REMOVAL OF FALSE POSITIVES FROM WHITE MATTER FIBER TRACTS

Abstract

The invention provides for a medical imaging system (100, 400), comprising: The execution of the machine executable instructions (112) causes a processor (104) to: receive (200) a set of input white matter fiber tracts (118): receive (202) the label from a discriminator neural network (116) in response to inputting the set of input w hue matter fiber tracts, generate (204) an optimized feature vector (122) using the set of input white matter fiber tracts and a generator neural network ((114) if the label indicates anatomically incorrect; receive (206) the set of generated white matter fiber tracts from the generator neural network in response to inputting the optimized feature vector, and construct (208) a false positive subset (126) of the set of input white matter fiber tracts using the generated set of white matter fiber tracts.

Claims

1. A medical imaging system comprising: a memory storing machine executable instructions and a generative adversarial neural network, wherein the generative adversarial neural network comprises a generator neural network and a discriminator neural network, wherein the generator neural network is configured for outputting a set of generated white matter fiber tracts in response to inputting a feature vector, wherein the discriminator neural network is configured for outputting a label in response to inputting a set of input white matter fiber tracts, wherein the label indicates the set of input white matter fiber tracts as anatomically correct or anatomically incorrect; a processor for controlling the medical imaging system, wherein execution of the machine executable instructions further causes the processor to: construct the set of input white matter fiber tracts using a diffusion tensor image of a region of interest descriptive of at least part of a brain according to a fiber tractography algorithm; receive the label from the discriminator neural network in response to inputting the set of input white matter fiber tracts; generate an optimized feature vector using the set of input white matter fiber tracts and the generator neural network if the label indicates anatomically incorrect, wherein the optimized feature vector is calculated using any one of the following: using a search algorithm to iteratively modify elements of the optimized feature vector, and backpropagation through the generator neural network such that the set of generated white matter fiber tracts match the set of input white matter tracts as closely as possible; receive the set of generated white matter fiber tracts from the generator neural network in response to inputting the optimized feature vector; construct a false positive subset of the set of input white matter fiber tracts using the generated set of white matter fiber tracts; and provide a set of corrected fiber tracts, by removing the false positive subset from the set of input white matter fiber tracts.

2. The medical imaging system of claim 1, wherein execution of the machine executable instructions further causes the processor to: render the set of input fiber tracts on a display; and indicate the false positive subset on the display and/or indicate an anatomically correct subset on the display using the false positive subset.

3. The medical imaging system claim 1, wherein the set of corrected fiber tracts is provided in response to receiving a signal from a user interface.

4. The medical imaging system of claim 1, wherein execution of the machine executable instructions further causes the processor to: receive the label from the discriminator neural network in response to inputting the set of corrected white matter fiber tracts; generate the optimized feature vector using the set of corrected white matter fiber tracts and the generator neural network if the label for the corrected white matter fiber tracts indicates anatomically incorrect; receive the set of generated white matter fiber tracts from the generator neural network in response to reinputting the optimized feature vector for the set of corrected white matter fiber tracts; and divide the set of white matter fiber tracts into a false positive subset and an anatomically correct subset using the generated set of white matter fiber tracts the set of corrected white matter fiber tracts.

5. The medical imaging system of claim 1, wherein the false positive subset of the set of input white matter fiber tracts is constructed by comparing each of the set of input white matter fiber tracts to the set of generated white matter fiber tracts.

6. The medical imaging system of claim 5, wherein each of the set of input white matter fiber tracts has end points and a path, wherein each of the set of generated white matter fiber tracts has end points and a path, wherein comparing each of the set of input white matter fiber tracts to the set of generated white matter fiber tracts comprises at least one of the following: comparing the end points of the set of input white matter fiber tracts to the end points of the set of generated white matter fiber tracts; and comparing the path of the set of input white matter fiber tracts to the path of the set of generated white matter fiber tracts.

