of rotation. Therefore, stage parameters can be adjusted. Adjusting stage parameters may include adjusting the stage height relative to the gantry to prevent any patient collisions with the gantry. Additionally, stage parameters can be adjusted to ensure that the patient is centered within the gantry once the scan begins.

[0053] Once the patient is properly positioned on the scanning stage at point 402, the spatial parameters of the stage are determined at point 404. For example, the positions of the four corners of the stage in the xyz plane of the Cartesian world coordinate system can be determined. As an exemplary example, the first corner can be located at [x1, y1, z1], the second corner at [x2, y2, z2], the third corner at [x3, y3, z3], and the fourth corner at [x4, y4, z4]. In one implementation, these corners can be defined as the upper left corner of the stage (e.g., the left corner of the stage closest to the gantry), the upper right corner of the stage (e.g., the right corner of the stage closest to the gantry), the lower left corner of the stage (e.g., the left corner of the stage furthest from the gantry), and the lower right corner of the stage (e.g., the right corner of the stage furthest from the gantry). The x, y, and z values ​​for each corner can then be determined. For example, the lower left corner of the workbench can be located at [-400, -78, -2469], the upper left corner at [-400, -78, -469], the lower right corner at [400, -78, -2469], and the upper right corner at [400, -78, -469]. The determined workbench spatial parameters can be input into the point cloud algorithm at position 414, as will be further described below.

[0054] At position 406, the depth camera is activated, which may be vertically positioned above the scanning stage. Activating the depth camera may include powering it on from a "off" state or a low-power mode. Activating the camera may also include selecting initial imaging settings, such as exposure, depth of focus, frame rate, etc. Exposure settings, such as aperture, shutter speed, and ISO, may be selected based on scanning chamber conditions (e.g., lighting conditions, presence of reflective surfaces, etc.). In some examples, the exposure settings may be repeatedly adjusted as images are acquired.

[0055] At 408, the depth camera acquires a depth frame. As an example, the depth camera can use a modulated light source to illuminate the patient (collectively referred to as the scene) on a scanning table and use a ToF sensor located within the depth camera to observe the reflected light. The duration elapsed between illumination and reflection is measured and converted into distance. The light source can include a solid-state laser or an array of light-emitting diodes (LEDs) operating in, for example, the near-infrared range (approximately 850 nm) invisible to the human eye. An imaging sensor designed to respond to light of the same wavelength emitted by the light source receives the light and converts the photon energy into an electric current. The light entering the sensor can have an ambient component and a reflection component, with distance (depth) information embedded only in the reflection component. In a ToF sensor, the distance is measured for each pixel in a 2D addressable array, resulting in a depth map or depth frame (e.g., a set of 3D points, where each point is also referred to as a voxel). As another example, the depth camera can acquire stereo images (e.g., via two or more spaced-apart image sensors), thus obtaining a 3D depth frame each time an image is acquired. At 414, the depth frame can be fed into a point cloud algorithm, as described below.

[0056] At position 410, camera calibration for depth camera 114 is performed. Camera calibration here refers to extracting the camera's intrinsic and extrinsic parameters using a 2D pattern (such as a checkerboard) so that all data points lie in a single plane and the z-components of these data points are zero in world space coordinates. Extrinsic parameters refer to the camera's position and orientation in the world, while intrinsic parameters refer to the relationship between pixel coordinates and camera coordinates. Once the intrinsic and extrinsic parameters are extracted, these matrices are multiplied by the original point cloud matrix to obtain the camera-to-gantry coordinate transformation. This matrix multiplication can be as follows:

[0057]

[0058] The first matrix represents the extrinsic camera parameters, the second matrix represents the scene coordinates, and the third matrix represents the intrinsic camera parameters. The third or intrinsic matrix may contain two intrinsic parameters, which cover the focal length (e.g., F0). c and C c ).

[0059] Once the camera is successfully calibrated, algorithm 400 can proceed to 412, where the camera is configured based on the calibration at 410. For example, the scale factor, rotation factor, and translation factor in the x and y directions can be determined based on the camera calibration so that the image acquired by the camera is appropriately scaled to world space. For example, the camera configuration can compensate for tilt or rotation in a depth camera mount. The camera calibration (including one or more of the scale factor, rotation factor, and translation factor mentioned above) can also be input into the point cloud algorithm at 414.

[0060] At 414, the point cloud algorithm can use data from inputs at 404, 408, and 412 to generate a filtered array of point clouds in gantry coordinates, which is output at 416. As an example, the point cloud algorithm can render the depth frame acquired at 408 into 3D space as a set of points or a point cloud. As another example, alternatively, the point cloud algorithm at 414 can filter out voxel noise and outliers from the depth frame data generated at 408, and then render the filtered data into a point cloud aligned with the gantry parameters determined at 404 based on the camera configuration determined at 412. The filtered point cloud array in gantry coordinates generated by the point cloud algorithm can be output at 416 and can then be used for 3D patient structure estimation to determine appropriate patient localization prior to radiographic imaging.

[0061] While using 3D patient structure estimation prior to radiographic imaging has reduced the number of rejected and duplicate scans compared to manual patient localization alone, accurately generating 3D models of patient structures using 3D depth cameras is not entirely robust. In practice, due to hardware limitations and scene structure (such as depth shading and / or the effects of materials on reflection, refraction, or IR absorption), the 3D information obtained from depth frames may be insufficient to provide accurate patient structure estimates. For example, differential illumination and surface reflectivity within the radiographic scanning chamber can cause holes or a lack of depth information in the acquired depth frames, as will be discussed below relative to... Figure 5 The use of depth frame data containing holes in the generation of 3D patient structure estimates can lead to inaccurate or incomplete representations of patient position and orientation. Technicians may misinterpret these inaccuracies or incomplete representations as falling within the parameters proposed by the radiologist. Therefore, even when using a 3D depth camera to aid in proper patient positioning, the patient may still be misaligned compared to a prescribed approach based on the selected CID, due to insufficient depth information used in the generation of the 3D patient structure estimate as interpreted by the technician.

[0062] therefore, Figures 4A to 4D A flowchart illustrating method 500 for identifying and counteracting the effects of variable illumination and reflection on patient structure estimation is provided. Method 500 and the remaining methods included herein can be generated by computing devices stored in the imaging system (such as...). Figure 2 The imaging system 200 executes computer-readable instructions in a non-transitory memory of a computing device 216, which is communicatively coupled to a depth camera, such as... Figure 1 and Figure 2 Depth camera 114.

