Localization of fibrous neural structures

Abstract

A data processing method for determining a path of a neural fibre in a patient, comprising the steps of: a) acquiring an atlas dataset representing an atlas of a fibrous structure comprising the neural fibre b) acquiring a nerve indicating dataset comprising information suitable for identifying the neural fibre in the patient c) calculating a matched atlas dataset by registering the atlas dataset with the nerve indicating dataset d) obtaining a generic path of the neural fibre from the matched atlas dataset e) defining a constraining volume in the patient around the generic path, the constraining volume having at least two end surfaces on which the generic path ends and f) determining the path of the neural fibre between end surfaces using a probabilistic approach, wherein the determined path lies completely within the constraining volume.

Claims

1. A method for determining a path of a neural fibre in a patient, comprising the steps of: a) acquiring an atlas dataset representing an atlas of a fibrous structure comprising the neural fibre; b) acquiring a nerve indicating dataset comprising information corresponding to the patient and suitable for identifying the neural fibre in the patient; c) calculating a matched atlas dataset by registering the atlas dataset with the nerve indicating dataset; d) obtaining a generic path of the neural fibre from the matched atlas dataset; e) defining a constraining volume in the patient around the generic path, the constraining volume having at least two end surfaces on which the generic path ends and the constraining volume is tubular and centered about the generic path; and f) determining the path of the neural fibre between end surfaces of the constraining volume by: selecting a seed point on one of the end surfaces as a current point; iteratively calculating a path of the neural fibre until the path ends on or extends through another surface of the constraining volume, wherein for the current point, each iteration includes: storing the current point as a point on the path of the neural fibre; calculating a path vector starting from the current point and having a length and a direction based on data from the nerve indicating dataset corresponding to the current point; and using an end point of the path vector as the current point in a next iteration; and outputting the iteratively determined path of the neural fibre only when the path ends on or extends through an end surface of the constraining volume.

2. The method according to claim 1, wherein the nerve indicating dataset is a diffusion tensor imaging dataset or a constructive interference in steady state dataset.

3. The method according to claim 1, wherein a plurality of paths is determined by repeating step f).

4. The method according to claim 3, further comprising the step of assigning a score to each of the determined paths.

5. The method according to claim 4, comprising the step of discarding all paths with a score below a predetermined score threshold.

6. The method according to claim 4, wherein the score is calculated depending on the nerve indicating dataset.

7. The method according to claim 4, wherein the score is calculated depending on a path property dataset representing information about known properties of the neural fibre.

8. The method according to claim 1, wherein calculating the matched atlas dataset in step c) comprises registering the atlas dataset with an intermediate modality atlas dataset to obtain an intermediate atlas dataset, registering the intermediate modality atlas dataset with an intermediate modality patient dataset to obtain a first transformation rule, registering the intermediate modality patient dataset with the nerve indicating dataset to obtain a second transformation rule and matching the intermediate atlas dataset to the nerve indicating dataset using the first and second transformation rules.

9. A method for determining a path of a neural fibre in a patient, comprising the steps of: a) acquiring an atlas dataset representing an atlas of a fibrous structure comprising the neural fibre; b) acquiring a nerve indicating dataset comprising information corresponding to the patient and suitable for identifying the neural fibre in the patient; c) calculating a matched atlas dataset by registering the atlas dataset with the nerve indicating dataset; d) obtaining a generic path of the neural fibre from the matched atlas dataset; e) defining a constraining volume in the patient around the generic path, the constraining volume having at least two end surfaces on which the generic path ends and the constraining volume is tubular and centered about the generic path; and f) determining the path of the neural fibre between end surfaces of the constraining volume by: selecting a sequence of points for which data exist in the nerve indicating dataset, wherein the number of points in the sequence is below a predetermined number of points, the start point and the endpoint of the sequence are located on or behind end surfaces and all other points are located within the constraining volume.

10. The method according to claim 9, wherein the distance of two consecutive points in the sequence is below a predetermined distance threshold.

11. The method according to claim 9, wherein a plurality of paths is determined by repeating step f).

12. The method according to claim 11, further comprising the step of assigning a score to each of the determined paths.

13. The method according to claim 12, comprising the step of discarding all paths with a score below a predetermined score threshold.

14. The method according to claim 12, wherein the score is calculated depending on the nerve indicating dataset.

15. The method according to claim 12, wherein the score is calculated depending on a path property dataset representing information about known properties of the neural fibre.

