Method and apparatus for classifying structures in an image

Abstract

A system and method for segmenting selected data is disclosed. In various embodiments, automatic segmentation of fiber tracts in image data can be performed. The automatic segmentation can allow identification of specific fiber tracts in the image.

Description

[0001] Cross-reference to related applications

[0002] This application includes subject matter similar to that disclosed in U.S. Patent Application 16 / 862,840 (Attorney's File No. 5074A-000211-US) and U.S. Patent Application 16 / 862,882 (Attorney's File No. 5074A-000212-US). The entire disclosure of each of the foregoing applications is incorporated herein by reference. Technical Field

[0003] This disclosure relates to a tracer imaging, and more particularly to a tracer imaging classification method and system. Background Technology

[0004] This section provides background information in connection with this disclosure, which is not necessarily prior art.

[0005] Imaging techniques can be used to acquire image data of a subject. This image data may include one or more types of data used to visualize one or more structures of the subject. These structures may include the subject's external or internal structures. In various systems, for example, selective imaging techniques can be used to acquire images of the subject's internal structures.

[0006] Image data can be used to generate images that are displayed for user viewing. These images can be displayed on selected systems, such as display devices, for visual inspection by the user. Generally, users can view the images to help perform procedures on the subject. Summary of the Invention

[0007] This section provides a general overview of this disclosure and is not a full disclosure of the complete scope or all features of this disclosure.

[0008] This method can be used to identify and / or classify various structures in data, such as image data. These structures can be identified as fibers or bundles in the brain of a subject (such as a human subject). However, it should be understood that the process can be used to identify bundles in any suitable subject. Generally, these bundles are neuronal bundles that include white matter axons. This method can be used to identify various bundles in the brain to aid in various procedures, such as tumor resection, implant placement (e.g., deep brain stimulation leads), or other appropriate procedures.

[0009] In various implementations, this process can be used to identify neuronal bundles even in the presence of distortions in standard anatomical structures. For example, tumors can grow in the brain, which can affect neuronal bundles that are normally understood or present. Therefore, this process can help identify neuronal bundles even in the presence of tumors or other abnormal anatomical structures in image data.

[0010] Neuronal bundles can then be displayed for various purposes. These bundles can be used to identify various features and / or structures of the brain to aid in procedures. Furthermore, images can be used to assist in the execution of procedures such as selecting a subject for tumor resection or implant placement. Therefore, image data can be used to identify bundles within the image data, and these bundles and / or other appropriate features can be displayed in one or more images using a display device.

[0011] Further areas of applicability will become apparent from the description provided herein. The descriptions and specific examples in this overview are intended for illustrative purposes only and are not intended to limit the scope of this disclosure. Attached Figure Description

[0012] The accompanying drawings described herein are for illustrative purposes only, representing selected embodiments and not all possible specific implementations, and are not intended to limit the scope of this disclosure.

[0013] Figure 1 It is an environmental view of the imaging system and the patient;

[0014] Figure 2 This is an environmental view of the regulations room;

[0015] Figure 3 It is a flowchart of methods for segmenting image data according to various implementation schemes;

[0016] Figure 4 It is a method for segmenting image data based on various implementation schemes;

[0017] Figure 5 It is a flowchart of methods for segmenting image data according to various implementation schemes;

[0018] Figure 6A This is a flowchart of a method for training a system to segment image data according to various implementation schemes; and

[0019] Figure 6B It is a flowchart of a method for segmenting image data using a system, according to various implementation schemes.

[0020] In several views of all the accompanying drawings, the corresponding reference numerals indicate the corresponding parts. Detailed Implementation

[0021] Exemplary embodiments will now be described more fully with reference to the accompanying drawings.

[0022] The system can be used to acquire image data of the examinee. For example, such as... Figure 1The image is schematically illustrated. Generally, a subject can be imaged using a selected imaging system 20. The imaging system 20 can be any suitable imaging system for acquiring selected image data of the subject 24, such as a magnetic resonance imaging (MRI) scanner. MRI can operate in various ways or modes, including diffusion-weighted data acquisition methods. In diffusion-weighted techniques, the diffusion or movement of water can be determined or estimated to help determine the bundle, as discussed further herein. The imaging system 20 may include a selected support 28 for supporting the subject 24 during movement through the imaging system 20. The imaging system may include components commonly known to those skilled in the art. However, it should be understood that any suitable MRI imaging system can be used as imaging system 20 or any other suitable imaging system. However, the imaging system 20 can be operated to generate image data of the subject 24.

[0023] Continue to refer to Figure 1 And refer to other sources Figure 2 After acquiring data using imaging system 20 (which may include anatomical image data and / or diffusion-weighted data) and / or during acquisition (e.g., using a suitable imaging system), the patient 24 may be moved or positioned in a selected operating room for the execution of procedures. During or in planning the execution of procedures on the patient 24, an image 32 of the patient 24 may be displayed on a display device 36 of a selected workstation or system 40. The display of image 32 may be used for various reasons, such as preparing for procedures on the patient 24, assisting in the execution of procedures on the patient 24, or for other appropriate purposes. As discussed herein, for example, image 32 may include, for example, an illustration of bundles identified in image 32 to be displayed on display device 36 and / or image data acquired using imaging system 24. Bundles may include bundles of fiber neurons in the brain of the patient 24. Bundles may also include other suitable bundles, such as fiber bundles in the spinal cord. Anatomical or known fiber bundles may also be referred to as given fiber bundles.

[0024] Workstation or processor system 40 may include various parts or features, such as input device 44, processor section or module 48, and memory 52. ​​Selected connections, such as wired or wireless connection 54, may connect processor 48 and / or memory 52 to display device 36 and / or input device 44. Thus, selected users, such as user 58, may provide input to system 40 or select input and / or view a view of display device 36, such as image 32.

[0025] In various implementations, image 32 may be registered to subject 24 for selected procedures. For example, tumor resection, implant placement, etc., may be performed on subject 24. However, it should be understood that subject 24 may be an inanimate object and other suitable procedures, such as part replacement or connection, may be performed. Imaging and registration may occur in suitable manners, such as those disclosed in U.S. Patent 9,311,335, issued April 12, 2016, which is incorporated herein by reference. Briefly, registration may include translation between subject space defined relative to subject 24 and image space defined by image 32. Coordinate systems may be developed for the two spaces, and translation maps may be created between the two spaces.