[0063] Method 500 may begin at 502. At 502, method 500 may include positioning the patient for scanning, as previously relative to... Figure 3For example, a technician can position the patient on the stage of a CT imaging system or another radiological imaging system based on the CID selected by the radiologist for the examination. The patient can be positioned supine, prone, or lateral, head-first or feet-first relative to the gantry of the imaging system, with limb positions adjusted to the desired posture. Furthermore, the stage height (e.g., vertical position) and position can be adjusted to align the patient's center of mass with the center of the gantry for dose optimization during the scan. As an example, the technician can manually adjust the stage height and position by inputting commands to a stage motor controller, which in turn actuates the stage motors accordingly. As another example, the computing device can adjust the stage height and position by sending commands to the stage motor controller to adjust the vertical and / or lateral position of the stage by the motors based on pre-programmed instructions corresponding to the selected CID. In other examples, both the technician and the computing device can adjust the stage height and position. For example, a technician can make coarser adjustments to the height and position of the worktable, and then the calculation equipment can refine the height and position of the worktable so that the patient's center of mass is aligned with the center of the gantry.

[0064] At position 504, a depth camera can be used to capture 3D depth, IR, thermal, and / or red, green, and blue (RGB) images of the patient's location. For example, depending on the configuration of the depth camera and one or more types of sensors included, only one of the 3D depth, IR, thermal, and RGB images may be captured, or more than one of these images may be captured. By capturing 3D depth images, IR images, thermal images, and / or RGB (e.g., color) images, distribution information related to the patient and workbench positioning / orientation within the scene can be determined. This distribution information within the captured images can then be graphically summarized and displayed via a histogram. As an example, histogram data from the depth image can be used to identify holes, underexposed and overexposed areas within the depth image, and reflections within the depth image. Alternatively or additionally, histogram data from the color image can be used to identify areas of poor lighting (e.g., dark / bright areas or spots within the image). Histogram data from both the color and depth images can be used individually or in combination to determine optimal camera exposure. An example of capturing a depth image of a scene is shown below. Figure 5 An example of capturing a thermal image of a scene is shown below. Figure 6 As shown in the text, and also described below.

[0065] At point 506, the illumination and lighting distribution in the image captured at point 504 can be monitored via histogram data to determine the region of interest (ROI). The ROI includes at least the patient's body and the worktable on which the patient is positioned. Histogram data can provide a graphical representation of the tonal range of the image captured at point 504 based solely on the intensity of the brightness or luminance of each pixel within each image (e.g., hue is not considered). The tonal range within the histogram can be represented from left to right, where the left side is black hue / chroma, progressively passing through the middle midtones to the highlights on the right. The magnitude or volume of each hue within the tonal range of the image can be represented by the height of a separate line corresponding to each hue or set of hues present within the captured image. For example, lower regions of the histogram (e.g., valleys) indicate low volumes of these hues within the image, while higher regions of the histogram (e.g., peaks) indicate high volumes of these hues. Therefore, the balance and height of peaks in the histogram are indicators of tonal range and tonal balance. Therefore, histogram data can be monitored based on tone distribution to determine the illumination distribution (e.g., overexposure or underexposure) in the captured image at 504.

[0066] At point 508, the histogram data generated and monitored at point 506 can be analyzed to identify poorly lit areas, highly reflective areas, and / or poorly or underexposed camera areas. For example, if the main body of the histogram is skewed to the right, it may indicate that the image captured at point 504 is overexposed. Alternatively, if the main body of the histogram is skewed to the left, it may indicate that the image captured at point 504 is underexposed, underexposed, or poorly lit. Highly reflective areas within the captured image can be represented by one or more extreme peaks on the right edge of the histogram. Thus, analyzing the histogram data can identify poorly lit areas, highly reflective areas, and / or poorly or underexposed camera areas.

[0067] At point 510, preliminary calculations of the patient's physique are performed based on the captured images. The patient's physique may include an indication of patient thickness and / or patient shape. In one embodiment, a patient thickness estimate may be determined using depth camera images by applying an algorithm that extracts only the volume of the patient lying on the scanning table and multiplies it by a fixed value of density, where color and depth gradients are used for patient segmentation (e.g., distinguishing the patient from the scanning table to estimate patient thickness and physique).

[0068] At 512, it is determined whether the patient's body size is larger than a threshold. This threshold may be a predetermined size threshold used to distinguish between two different techniques that perform minimization and correction of illumination and reflection variations in the captured image: bounding exposure depth imaging (BEDI) and correction based on the illumination variation coefficient (CoIV). If the patient's body size is not larger than the threshold (e.g., the patient's body size is smaller than the threshold), method 500 proceeds to 513 (see [link to method]). Figure 4BThis allows for the determination of the histogram standard deviation (SD) and CoIV for both depth and color. Therefore, CoIV-based corrections can be used when the patient is small. CoIV can be a scale of illumination uniformity within a scene, measured from histogram data generated from images of the scene. The SD of the histogram data can be determined based on the data mean. CoIV can be defined as the ratio of the SD of all measured illumination values ​​to the mean, or a measure of relative illumination variability. A higher CoIV indicates a greater level of dispersion around the mean. Alternatively, the histogram SD measures illumination variability proportional to the mean or average of the histogram dataset. In one implementation, CoIV can be determined using algorithms that consider the direction of light in the illumination scene and the discrete posture of the patient, such as:

[0069]

[0070] Where α is the surface albedo; I is the direction. The illumination intensity is s; s is the normal to the patient's surface; the 3D patient position represents the patient's x-axis, y-axis, and z-axis; and the posture can be represented in eight different categories, including posterior-anterior (PA), anterior-posterior (AP), left side, right side, prone, supine, head-first, and foot-first.

[0071] As an example, histogram data from a depth image and histogram data from a color image can be combined before calculating CoIV and SD. As another example, separate CoIV and SD values ​​can be calculated based on depth and color histogram data. For instance, a first CoIV and a first SD can be determined based on histogram data from the depth image, and a second CoIV and a second SD can be determined based on histogram data from the color image. In some examples, the first CoIV can be combined with the second CoIV, and the first SD can be combined with the second SD (e.g., by averaging).