7. The medical imaging system of claim 6, wherein comparing the end points of the set of input white matter fiber tracts to the end points of the set of generated white matter fiber tracts comprises any one of the following: generate predetermined endpoint volumes surrounding endpoints of one of the set of generated white matter fiber tracts, and test if endpoints of the each of the set of input white matter fiber tracts are both within the predetermined endpoint volumes of the one of the set of generated white matter fiber tracts; and calculate an endpoint distance between endpoints of the one of the set of generated white matter fiber tracts and the each of the set of input white matter fiber tracts, input the endpoint distance into an endpoint measure function to at least partially provide the comparison.

8. The medical imaging system of claim 6, wherein comparing the path of the set of input white matter fiber tracts to the path of the set of generated white matter fiber tracts comprises any one of the following: generate predetermined path volume surrounding paths of one of the set of generated white matter fiber tracts, and test if the path of the each of the set of input white matter fiber tracts are within the predetermined path volume of the one of the set of generated white matter fiber tracts; and calculate input path distances between the path of the one of the set of generated white matter fiber tracts and the path of the set of input white matter fiber tracts, input the input path distances into a path distance measure function to at least partially provide the comparison.

9. The medical imaging system of claim 1, wherein execution of the machine executable instructions further causes the processor to reconstruct the diffusion tensor image for the region of interest using diffusion weighted magnetic resonance imaging data.

10. The medical imaging system of claim 9, wherein the medical imaging system further comprises a magnetic resonance imaging system configured for acquiring the diffusion weighted magnetic resonance imaging data from an imaging zone wherein the memory further comprises pulse sequence commands configured to control the magnetic resonance imaging system to acquire the diffusion weighted magnetic resonance imaging data for the region of interest according to a diffusion weighted magnetic resonance imaging protocol, wherein execution of the machine executable instructions further cause the processor to acquire (500) the diffusion weighted magnetic resonance imaging data by controlling the magnetic resonance imaging system with the pulse sequence commands.

11. The medical imaging system of claim 1, wherein execution of the machine executable instructions further causes the processor to: initialize the generator neural network and the discriminator neural network; receive training data, wherein the training data comprises sets of ground truth white matter fiber tracts; and training the discriminator neural network and the generator neural network according to a generative adversarial neural network training algorithm, wherein after training the generator neural network is configured for configured for outputting a set of generated white matter fiber tracts in response to inputting a feature vector, wherein after training the discriminator neural network is configured for outputting a label in response to inputting a set of input white matter fiber tracts, wherein the label indicates the set of input white matter fiber tracts as anatomically correct or anatomically incorrect.

12. The medical imaging system of claim 1, wherein the generative adversarial neural network comprising the generator neural network and the discriminator neural network is trained according to the method of claim 13.

13. (canceled)

14. A method of operating a medical imaging system using a generative adversarial neural network, wherein the generative adversarial neural network comprises a generator neural network and a discriminator neural network, wherein the generator neural network is configured for outputting a set of generated white matter fiber tracts in response to inputting a feature vector, wherein the discriminator neural network is configured for outputting a label in response to inputting a set of input white matter fiber tracts, wherein the label indicates the set of input white matter fiber tracts as anatomically correct or anatomically incorrect, wherein the method comprises: construct the set of input white matter fiber tracts using a diffusion tensor image a region of interest descriptive of at least part of a brain according to a fiber tractography algorithm; receiving label from the discriminator neural network in response to inputting the set of input white matter fiber tracts; generating an optimized feature vector using the set of input white matter fiber tracts and the generator neural network if the label indicates anatomically incorrect, werein the optimized feature vector is calculated using any one of the following: using a search algorithm to iteratively modify elements of the optimized feature vector; and backpropagation through the generator neural network receiving the set of generated white matter fiber tracts from the generator neural network in response to inputting the optimized feature vector; constructing false positive subset of the set of input white matter fiber tracts using the generated set of white matter fiber tracts; provide a set of corrected fiber tracts by removing the false positive subset from the set of input white matter fiber tracts.

