Source code for dipy.tracking.metrics

"""Metrics for tracks, where tracks are arrays of points"""

import numpy as np
from scipy.interpolate import splev, splprep

from dipy.testing.decorators import warning_for_keywords


[docs] def winding(xyz): """Total turning angle projected. Project space curve to best fitting plane. Calculate the cumulative signed angle between each line segment and the previous one. Parameters ---------- xyz : array-like shape (N,3) Array representing x,y,z of N points in a track. Returns ------- a : scalar Total turning angle in degrees. """ U, s, V = np.linalg.svd(xyz - np.mean(xyz, axis=0), 0) proj = np.dot(U[:, 0:2], np.diag(s[0:2])) turn = 0 for j in range(len(xyz) - 1): v0 = proj[j] v1 = proj[j + 1] v = np.dot(v0, v1) / (np.linalg.norm(v0) * np.linalg.norm(v1)) v = np.clip(v, -1, 1) tmp = np.arccos(v) turn += tmp return np.rad2deg(turn)
[docs] @warning_for_keywords() def length(xyz, *, along=False): """Euclidean length of track line This will give length in mm if tracks are expressed in world coordinates. Parameters ---------- xyz : array-like shape (N,3) array representing x,y,z of N points in a track along : bool, optional If True, return array giving cumulative length along track, otherwise (default) return scalar giving total length. Returns ------- L : scalar or array shape (N-1,) scalar in case of `along` == False, giving total length, array if `along` == True, giving cumulative lengths. Examples -------- >>> from dipy.tracking.metrics import length >>> xyz = np.array([[1,1,1],[2,3,4],[0,0,0]]) >>> expected_lens = np.sqrt([1+2**2+3**2, 2**2+3**2+4**2]) >>> length(xyz) == expected_lens.sum() True >>> len_along = length(xyz, along=True) >>> np.allclose(len_along, expected_lens.cumsum()) True >>> length([]) 0 >>> length([[1, 2, 3]]) 0 >>> length([], along=True) array([0]) """ xyz = np.asarray(xyz) if xyz.shape[0] < 2: if along: return np.array([0]) return 0 dists = np.sqrt((np.diff(xyz, axis=0) ** 2).sum(axis=1)) if along: return np.cumsum(dists) return np.sum(dists)
[docs] def bytes(xyz): """Size of track in bytes. Parameters ---------- xyz : array-like shape (N,3) Array representing x,y,z of N points in a track. Returns ------- b : int Number of bytes. """ return xyz.nbytes
[docs] def midpoint(xyz): """Midpoint of track Parameters ---------- xyz : array-like shape (N,3) array representing x,y,z of N points in a track Returns ------- mp : array shape (3,) Middle point of line, such that, if L is the line length then `np` is the point such that the length xyz[0] to `mp` and from `mp` to xyz[-1] is L/2. If the middle point is not a point in `xyz`, then we take the interpolation between the two nearest `xyz` points. If `xyz` is empty, return a ValueError Examples -------- >>> from dipy.tracking.metrics import midpoint >>> midpoint([]) Traceback (most recent call last): ... ValueError: xyz array cannot be empty >>> midpoint([[1, 2, 3]]) array([1, 2, 3]) >>> xyz = np.array([[1,1,1],[2,3,4]]) >>> midpoint(xyz) array([ 1.5, 2. , 2.5]) >>> xyz = np.array([[0,0,0],[1,1,1],[2,2,2]]) >>> midpoint(xyz) array([ 1., 1., 1.]) >>> xyz = np.array([[0,0,0],[1,0,0],[3,0,0]]) >>> midpoint(xyz) array([ 1.5, 0. , 0. ]) >>> xyz = np.array([[0,9,7],[1,9,7],[3,9,7]]) >>> midpoint(xyz) array([ 1.5, 9. , 7. ]) """ xyz = np.asarray(xyz) n_pts = xyz.shape[0] if n_pts == 0: raise ValueError("xyz array cannot be empty") if n_pts == 1: return xyz.copy().squeeze() cumlen = np.zeros(n_pts) cumlen[1:] = length(xyz, along=True) midlen = cumlen[-1] / 2.0 ind = np.where((cumlen - midlen) > 0)[0][0] len0 = cumlen[ind - 1] len1 = cumlen[ind] Ds = midlen - len0 Lambda = Ds / (len1 - len0) return Lambda * xyz[ind] + (1 - Lambda) * xyz[ind - 1]