[0072] At 513, method 500 determines whether the calculated CoIV is greater than the SD of the histogram data for both the color image and the depth image. If the CoIV is less than the SD of the histogram data for both the color image and the depth image, it indicates that the data points within the histogram are distributed over a larger range of values, which may indicate the presence of areas with high reflectivity in the scene that can cause holes within the 3D depth frame. In contrast, if the CoIV is greater than the SD of the histogram data for both the color image and the depth image, the illumination dispersion occurs over a larger range, thus indicating that the image may be poorly illuminated.

[0073] If, for both the depth image and the color image, CoIV is not greater than the SD of the histogram data, then method 500 continues to 517 and includes disabling automatic exposure. For example, the automatic exposure settings of the depth camera can be disabled to reduce or eliminate image artifacts caused by highly reflective areas within the scene. Once automatic exposure is disabled, a new 3D depth image, IR image, thermal image, or RGB image of the patient's location can be captured using the depth camera. Method 500 can then proceed to 531 (see...). Figure 4C ), as will be described below.

[0074] If CoIV is greater than the histogram's SD, method 500 can continue to 519, and the depth camera's automatic exposure can be enabled, and the exposure settings can be determined and applied to capture a new image. Specifically, the automatic exposure settings can be enabled and set based on CoIV and histogram data, as indicated at 521. In one embodiment, the automatic exposure settings can be determined by multiplying the histogram's SD by CoIV and a scale factor to help extract noise from the scene. The scale factor refers to the conversion of the gantry's physical dimensions to their corresponding dimensions in the camera image. The depth camera's gain (which controls the amplification of the signal from the camera sensor) can be set to 1 / CoIV. Once automatic exposure is enabled and set based on the CoIV and histogram SD determined at 513, a new 3D depth image, IR image, thermal image, or RGB image of the patient's location can be captured using the depth camera, and method 500 can continue to 531. However, in other embodiments, CoIV-based correction can be used in conjunction with BEDI, which will be described below.

[0075] Return to 512 ( Figure 4AIf the patient's size is larger than the threshold, method 500 proceeds to 513 and includes combining depth image information from the surrounding exposure settings. Therefore, BEDI can be applied when the patient's size is larger than the threshold, or optionally in conjunction with CoIV-based correction. BEDI attempts to fill in any holes contained within the depth regions of the captured image using exposure settings ranging from underexposed to overexposed (e.g., exposure settings that surround the original exposure of the image). For example, a depth image of the scene (e.g., depth image N) can be obtained using the most suitable exposure based on the current lighting conditions. Next, an overexposed depth image (e.g., exposure settings relative to depth image N; depth image N+1) and an underexposed depth image (e.g., exposure settings relative to depth image N; depth image N-1) can be captured. Regions lacking depth information can then be identified by comparing the original depth image with the overexposed and underexposed depth images (e.g., comparing depth image N with depth images N+1 and N-1). The missing depth information in the original depth image (e.g., depth image N) can then be filled using data from underexposed and / or overexposed images (e.g., depth image N+1 and depth image N-1).

[0076] At 514, method 500 includes determining whether the histogram has undesirable illumination on the RGB. Various aspects of the histogram, including illumination on the RGB histogram (e.g., histogram data from a color image), can be analyzed to determine the correction to be applied. If the RGB histogram does not have undesirable illumination, method 500 can proceed to 520, as described below.

[0077] If the RGB histogram is determined to indicate poor lighting, method 500 proceeds to 516, and histogram equalization is performed. Histogram equalization increases the contrast of the captured image by changing pixel values ​​under the guidance of the image's intensity histogram, thereby effectively modifying the dynamic range of each image. As previously described at 506, a histogram is a graphical representation of the intensity distribution of an image, representing the number of pixels for each intensity value considered. For RGB images, each of the R, G, and B components has a separate table entry. Histogram equalization cannot be applied to the R, G, and B components individually because it can cause significant changes in the image's color balance. Therefore, in one embodiment, histogram equalization can be performed via a nonlinear mapping that redistributes the intensity values ​​in the input image so that the resulting image contains a uniform distribution of intensity, thereby producing a flat (or near-flat) histogram. This mapping operation can be performed using a lookup table stored in the non-transitory memory of a computing device. In another implementation, an RGB histogram equalization method can be applied, which utilizes the correlation between color components and is enhanced by multi-layer smoothing techniques borrowed from statistical language engineering. In other implementations, the RGB image can be converted to another color space (e.g., Hue, Saturation, and Value (HSV) color space or Hue, Saturation, and Lightness (HSL) color space) before performing histogram equalization.

[0078] At 518, the RGB histogram processed by histogram equalization at 516 can be analyzed to ensure uniform intensity and contrast levels within the histogram. In one embodiment, an algorithm stored in the non-transitory memory of a computing device determines whether the processed histogram is flat (e.g., whether pixels are uniformly distributed across the entire intensity range). After equalization, the histogram may not be perfectly flat due to the characteristics of some intensity values ​​that may exist within the image, but these values ​​can be more evenly distributed. If it is determined that the processed histogram does not have uniform intensity and contrast levels, the camera exposure settings can be adjusted, a new RGB image can be captured, and subsequent histogram data can be equalized to ensure the histogram is flat. Once a processed RGB histogram with uniform intensity and contrast levels has been generated, method 500 can proceed to 520.

[0079] At 520, method 500 determines whether the histogram data of the image captured by the depth camera includes reflection noise (e.g., one or more extreme peaks on the right edge of the graph). If the histogram does not contain reflection noise, method 500 may proceed to 526, as described below. If the histogram data does reflect reflection noise, method 500 proceeds to 522, and the reflection intensity value of the reflection noise in the spatial region may be calculated. In one embodiment, an algorithm stored in non-transitory memory of a computing device may be used to calculate the intensity value of the pixel corresponding to the reflection noise (identified via histogram) in the spatial region of the captured image. The spatial region may include the patient, a hospital gown or blanket being used by the patient, a scanning table, the floor surrounding the scanning table, medical equipment surrounding the scanning table, a support pillar supporting the patient's position on the scanning table, and / or any other object within the captured scene. In another example, additionally or alternatively, the spatial region may be defined by a grid system and not based on the composition of the region. Once the intensity values ​​of the noise / reflection regions within the image have been determined, the tolerance level of these reflection regions may be adjusted at 524. In one implementation, the tolerance level can be determined using Euclidean distance. For example, the Euclidean distance can be determined between an image point with high reflectivity noise and surrounding image points to determine and set a reflectivity tolerance level that prevents the surrounding image points from being skewed by reflectivity noise, and to ensure that reflectivity saturation (which can cause holes in subsequent 3D depth frame acquisition) does not occur.