15. A computer program product comprising machine executable instructions for execution by a processor controlling a medical imaging system, wherein the computer program product further comprises a generative adversarial neural network, wherein the generative adversarial neural network comprises a generator neural network and a discriminator neural network, wherein the generator neural network is configured for outputting a set of generated white matter fiber tracts in response to inputting a feature vector, wherein the discriminator neural network is configured for outputting a label in response to inputting a set of input white matter fiber tracts, wherein the label indicates the set of input white matter fiber tracts as anatomically correct or anatomically incorrect, wherein execution of the machine executable instructions further causes the processor to: construct the set of input white matter fiber tracts using a diffusion tensor image of a region of interest descriptive of at least part of a brain according to a fiber tractography algorithm; receive the label from the discriminator neural network in response to inputting the set of input white matter fiber tracts; generate an optimized feature vector using the set of input white matter fiber tracts and the generator neural network if the label indicates anatomically incorrect, werein the optimized feature vector is calculated using any one of the following: using a search algorithm to iteratively modify elements of the optimized feature vector; and backpropagation through the generator neural network such that the set of generated white matter fiber tracts match the set of input white matter tracts as closely as possible; receive the set of generated white matter fiber tracts from the generator neural network in response to inputting the optimized feature vector; construct a false positive subset of the set of input white matter fiber tracts using the generated set of white matter fiber tracts; and provide a set of corrected fiber tracts by removing the false positive subset from the set of input white matter fiber tracts.

Description

BRIEF DESCRIPTION OF THE DRAWINGS

[0060] In the following preferred embodiments of the invention will be described, by way of example only, and with reference to the drawings in which:

[0061] FIG. 1 illustrates an example of a medical imaging system;

[0062] FIG. 2 shows a flow chart which illustrates a method of operating the medical imaging system of FIG. 1;

[0063] FIG. 3 shows a flow chart which illustrates a further method of operating the medical imaging system of FIG. 1;

[0064] FIG. 4 illustrates a further example of a medical imaging system;

[0065] FIG. 5 shows a flow chart which illustrates a method of operating the medical imaging system of FIG. 4;

[0066] FIG. 6 illustrates a method of comparing white matter fiber tracts; and

[0067] FIG. 7 illustrates a further method of comparing white matter fiber tracts.

DETAILED DESCRIPTION OF THE EMBODIMENTS

[0068] Like numbered elements in these figures are either equivalent elements or perform the same function. Elements which have been discussed previously will not necessarily be discussed in later figures if the function is equivalent.

[0069] FIG. 1 illustrates an example of a medical imaging system 100. In this example the medical imaging system comprises a computer 102. The computer 102 may be one or more networked computers. The computer 102 comprises a processor 104. The processor 104 is intended to be representative and may represent one or more computing cores as well as one or more processors 104 that are distributed within different computer systems. The processor 104 is shown as being connected to a hardware interface 106 that may enable the processor 104 to communicate with other computer systems as well as control various components of the medical imaging system 100. The processor 104 is further shown as being connected to a user interface 108 which may be a display and/or graphical user interface. The processor 104 is also shown as being in communication with a memory 110. The memory 110 may be any type of memory from which the processor 104 can access. The memory 110 may be any combination of memory which is accessible to the processor 104. This may include such things as main memory, cached memory, and also non-volatile memory such as flash RAM, hard drives, or other storage devices. In some examples the memory 110 may be considered to be a non-transitory computer-readable medium.

[0070] The memory 110 is shown as containing a set of machine-executable instructions 112 which enable the processor 104 to control other components of the medical imaging system 100 as well as to perform various data processing and analysis tasks. The memory 110 is further shown as containing a generator neural network 114 and a discriminator neural network 116. The generator neural network 114 and the discriminator neural network 116 form a generative adversarial neural network that may have been trained together. The memory 110 is further shown as containing a set of input white matter fiber tracts 118 that may have been received via data analysis or from a different medical imaging or computer system.