[docs] def center_of_mass(xyz): """Center of mass of streamline Parameters ---------- xyz : array-like shape (N,3) array representing x,y,z of N points in a track Returns ------- com : array shape (3,) center of mass of streamline Examples -------- >>> from dipy.tracking.metrics import center_of_mass >>> center_of_mass([]) Traceback (most recent call last): ... ValueError: xyz array cannot be empty >>> center_of_mass([[1,1,1]]) array([ 1., 1., 1.]) >>> xyz = np.array([[0,0,0],[1,1,1],[2,2,2]]) >>> center_of_mass(xyz) array([ 1., 1., 1.]) """ xyz = np.asarray(xyz) if xyz.size == 0: raise ValueError("xyz array cannot be empty") return np.mean(xyz, axis=0)
[docs] @warning_for_keywords() def magn(xyz, *, n=1): """magnitude of vector""" mag = np.sum(xyz**2, axis=1) ** 0.5 imag = np.where(mag == 0) mag[imag] = np.finfo(float).eps if n > 1: return np.tile(mag, (n, 1)).T return mag.reshape(len(mag), 1)
[docs] def frenet_serret(xyz): r"""Frenet-Serret Space Curve Invariants Calculates the 3 vector and 2 scalar invariants of a space curve defined by vectors r = (x,y,z). If z is omitted (i.e. the array xyz has shape (N,2)), then the curve is only 2D (planar), but the equations are still valid. Similar to https://www.mathworks.com/matlabcentral/fileexchange/11169-frenet In the following equations the prime ($'$) indicates differentiation with respect to the parameter $s$ of a parametrised curve $\mathbf{r}(s)$. - $\mathbf{T}=\mathbf{r'}/|\mathbf{r'}|\qquad$ (Tangent vector)} - $\mathbf{N}=\mathbf{T'}/|\mathbf{T'}|\qquad$ (Normal vector) - $\mathbf{B}=\mathbf{T}\times\mathbf{N}\qquad$ (Binormal vector) - $\kappa=|\mathbf{T'}|\qquad$ (Curvature) - $\mathrm{\tau}=-\mathbf{B'}\cdot\mathbf{N}$ (Torsion) Parameters ---------- xyz : array-like shape (N,3) array representing x,y,z of N points in a track Returns ------- T : array shape (N,3) array representing the tangent of the curve xyz N : array shape (N,3) array representing the normal of the curve xyz B : array shape (N,3) array representing the binormal of the curve xyz k : array shape (N,1) array representing the curvature of the curve xyz t : array shape (N,1) array representing the torsion of the curve xyz Examples -------- Create a helix and calculate its tangent, normal, binormal, curvature and torsion >>> from dipy.tracking import metrics as tm >>> import numpy as np >>> theta = 2*np.pi*np.linspace(0,2,100) >>> x=np.cos(theta) >>> y=np.sin(theta) >>> z=theta/(2*np.pi) >>> xyz=np.vstack((x,y,z)).T >>> T,N,B,k,t=tm.frenet_serret(xyz) """ xyz = np.asarray(xyz) n_pts = xyz.shape[0] if n_pts == 0: raise ValueError("xyz array cannot be empty") dxyz = np.gradient(xyz)[0] ddxyz = np.gradient(dxyz)[0] # Tangent T = np.divide(dxyz, magn(dxyz, n=3)) # Derivative of Tangent dT = np.gradient(T)[0] # Normal N = np.divide(dT, magn(dT, n=3)) # Binormal B = np.cross(T, N) # Curvature k = magn(np.cross(dxyz, ddxyz), n=1) / (magn(dxyz, n=1) ** 3) # Torsion # (In matlab was t=dot(-B,N,2)) t = np.sum(-B * N, axis=1) # return T,N,B,k,t,dxyz,ddxyz,dT return T, N, B, k, t
[docs] def mean_curvature(xyz): """Calculates the mean curvature of a curve Parameters ---------- xyz : array-like shape (N,3) array representing x,y,z of N points in a curve Returns ------- m : float Mean curvature. Examples -------- Create a straight line and a semi-circle and print their mean curvatures >>> from dipy.tracking import metrics as tm >>> import numpy as np >>> x=np.linspace(0,1,100) >>> y=0*x >>> z=0*x >>> xyz=np.vstack((x,y,z)).T >>> m=tm.mean_curvature(xyz) #mean curvature straight line >>> theta=np.pi*np.linspace(0,1,100) >>> x=np.cos(theta) >>> y=np.sin(theta) >>> z=0*x >>> xyz=np.vstack((x,y,z)).T >>> _= tm.mean_curvature(xyz) #mean curvature for semi-circle """ xyz = np.asarray(xyz) n_pts = xyz.shape[0] if n_pts == 0: raise ValueError("xyz array cannot be empty") dxyz = np.gradient(xyz)[0] ddxyz = np.gradient(dxyz)[0] # Curvature k = magn(np.cross(dxyz, ddxyz), n=1) / (magn(dxyz, n=1) ** 3) return np.mean(k)
[docs] def mean_orientation(xyz): """ Calculates the mean orientation of a curve Parameters ---------- xyz : array-like shape (N,3) array representing x,y,z of N points in a curve Returns ------- m : float Mean orientation. """ xyz = np.asarray(xyz) n_pts = xyz.shape[0] if n_pts == 0: raise ValueError("xyz array cannot be empty") dxyz = np.gradient(xyz)[0] return np.mean(dxyz, axis=0)