[0080] At 526, method 500 includes determining whether histogram data of an image captured by a depth camera has areas of partial or poor exposure. Areas of partial or poor exposure can be caused by dim lighting conditions in the scanning chamber and / or camera exposure settings (e.g., underexposure). For example, images captured in a dark scanning chamber and with the depth camera's auto exposure off (e.g., relative to...) Figure 7 (Further illustrated) Images of scenes that are poorly exposed due to lack of light during exposure. In other examples, underexposed or poorly exposed areas may occur due to the camera's automatic exposure settings. In other examples, underexposed or poorly exposed areas may be the result of camera positioning relative to the scene combined with exposure settings and / or lighting conditions within the scanning room.

[0081] If the histogram data does not show any underexposed or poorly exposed areas, method 500 can proceed to 530, as described below. If the histogram data does reflect underexposed or poorly exposed areas (e.g., one or more extreme peaks on the left side of the graph), method 500 proceeds to 528, and the intensity range of the underexposed or poorly exposed areas in the 3D depth data can be calculated. For example, the computing device can identify underexposed or poorly exposed areas based on the location of valleys within the histogram data, and can also determine the intensity range of those valleys.

[0082] At 530, the camera exposure settings can be automatically adjusted. In one embodiment, a lookup table stored in the non-transitory memory of a computing device can be used to automatically adjust the exposure settings. This lookup table is programmed to have exposure settings indexed according to intensity ranges identified in histogram data (e.g., such as at 518, 522, and / or 528). In some examples, in response to the identification of no poor lighting, no reflection noise, and no half-exposed or poorly exposed areas on the histogram, inputting an intensity range may cause the lookup table to output the same exposure settings, since the exposure settings may already be suitable for the given lighting conditions. As another example, in response to the identification of poor lighting (such as due to dim lighting conditions) at 514, the exposure settings can be adjusted to increase the exposure. As yet another example, in response to the identification of reflection noise (such as due to bright lighting conditions) on the histogram at 520, the exposure settings can be adjusted to decrease the exposure. Additionally, in some examples, the exposure settings may additionally or alternatively include exposure bracketing, wherein the selected exposure settings result in the automatic capture of additional images at both lower and higher exposure settings during the same acquisition.

[0083] At point 531, a camera-to-world coordinate transformation can be performed so that the positions of objects within the scene can be described independently of the camera position (e.g., based on point positions [x, y, z] in the world coordinate system). In one implementation, the transformation from camera to world coordinates can be given by the following equation:

[0084]

[0085] The first matrix represents the object's coordinates in world coordinates, the second matrix represents the same object's coordinates in camera coordinates, R is the rotation matrix, and T is the transformation matrix. The extrinsic parameters R and T can be obtained during camera calibration.

[0086] At point 532, an original 3D point cloud can be generated using images captured from a depth camera. The original 3D point cloud refers to a set of data points defined by a 3D world coordinate system. In one implementation, the depth camera can use a modulated light source to illuminate the scene (e.g., a patient on a scanning table) and use a ToF sensor located within the camera to observe the reflected light to generate the original 3D point cloud. In another implementation, a computing device can extract the original 3D point cloud data from stereo camera images. For example, an algorithm can be applied to a pair of captured stereo images to generate an original 3D point cloud based on differences between matching features in the right and left images.

[0087] At position 534, isosurface volume extraction can be performed on the raw 3D point cloud generated at position 532 to detect the patient's body shape / orientation / pose. An isosurface is a surface representing constant-value points within a spatial volume, thus allowing the extraction of 3D patient structure from the raw 3D point cloud. This is performed to identify holes or loss of depth information due to underexposure in the depth frame / image. Additionally, by extracting the patient's raw body shape prior to the filtering operation, data related to the patient's periphery and possible patient movement can be detected and used for subsequent filtering at position 536. An algorithm stored as executable instructions in a computing device can be used to determine the isosurface. In one implementation, the algorithm can use a voxel representation of volume, thus considering each data point as a vertex of a geometric primitive (such as a cube or tetrahedron). These primitives or cells subdivide the volume and provide useful abstractions for calculating the isosurface. For example, the isosurface volume can be extracted by converting the depth frame or depth values ​​into a 3D point cloud or mesh model. A frame can be converted into a 3D volumetric isosurface representation with dimensions in the x, y, and z directions using a vertex shader and a moving cube algorithm, allowing polygons to fit into the 3D point cloud data. This rendering technique will fit and reconstruct the patient's body shape in 3D world coordinates. In another implementation, an algorithm combining aspects of both geometric decomposition techniques and cross-spatial algorithms can be used to extract the isosurface.

[0088] At position 536, voxel filtering can be performed to reduce the density of the 3D point cloud and accelerate subsequent computations (e.g., the generation of 3D patient structure estimates can occur in less than 100 ms). In one implementation, a voxel lattice filter can be used to return a processed point cloud with a smaller number of points that accurately represent the input point cloud as a whole. The voxel lattice filter downsamples the data by taking the spatial average of the points in the cloud and adjusting the undersampling rate by setting the voxel size along each dimension. Any points located within the boundaries of each voxel are assigned to that voxel and will be combined into an output point (e.g., point clustering). In another implementation, a pass-through filter can be applied to generate the processed point cloud. The pass-through filter passes the input points through constraints that remove infinite points and any points located outside a specified field.

[0089] At point 538, the processed 3D point cloud can be segmented so that only points of interest remain within the scene. Points of interest in this paper include the workbench and the patient's body positioned on it. Segmentation is the process of grouping the point cloud into multiple homogeneous regions with similar characteristics (e.g., labeling each measurement in the point cloud so that points belonging to the same surface or region are given the same label). Object recognition and classification are the steps of labeling these regions. Once these objects are extracted and classified, it is possible to remove noisy and unwanted objects. For example, segmentation can be combined with object recognition and classification to remove points in the processed 3D point cloud that are associated with equipment (such as life-saving devices) around the workbench, such as relative to... Figure 6 Further description. Other objects in the scan room that can be segmented from the scene may include chairs, benches, shelves, trolleys, and / or other various medical equipment.