[0071] The memory 110 is further shown as comprising a label 120 that was received in response to inputting the set of input white matter fiber tracts into the discriminator neural network 116. The discriminator neural network is configured or trained for outputting a label in response to inputting the set of input white matter fiber tracts. The label indicates whether the set of input white matter fiber tracts are anatomically correct or anatomically incorrect. Anatomically incorrect may be interpreted as comprising false positive white matter fiber tracts.

[0072] The memory 110 is further shown as containing an optimized feature vector 122. The optimized feature vector 122 may be constructed by finding a feature vector that when input into the generator neural network 114 produces a set of generated white matter tracts 124 that closely match the set of input white matter tracts 118. For example, an efficient way of doing this would be to use back propagation and start with the set of generated white matter tracts 124 holding the various nodes and factors within the generator neural network 114 constant and then calculating using the back propagation the optimized feature vector 122.

[0073] The memory 110 is further shown as containing a false positive subset 126. The false positive subset is a subset of the set of input white matter tracts 118 that has been identified as being likely false positives. This may be accomplished by for example comparing the set of generated white matter tracts 124 to the set of input white matter tracts 118. Tracts which do not match a starting and ending location as well as a path of the set of generated white matter tracts may be identified as being false positives. The memory 110 is shown as further containing a set of corrected white matter tracts 128 that was formed by removing the false positive subset 126 from the set of input white matter tracts 118.

[0074] FIG. 2 shows a flowchart which illustrates a method of operating the medical imaging system 100 of FIG. 1. First in step 200 the set of input white matter tracts 118 are received. Next in step 202 the label 120 is received from the discriminator neural network 116 in response to inputting the set of input white matter fiber tracts 118. Then in step 204 the optimized feature vector 122 is generated using the set of input white matter fiber tracts 118 and the generator neural network 114 if the label indicates anatomically incorrect. This for example may be performed by performing back propagation and using the input white matter tracts 118 as the output.

[0075] Next in step 206 the set of generated white matter tracts 124 is received from the generator neural network 114 in response to inputting the optimized feature vector 122. Then in step 208 a false positive subset 126 of the set of input white matter tracts 118 is constructed using the generated set of white matter fiber tracts 124 and the set of input white matter tracts 118. Then in step 210 the set of input fiber tracts are optionally rendered on a display. The user interface 108 may for example comprise a display. In step 212 the false positive subset 126 is indicated on the display and/or an anatomically correct subset is indicated on the display. This may be performed by using the false positive subset 126. For example, the false positive subset may be given a different color or highlighting than other fiber tracts. Finally, in step 214, the set of corrected white matter fiber tracts 128 may be provided by removing the false positive subset 126 from the set of input white matter tracts 118.

[0076] FIG. 3 shows another flowchart which illustrates an alternative method of operating the medical imaging system 100 of FIG. 1. The method in FIG. 3 is similar to the method illustrated in FIG. 2 with the addition of several additional steps. In FIG. 3 the method starts with steps 200 and 202 as was illustrated in FIG. 2. After step 202 the method then proceeds to step 300 which is a question box. In this box the question is ‘is the label indicating anatomically incorrect’. If the answer is no then the method proceeds to step 302 which is to render on the display the set of input fiber tracts and the method ends.

[0077] When the method reaches step 302 there is no need to proceed further because the set of input white matter fiber tracts do not have the anatomically incorrect label, which is to say the discriminator neural network did not identify any tracts as being abnormal or false positive. If the answer to the question in box 300 is yes, then the method proceeds to steps 204-212 as is illustrated in FIG. 2. The method then proceeds to question box 304. The question in box 304 is ‘should the false positive step be removed’. If the answer is no then the method proceeds to step 302 again and the method ends. If the answer is yes then the method proceeds to step 214 and the set of corrected fiber tracts is provided by removing the false positive subset from the set of input white matter tracts.