16. The method according to claim 9, wherein calculating the matched atlas dataset in step c) comprises registering the atlas dataset with an intermediate modality atlas dataset to obtain an intermediate atlas dataset, registering the intermediate modality atlas dataset with an intermediate modality patient dataset to obtain a first transformation rule, registering the intermediate modality patient dataset with the nerve indicating dataset to obtain a second transformation rule and matching the intermediate atlas dataset to the nerve indicating dataset using the first and second transformation rules.

17. A computer program embodied on a non-transitory computer readable medium which, when running on a computer or when loaded onto a computer, causes the computer to perform a data processing method for determining a path of a neural fibre in a patient, comprising the steps of: a) acquiring an atlas dataset representing an atlas of a fibrous structure comprising the neural fibre; b) acquiring a nerve indicating dataset comprising information corresponding to the patient and suitable for identifying the neural fibre in the patient; c) calculating a matched atlas dataset by registering the atlas dataset with the nerve indicating dataset; d) obtaining a generic path of the neural fibre from the matched atlas dataset; e) defining a constraining volume in the patient around the generic path, the constraining volume having at least two end surfaces on which the generic path ends and the constraining volume is tubular and centered about the generic path; and f) determining the path of the neural fibre between end surfaces of the constraining volume by: selecting a seed point on one of the end surfaces as a current point; iteratively calculating a path of the neural fibre until the path ends on or extends through another surface of the constraining volume, wherein for the current point, each iteration includes: storing the current point as a point on the path of the neural fibre; calculating a path vector starting from the current point and having a length and a direction based on data from the nerve indicating dataset corresponding to the current point; and using an end point of the path vector as the current point in a next iteration; and outputting the iteratively determined path of the neural fibre only when the path ends on or extends through an end surface of the constraining volume.

18. A computer program embodied on a non-transitory computer readable medium which, when running on a computer or when loaded onto a computer, causes the computer to perform a data processing method for determining a path of a neural fibre in a patient, comprising the steps of: a) acquiring an atlas dataset representing an atlas of a fibrous structure comprising the neural fibre; b) acquiring a nerve indicating dataset comprising information corresponding to the patient and suitable for identifying the neural fibre in the patient; c) calculating a matched atlas dataset by registering the atlas dataset with the nerve indicating dataset; d) obtaining a generic path of the neural fibre from the matched atlas dataset; e) defining a constraining volume in the patient around the generic path, the constraining volume having at least two end surfaces on which the generic path ends and the constraining volume is tubular and centered about the generic path; and f) determining the path of the neural fibre between end surfaces of the constraining volume by: selecting a sequence of points for which data exist in the nerve indicating dataset, wherein the number of points in the sequence is below a predetermined number of points, the start point and the endpoint of the sequence are located on or behind end surfaces and all other points are located within the constraining volume.

Description

(1) The present invention shall be explained in more detail with reference to the accompanying Figures. These Figures show:

(2) FIG. 1 a medical diagnostic system

(3) FIG. 2 matching an atlas to DTI data

(4) FIG. 3 a constraining volume around a generic fibre

(5) FIG. 4 DTI tensors for crossing fibres

(6) FIG. 5 DTI tensors within a constraining volume

(7) FIG. 6 a first approach for determining a path through the constraining volume and

(8) FIG. 7 a second approach for determining a path through the constraining volume.

(9) FIG. 1 shows a medical diagnostic system 1 for performing a data processing method for determining a path of a neural fibre in a patient. The medical diagnostic system 1 comprises a computer 2 connected to an input device 6, such as a keyboard, a mouse or a touch screen, and an output device 7, such as a monitor or any other display device. The computer 2 is further connected to a storage device 8 and an MRI imaging apparatus 9. The computer 2 comprises a central processing unit 3, and interface 4 and a memory 5. The computer 2 is connected to the storage 8 and the imaging apparatus 9 via the interface 4. The central processing unit 3 performs the method described herein using data acquired from the imaging apparatus 9 and/or the storage 8. The program for performing the method is stored in memory 5. Optionally further stored in the memory 5 are data acquired from the imaging apparatus 9 and/or the storage 8.

(10) The computer first acquires an atlas dataset representing an atlas of a fibrous structure comprising the neural fibre and a diffusion tensor imaging (DTI) dataset as an example of a nerve indicating dataset comprising information suitable for identifying the neural fibre in the patient. The central processing unit 3 then calculates a matched atlas dataset by registering the atlas dataset with the DTI dataset. This is shown in detail in FIG. 2. In its upper left, FIG. 2 shows an atlas dataset A, which in the present exemplary embodiment only represents a single neural fibre having a branching structure, such that the path of the neural fibre has the shape of an Y.