[0090] In one implementation, an edge-based segmentation algorithm can be used to remove noise and unwanted objects from a processed 3D point cloud. Edge-based segmentation algorithms have two main stages: edge detection that delineates the boundaries of different regions, followed by grouping points within the boundaries to produce the final segments. Edges are defined by points where changes in local surface properties exceed a given threshold. In another implementation, segmentation can be performed using a model-fitting algorithm based on the decomposition of man-made objects into geometric primitives (e.g., planes, cylinders, spheres). For example, a model-fitting algorithm can extract shapes by constructing candidate shape primitives using a minimum number of randomly selected data points. The candidate shapes are examined against all points in the dataset to determine a value representing the number of points for the best fit.

[0091] At position 540, post-processing of the processed 3D point cloud can be performed to further refine unwanted points and noise that could negatively impact 3D patient structure estimation. This can be based on position 524 (see [reference]). Figure 4A The parameters for post-processing are set by determining the reflective and non-reflective regions at a given location. In one implementation, post-processing can be performed via hypervoxarization, which oversegments the point cloud by grouping the points into homogeneous fragments called hypervoxels based on various attributes (e.g., normal, color, intensity, shape). Hypervoxarization can begin with normal (regularly spaced) voxels grouping these points into a 3D mesh, and then, for each voxel, iteratively clustering neighboring points with similar attributes to form hypervoxels with irregular shapes. In another implementation, hypervoxarization can be combined with Euclidean clustering by using 3D mesh subdivision with a fixed-width bounding box or, more generally, an octree data structure.

[0092] At point 542, the post-processed 3D point cloud can be overlaid on the original 3D point cloud, and the Hausdorff distance is used to determine the offset between the two point clouds. The Hausdorff distance measures how close each point in the model set is to a point in the image set, and vice versa. Therefore, this distance can be used to determine the similarity between two objects overlaid on each other. For example, the post-processed point cloud can be considered as a model set, and the original 3D point cloud generated at point 532 can be considered as an image set, where the offset between the two is determined based on the Hausdorff distance using an algorithm stored in the non-transitory memory of the computing device.

[0093] In 544 (see Figure 4D At point 542, method 500 determines whether the maximum value of the determined Hausdorff distance at point 542 is greater than one in any of the x, y, and z directions (e.g., defined by the patient's x, y, and z axes). The closer the determined Hausdorff distance is to zero, the more similar the two point clouds are to each other. Alternatively, if the maximum determined value is greater than 1 in any of the x, y, and z directions, it may indicate a depth error associated with depth camera calibration.

[0094] If the Hausdorff distance is no greater than one, method 500 can continue to 556, and post-processed point clouds can be used to perform scan result prediction, scan range overlay, and scan keypoint calibration, as will be discussed below relative to... Figure 9 Further description. Then method 500 can end.

[0095] If the Hausdorff distance is greater than 1, method 500 can continue to 548, where an internal camera calibration can be performed to address depth errors. Once the depth camera has been recalibrated, method 500 can continue at 550, where a new depth frame can be captured by repeating the workflow that began at 504, and a new raw 3D point cloud and a post-processed 3D point cloud can be generated using this new depth frame.

[0096] At 552, method 500 can examine patient motion by determining the offset between the new post-processed 3D point cloud and the original 3D point cloud of the previous depth frame. This offset refers to the change in patient position that occurs between acquiring the previous depth frame and the new (e.g., the currently acquired) depth frame. As previously described at 542, this offset can be determined by overlaying the new post-processed 3D point cloud onto the original 3D point cloud (generated at 532 by the first depth frame acquired in method 500) and using Hausdorff distance.

[0097] At 554, method 500 determines whether the offset determined at 552 is greater than 1 in any of the x, y, and z directions. If the offset is not greater than 1, it indicates that the patient has not moved, and method 500 can proceed to 556, where the post-processed point cloud can be used to perform scan result prediction, scan range overlay, and scan key point calibration, and then method 500 can terminate.

[0098] If the offset is greater than 1, it indicates that patient movement has occurred, and method 500 can continue to 558, where a technician can be alerted to reposition and reorient the patient. This alert can be a notification issued by the computing device in response to an offset greater than 1 determined at 554. Method 500 can then terminate. For example, method 500 can be repeated once the technician has repositioned and reoriented the patient and resolved the discrepancy related to patient movement / mobility.

[0099] The implementation of Method 500 allows the extraction of 3D point clouds containing the full range of depth information of patient structures, without relying on depth camera exposure settings and lighting conditions in the scanning room. Figure 5 Several exemplary representations 600 of variable scan room conditions are shown, from which depth frames can be captured and subsequently used to generate 3D point clouds of patient structures. A 2D image of the patient is shown in the first column 602, a tonal depth image determined from the 2D image is shown in the second column 604, and a 3D point cloud of the patient determined from the tonal depth image is shown in the third column 606. Each row represents data acquired during different lighting conditions. For example, a 2D image, tonal depth image, and 3D point cloud from a first lighting condition are shown in the first row 608; a 2D image, tonal depth image, and 3D point cloud from a second lighting condition are shown in the second row 610; a 2D image, tonal depth image, and 3D point cloud from a third lighting condition are shown in the third row 622; and a 2D image, tonal depth image, and 3D point cloud from a fourth lighting condition are shown in the fourth row 624.

[0100] The first lighting conditions shown in row 608 include conditions where all lights in the scanning room are on (e.g., bright lighting conditions) and the automatic exposure of the depth camera is turned on. This creates highly reflective areas 612 on either side of the patient on the scanning table, as shown in the 2D image (first column 602). These highly reflective areas cause a loss of depth information in the tonal depth image, as shown in the second column 604 of row 608. This loss of depth information, or depth holes, can be seen as black areas throughout the frame. If not corrected, these depth holes can lead to inaccurate 3D patient outcome estimates. For example, a depth hole 616 can be seen along the left lateral side of the patient's lower leg in the tonal depth image of row 608 (second row 604). Method 500 can be used to fill in the depth holes 616 to generate a 3D point cloud containing the full range of depth information related to the patient's structure, as shown in the 3D point cloud of row 608 (third column 606).

[0101] Similarly, the second illumination condition (second row 610) includes partially illuminating the scanning chamber and turning off the automatic exposure of the depth camera, thereby creating a highly reflective area 618 on the floor to the left of the patient on the scanning table (see the 2D image in first column 602). The highly reflective area 618 can cause a loss of depth information, such as the depth hole 620 at the left outer edge of the patient's knee in the tonal depth image (second row 604) of second row 610. Using... Figures 4A to 4D The method 500 can still generate a complete 3D point cloud (third column 606) under the second lighting conditions (second row 610).