[0078] Then the method proceeds to step 306. In step 306 the set of corrected fiber tracts is then set to the set of input white matter tracts and the method then proceeds back to step 200. The method if therefore iterative. The method proceeds until either the discriminator neural network recognizes all of the white matter fiber tracts as being anatomically correct or until in step 304 a decision is made not to remove any more false positive white matter fiber tracts.

[0079] FIG. 4 illustrates a further example of a medical imaging system 400. In this example the medical imaging system further comprises a magnetic resonance imaging system 402. The magnetic resonance imaging system 402 comprises a magnet 404. The magnet 404 is a superconducting cylindrical type magnet with a bore 406 through it. The use of different types of magnets is also possible; for instance it is also possible to use both a split cylindrical magnet and a so called open magnet. A split cylindrical magnet is similar to a standard cylindrical magnet, except that the cryostat has been split into two sections to allow access to the iso-plane of the magnet, such magnets may for instance be used in conjunction with charged particle beam therapy. An open magnet has two magnet sections, one above the other with a space in-between that is large enough to receive a subject: the arrangement of the two sections area similar to that of a Helmholtz coil. Open magnets are popular, because the subject is less confined. Inside the cryostat of the cylindrical magnet there is a collection of superconducting coils. Within the bore 406 of the cylindrical magnet 404 there is an imaging zone 108 where the magnetic field is strong and uniform enough to perform magnetic resonance imaging. A region of interest 409 is shown within the imaging zone 408. The magnetic resonance data that is acquired typically acquried for the region of interest. A subject 418 is shown as being supported by a subject support 420 such that at least a portion of the subject 418 is within the imaging zone 408 and the region of interest 409.

[0080] Within the bore 406 of the magnet there is also a set of magnetic field gradient coils 410 which is used for acquisition of preliminary magnetic resonance data to spatially encode magnetic spins within the imaging zone 408 of the magnet 404. The magnetic field gradient coils 410 connected to a magnetic field gradient coil power supply 412. The magnetic field gradient coils 410 are intended to be representative. Typically magnetic field gradient coils 410 contain three separate sets of coils for spatially encoding in three orthogonal spatial directions. A magnetic field gradient power supply supplies current to the magnetic field gradient coils. The current supplied to the magnetic field gradient coils 410 is controlled as a function of time and may be ramped or pulsed.

[0081] Adjacent to the imaging zone 408 is a radio-frequency coil 414 for manipulating the orientations of magnetic spins within the imaging zone 408 and for receiving radio transmissions from spins also within the imaging zone 408. The radio frequency antenna may contain multiple coil elements. The radio frequency antenna may also be referred to as a channel or antenna. The radio-frequency coil 414 is connected to a radio frequency transceiver 416. The radio-frequency coil 414 and radio frequency transceiver 416 may be replaced by separate transmit and receive coils and a separate transmitter and receiver. It is understood that the radio-frequency coil 414 and the radio frequency transceiver 416 are representative. The radio-frequency coil 414 is intended to also represent a dedicated transmit antenna and a dedicated receive antenna. Likewise the transceiver 116 may also represent a separate transmitter and receivers. The radio-frequency coil 414 may also have multiple receive/transmit elements and the radio frequency transceiver 416 may have multiple receive/transmit channels. For example if a parallel imaging technique such as SENSE is performed, the radio-frequency could 414 will have multiple coil elements.

[0082] In this example the subject, 118 is positioned such that the subject's head region is within the region of interest 109 for imaging the brain.

[0083] The transceiver 416 and the gradient controller 412 are shown as being connected to the hardware interface 106 of the computer system 101.