[0026] In various implementations, for registration, portions of the subject 24 may be identified by user 52, and identical portions may be identified in image 32. Registration between the subject space and image space can then be performed by generating or determining a translation map between the subject space and image space based on the identified similar or identical points (also referred to as reference points). Processor module 48 can execute instructions to perform translation and registration from subject space to image space.

[0027] Following registration, a navigation system 70, which may be incorporated into a workstation or processor system 40, may selectively perform translation and registration to aid in the execution of procedures. Once registration occurs, a dynamic reference frame or patient reference device 74 may be attached to or secured to the subject 24. The DRF 74 may include a tracking device or portion thereof that utilizes the navigation system 70 to track and determine the movement of the subject 24 and maintain registration with image 32. Instrument tools 76 may also utilize selected tracking devices, such as a tracking device 78 associated with (e.g., attached to) the instrument, for tracking purposes in procedures. The instrument may be at least one of implantable deep brain stimulation (DBS) leads. In addition to the instrument, selected procedures may be tracked and / or determined, such as calculating activation levels in the DBS leads and / or calculating thermal ablation estimates for ablation. In various embodiments, a locator 82 may generate a field 86 sensed by the tracking device 78 and / or the DRF 74 to allow tracking using the navigation system 70. A controller, such as a coil array controller 90, may receive signals from and / or control the positioner 82 and / or receive signals from the tracking device 78 and transmit those signals to a workstation processor module 48. The processor module 48 may then execute instructions to determine the position of the tracking device 78 relative to the subject 24, allowing the instrument 76 to navigate relative to the subject 74. Therefore, a graphical representation 94 of the tracked portion (e.g., instrument 76) may be displayed on a display device 36 relative to an image 32 to show the real-time positioning of the instrument 76 relative to the subject 24 by displaying the graphical representation 94 of the instrument relative to the image 32. Generally, a tracking system is capable of tracking and displaying the position or pose of the tracked portion. This positioning may include three-dimensional position and one or more dimensional orientations. Generally, at least four degrees of freedom information is included, comprising at least six degrees of freedom or any selected quantity.

[0028] As discussed further herein, image 32 may include bundles of selected image features, such as the brain, and graphical representation 94 may be displayed relative to these selected image features. Thus, defining or identifying the bundles to be performed can help execute the selected procedures, as discussed further herein.

[0029] As discussed above, imaging system 20 can be used to generate image data of subject 24 for selected procedures. The image data can be processed according to procedures (including one or more procedures), as further discussed herein. After one or more selected procedures, the image data can be used to generate one or more images to be displayed on display device 36. Furthermore, various procedures can be used to identify features that can also be output by selected processing systems (such as processor module 48) as discussed herein.

[0030] According to various embodiments, this document discloses systems and methods for identifying (such as by segmentation) portions of an image or data. For example, one or more fiber tracts or bundles may be segmented. A fiber bundle may include one or more fibers, which are segmented by the system to represent a bundle of neurons in the subject's brain. Fibers or lines include individual lines generated by a selected system (e.g., an algorithm), which may be one or a portion of a set of lines, intended to represent a physical bundle of neurons in the subject's brain. White matter bundles or anatomical bundles are physical tracts in the subject's brain that transmit electrical signals from one region of the brain to another, and are intended to be represented by fiber bundles. More than one fiber bundle and white matter bundle may exist in any given data. A given fiber bundle or white matter bundle may be a predetermined or named bundle.

[0031] Original Reference Figure 3 And continue to refer to Figure 1 and Figure 2 A method for identifying or segmenting fiber bundles or plexuses is illustrated in process 100. This process can be used for identification, such as by segmenting selected fiber bundles, which may also be referred to as given fiber bundles. Fiber bundles may be brain fiber bundles and are also referred to as given fiber neuron bundles. Segmentation may include segmenting a given bundle from other bundles and / or other data such as image data (i.e., classifying which fibers belong to a given neuron bundle). It should be understood that segmentation can occur according to various embodiments (including combinations of various embodiments), as discussed herein.

[0032] Initially, procedure 100 may begin in start box 104. The start in start box 104 may include appropriate measures or steps, such as assessing the patient, positioning the patient or subject for imaging, evaluating the subject, or assessing possible treatment or diagnosis for the subject. However, it should be understood that any appropriate procedure may occur in start box 104. After procedure 100 is started in start box 104, diffusion-weighted gradient images or appropriate data may be accessed or retrieved in box 108. It should be understood that procedure 100 may include, for example, acquiring image data using imaging system 20 in box 108.

[0033] In various implementations, image data may be acquired at any appropriate time and stored in a selected manner to aid in the execution of procedures. Therefore, accessing or retrieving image data in box 108 may include: acquiring image data of the subject using an imaging system such as imaging system 20 as discussed above and / or retrieving previously acquired image data. For example, image data may be acquired in a first case, such as during an initial diagnostic or testing procedure, and may be used in procedure 100. Therefore, image data may be accessed in box 108 rather than acquiring new image data. Additionally or alternatively, image data of the subject 24 may be acquired during the procedure or at the procedure location (e.g., in an operating room for resection or implantation).

[0034] After accessing the image data in box 108, the determination of the selected tracer imaging may occur in box 112. The determination of the selected tracer imaging in box 112 may include full-image tracer imaging, which includes identifying all possible fiber strands or bundles based on one or more of various known techniques given the acquired image data. Various techniques may include identifying or attempting to identify all possible bundles using or through a selected algorithm and / or by a user. For example, tracer imaging may be performed according to appropriate techniques such as appropriate algorithmic techniques. For example, tracer imaging may utilize a diffusion tensor imaging tracer imaging system (such as those sold by Medtronic that include…) The tracer imaging system of the image-guided surgical system is used to perform the procedure. Other suitable tracer imaging techniques may include the tracer imaging techniques described in U.S. Patent 9,311,335, published April 12, 2016.

[0035] In various implementations, selected tracer imaging may be smaller than or may include less than full image tracer imaging, as a supplement to or alternative to whole or complete image tracer imaging (e.g., whole brain tracer imaging). Therefore, selected tracer imaging may include a subset of full image tracer imaging. For example, only between selected regions of interest (e.g., half of the brain or the brainstem). Thus, as discussed herein, selected tracer imaging generated for a segmentation / recognition process may include full image tracer imaging and / or only a subset of full image tracer imaging. Furthermore, while full image tracer imaging may be generated, only a subset of fibers or bundles may be analyzed for segmentation. Therefore, unless otherwise specifically stated, the discussion herein of image tracer imaging or equivalents in box 112 is to be understood as including full image tracer imaging, selected image tracer imaging, or combinations thereof.