(11) The computer 2 also acquires an intermediate modality atlas dataset and an intermediate modality patient dataset. In the present embodiment, the intermediate modality is MRI. The intermediate modality atlas dataset, referred to as A-MRI in FIG. 2, represents MRI data corresponding to the same (generic) structure or body as the atlas A. The intermediate modality patient dataset, referred to in FIG. 2 as P-MRI, refers to the same patient as the DTI dataset.

(12) In a first step, the atlas dataset A is registered with the MRI atlas dataset A-MRI in order to obtain an intermediate atlas dataset IA. The intermediate atlas dataset IA is thus aligned with the MRI atlas dataset A-MRI. As an alternative, the atlas dataset can already be registered with the MRI atlas dataset.

(13) In a second step, the MRI atlas dataset A-MRI is registered with the MRI patient dataset P-MRI in order to obtain a first transformation rule f1 which maps the MRI atlas dataset to the MRI patient dataset. In a third step, the MRI patient dataset P-MRI is registered with the DTI dataset in order to obtain a second transformation rule f2. This second transformation rule f2 maps the MRI patient dataset to the DTI dataset. By combining the first transformation rule f1 and the second transformation rule f2, the intermediate atlas dataset IA can be directly mapped to the DTI dataset, resulting in a matched atlas dataset MA. In other words, the atlas is now mapped to the status of the patient as represented by the DTI dataset. In particular, the atlas dataset A is mapped into a reference or coordinate system of the patient. The matched atlas dataset then represents an approximation of the fibrous structure in the patient.

(14) The central processing unit 3 then obtains a generic path P.sub.G of the neural fibre from the matched atlas dataset MA. In this exemplary embodiment, the generic path P.sub.G shown in FIG. 3 represents only a part of the full length of the neural fibre. In this example, the generic path P.sub.G is simply copied from the matched atlas dataset MA.

(15) The central processing unit 3 then defines a constraining volume in the patient around the generic path P.sub.G. The constraining volume is labeled V in FIG. 3. The constraining volume V has at least two end surfaces on which the generic path ends. If neural fibre continues outside the constraining volume V, then only the part of the neural fibre inside the constraining volume V is represented by the generic path P.sub.G in this document.

(16) FIG. 4 shows a part of a DTI dataset for two crossing neural fibres. Diffusion tensors of the DTI are represented by circles and ellipses. The shape of the representation of a tensor indicates the directional distribution of the tensor, such as a Bingham distribution. In general, the directional distribution is three-dimensional. For a particular direction, the directional distribution provides the probability with which the neural fibre runs in this particular direction.

(17) For simplifying the illustration, the directional distribution is depicted in two dimensions in FIG. 4. Along a single fibre, the tensor typically represents a preferred direction, resulting in a probability distribution of the directions having a clear maximum. In the upper graph in FIG. 4, the probability p(d) for a particular direction d of the neural fibre is plotted over the direction d, given as an angle between /2 and /2. In an area where the two neural fibres are crossing, the tensor is represented by a circle because the tensors do not provide a preferred direction. Instead, the probability p(d) is uniformly distributed over the whole range from /2 to /2. In this area, a fibre tracking algorithm based on DTI data will probably fail.

(18) FIG. 5 shows a sectional view through the constraining volume V having two end surfaces ES1 and ES2 together with a part of the DTI dataset. In particular, twelve tensors of the DTI dataset are shown. The tensors are denoted by T in combination with a two-digit index. In the representation shown in FIG. 5, the first digit of the index represents the row and the second digit of the index represents the column of a tensor T. The central processing unit 3 then determines the path of the neural fibre between the end surfaces ES 1 and ES2 using a probabilistic approach, wherein the determined path lies completely within the constraining volume V. Two approaches for determining the path of the neural fibre are explained with reference to FIGS. 6 and 7 which are based on the constraining volume V and the diffusion tensors shown in FIG. 5.

(19) The first explained with reference to FIG. 6 starts with a starting point S1 on an end surface S1 of the constraining volume V. Since the DTI dataset does not comprise a tensor at this point S1, the central processing unit 3 interpolates other tensors, resulting in an interpolated tensor T.sub.I1. From the directional distribution corresponding the interpolated tensor T.sub.I1, the central processing unit 3 calculates a path vector having a random direction. The length of a path vector can be constant, inversely proportional to the probability of the direction of the path vector or calculated using another probability distribution. In the present example, the end point of the path vector starting at the point S1 is the point, or location, of the tensor T.sub.42. The CPU 3 then calculates a new path vector starting at the point of the tensor T.sub.42 based on the directional distribution represented by the tensor T.sub.42. The endpoint of this path vector is the point, or location, of the tensor T.sub.32.