[0084] The memory 110 is further shown as containing pulse sequence commands that are configured to control the magnetic resonance imaging system 402 to acquire diffusion weighted magnetic resonance imaging data for the region of interest 409 according to a diffusion weighted magnetic resonance imaging protocol. The memory is further shown as containing diffusion weighted magnetic resonance imaging data 432 that was acquired by controlling the magnetic resonance imaging system 402 with the pulse sequence commands 430. The diffusion weighted magnetic resonance imaging data 432 was acquired at a variety of different directions. The memory 110 further shows a diffusion tensor image 434 that was reconstructed from the diffusion weighted magnetic resonance imaging data 432. The memory 110 is shown as further containing a fiber tractography algorithm 436 that takes the diffusion tensor image 434 as input and then outputs the set of input white matter tracts 118.

[0085] FIG. 5 shows a flowchart which illustrates a method of operating the medical imaging system 400 of FIG. 4. The method in FIG. 5 is similar to the method illustrated in FIG. 2. The method starts with step 500 which is to acquire the diffusion weighted magnetic resonance imaging data 432 by controlling the magnetic resonance imaging system 402 with the pulse sequence commands 430. Next in step 502, the diffusion tensor image 434 is reconstructed from the diffusion weighted magnetic resonance imaging data 432. Next in step 504 the set of input white matter fiber tracts 118 is constructed from the diffusion tensor image 434 using the fiber tractography algorithm 434.

[0086] FIGS. 6 and 7 illustrate two alternative ways of determining if an input white matter fiber tract is a false positive or not. First in FIG. 6 there is a generated white matter fiber tract 600. Also shown are two input white matter fiber tracts 602 and 604. Each of the white matter fiber tracts has an end point 606. Around the generated white matter fiber tract 600 are predetermined end point volumes 608 which are drawn about the end points 606. There is also a predetermined path volume 610 drawn about the path of the generated white matter fiber tract 600. It can be seen that the end points of the input white matter fiber tract 602 are both within the two different predetermined end point volumes 608. The path of the input white matter fiber tract 602 is also completely within the predetermined path volume 610. The input white matter fiber tract 602 is then determined to be in agreement with the generated white matter fiber tract 600. The other input white matter fiber tract 604 is however completely outside of the predetermined end point volume 608 and the predetermined path volume 610. The input white matter fiber tract 604 is therefore a false positive 612 and would be assigned to the false positive subset 126.

[0087] FIG. 7 shows an alternative to the scheme in FIG. 6. In this example the fiber tracts 600 and 604 are illustrated again. Instead of drawing the volumes 608 and 610 it is illustrated that there is an end point distance 700 between the end points 606. A number of path distances 702 are also indicated. They measure the distance between the path of the generated white matter fiber tract 600 and the input white matter fiber tract 604 periodically. The end point distance 700 and the various path distances 702 can be put into a mathematical measure to provide a measure of how closely the input white matter fiber tract 604 matches the generated white matter fiber tract 600. The value of this measure can then be discriminated to determine if the input white matter fiber tract 604 is a false positive or not.

[0088] Neuronal fiber tract networks (sets of white matter fiber tracts) in the brain can reveal important clinical information for studying neurological diseases via non-invasive diffusion magnetic resonance imaging (dMRI) technology. Fiber tractography is an algorithmic process for reconstructing structural fiber tracts from dMRI, in vivo. However, the anatomical accuracy of fiber tractography is difficult to measure quantitatively without histological ground truths. In particular, studies using phantom or ex vivo data have shown a prevalence for false positive fiber tracts, which can negatively impact clinical studies used to compare detailed anatomies of disease populations in search of biomarkers. This invention is a deep learning-based framework to detect, visualize, and remove false positive fiber tracts from tractography networks to provide confidence in anatomical accuracy of fiber tractography results for later use in clinical studies.

[0089] Neuronal fiber tract networks in the brain can reveal important clinical information for studying neurological diseases via non-invasive diffusion magnetic resonance imaging (dMRI) technology. Fiber tractography is an algorithmic process for reconstructing structural fiber tracts from dMRI, in vivo.