[0036] After the selected tracer image is determined in box 112, the selected tracer image can be output in box 116. In various embodiments, the image may include an image of the brain of the subject 24, such as a human subject. Thus, complete image tracer imaging may include tracer imaging of the whole brain. As discussed above, the selected tracer image may be part of a tracer imaging of the whole brain. However, it should be understood that the brain or image tracer image output in box 116 may simply include the determination of all possible connections of the bundles. As those skilled in the art will understand, tracer imaging may include the determination of possible or average movement or movement trends of selected substances such as water. Thus, the tracer image output in box 116 may simply be all possible bundles or connections and / or selected bundles identified within the accessed image data 108. Thus, the tracer image output in box 116 may not fully or accurately define or identify specific anatomical neuronal bundles within the subject.

[0037] The output tracer image can then be segmented to output or identify selected or given fiber bundles, such as neuronal bundles (e.g., corticospinal tract, optic tract, etc.), by invoking a trained classification system in box 120. The trained classification system may include a random forest classification system trained to use various features such as fiber points and image features to classify whether fibers in the selected tracer image belong to a specific neuronal bundle. As discussed above, brain tracer imaging may include any identified bundles or connections where material such as water may be present within the image, either possibly or evenly diffused. Therefore, segmentation of the image tracer image may attempt to identify or classify various possible bundles (such as the corticospinal tract (CST), orbital or ocular tract, or other appropriate anatomical or neural bundles) that are associated with certain anatomical functions or features.

[0038] The classification system can be trained using a variety of parameters that can be used as learning parameters for different bundles. The parameters of the features can include various parameters of features that can be identified and evaluated within the image-tracing imaging. Features can include the identified points and their positions on the bundles in the image-tracing imaging; line segments between points and / or vectors from each point; fractional anisotropy (FA) at each point; diffusion-coded color (DEC) value at each point; curvature at each point, such as the curvature of the line segment to the next point; comparison with atlases of known fiber bundles or known brain regions at the identified points or a selected number of points. Atlases can be generally known atlases of selected fiber bundles or known brain regions (e.g., brain neuron bundle atlases); the distance or length of the bundle from the corresponding start region to the end region; additional or other image properties at each point and / or at anatomical gradient values ​​within the image. These features can be trained using multiple data sets, such as data from previous classifications. Previously classified data can be classified by a reading expert (e.g., a neurosurgeon).

[0039] The trained classification system can then be used to classify the bundles within image tracing imaging 116. For example, instructions can be stored and retrieved from memory 52, which includes one or more of the features described above. The classification system can then be used to evaluate the image tracing imaging to identify a given bundle. The segmented bundle may also be referred to as the identified or segmented bundle.

[0040] The trained classification system invoked in box 120 is used to segment a given fiber bundle. Automatic segmentation of the image-tracing imaging can then occur in box 130. As discussed above, the system may include a workstation or processor system 40, which may include a processor module 48. The processor module 48 may communicate with a memory 52. ​​The memory 52 may include accessed image data 108 and the classification system invoked in box 120.

[0041] Processor module 48 utilizes a trained classification system to perform or evaluate accessed data from box 108 on brain tracking imaging to automatically segment fiber tracts, such as neural tracts or nerve fibers, in image tracking imaging within box 130. Automatic segmentation can occur by applying a classification system to selected image tracking imaging. Automatic segmentation applies classification parameters to selected tracking imaging and thus can generate or determine segmentation of the selected image (including complete or subset segmentation). In various embodiments, automatic segmentation can identify selected features (such as regions of interest (ROIs) (e.g., possible start and end regions) or other brain regions through which the tract passes or does not pass) and classify fibers using these selected features. When classifying tracts, the classification system can identify a given fiber tract that can be a neuronal tract, i.e., an anatomical feature tract. Anatomical features can be pre-selected by a user (such as user 78) or can be identified by the system, such that all possible tracts are segmented and identified.

[0042] Automatic segmentation in box 130 can be performed by executing instructions using processor module 48 and evaluating selected image tracer imaging using a invoked classification system. Therefore, automatic segmentation can determine or identify the segmented fiber bundles that can be output in box 140. Outputting the segmented fiber bundles in box 140 may include: outputting all fiber bundles with selected confidence intervals identified as selected bundles based on the classification system invoked in box 120. The output in box 140 may include only those bundles selected for identification by user 78 or other suitable systems, and / or may include identification and output of all possible bundles from classification system 120. Therefore, classification system 120 can be trained to identify multiple bundles (e.g., CST, visual bundle, auditory bundle, etc.), and all trained bundles can be segmented and output in box 140. Therefore, output 140 may include all bundles trained by the system and / or include only selected bundles based on user 78's selection or other suitable selection.

[0043] Method 100 can then end in box 150. Ending in box 150 allows for further analysis of the selected procedure 100 to be used for the procedure to be performed. As discussed above, the selected segmented (i.e., identified) bundles can be used for selected procedures such as tumor resection, implant placement, or other appropriate determinations. Thus, after or within the ending box 150, user 78 can identify specific locations or bundles relative to or near the tumor for tumor resection. Furthermore, user 78 can identify bundles within subject 24 relative to the tumor or other physiological abnormalities that may affect (e.g., move) the bundles, based on generally known or recognized bundles, to aid in implant placement within subject 24, etc.

[0044] In various embodiments, the segmented fiber bundles output from block 140 can be displayed on display device 36, such as an image relative to subject 24 (e.g., a 3D MRI of subject 24), to aid in protocol execution. As discussed above, image 32 can be registered to subject 24. Therefore, the segmented fiber bundles output from block 140 can be superimposed on image 32 to aid in protocol execution on subject 24. Consequently, the icon or graphical representation 94 of instrument 76 can also be displayed relative to the segmented fiber bundles output from block 140. It should be understood that the output segmented fiber bundles can also be displayed as a graphical representation superimposed on an image generated from image data of subject 24 and displayed as image 32.

[0045] Therefore, method 100 can be used to automatically segment a given fiber tract in the brain of, for example, subject 24. The segmented fiber tract can be used to aid in procedures, such as in closing box 150. This procedure can be a navigation procedure for the implantation of leads (e.g., deep brain stimulation leads), tumor resection, etc. Automatic segmentation can be performed by processor module 48 by evaluating image tracing imaging using a invoked classification system.