[docs] def generate_combinations(items, n): """Combine sets of size n from items Parameters ---------- items : sequence n : int Returns ------- ic : iterator Examples -------- >>> from dipy.tracking.metrics import generate_combinations >>> ic=generate_combinations(range(3),2) >>> for i in ic: print(i) [0, 1] [0, 2] [1, 2] """ if n == 0: yield [] elif n == 2: # if n=2 non_recursive for i in range(len(items) - 1): for j in range(i + 1, len(items)): yield [i, j] else: # if n>2 uses recursion for i in range(len(items)): for cc in generate_combinations(items[i + 1 :], n - 1): yield [items[i]] + cc
[docs] @warning_for_keywords() def longest_track_bundle(bundle, *, sort=False): """Return longest track or length sorted track indices in `bundle` If `sort` == True, return the indices of the sorted tracks in the bundle, otherwise return the longest track. Parameters ---------- bundle : sequence of tracks as arrays, shape (N1,3) ... (Nm,3) sort : bool, optional If False (default) return longest track. If True, return length sorted indices for tracks in bundle Returns ------- longest_or_indices : array longest track - shape (N,3) - (if `sort` is False), or indices of length sorted tracks (if `sort` is True) Examples -------- >>> from dipy.tracking.metrics import longest_track_bundle >>> import numpy as np >>> bundle = [np.array([[0,0,0],[2,2,2]]),np.array([[0,0,0],[4,4,4]])] >>> longest_track_bundle(bundle) array([[0, 0, 0], [4, 4, 4]]) >>> longest_track_bundle(bundle, sort=True) #doctest: +ELLIPSIS array([0, 1]...) """ alllengths = [length(t) for t in bundle] alllengths = np.array(alllengths) if sort: ilongest = alllengths.argsort() return ilongest else: ilongest = alllengths.argmax() return bundle[ilongest]
[docs] def intersect_sphere(xyz, center, radius): """If any segment of the track is intersecting with a sphere of specific center and radius return True otherwise False Parameters ---------- xyz : array, shape (N,3) representing x,y,z of the N points of the track center : array, shape (3,) center of the sphere radius : float radius of the sphere Returns ------- tf : {True, False} True if track `xyz` intersects sphere >>> from dipy.tracking.metrics import intersect_sphere >>> line=np.array(([0,0,0],[1,1,1],[2,2,2])) >>> sph_cent=np.array([1,1,1]) >>> sph_radius = 1 >>> intersect_sphere(line,sph_cent,sph_radius) True Notes ----- The ray to sphere intersection method used here is similar with https://paulbourke.net/geometry/circlesphere/ https://paulbourke.net/geometry/circlesphere/source.cpp we just applied it for every segment neglecting the intersections where the intersecting points are not inside the segment """ center = np.array(center) # print center lt = xyz.shape[0] for i in range(lt - 1): # first point x1 = xyz[i] # second point x2 = xyz[i + 1] # do the calculations as given in the Notes x = x2 - x1 a = np.inner(x, x) x1c = x1 - center b = 2 * np.inner(x, x1c) c = ( np.inner(center, center) + np.inner(x1, x1) - 2 * np.inner(center, x1) - radius**2 ) bb4ac = b * b - 4 * a * c # print 'bb4ac',bb4ac if abs(a) < np.finfo(float).eps or bb4ac < 0: # too small segment or # no intersection continue if bb4ac == 0: # one intersection point p mu = -b / 2 * a p = x1 + mu * x # check if point is inside the segment # print 'p',p if np.inner(p - x1, p - x1) <= a: return True if bb4ac > 0: # two intersection points p1 and p2 mu = (-b + np.sqrt(bb4ac)) / (2 * a) p1 = x1 + mu * x mu = (-b - np.sqrt(bb4ac)) / (2 * a) p2 = x1 + mu * x # check if points are inside the line segment # print 'p1,p2',p1,p2 if np.inner(p1 - x1, p1 - x1) <= a or np.inner(p2 - x1, p2 - x1) <= a: return True return False