[0102] The third lighting condition (row 3, 622) involves dim lighting in the scanning chamber and automatic exposure of the depth camera. Due to this dim lighting, there are no highly reflective areas in the 2D image (row 1, 602). Furthermore, there are no noticeable holes in the resulting tonal depth image (row 2, 604). Therefore, compared to the first and second lighting conditions, the tonal depth image undergoes reduced correction and processing during the generation of the 3D point cloud (column 3, 606) under the third lighting condition (row 3, 622).

[0103] The fourth lighting condition (row 4, 624) includes no light in the scanning chamber and automatic exposure of the depth camera turned off. However, even in poor lighting, the resulting tonal depth image (column 2, 604) does not have holes lacking depth information. For example, the histogram of the tonal depth image can be adjusted to generate the 3D point cloud shown in row 3, 606. Therefore, it can be used... Figures 4A to 4D Method 500 accurately generates 3D point clouds of patient structures under varying lighting conditions and different depth camera settings by processing the captured image according to its own characteristics (such as poor lighting, high reflectivity, etc.).

[0104] Figure 6 An example 700 illustrates how the processed 3D point cloud generated in method 500 can be segmented so that only points of interest are retained within the scene. As a non-limiting example, view 702 of the scene shows medical equipment 708 surrounding a patient positioning workbench. When a thermal image 704 of the scene is captured, object recognition and classification can be used to identify the medical equipment 708 within the thermal image 704. Once the medical equipment 708 has been identified and classified, it can be... Figure 4C The points are segmented as described above so that the subsequently generated 3D point cloud 706 contains only information related to the patient's orientation, posture, and position on the scanning table. For example, the 3D point cloud 706 does not show the structure of the medical device 708.

[0105] Figure 7 It shows the use of Figures 4A to 4D Example 800 presents the method for determining 3D patient structures under dim lighting conditions. Dim lighting conditions can occur when the patient examination scene is partially illuminated and / or illuminated using lamps of low intensity. Furthermore, the patient is covered with a blanket, as shown in 2D image 802. Using methods such as... Figures 4A to 4D The method 500 generates a tonal depth image 804 of a patient covered with a blanket from a 2D image 802 (e.g., under dim lighting conditions), and then uses this tonal depth image to generate a raw 3D point cloud 806 from which an accurately filtered and processed 3D point cloud 808 can be extracted. The filtered and processed 3D point cloud 808 can then be superimposed on the raw 3D point cloud 806, as shown in superposition 810. Superposition 810 can then be used to perform patient-scan collision prediction, scan range overlay, and scan keypoint calibration, because superposition 810 distinguishes the patient structure from the rest of the patient examination scene (including the blanket, the patient positioning table, and gantry holes).

[0106] Figures 8A to 8C It is shown in a series of consecutive images 900 as relative to Figures 4A to 4D Various aspects of the method 500. Figure 8A Image 902 is a 2D image of the patient positioned on the scanning stage, which can be acquired at 504 of method 500. Figure 8A Image 904 is a depth image of the patient shown in image 902, which can be captured using a depth camera at 532 of method 500. Figure 8A Image 906 is a raw 3D point cloud that can be generated and used to perform camera-to-world coordinate transformation on the depth camera at point 531 of method 500. Figure 8A Image 908 is the original 3D point cloud of the patient's location captured at 532 in method 500 after the depth camera was transformed with world coordinates. Figure 8BImage 910 can be generated after isosurface volume extraction is performed on image 908 at point 534 of method 500. Figure 8B Image 912 shows image 910 after voxel filtering is performed at 536 of method 500 to reduce the density of the 3D point cloud. Figure 8B Image 914 shows image 912 after the unwanted region was segmented at 538 in method 500. Figure 8B Image 916 is the raw 3D point cloud, which can be overlaid at 542 of method 500 with the segmented, processed 3D point cloud shown in image 914 to determine potential offsets based on patient movement or camera calibration. Figure 8C Images 918, 920, and 922 show different angles of the post-processed 3D point cloud generated using method 500, which can be used to determine 3D patient structure estimates prior to medical imaging.

[0107] As mentioned above, 3D patient structure estimation can be used for scan result prediction, scan range overlay, and scan key point calibration. This scan key point calibration can also be used to determine whether the patient is in the required scan pose. Therefore, Figure 9 An exemplary method 1000 is provided for analyzing 3D patient structure estimates to perform pose prediction. As an example, this can be achieved using a computing device (e.g., Figure 2 The computing device 216) as Figures 4A to 4D Method 500 (e.g., at 556) is performed in part before and / or during medical imaging of method 1000. A CT imaging system is used, by way of example, in which a patient is positioned on a stage capable of moving relative to a gantry aperture.

[0108] At point 1002, method 1000 includes receiving a superposition of the post-processed 3D point cloud and the original 3D point cloud. For example, a computing device can use... Figures 4A to 4D Method 500 iteratively adjusts the depth camera settings and / or corrects the depth frames determined by the images captured by the depth camera until a post-processed 3D point cloud is obtained that does not include holes in the depth information.

[0109] At 1004, method 1000 includes determining patient body shape and patient posture from a post-processed 3D point cloud. In this example, patient posture includes both the patient's position and orientation on the worktable. For example, the post-processed 3D point cloud can be used to determine whether the patient's orientation relative to the gantry is head-first or feet-first, and whether the patient is in a supine, prone, or lateral position. Additionally, the positioning of the patient's limbs can be determined (e.g., arms crossed over the chest, right leg bent towards the chest, left arm extended and raised, etc.). In one embodiment, a trained classifier stored in non-transitory memory of a computing device can be used to classify patient body shape and patient posture to analyze the post-processed 3D point cloud based on anatomical key points and body regions. For example, a first classifier can identify patient orientation (e.g., foot-first or head-first), a second classifier can identify patient position (e.g., supine, prone, lateral), a third classifier can identify limb posture (e.g., both legs straight, left arm crossed over chest), and a fourth classifier can estimate the position of internal organs relative to the gantry and scanning stage (e.g., heart centered relative to the gantry / scanning stage), etc. In another embodiment, the post-processed 3D point cloud can be parsed or segmented from body regions (e.g., head, torso) using global constraints (e.g., patient's height, weight, width), and the anatomical features within each region are further defined based on body boundaries.