[0046] Steering Reference Figure 4 And continue to refer to Figure 1 and Figure 2 Method 200 illustrates a procedure for identifying or segmenting neuronal fiber bundles. In addition to the features discussed above with respect to Method 100, Method 200 may include additional features and / or may include steps or procedures similar to or identical to those discussed above. Therefore, similar procedures will be included with similar or identical reference numerals supplemented by superscript symbols ('), and will not be discussed in detail here. It should be understood that the procedures for such identification may be identical to those discussed above, and / or may be similar and include supplementary minor annotations related to the specific method shown in Method 200.

[0047] Initially, method 200 may begin in start box 204, which is similar to start box 104 as discussed above. Method 200 may then proceed to access and / or invoke a diffusion-weighted gradient image in box 108'. The accessed or invoked diffusion-weighted gradient images may be similar to those discussed above in box 108. The diffusion-weighted image may then have a selected tracer image determined in box 112'. This selected image tracer image may be similar to the image tracer image in box 112 as discussed above. Furthermore, the image tracer image may then be output in box 116'. Similarly, the output of the selected image tracer image may be similar to the image output in box 116 as discussed above.

[0048] The process of accessing or calling the output of the diffusion-weighted image in box 108', the image tracing imaging in box 112', and the image tracing imaging in box 116' is similar to the process discussed above. Therefore, the details of the process will not be repeated here and can be referred to above.

[0049] However, process 200 may also include additional and / or alternative processes, including generating noise diffusion tracer imaging based on selected tracer imaging from box 112'. As discussed herein, various additional inputs can be used to generate noise diffusion tracer imaging. Thus, noise diffusion tracer imaging may include and / or may be the result of a selected tracer imaging algorithm that identifies possible bundles due to diffusion weighting information collected in the image data accessed in box 108'.

[0050] Method 200 may also include a second or secondary input path or portion 220. In various embodiments, the sub-process may be an input to the diffusion-weighted gradient image accessed in block 108'. The secondary input may help identify regions of interest (ROIs), such as within the brain, at least within a selected portion (including the image). Input sub-process 220 may include an MR image (which may include a FA image) accessible in block 224. The MR image accessed in block 224 may be an image acquired using the diffusion-weighted gradient image in block 108'. As those skilled in the art will understand, the diffusion-weighted gradient image may be acquired during the process of acquiring the image using an MRI system. Therefore, the MR image may be acquired substantially simultaneously with the diffusion-weighted gradient image accessed in block 108'.

[0051] Subprocess 220 may also include the ability to invoke a trained machine learning system, such as an artificial neural network (ANN), in box 226. This ANN may include a convolutional neural network (ANN, e.g., a CNN). The trained ANN may be trained using appropriate image data, such as a training set of image data of the same region or portion imaged in the MR image accessed in box 224. For example, the ANN may be trained using classified or identified regions (such as ROIs within the accessed MR image). Various ROIs may include anatomical or feature regions or constructs relating to the start and end regions of a known or identified given fiber tract. They may also include regions through which the known tract may or may not pass. Selected ROIs may include the start region of the corticospinal tract in the brainstem (e.g., a portion of the cerebral peduncle) and the end region of the corticospinal tract in the central anterior and postcentral gyri.

[0052] Therefore, a trained ANN trained on previously generated image data can be stored along with appropriate weights for the individual neurons in the ANN to identify ROIs' in new images. For example, the MR image accessed in box 224 could be an MR image of subject 24. Thus, the trained ANN can be used in box 230 to identify ROIs' from the accessed MR image in box 224. The application or identification of ROIs' in the MR image in box 230 can help identify ROIs' of the start and end regions of neuronal bundles. The identified ROIs' can help identify a given or selected neuronal bundle to aid in segmenting the neuronal bundle in method 200. The segmented brain regions or ROIs' can be output in box 240. The output segmented brain regions can include start or end ROIs' such as those identified in the training image data of the CNN, or any suitable brain regions. They can also include regions known to be traversed or not traversed by the bundle. Other suitable brain regions can include selected anatomical or physiological brain regions such as the precentral and postcentral gyri, cerebral peduncles, primary visual cortex, and lateral geniculate nucleus.

[0053] As discussed above, in method 100, the classification system can be used to automatically identify (also referred to as classify) bundles (e.g., bundles of neurons) from the diffusion-weighted gradient image. Therefore, method 200 can also invoke a trained classification system in box 120'. The invoked trained classification system may be similar to the classification system discussed above in box 120 of method 100, but it may have inputs from subprocess 220.

[0054] Subprocess 220 can also be, or alternatively, input to the trained classification system invoked in box 120'. Therefore, subprocess 220 can be used as input to process 200. Subprocess 220 allows segmentation to include selected information about the ROI or the segmentation itself.

[0055] In box 130, the invoked trained classification system can be used to automatically identify or segment image bundles. Automatic identification of image bundles can be similar to the automatic identification of image bundles discussed above in box 130. As shown in method 200, subprocess 220 can also be a separate input to the automatic segmentation in box 130'. Therefore, the segmented region of the output in box 240 can also be, or alternatively, an input to the automatic segmentation in box 130'.

[0056] Segmentation of a given neuronal bundle in box 130' can be performed automatically by executing a trained classification system invoked in box 120'. Identification or segmentation of selected regions in an image (such as the brain) from box 240 can help or provide the start and end points of the neuronal bundle or other appropriate regions in the image. Other image regions can help the classification system identify and classify the neuronal bundle. As discussed above, neuronal bundles can be identified based on various characteristics, and selected start and end regions (or other regions through which the bundle passes or does not pass) can help identify the neuronal bundle as a fiber bundle in the image. However, in box 130', identification and segmentation of selected regions in the image can aid in the identification and segmentation of a given neuronal bundle.

[0057] Therefore, method 200 may include various alternative or additional processes. For example, method 200 may automatically segment the image in box 130' using input from subprocess 220 for identifying ROIs in the image.

[0058] The segmented neuronal bundles can then be output in box 140', similar to the output in box 140 discussed above. The process 200 can then terminate in box 150', also similar to the termination in box 150 above. Therefore, the output of the segmented neuronal bundles can include any appropriate output, and the termination of process 200 can include various additional or separate procedures, as discussed above. However, as discussed above, the method 200, which can also be executed using processor module 48, can include segmentation of a selected image region and the output of the selected image region from the transmission process in box 240, or input 220 for assisting in the automatic identification of neuronal bundles in the accessed diffusion-weighted image from box 108'.