[docs] def inside_sphere(xyz, center, radius): r"""If any point of the track is inside a sphere of a specified center and radius return True otherwise False. Mathematically this can be simply described by $|x-c|\le r$ where $x$ a point $c$ the center of the sphere and $r$ the radius of the sphere. Parameters ---------- xyz : array, shape (N,3) representing x,y,z of the N points of the track center : array, shape (3,) center of the sphere radius : float radius of the sphere Returns ------- tf : {True,False} Whether point is inside sphere. Examples -------- >>> from dipy.tracking.metrics import inside_sphere >>> line=np.array(([0,0,0],[1,1,1],[2,2,2])) >>> sph_cent=np.array([1,1,1]) >>> sph_radius = 1 >>> inside_sphere(line,sph_cent,sph_radius) True """ return (np.sqrt(np.sum((xyz - center) ** 2, axis=1)) <= radius).any()
[docs] def inside_sphere_points(xyz, center, radius): r"""If a track intersects with a sphere of a specified center and radius return the points that are inside the sphere otherwise False. Mathematically this can be simply described by $|x-c| \le r$ where $x$ a point $c$ the center of the sphere and $r$ the radius of the sphere. Parameters ---------- xyz : array, shape (N,3) representing x,y,z of the N points of the track center : array, shape (3,) center of the sphere radius : float radius of the sphere Returns ------- xyzn : array, shape(M,3) array representing x,y,z of the M points inside the sphere Examples -------- >>> from dipy.tracking.metrics import inside_sphere_points >>> line=np.array(([0,0,0],[1,1,1],[2,2,2])) >>> sph_cent=np.array([1,1,1]) >>> sph_radius = 1 >>> inside_sphere_points(line,sph_cent,sph_radius) array([[1, 1, 1]]) """ return xyz[(np.sqrt(np.sum((xyz - center) ** 2, axis=1)) <= radius)]
[docs] @warning_for_keywords() def spline(xyz, *, s=3, k=2, nest=-1): """Generate B-splines as documented in https://scipy-cookbook.readthedocs.io/items/Interpolation.html The scipy.interpolate packages wraps the netlib FITPACK routines (Dierckx) for calculating smoothing splines for various kinds of data and geometries. Although the data is evenly spaced in this example, it need not be so to use this routine. Parameters ---------- xyz : array, shape (N,3) array representing x,y,z of N points in 3d space s : float, optional A smoothing condition. The amount of smoothness is determined by satisfying the conditions: sum((w * (y - g))**2,axis=0) <= s where g(x) is the smoothed interpolation of (x,y). The user can use s to control the tradeoff between closeness and smoothness of fit. Larger satisfying the conditions: sum((w * (y - g))**2,axis=0) <= s where g(x) is the smoothed interpolation of (x,y). The user can use s to control the tradeoff between closeness and smoothness of fit. Larger s means more smoothing while smaller values of s indicate less smoothing. Recommended values of s depend on the weights, w. If the weights represent the inverse of the standard-deviation of y, then a: good s value should be found in the range (m-sqrt(2*m),m+sqrt(2*m)) where m is the number of datapoints in x, y, and w. k : int, optional Degree of the spline. Cubic splines are recommended. Even values of k should be avoided especially with a small s-value. for the same set of data. If task=-1 find the weighted least square spline for a given set of knots, t. nest : None or int, optional An over-estimate of the total number of knots of the spline to help in determining the storage space. None results in value m+2*k. -1 results in m+k+1. Always large enough is nest=m+k+1. Default is -1. Returns ------- xyzn : array, shape (M,3) array representing x,y,z of the M points inside the sphere Examples -------- >>> import numpy as np >>> t=np.linspace(0,1.75*2*np.pi,100)# make ascending spiral in 3-space >>> x = np.sin(t) >>> y = np.cos(t) >>> z = t >>> x+= np.random.normal(scale=0.1, size=x.shape) # add noise >>> y+= np.random.normal(scale=0.1, size=y.shape) >>> z+= np.random.normal(scale=0.1, size=z.shape) >>> xyz=np.vstack((x,y,z)).T >>> xyzn=spline(xyz,s=3,k=2,nest=-1) >>> len(xyzn) > len(xyz) True See Also -------- scipy.interpolate.splprep scipy.interpolate.splev """ # find the knot points tckp, u = splprep([xyz[:, 0], xyz[:, 1], xyz[:, 2]], s=s, k=k, nest=nest) # evaluate spline, including interpolated points xnew, ynew, znew = splev(np.linspace(0, 1, 400), tckp) return np.vstack((xnew, ynew, znew)).T