[0110] At 1006, method 1000 includes comparing the determined patient pose with the desired patient pose. The desired patient pose may be determined based on, for example, a received examination CID, which specifies the scanning protocol to be used and the desired patient pose for performing the examination. As an example, the determined patient pose may be classified using a trained classifier as described above, and then compared directly with category information outlining the desired patient pose. In this example, each category of the determined patient pose may be compared with the corresponding category of the desired patient pose, and a match or mismatch may be determined. As another example, additionally or alternatively, model fitting may be used to perform a coarse alignment between the determined patient pose and the desired patient pose. For example, important anatomical key points within an outline of body segments (e.g., head, trunk, pelvis, thigh, calf) determined using a trained classifier may be coarsely aligned with a segmented drawing structural model of the desired patient pose.

[0111] At 1008, method 1000 includes determining whether the determined patient posture matches the desired patient posture. As an example, if no category of the determined patient posture matches the corresponding category of the desired patient posture (e.g., at least one mismatch exists), it can be inferred that the determined patient posture does not match the desired patient posture. Alternatively, if no mismatch exists (e.g., all categories between the determined posture and the desired posture match), it can be inferred that the determined patient posture matches the desired patient posture. As yet another example, additionally or alternatively, the algorithm may analyze a coarse alignment of the determined patient posture and the desired patient posture to determine whether the two postures match.

[0112] If the determined patient posture does not match the required patient posture, method 1000 proceeds to 1014 and includes alerting the technician to reposition and reorient the patient. This can be achieved via a computing device (such as...) Figure 2 The computing device 216) or a user display (e.g., communicatively coupled to the computing device) is either a computing device or a user display that is communicatively coupled to the computing device. Figure 2 The display 232) issues a notification to alert the technician to reposition and reorient the patient. For example, the computing device may output an auditory or visual warning. In at least some examples, the warning may include an indication that the patient's posture does not match the required posture, as well as instructions for repositioning the patient. Method 1000 may then terminate.

[0113] Returning to 1008, if the determined patient pose does indeed match the desired patient pose, method 1000 proceeds to 1010 and includes performing scan outcome prediction based on the patient's body shape relative to the stage coordinates. Scan outcome prediction may include determining whether any potential patient-gantry contact may occur when the patient and scanning stage move into the aperture, and predicting potential contact once scanning begins. In some examples, scan outcome prediction may also include identifying the start and end ranges of the scan by overlaying scans on the patient's body shape. In one implementation, an algorithm may be used to perform scan outcome prediction to determine how many points of the determined patient location extend beyond the aperture boundary and how many points remain within the aperture boundary, using determined patient structure estimates (e.g., patient body shape and pose) as input.

[0114] At 1012, it is determined whether a scanning problem is predicted. A scanning problem may include patient positioning that would result in physical contact between the patient and the gantry aperture once the scan begins. Alternatively or additionally, a scanning problem may include the start and end ranges of the scan overlay not being aligned with the CID of the examination. For example, even if the patient is positioned correctly, small adjustments to limb position may result in clearer images of the anatomical structures of interest. If a scanning problem is predicted, method 1000 proceeds to 1014 and includes alerting the technician to reposition and reorient the patient, as described above. If a scanning problem is not predicted, method 1000 proceeds to 1016 and includes initiating the patient scanning protocol. Method 1000 may then terminate.

[0115] Thus, once the analysis of patient structures determines that the patient is properly positioned, patient scanning can begin. Patient structures can be determined using data from a depth camera, where depth holes are compensated for by correcting the depth image and / or adjusting the depth camera settings. Therefore, the accuracy of patient imaging can be improved, while reducing the incidence of rejection and repeat scans, thereby reducing the amount of time before a diagnosis can be made. Furthermore, patient structures can be determined with the same increased accuracy under variable illumination conditions.

[0116] The effect of applying dynamic correction to depth images of a patient positioned on the worktable of a medical imaging system is to increase the accuracy of patient pose estimation, thereby increasing the accuracy of scans performed by the medical imaging system by ensuring that the patient is in the required pose for scanning.

[0117] As used herein, elements or steps listed in the singular and beginning with the word "a" or "an" should be understood to not exclude a plurality of said elements or steps unless such exclusion is explicitly stated. Furthermore, references to "an embodiment" of the invention are not intended to be construed as excluding the existence of additional embodiments that also include the referenced features. Moreover, unless explicitly stated to the contrary, embodiments that "comprise," "include," or "have" elements or multiple elements having a particular characteristic may include additional such elements that do not have that characteristic. The terms "comprise" and "in..." are used as concise linguistic equivalents to the corresponding terms "comprising" and "wherein". Furthermore, the terms "first," "second," and "third," etc., are used merely as notations and are not intended to impose numerical requirements or a particular order of position on their objects.

[0118] This written description uses examples to disclose the invention, including the best mode, and also enables those skilled in the art to practice the invention, including making and using any device or system and performing any included methods. While the examples provided herein relate to medical applications, the scope of this disclosure covers non-destructive testing in industrial, biomedical, and other fields. The patentable scope of the invention is defined by the claims and may include other examples that would occur to those skilled in the art. Such other examples are intended to fall within the scope of the claims if they have structural elements that are not indistinguishable from the literal language of the claims, or if they include equivalent structural elements that differ only slightly from the literal language of the claims.

Claims

1. A method for a medical imaging system, the method comprising: Depth images of a patient positioned on the worktable of the medical imaging system are acquired via a depth camera; The depth image is corrected based on histogram data from the depth image; The patient's three-dimensional structure is extracted based on the corrected depth image; Correcting the depth image based on the histogram data from the depth image includes performing a correction based on an illumination variation coefficient, and further based on histogram data from the patient's color image. The correction based on the illumination variation coefficient includes: Determine the standard deviation of the combined histogram data from both the depth image and the color image, as well as the illumination variation coefficient of the combined histogram data; The automatic exposure settings of the depth camera are turned off in response to the illumination variation coefficient of the combined histogram data being greater than the standard deviation of the combined histogram data; and The automatic exposure setting of the depth camera is activated in response to the illumination variation coefficient of the combined histogram data not being greater than the standard deviation of the combined histogram data, wherein the automatic exposure setting is adjusted based on the illumination variation coefficient of the combined histogram data, the standard deviation of the combined histogram data, and a scale factor that correlates the coordinates of the depth image with the coordinates of the worktable.

2. The method according to claim 1, further comprising: The patient's posture is determined from the extracted three-dimensional structure of the patient; The determined posture of the patient is compared with the required patient posture; as well as An alert is output in response to the patient's determined posture not matching the desired patient posture.