[0059] Steering Reference Figure 5 And continue to refer to Figure 1 and Figure 2 Methods and procedures 300 for segmenting fiber bundles are discussed. Method 300 may include various parts or features similar to those discussed above in methods 100 and 200. Similar boxes will be identified by similar numbers supplemented with double superscript symbols (”), and will not be discussed in detail again here. Therefore, method 300 can be used to segment (including identify) a given neuronal bundle.

[0060] Initially, method 300, similar to methods 100 and 200 discussed above, can be executed by a selected processor (such as processor module 48) via instructions in a selected algorithm or program. Therefore, it should be understood that method 300 can be executed by a workstation or processor system 40. Method 300 may be included in an algorithm that can be executed or implemented by a processor to assist and / or efficiently implement selected procedures, as discussed further herein.

[0061] Therefore, method 300 can begin in start box 304. The start in box 304 can be similar to the start discussed above, such as in box 104. Therefore, various operations can be used to begin process 300. After starting in box 304, the diffusion-weighted gradient image data can be invoked or accessed in box 108". After accessing the invoked diffusion-weighted image, image tracing imaging can be performed in box 112". This image tracing imaging can be similar to the full image tracing imaging discussed above in boxes 112 and 112'. Then, in box 116", the image tracing imaging can be output as an image (e.g., brain) tracing imaging.

[0062] However, process 300 may also include additional and / or alternative processes, including generating noise diffusion tracer imaging based on selected tracer imaging from box 112". As discussed herein, various additional inputs can be used to generate noise diffusion tracer imaging. Therefore, noise diffusion tracer imaging may include and / or may be the result of a selected tracer imaging algorithm that identifies possible bundles due to diffusion weighting information collected in the image data accessed in box 108".

[0063] As discussed above in method 200, a selected sub-method or procedure 220” can be used to help identify various ROIs in the image. Sub-procedure 220” may be similar to the sub-procedures discussed above and will be briefly described here. Therefore, the MR image can be accessed or invoked in box 224”. As discussed above with respect to box 226, a trained ANN can be invoked in box 226”. The trained ANN may be similar to the ANN discussed above in box 226 and can therefore be trained using the selected image to identify various ROIs in the image.

[0064] In box 230", the trained CNN can be applied to evaluate the accessed image. Then, in box 240", the segmented regions, such as selected ROIs including the start and end of selected fiber bundles, can be output. The output segmented ROIs can then be fed into noise diffusion in box 210" to help identify selected regions for classification or to identify specific or given neuronal bundles in the image.

[0065] Method 300 may include invoking a trained machine learning system, such as one or more artificial neural networks (ANNs) (e.g., generative adversarial networks), in box 320. The trained machine learning system may include an ANN trained to segment bundles of given neurons in image tracing imaging. The trained machine learning system may be trained in any suitable manner and includes appropriate weights used to adjust the inputs to an activation function to provide an appropriate output. Generally, weights may be applied to various inputs used to compute the activation function. At appropriate values, the activation function is activated to provide an output that helps determine selected features. Thus, the trained machine learning system may be trained based on selected inputs, such as the inputs to the identified bundles. The identified bundles may be identified by a selected user (e.g., a neurosurgeon) who identifies bundles in the training data. The machine learning system may then be trained based on the training data and then saved for application to additional input data, which may include accessed images in whole or in image tracing imaging, as discussed above. However, it should be understood that various machine learning systems may include semi-supervised or unsupervised methods.

[0066] Therefore, after invoking the trained machine learning system in box 320, automatic segmentation of a given bundle of neural fibers can be performed in box 330. Automatic segmentation in box 330 uses the trained machine learning system to identify specific or given bundles in the image. Similarly, automatic segmentation in box 330 can be performed by processor module 48, which includes executing instructions from the invoked trained ML system from box 320 and other selected inputs.

[0067] Additionally, subprocess 220” may optionally provide input to the trained machine learning system in box 320 for automatic segmentation 330. In other words, the segmented ROI’ output in box 240” may or may not be input to the selected segmentation (automatic segmentation in box 330) from image tracing imaging in box 116”. Therefore, automatic segmentation may or may not be performed with the aid of the segmented ROI from subprocess 220”.

[0068] Following the automatic segmentation in box 320, the automatically segmented fiber bundles of a given neuronal bundle can be output in box 140". This output may be similar to the output discussed above in boxes 140 and 140'. Method 300 can then end in box 150". Similarly, as discussed above, the end of method 300 in box 150" can lead to various additional procedures or processes, such as using the segmented neuronal bundles for selected procedures, such as tumor resection, lead implantation, or other appropriate procedures.

[0069] Various methods, including methods 100, 200, and 300 discussed above, include general features that can be included in the algorithm for segmenting fiber bundles in a selected image dataset. As discussed above, these methods can include various features or processes that can be implemented by a selected system (such as processor module 48) when executing the selected method. Therefore, reference... Figures 3 to 5 And refer to other sources. Figure 6A Algorithm or method 500 is shown to aid in training a system for automatic segmentation, as discussed above. Method 500 may be incorporated into the processes discussed above, including methods 200, 300, and 400, to generate or allow segmentation of a given fiber bundle according to a selected method, as discussed above. In various embodiments, one or more processes of method 500 may be included in methods 100, 200, and 300 discussed above.

[0070] Therefore, in general, method 500 may include accessing a diffusion gradient image in block 510. The gradient diffusion image accessed in block 510 may be similar to those discussed above. The accessed gradient diffusion image may then be processed to determine tracer imaging in the accessed image 510 in block 514. The tracer imaging processing in block 514 may also be similar to the tracer imaging processing discussed above, including various tracer imaging algorithms for identifying bundles or possible bundles within the image accessed in block 510. According to method 500 consistent with methods 200-400 discussed above, selected specific fibers may be identified in the tracer imaging.

[0071] In method 500, various processing can be performed on the tracer imaging, and processed image data can be output and / or accessed after or from the tracer imaging processing in block 514. For example, a Directional Coded Color (DEC) Fractional Anisotropy (FA) mapping can be generated in block 520. The DEC / FA image may include a DEC / FA image identified based on the accessed image data according to generally known techniques. Additionally, the B0 mask image in block 524 and the B0 image in block 528 can be accessed and applied to the tracer imaging processing. The B0 mask and image may include images in which diffusion information is not included. Furthermore, selected image (e.g., brain) tracer imaging, which may include whole-brain tracer imaging (WBT) in block 532, may also be output from the tracer imaging processing.