[docs] def startpoint(xyz): """First point of the track Parameters ---------- xyz : array, shape(N,3) Track. Returns ------- sp : array, shape(3,) First track point. Examples -------- >>> from dipy.tracking.metrics import startpoint >>> import numpy as np >>> theta=np.pi*np.linspace(0,1,100) >>> x=np.cos(theta) >>> y=np.sin(theta) >>> z=0*x >>> xyz=np.vstack((x,y,z)).T >>> sp=startpoint(xyz) >>> sp.any()==xyz[0].any() True """ return xyz[0]
[docs] def endpoint(xyz): """ Parameters ---------- xyz : array, shape(N,3) Track. Returns ------- ep : array, shape(3,) First track point. Examples -------- >>> from dipy.tracking.metrics import endpoint >>> import numpy as np >>> theta=np.pi*np.linspace(0,1,100) >>> x=np.cos(theta) >>> y=np.sin(theta) >>> z=0*x >>> xyz=np.vstack((x,y,z)).T >>> ep=endpoint(xyz) >>> ep.any()==xyz[-1].any() True """ return xyz[-1]
[docs] def arbitrarypoint(xyz, distance): """Select an arbitrary point along distance on the track (curve) Parameters ---------- xyz : array-like shape (N,3) array representing x,y,z of N points in a track distance : float float representing distance travelled from the xyz[0] point of the curve along the curve. Returns ------- ap : array shape (3,) Arbitrary point of line, such that, if the arbitrary point is not a point in `xyz`, then we take the interpolation between the two nearest `xyz` points. If `xyz` is empty, return a ValueError Examples -------- >>> import numpy as np >>> from dipy.tracking.metrics import arbitrarypoint, length >>> theta=np.pi*np.linspace(0,1,100) >>> x=np.cos(theta) >>> y=np.sin(theta) >>> z=0*x >>> xyz=np.vstack((x,y,z)).T >>> ap=arbitrarypoint(xyz,length(xyz)/3) """ xyz = np.asarray(xyz) n_pts = xyz.shape[0] if n_pts == 0: raise ValueError("xyz array cannot be empty") if n_pts == 1: return xyz.copy().squeeze() cumlen = np.zeros(n_pts) cumlen[1:] = length(xyz, along=True) if cumlen[-1] < distance: raise ValueError("Given distance is bigger than the length of the curve") ind = np.where((cumlen - distance) > 0)[0][0] len0 = cumlen[ind - 1] len1 = cumlen[ind] Ds = distance - len0 Lambda = Ds / (len1 - len0) return Lambda * xyz[ind] + (1 - Lambda) * xyz[ind - 1]
def _extrap(xyz, cumlen, distance): """Helper function for extrapolate""" ind = np.where((cumlen - distance) > 0)[0][0] len0 = cumlen[ind - 1] len1 = cumlen[ind] Ds = distance - len0 Lambda = Ds / (len1 - len0) return Lambda * xyz[ind] + (1 - Lambda) * xyz[ind - 1]
[docs] def principal_components(xyz): """We use PCA to calculate the 3 principal directions for a track Parameters ---------- xyz : array-like shape (N,3) array representing x,y,z of N points in a track Returns ------- va : array_like eigenvalues ve : array_like eigenvectors Examples -------- >>> import numpy as np >>> from dipy.tracking.metrics import principal_components >>> theta=np.pi*np.linspace(0,1,100) >>> x=np.cos(theta) >>> y=np.sin(theta) >>> z=0*x >>> xyz=np.vstack((x,y,z)).T >>> va, ve = principal_components(xyz) >>> np.allclose(va, [0.51010101, 0.09883545, 0]) True """ C = np.cov(xyz.T) va, ve = np.linalg.eig(C) return va, ve
[docs] def midpoint2point(xyz, p): """Calculate distance from midpoint of a curve to arbitrary point p Parameters ---------- xyz : array-like shape (N,3) array representing x,y,z of N points in a track p : array shape (3,) array representing an arbitrary point with x,y,z coordinates in space. Returns ------- d : float a float number representing Euclidean distance Examples -------- >>> import numpy as np >>> from dipy.tracking.metrics import midpoint2point, midpoint >>> theta=np.pi*np.linspace(0,1,100) >>> x=np.cos(theta) >>> y=np.sin(theta) >>> z=0*x >>> xyz=np.vstack((x,y,z)).T >>> dist=midpoint2point(xyz,np.array([0,0,0])) """ mid = midpoint(xyz) return np.sqrt(np.sum((xyz - mid) ** 2))