3. The method of claim 1, wherein the worktable is movable relative to the frame hole, and the method further comprises: The patient's body shape was determined from the extracted three-dimensional structure. The scan prediction is performed relative to the coordinates of the worktable based on the patient's determined body shape; as well as An alert is output in response to the scan prediction indicating contact between the patient and the gantry aperture.

4. The method of claim 1, wherein correcting the depth image based on the histogram data from the depth image comprises performing bracketed exposure depth imaging correction.

5. The method of claim 4, wherein performing the bracketing exposure depth imaging correction comprises at least one of: performing equalization of the histogram data, determining the tolerance level of the reflective regions in the depth image, and adjusting the exposure settings of the depth camera.

6. The method of claim 5, wherein performing the equalization of the histogram data responds to identifying poor lighting in the depth image based on the histogram data, determining the tolerance level of the reflective region in the depth image responds to identifying reflective noise in the depth image based on the histogram data, and adjusting the exposure settings of the depth camera based on the current intensity value and current exposure settings of the histogram data.

7. The method of claim 1, wherein extracting the three-dimensional structure of the patient based on the corrected depth image comprises: Generate a 3D point cloud based on the corrected depth image; as well as Perform isosurface volume extraction of the three-dimensional point cloud to extract the three-dimensional structure of the patient.

8. The method of claim 7, wherein the three-dimensional point cloud is an original three-dimensional point cloud, and the method further comprises: The original 3D point cloud is processed via voxel filtering and segmentation to isolate the patient's 3D structure from other objects in the acquired depth image; The processed 3D point cloud is superimposed on the original 3D point cloud; Hausdorff distance is used to determine the offset between the processed 3D point cloud and the original 3D point cloud; as well as In response to the Hausdorff distance being greater than one, a depth error is indicated and the calibration of the depth camera is adjusted before acquiring a new depth image of the patient.

9. A method for a medical imaging system, the method comprising: Based on histogram data from the acquired images, the exposure mode and gain of the depth camera located for acquiring images of the patient examination scene are adjusted. as well as The three-dimensional structure of the patient in the patient examination scene is determined based on the acquired images. Adjusting the exposure mode and gain of the depth camera includes adjusting the exposure mode between a first mode with auto exposure on and a second mode with auto exposure off, based on the peak distribution in the histogram data. Adjusting the exposure mode between the first mode and the second mode based on the peak distribution in the histogram data includes: Determine the illumination variation coefficient and standard deviation of the peak distribution in the histogram data; In response to the illumination variation coefficient being greater than the standard deviation, the exposure mode is adjusted to the first mode and the gain is set based on the illumination variation coefficient; and In response to the illumination variation coefficient not being greater than the standard deviation, the exposure mode is adjusted to the second mode.

10. The method of claim 9, further comprising: The patient's posture and body shape are determined based on the patient's identified three-dimensional structure; The determined posture of the patient is compared with the desired patient posture; The patient examination scenario is based on the patient's determined body shape to predict whether an examination problem will occur; as well as An alert is output in response to at least one of the determined patient posture mismatch and the predicted examination problem.

11. The method of claim 9, wherein adjusting the exposure mode and the gain of the depth camera based on the histogram data from the acquired image comprises: The tonal range and tonal balance of each image in the acquired images are determined based on the histogram data. Based on the tonal range and the intensity value of the tonal balance, identify overexposed and underexposed areas in the acquired image; as well as The exposure mode and the gain are adjusted based on the intensity value to compensate for the overexposed and underexposed areas.

12. A medical imaging system, comprising A rotatable frame having a centrally located hole therein; A worktable that is movable within the aperture and configured to position a subject for image data acquisition within the aperture; A depth camera, positioned to acquire a depth image of the subject on the worktable before entering the hole; as well as A computing device that stores executable instructions in non-transitory memory, wherein the executable instructions, when executed, cause the computing device to: Receive the depth image of the subject on the worktable from the depth camera; Identify the lighting conditions in the received depth image based on histogram data from the received depth image; The received depth image is corrected based on the identified lighting conditions; The posture of the subject on the worktable is identified based on the corrected depth image; as well as A warning is output in response to the subject's posture deviating from the posture required for image data acquisition. Correcting the received depth image based on the identified lighting conditions includes performing a correction based on an illumination variation coefficient, and further based on histogram data from the color image of the subject. The correction based on the illumination variation coefficient includes: Determine the standard deviation of the combined histogram data from both the depth image and the color image, as well as the illumination variation coefficient of the combined histogram data; The automatic exposure settings of the depth camera are turned off in response to the illumination variation coefficient of the combined histogram data being greater than the standard deviation of the combined histogram data; and The automatic exposure setting of the depth camera is activated in response to the illumination variation coefficient of the combined histogram data not being greater than the standard deviation of the combined histogram data, wherein the automatic exposure setting is adjusted based on the illumination variation coefficient of the combined histogram data, the standard deviation of the combined histogram data, and a scale factor that correlates the coordinates of the depth image with the coordinates of the worktable.

13. The medical imaging system of claim 12, wherein the illumination conditions include at least one of dim lighting and bright lighting, and the instruction causing the computing device to correct the received depth image based on the identified illumination conditions includes additional instructions stored in non-transitory memory, which, when executed, cause the computing device to: In response to detecting dim lighting, identify underexposed areas in the received depth image; and In response to the detection of bright illumination, a reflective region is identified in the received depth image.

14. The medical imaging system of claim 13, wherein the computing device includes additional instructions stored in a non-transitory memory, the additional instructions, when executed, causing the computing device to: In response to detecting dim lighting, increase the exposure setting of the depth camera; and When operating with the increased exposure settings, the underexposed areas in the received depth image are corrected based on the new depth image received from the depth camera.

15. The medical imaging system of claim 13, wherein the computing device includes additional instructions stored in a non-transitory memory, the additional instructions, when executed, causing the computing device to: In response to the detection of bright lighting, the exposure settings of the depth camera are reduced; and When operating with the reduced exposure settings, the reflective areas in the received depth image are corrected based on the new depth image received from the depth camera.

16. The medical imaging system of claim 12, wherein the instruction causing the computing device to identify the posture of the subject on the worktable based on the calibrated depth image includes additional instructions stored in a non-transitory memory, which, when executed, cause the computing device to: A three-dimensional point cloud of the subject on the worktable is generated based on the corrected depth image; Extract the structure of the subject on the worktable from the three-dimensional point cloud; and The structure of the subject on the workbench is compared with a posture classifier to identify the posture of the subject on the workbench.

Images