[0072] In box 536, the tracer imaging output can be optionally registered to one or more reference images. If registered, the registered image can also be transformed using DEC / FA. In box 536, registration to the reference space may include appropriate scaling and spatial transformation. Registration to the reference space allows determination of the location or various features of the identified image tracer imaging.

[0073] Negative bundles can be created or defined in box 540. Negative bundles are those fibers in image tracing imaging that are not part of the neuronal bundles the system is trained to classify / segment. Positive bundles are those fibers that belong to the neuronal bundles the system is trained to classify / segment, and they are created by trained experts (e.g., neuroradiologists). At 540, negative bundles can be generated or created, for example, by filtering bundles that are furthest from positive bundles and most different in shape from them. Various additional processing can be performed, such as enhancing negative and positive bundles, which can then be done in box 546. Enhancement can include random rotation, scaling, translation, and adding Gaussian and other types of noise to the fibers and image features.

[0074] As discussed above, the bundles in the data can then be analyzed and processed to train a selected segmentation method for a given bundle. Segmentation training can be performed using various techniques that can be performed in series or individually. Therefore, training can be conducted along at least two paths together or alternately to train the selected system.

[0075] In various implementations, a classification system can be trained so that various features of each line in box 560 can be computed by following path 564. These features may include various determinations or analyses of various bundles (which may also be referred to as lines) for further segmentation and identification of fiber bundles, as discussed above and further herein. For example, the number of points and the location of the points along the bundle lines may be determined. Vectors of each line segment between points may be determined, such as relative to each other and according to each point in the respective bundle. Fractional anisotropy (FA) values ​​may be determined or computed for each point. Additionally, orientation-encoded color information may be computed or determined for each point. Curvature at each point (such as each point in a neighboring bundle) may also be computed along each bundle in the bundle. In various implementations, atlases such as commonly known brain atlas images or models may also be identified or accessed, and comparisons with the structure of the identified bundles may be made. Appropriate brain atlases may include those included in open-source applications such as Freesurfer or FSL (created by the FMRIB group at the University of Oxford, UK). Additional calculations may include the length of the bundle between the start region or point and / or the end region or point. Therefore, the entire length of a bundle can be measured or calculated for each bundle in the image.

[0076] Inputs can also be made to the Region of Interest (ROI) subprocess 580. The ROI subprocess 580 can be similar to the ROI subprocess 220 discussed above. Therefore, anatomical image data can be accessed in box 584, and various regions, such as start and end regions, anatomical features, or structures, can be identified in box 588. The identification of various features can include applications of machine learning systems incorporating CNNs (such as those discussed above). CNNs can be trained using various data, and trained CNNs can be used to identify structures or regions in the accessed image data (such as the image data accessed in box 584). However, the ROI subprocess 580 can be used to identify ROIs to help determine or segment selected neuronal fibers. Inputs from subprocess 580 can include and / or enhance the computed features in box 560.

[0077] In box 592, subprocess 580 can also access or compare selected fiber bundle maps. The maps may include models of fiber bundles that can be used to analyze or process the bundles in the segmentation of box 514. Therefore, subprocess 580 can utilize selected fiber bundle maps.

[0078] During classification system training, a classification system for training can be invoked or generated in box 610. The training classification system may include appropriate classification systems, such as those discussed above that include random forces or other appropriate classification systems.

[0079] During training or at any selected time, the classification system can be used to classify and segment appropriate or given neuronal fibers in the accessed diffusion gradient image data in box 510. Segmentation can be performed automatically, as discussed above. Therefore, the tracer imaging can be classified using the invoked classification system in box 620. Based on the trained classification system, the classification can be similar to that discussed above.

[0080] After classifying or segmenting fiber bundles, the classified or segmented fiber bundles belonging to a given neuronal bundle can be output in box 630. The output can be any appropriate output, including those discussed above. The output may include a display of the segmented neuronal bundles, a save of the segmented neuronal bundles for additional analysis, or other appropriate output.

[0081] In various implementations, a second or alternative path 634 may be followed. Path 634 involves training a machine learning system for segmentation in box 636. The machine learning system may include an artificial neural network, such as a convolutional neural network. ML training in box 636 may be performed using input from the tracer imaging in box 514 and may include optional reference image registration. In various additional or alternative implementations, training ML may also include computed features from box 560. Thus, training ML in box 636 may include selected inputs, as discussed above. Therefore, training ML may also result in segmentation of fiber bundles in box 620.

[0082] Then, method 500 may output and / or save the trained system in block 634. In various embodiments, the output may be updated or verified (such as by adding training data) and / or the output may be acknowledged or checked as training progresses. Thus, in block 634, the trained system may be saved or stored at a selected time and / or when the system is properly trained.

[0083] Method 500 may end in box 640. The end of process 500 may be similar to the end of processes 200-400 as discussed above. Therefore, the output from box 630 can be used to help identify appropriate fiber bundles within a selected subject (such as subject 24). The segmented fiber bundles can be used to assist in various procedures, such as tumor resection, device implantation, or other appropriate procedures. Therefore, output 630 can be used to assist in the execution of procedures and / or the planning of procedures for selected subjects (such as human subjects for selected procedures).

[0084] Method 500 can be a training process for training a selected system (such as a classification or machine learning system for segmenting fiber bundles). The trained system can then be used at selected times to segment fiber bundles, for example, according to the methods discussed above. Generally, refer to... Figure 6B Method 700 can be used to segment data to segment fiber bundles.

[0085] Segmentation method 700 may be similar to training method 500, and similar or identical boxes will use the same reference numerals supplemented with superscript symbols ('). However, segmentation method 700 may include or utilize a trained system to segment fiber bundles, as discussed above. Thus, method 700 may include accessing a diffusion gradient image in box 510'. The gradient diffusion image accessed in box 510' may be similar to those discussed above. The accessed gradient diffusion image may then be processed to determine tracer imaging in the accessed image 510' in box 514'. The tracer imaging processing in box 514' may also be similar to the tracer imaging processing discussed above, including various tracer imaging algorithms for identifying bundles or possible bundles within the image accessed in box 510'. According to method 700, consistent with methods 200-400 discussed above, selected specific fibers may be identified in the tracer imaging.