[0090] The anatomical accuracy of fiber tractography is difficult to measure quantitatively without histological data of the true brain anatomy. Studies using phantom or ex vivo data have shown a prevalence for false positive fiber tracts, which are fiber tracts that are reconstructed but which are not anatomically correct, ie connections that should not exist. (False negatives are alternatively defined as fiber tracts that exist in the brain for which the tractography algorithm did not reconstruct. These are more difficult to determine because a lack of a connection could be the reason for a particular disease and therefore the algorithm is correct to not reconstruct it.) False positive fiber tracts can negatively impact clinical studies that are used to compare detailed anatomies of disease populations, creating false scientific findings. Current tractography algorithms are unable to effectively detect and remove false positives because there is no ground truth to correctly determine if it is a false positive.

[0091] Examples may provide a deep learning-based framework to detect, visualize, and remove false positive fiber tracts from tractography networks to provide confidence in anatomical accuracy of fiber tractography results for later use in clinical studies. There are three main elements of this invention.

[0092] The first element is a generative adversarial network (GAN) which is used to learn a distribution of normal anatomical network of fiber tracts for healthy subjects. Within the GAN, a generator network learns a representation for normal fiber tract networks and is used to generate synthetic fiber tract networks from this representation. In parallel a discriminator network is used to classify real and synthetically generated data as “real” (anatomically correct) or “fake” (anatomically incorrect), which updates the accuracy of the generator's representation and synthetic generation. Then, given a reconstructed fiber tract network from a new subject, the discriminator can determine if it is “real” or “fake”, ie anatomically correct or incorrect.

[0093] The second element is an explanation system to detect and visualize which fibers in the network are responsible for the “fake” classification, ie which fibers are false positives. This is done using a comparison algorithm which navigates the latent representation space (learned by the generator) to generate a similar or close network of fiber tracts with the “real” label. Then by subtraction we can display the differences between the two networks, revealing the set of false positive fiber tracts.

[0094] Finally, the third element is a system to remove false positive fiber tracts and provide confidence for the decision. Given a threshold of the difference between the two fiber tract networks, we can remove the false positive fiber tracts and provide additional confidence, by testing this new fiber tract network against the discriminator to classify again if it is “real” or “fake”. If the classification is “fake” again, a different removal threshold is used to remove more or fiber tracts or, the process of finding a similar synthetic fiber tract network in step 2 is repeated.

[0095] Some examples are constructed by firstly building or training the GAN. While GANs have been widely used for image data, GANs have not been attempted for fiber tract data. Fiber tract data is represented as a set of 3D coordinates, listed sequentially for each fiber tract. The GAN will be trained on this unique data type to learn a latent representation for the fiber tract network and an accompanying discriminator.

[0096] Then, given a new subject, the input for the GAN will be a fiber tract network represented by a sequential list of 3D coordinates for each tract and the output of the trained GAN will be a binary class label of “real” or “fake” for the entire dataset. If the new subject fiber tract network is classified as “fake”, a synthetic fiber tract network with label “real” will be generated by the GAN that is close to the input subject. Then, given a user input threshold, the difference between these networks will reveal fiber tracts that are responsible for the “fake” label prediction, ie false positives.

[0097] Once a set of false positive fiber tracts are located and visualized for a set of thresholds, we can remove these coordinates from the fiber tract network (set of input white matter fiber tracts). We will then re-classify this network using the GAN as “real” or “fake”. If it is classified as real, we are done. If it is classified again as “fake”, we can either augment the difference threshold and repeat, or generate a new synthetic fiber tract network that is close to this edited fiber tract network (with false positives removed), and repeat.

[0098] An additional application of examples is a process of generating phantom fiber tract datasets with ground truths that can be used for research purposes.