(20) In analogy, the central processing unit 3 calculates a new path vector. The end point of this new path vector does not coincide with a point, or location, of a tensor in the DTI dataset. The central processing unit 3 therefore interpolates a tensor T.sub.I2 from surrounding tensors of the DTI dataset and calculates a new path vector, the end point of which coincides with the point, or location, of the tensor T.sub.11. The central processing unit 3 then calculates a new path vector from the tensor T.sub.11 and starting at its location.

(21) This new path vector ends at a point for which no tensor exists in the DTI data, such that the central processing unit 3 interpolates a new tensor T.sub.I3. The central processing unit 3 then calculates a new path vector as described above. This new path vector extends through the opposite end surface ES2 of the constraining volume V. The central processing unit has thus found a valid path between the end surfaces ES1 and ES2. This path completely lies within the constraining volume V and is represented by the point S1, the locations of the tensors T.sub.I1, T.sub.32, T.sub.I2, T.sub.11 and T.sub.I3 as well as the point at which the last path vector intersects with the end surface ES2.

(22) As a preferred option, plurality of paths of the neural fibre between the end surfaces ES1 and ES2 is determined. For the same starting point S1, the paths will be different because the directions of the path vectors are calculated based on a directional distribution using a probabilistic approach. In addition, a plurality of start points on the first end surface ES1 can be used.

(23) A second approach for determining the path of the neural fibre is explained with reference to FIG. 7. This approach comprises selecting a sequence of points for which diffusion tensors exist in the DTI dataset. The start point and the end point of the sequence are located on or behind the end surfaces ES1 and ES2 and all other points are located within the constraining volume V.

(24) As can be seen from FIG. 7, there are no tensors in the DTI dataset for points lying on the end surfaces ES1 and ES2. The central processing unit 3 therefore uses points, or locations, of tensors T.sub.S and T.sub.E located behind the end surfaces ES1 and ES2, respectively, as start and end points of the sequence. A point is considered as lying behind an end surface if the straight connection between this point and the neighbouring point in the sequence passes through the end surface, but no other surface of the constraining volume V.

(25) The central processing unit 3 determines a first path P.sub.1 of the neural fibre as the sequence of points, or locations, of the tensors T.sub.S, T.sub.43, T.sub.31, T.sub.21, T.sub.12, T.sub.13 and T.sub.E. Preferably, the central processing unit 3 determines a plurality of paths, for example including a second path P.sub.2 of the neural fibre comprising the sequence of points, or locations, of the tensors T.sub.S, T.sub.43, T.sub.33, T.sub.23, T.sub.13 and T.sub.E.

(26) Preferably, in particular if a plurality of paths is determined, each path is assigned a score. Preferably, the score represents the likelihood that the corresponding path is the correct path of the neural fibre in the patient. The score has at least one of a data dependent part depending on the DTI dataset and a data independent part depending on a path property dataset representing information about known properties of the neural fibre.

(27) As explained with reference to FIGS. 6 and 7, the directions of the path vectors forming the path have a certain probability depending on the directional distribution corresponding to the tensors at the starting points of the respective path vectors. Preferably, the data dependent part of the score is calculated as the average of the probability of the selected path vector directions.

(28) This shall be explained with reference to the paths P.sub.1 and P.sub.2 shown in FIG. 7. The first path P.sub.1 branches from the point, or location, of the tensor T.sub.43 to the point, or location, of the tensor T.sub.31, while the path P.sub.2 branches to the point, or location, of the tensor T.sub.33. It is assumed that the directional distribution corresponding to the tensor T.sub.43 is the one shown in the upper graph in FIG. 4. This means that the probability for the path vector starting at the point, or location, of the tensor T.sub.43 pointing to the point of the tensor T.sub.33 is significantly higher than the one pointing to the point of the tensor T.sub.31. In analogy, all other path vectors are analyzed. As a result, the score of the path P.sub.2 is higher than the score of the path P.sub.1.

(29) Several information about known properties of the neural fibre can be used for the data independent part of the score. For the neural fibre of this exemplary embodiment, it is known that it has a prevalent curvature to the right. Starting at the point, or location, of the tensor T.sub.43, path P.sub.2 continues to the right, while P.sub.1 proceeds to the left. This results in a higher data independent part of the score for P.sub.2 than for P.sub.1.

(30) Another known property of the particular neural fibre of the present exemplary embodiment is that sharp turns of the path are very unlikely. However, the path P.sub.1 branching from the point of the tensor T.sub.43 to the point of the tensor T.sub.31 has a strong directional change relative to the incoming path vector compared to a significantly smaller directional change for the path vector of the path P.sub.2 pointing to the point of the tensor T.sub.33. Based on the known property of the smoothness of the neural fibre, path P.sub.1 has a lower data independent part of the score than path P.sub.2.