[0086] In method 700, various processing can be performed on the tracer imaging, and processed image data can be output and / or accessed after or from the tracer imaging processing in box 514'. For example, a Directional Coded Color (DEC) Fractional Anisotropy (FA) mapping can be generated in box 520'. The DEC / FA image may include a DEC / FA image identified based on the accessed image data according to generally known techniques. Additionally, the B0 mask image in box 524' and the B0 image in box 528' can be accessed and applied to the tracer imaging processing. The B0 mask and image may include images in which diffusion information is not included. Furthermore, selected image (e.g., brain) tracer imaging, which may include whole-brain tracer imaging (WBT) in box 532', may also be output from the tracer imaging processing.

[0087] In box 536', the tracer imaging output can be optionally registered to one or more reference images. If registered, the registered image can also be transformed using DEC / FA. In box 536', registration to the reference space may include appropriate scaling and spatial transformation. Registration to the reference space allows determination of the location or various features of the identified image tracer imaging.

[0088] In various implementations, the classification system can be used to segment fiber bundles, thus allowing the calculation of various features for each line in box 560' by following path 564'. These features may include various determinations or analyses of the various bundles (which may also be referred to as lines) for further segmentation and identification of the fiber bundles, as discussed above and further herein. For example, the number of points along the bundle lines and the location of the points may be determined. Vectors between each line segment may be determined, such as relative to each other and based on each point in the respective bundle. Fractional anisotropy (FA) values ​​may be determined or calculated for each point. Additionally, orientation-coded color information may be calculated or determined for each point. Curvature at each point (such as each point in a neighboring bundle) may also be calculated along each bundle in the bundle. In various implementations, atlases such as commonly known brain atlas images or models may also be identified or accessed, and comparisons with the structure of the identified bundles may be made. Appropriate brain atlases may include those included in open-source applications such as Freesurfer or FSL (created by the FMRIB group at the University of Oxford, UK). Additional calculations may include the length of the bundle between the start region or point and / or the end region or point. Therefore, the entire length of a bundle can be measured or calculated for each bundle in the image.

[0089] Inputs can also be made to the Region of Interest (ROI) sub-procedure 580'. The ROI sub-procedure 580' can be similar to the ROI sub-procedure 220 discussed above. Therefore, anatomical image data can be accessed in box 584', and various regions, such as start and end regions, anatomical features, or structures, can be identified in box 588'. The identification of various features can include applications of machine learning systems incorporating CNNs (such as those discussed above). CNNs can be trained using various data, and trained CNNs can be used to identify structures or regions in the accessed image data (such as the image data accessed in box 584'). However, the ROI sub-procedure 580' can be used to identify ROIs to help determine or segment selected neuronal fibers. Inputs from sub-procedure 580' can include and / or enhance the computed features in box 560'.

[0090] In box 592', subprocess 580' can also access or compare selected fiber bundle maps. Maps may include models of fiber bundles that can be used to analyze or process the bundles in the segmentation of box 514'. Therefore, subprocess 580' can utilize selected fiber bundle maps.

[0091] In the classification system, a trained classification system can be invoked in box 610'. The trained classification system may include appropriate classification systems, such as those discussed above that include random forces or other appropriate classification systems.

[0092] The trained classification system can be used to classify and segment appropriate or given neuronal fibers in the accessed diffusion gradient image data in box 510. Segmentation can be performed automatically, as discussed above. Therefore, the tracer imaging can be classified in box 620' using the invoked classification system. Based on the trained classification system, the classification can be similar to that discussed above.

[0093] After classifying or segmenting fiber bundles, the classified or segmented fiber bundles belonging to a given neuronal bundle can be output in box 630'. The output can be any appropriate output, including those discussed above. The output may include a display of the segmented neuronal bundles, a save of the segmented neuronal bundles for additional analysis, or other appropriate output.

[0094] In various implementations, a second or alternative path 634' may be followed. Path 634' relates to a trained machine learning system for segmentation in box 636'. The machine learning system may include artificial neural networks, such as convolutional neural networks. ML training in box 636' may be performed using input from tracer imaging in box 514' and may include optional reference image registration. In various additional or alternative implementations, training ML may also include computed features from box 560'. Thus, training ML in box 636' may include selected inputs, as discussed above. Therefore, training ML may also result in segmentation of fiber bundles in box 620'.

[0095] Method 700 may also provide outputs in box 630'. Outputs from box 630' can be used to help identify appropriate fiber bundles within a selected subject (such as subject 24). The segmented fiber bundles can be used to assist in various procedures, such as tumor resection, device implantation, or other appropriate procedures. In various embodiments, the instrument can be navigated relative to the fiber bundle. Additionally, ablation planning and tracking can be performed. Therefore, output 630 can be used to assist in the execution of procedures and / or the planning of procedures for a selected subject (such as a human subject for a selected procedure). Process 700 may then also terminate in box 640', similar to the termination of processes 200-400 as discussed above.

[0096] In light of the above, the selected system can automatically segment appropriate or given fiber bundles in the accessed diffusion gradient image data. In various embodiments, the data can be data related to nerves or neural plexuses (e.g., brain data), and therefore, a bundle can also be referred to as a fiber neuron bundle in a brain image. Automatic segmentation can be performed by a selected processor module to identify fiber bundles within a selected subject. Identifiable fiber bundles can include any appropriate fiber bundle, such as the optic radiation, arcuate fasciculus (AF), superior longitudinal fasciculus (SLF), corticospinal tract (CST), frontooblique fasciculus (FAT), fornix tract, dentate-rubracothalamic tract (DRT), or any other appropriate bundle. However, automatic segmentation can allow for minimal or no user intervention and identify relevant fiber bundles within the subject.

[0097] Exemplary embodiments are provided to make this disclosure thorough and to fully communicate the scope of this disclosure to those skilled in the art. Numerous specific details, such as examples of particular components, apparatus, and methods, are set forth to provide a thorough understanding of embodiments of this disclosure. It will be apparent to those skilled in the art that specific details are not required, that exemplary embodiments may be embodied in many different forms, and should not be construed as limiting the scope of this disclosure. In some exemplary embodiments, well-known processes, well-known apparatus structures, and well-known techniques are not described in detail.

[0098] In one or more examples, the described techniques may be implemented in hardware, software, firmware, or any combination thereof. If implemented in software, the functionality may be stored as one or more instructions or code on a computer-readable medium and executed by a hardware-based processing unit. Computer-readable media may include non-transitory computer-readable media, which correspond to tangible media such as data storage media (e.g., RAM, ROM, EEPROM, flash memory, or any other medium that can be used to store desired program code in the form of instructions or data structures and is accessible by a computer).