[0099] Examples may provide for accurate pre-processing of fiber tract networks for clinical research and diagnostic studies related to neurological and psychiatric diseases and disorders.

[0100] While the invention has been illustrated and described in detail in the drawings and foregoing description, such illustration and description are to be considered illustrative or exemplary and not restrictive; the invention is not limited to the disclosed embodiments.

[0101] Other variations to the disclosed embodiments can be understood and effected by those skilled in the art in practicing the claimed invention, from a study of the drawings, the disclosure, and the appended claims. In the claims, the word “comprising” does not exclude other elements or steps, and the indefinite article “a” or “an” does not exclude a plurality. A single processor or other unit may fulfill the functions of several items recited in the claims. The mere fact that certain measures are recited in mutually different dependent claims does not indicate that a combination of these measured cannot be used to advantage. A computer program may be stored/distributed on a suitable medium, such as an optical storage medium or a solid-state medium supplied together with or as part of other hardware, but may also be distributed in other forms, such as via the Internet or other wired or wireless telecommunication systems. Any reference signs in the claims should not be construed as limiting the scope.

LIST OF REFERENCE NUMERALS

[0102] 100 medical imaging system [0103] 102 computer [0104] 104 processor [0105] 106 hardware interface [0106] 108 user interface [0107] 110 memory [0108] 112 machine executable instructions [0109] 114 generator neural network [0110] 116 discriminator neural network [0111] 118 set of input white matter tracts [0112] 120 label [0113] 122 optimized feature vector [0114] 124 set of generated white matter tracts [0115] 126 false positive subset [0116] 128 set of corrected white matter tracts [0117] 200 receive the set of input white matter fiber tracts [0118] 202 receive the label from the discriminator neural network in response to inputting the set of input white matter fiber tracts [0119] 204 generate an optimized feature vector using the set of input white matter fiber tracts and the generator neural network if the label indicates anatomically incorrect [0120] 206 receive the set of generated white matter fiber tracts from the generator neural network in response to inputting the optimized feature vector [0121] 208 construct a false positive subset of the set of input white matter fiber tracts using the generated set of white matter fiber tracts [0122] 210 render the set of input fiber tracts on a display [0123] 212 indicate the false positive subset on the display and/or indicate an anatomically correct subset on the display using the false positive subset [0124] 214 provide a set of corrected fiber tracts by removing the false positive subset from the set of input white matter fiber tracts [0125] 300 Is the label anatomically incorrect? [0126] 302 end [0127] 304 Remove set of false positive subset from the set of input white matter fiber tracts? [0128] 306 Set the set of input white matter fiber tracts equal to the set of corrected white matter fiber tracts [0129] 400 medical imaging system [0130] 402 magnetic resonance imaging system [0131] 404 magnet [0132] 406 bore of magnet [0133] 408 imaging zone [0134] 409 region of interest [0135] 410 magnetic field gradient coils [0136] 412 magnetic field gradient coil power supply [0137] 414 radio-frequency coil [0138] 416 transceiver [0139] 418 subject [0140] 420 subject support [0141] 430 pulse sequence commands [0142] 432 diffusion weighted magnetic resonance imaging data [0143] 434 diffusion tensor image [0144] 436 fiber tractography algorithm [0145] 500 acquire the diffusion weighted magnetic resonance imaging data by controlling the magnetic resonance imaging system with the pulse sequence commands [0146] 502 reconstruct the diffusion tensor image for the region of interest using diffusion weighted magnetic resonance imaging data [0147] 504 construct the set of input white matter fiber tracts using a diffusion tensor image of a region of interest descriptive of at least part of a brain according to a fiber tractography algorithm [0148] 600 generated white matter fiber tract [0149] 602 input wite matter fiber tract [0150] 604 input white matter fiber tract [0151] 606 endpoint [0152] 608 predetermined endpoint volume [0153] 610 predetermined path volume [0154] 612 false positive [0155] 700 endpoint distance [0156] 702 path distance