[0099] The instructions can be executed by one or more processors, such as one or more digital signal processors (DSPs), general-purpose microprocessors, graphics processing units (GPUs), application-specific integrated circuits (ASICs), field-programmable arrays (FPGAs), or other equivalent integrated or discrete logic circuits. Therefore, the term "processor" as used herein can refer to any of the foregoing structures or any other physical structures suitable for implementing the described techniques. Furthermore, this technique can be fully implemented in one or more circuit or logic elements.

[0100] The foregoing description of embodiments has been provided for illustrative and descriptive purposes. The foregoing description is not intended to be exhaustive or limiting of this disclosure. Individual elements or features of a particular embodiment are generally not limited to that particular embodiment, but are interchangeable and may also be used in selected embodiments where applicable, even if not specifically shown or described. The same element or feature may be varied in many ways. Such variations should not be considered as departing from this disclosure, and all such modifications are intended to be included within the scope of this disclosure.

[0101] It should be understood that the various aspects disclosed herein can be combined in combinations different from those specifically given in the specification and drawings. It should also be understood that, depending on the example, certain actions or events of any process or method described herein may be performed in a different order, or may be completely added, combined, or omitted (e.g., performing the described technique may not require all the described actions or events). Furthermore, although for clarity some aspects of this disclosure are described as being performed by a single module or unit, it should be understood that the techniques of this disclosure can be performed by combinations of units or modules associated with, for example, a medical device.

Claims

1. A method for segmenting and identifying a given bundle of fiber neurons in an image that is associated with certain anatomical functions or features, the method comprising: Access selected data including diffusion-weighted gradient images; Evaluate the selected data accessed to determine image tracing imaging that includes multiple beams; At least a plurality of bundles are evaluated using a trained machine learning algorithm, which is trained to segment and identify bundles for a given fiber neuron bundle in a determined image tracing imaging. Based at least on the evaluation of at least the sub-bundles of the plurality of bundles, determine whether at least one of the evaluated sub-bundles is a given fiber neuron bundle associated with certain anatomical functions or features and identify the given fiber neuron bundle; as well as When at least one of the evaluated bundles is identified as the given fiber neuron bundle, at least one of the bundles is output as the given fiber neuron bundle.

2. The method of claim 1, wherein evaluating the accessed selected data based on a first criterion to determine the image tracing imaging comprising the plurality of beams comprises: Determine the anisotropy of water within the selected data; as well as The bundle of images passing through the selected data is determined based on the determined anisotropy.

3. The method of claim 1, wherein evaluating the plurality of sub-bundles comprises at least one of the following: Determine the points along each of the sub-bundles; Assess the score anisotropy at each of the identified points; Evaluate the diffusion-coded color at each of the identified points; Determine the distance from the start region of each bundle to the end region of each bundle; as well as Determine the curvature of the bundle at each of the determined points.

4. The method according to claim 1, further comprising: The selected region of interest is determined relative to the selected data accessed by the invoked trained convolutional neural network.

5. The method according to claim 4, further comprising: The evaluation of selected data accessed to determine image tracing imaging includes evaluating the entire image to determine multiple beams.

6. The method according to claim 5, further comprising: Invoke the trained convolutional neural network; Access anatomical image data; as well as The trained convolutional neural network invoked is used to determine at least one region of interest in the accessed image data.

7. The method of claim 6, wherein determining the selected region of interest relative to the accessed selected data using the invoked trained convolutional neural network comprises: Identify at least one region of interest in the selected data accessed, specifically the image data accessed.

8. The method of claim 7, wherein the at least one region of interest comprises: At least one of the start region or the end region of the given fiber neuron bundle.

9. The method according to claim 7, further comprising: Identify at least one of the sub-bundles that interacts with at least one identified region of interest; The evaluation of at least one of the plurality of bundles includes evaluating only the determined at least one bundle.

10. The method of claim 4, further comprising: Invoke the trained convolutional neural network; Access anatomical image data; as well as The trained convolutional neural network invoked is used to determine at least one region of interest in the accessed image data.

11. The method of claim 10, wherein determining the selected region of interest relative to the accessed selected data using the invoked trained convolutional neural network comprises: Identify at least one region of interest in the selected data accessed, specifically the image data accessed.

12. The method of claim 11, wherein evaluating at least one of the sub-bundles among the plurality of bundles using the trained machine learning algorithm comprises: The sub-bundles are evaluated based on at least one region of interest identified in the accessed image data.

13. The method according to any one of claims 1 to 12, wherein the given fiber neuron bundle is at least one of the corticospinal tract, optic tract, fronto-oblique tract, dentate-red thalamic tract, fornix tract, or a combination thereof.

14. The method according to any one of claims 1 to 12, further comprising: A machine learning algorithm for identifying the given fiber neuron bundle is trained using training data including at least those identified therein to generate a trained machine learning algorithm.

15. A system configured to segment and identify a given bundle of fiber neurons in an image of a subject that is associated with certain anatomical functions or features, the system comprising: Processor system, the processor system being configured to execute instructions to: Access selected data from the subject, including diffusion-weighted gradient images; The selected data accessed are evaluated based on a first criterion to determine image tracing imaging comprising multiple beams; At least a plurality of bundles are evaluated using a trained machine learning algorithm, which is trained to segment and identify bundles for a given fiber neuron bundle in a determined image tracing imaging. Based at least on the evaluation of at least the sub-bundles of the plurality of bundles, determine whether at least one of the evaluated sub-bundles is a given fiber neuron bundle associated with certain anatomical functions or features and identify the given fiber neuron bundle; as well as When at least one of the evaluated bundles is identified as the given fiber neuron bundle, at least one of the bundles is output as the given fiber neuron bundle.

16. The system of claim 15, wherein the processor system is configured to execute additional instructions to determine a selected region of interest relative to the accessed selected data using the invoked trained convolutional neural network; The determination of whether at least one of the evaluated sub-bundles is a given fiber neuron bundle is further based on the determination of the selected region of interest.

17. The system of claim 16, further comprising: A memory system configured to store the instructions.

18. The system of claim 15, further comprising: A display device configured to display an image of the subject and a given bundle of output neurons.

19. The system of claim 15, further comprising: An imaging system configured to acquire a diffusion-weighted gradient image of the subject.

20. The system of claim 15, further comprising: The navigation system includes a tracking system and a tracking device; The instrument is operable to navigate relative to a given output neuron bundle within the navigation system.

Images