# Copyright (C) 2015 Simon Biggs
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
# http://www.apache.org/licenses/LICENSE-2.0
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
"""Model insert factors and parameterise inserts as equivalent ellipses."""
from pymedphys._imports import numpy as np
from pymedphys._imports import scipy, shapely
[docs]def spline_model(
width_test, ratio_perim_area_test, width_data, ratio_perim_area_data, factor_data
):
"""Return the result of the spline model.
The bounding box is chosen so as to allow extrapolation. The spline orders
are two in the width direction and one in the perimeter/area direction. For
justification on using this method for modelling electron insert factors
see the *Methods: Bivariate spline model* section within
<http://dx.doi.org/10.1016/j.ejmp.2015.11.002>.
Parameters
----------
width_test : np.ndarray
The width point(s) which are to have the electron insert factor
interpolated.
ratio_perim_area_test : np.ndarray
The perimeter/area which are to have the electron insert factor
interpolated.
width_data : np.ndarray
The width data points for the relevant applicator, energy and ssd.
ratio_perim_area_data : np.ndarray
The perimeter/area data points for the relevant applicator, energy and
ssd.
factor_data : np.ndarray
The insert factor data points for the relevant applicator, energy and
ssd.
Returns
-------
result : np.ndarray
The interpolated electron insert factors for width_test and
ratio_perim_area_test.
"""
bbox = [
np.min([np.min(width_data), np.min(width_test)]),
np.max([np.max(width_data), np.max(width_test)]),
np.min([np.min(ratio_perim_area_data), np.min(ratio_perim_area_test)]),
np.max([np.max(ratio_perim_area_data), np.max(ratio_perim_area_test)]),
]
spline = scipy.interpolate.SmoothBivariateSpline(
width_data, ratio_perim_area_data, factor_data, kx=2, ky=1, bbox=bbox
)
return spline.ev(width_test, ratio_perim_area_test)
def _single_calculate_deformability(x_test, y_test, x_data, y_data, z_data):
"""Return the result of the deformability test for a single test point.
The deformability test applies a shift to the spline to determine whether
or not sufficient information for modelling is available. For further
details on the deformability test see the *Methods: Defining valid
prediction regions of the spline* section within
<http://dx.doi.org/10.1016/j.ejmp.2015.11.002>.
Parameters
----------
x_test : float
The x coordinate of the point to test
y_test : float
The y coordinate of the point to test
x_data : np.ndarray
The x coordinates of the model data to test
y_data : np.ndarray
The y coordinates of the model data to test
z_data : np.ndarray
The z coordinates of the model data to test
Returns
-------
deformability : float
The resulting deformability between 0 and 1
representing the ratio of deviation the spline model underwent at
the point in question by introducing an outlier at the point in
question.
"""
deviation = 0.02
adjusted_x_data = np.append(x_data, x_test)
adjusted_y_data = np.append(y_data, y_test)
bbox = [
min(adjusted_x_data),
max(adjusted_x_data),
min(adjusted_y_data),
max(adjusted_y_data),
]
initial_model = scipy.interpolate.SmoothBivariateSpline(
x_data, y_data, z_data, bbox=bbox, kx=2, ky=1
).ev(x_test, y_test)
pos_adjusted_z_data = np.append(z_data, initial_model + deviation)
neg_adjusted_z_data = np.append(z_data, initial_model - deviation)
pos_adjusted_model = scipy.interpolate.SmoothBivariateSpline(
adjusted_x_data, adjusted_y_data, pos_adjusted_z_data, kx=2, ky=1
).ev(x_test, y_test)
neg_adjusted_model = scipy.interpolate.SmoothBivariateSpline(
adjusted_x_data, adjusted_y_data, neg_adjusted_z_data, kx=2, ky=1
).ev(x_test, y_test)
deformability_from_pos_adjustment = (pos_adjusted_model - initial_model) / deviation
deformability_from_neg_adjustment = (initial_model - neg_adjusted_model) / deviation
deformability = np.max(
[deformability_from_pos_adjustment, deformability_from_neg_adjustment]
)
return deformability
[docs]def calculate_percent_prediction_differences(
width_data, ratio_perim_area_data, factor_data
):
"""Return the percent prediction differences.
Calculates the model factor for each data point with that point removed
from the data set. Used to determine an estimated uncertainty for
prediction.
Parameters
----------
width_data : np.ndarray
The width data points for a specific
applicator, energy and ssd.
ratio_perim_area_data : np.ndarray
The perimeter/area data points for
a specific applicator, energy and ssd.
factor_data : np.ndarray
The insert factor data points for a specific
applicator, energy and ssd.
Returns
-------
percent_prediction_differences : np.ndarray
The predicted electron insert factors for each data point
with that given data point removed.
"""
predictions = [
spline_model_with_deformability(
width_data[i],
ratio_perim_area_data[i],
np.delete(width_data, i),
np.delete(ratio_perim_area_data, i),
np.delete(factor_data, i),
)
for i in range(len(width_data))
]
percent_prediction_differences = 100 * (factor_data - predictions) / factor_data
return percent_prediction_differences
def shapely_insert(x, y):
"""Return a shapely object from x and y coordinates."""
return shapely.geometry.Polygon(np.transpose((x, y)))
def search_for_centre_of_largest_bounded_circle(x, y, callback=None):
"""Find the centre of the largest bounded circle within the insert."""
insert = shapely_insert(x, y)
boundary = insert.boundary
centroid = insert.centroid
furthest_distance = np.hypot(
np.diff(insert.bounds[::2]), np.diff(insert.bounds[1::2])
)
def minimising_function(optimiser_input):
x, y = optimiser_input
point = shapely.geometry.Point(x, y)
if insert.contains(point):
edge_distance = point.distance(boundary)
else:
edge_distance = -point.distance(boundary)
return -edge_distance
x0 = np.squeeze(centroid.coords)
niter = 200
T = furthest_distance / 3
stepsize = furthest_distance / 2
niter_success = 50
output = scipy.optimize.basinhopping(
minimising_function,
x0,
niter=niter,
T=T,
stepsize=stepsize,
niter_success=niter_success,
callback=callback,
)
circle_centre = output.x
return circle_centre
def calculate_width(x, y, circle_centre):
"""Return the equivalent ellipse width."""
insert = shapely_insert(x, y)
point = shapely.geometry.Point(*circle_centre)
if insert.contains(point):
distance = point.distance(insert.boundary)
else:
raise ValueError("Circle centre not within insert")
return distance * 2
def calculate_length(x, y, width):
"""Return the equivalent ellipse length."""
insert = shapely_insert(x, y)
length = 4 * insert.area / (np.pi * width)
return length
[docs]def parameterise_insert(x, y, callback=None):
"""Return the parameterisation of an insert given x and y coords."""
circle_centre = search_for_centre_of_largest_bounded_circle(x, y, callback=callback)
width = calculate_width(x, y, circle_centre)
length = calculate_length(x, y, width)
return width, length, circle_centre
[docs]def visual_alignment_of_equivalent_ellipse(x, y, width, length, callback):
"""Visually align the equivalent ellipse to the insert."""
insert = shapely_insert(x, y)
unit_circle = shapely.geometry.Point(0, 0).buffer(1)
initial_ellipse = shapely.affinity.scale(
unit_circle, xfact=width / 2, yfact=length / 2
)
def minimising_function(optimiser_input):
x_shift, y_shift, rotation_angle = optimiser_input
rotated = shapely.affinity.rotate(
initial_ellipse, rotation_angle, use_radians=True
)
translated = shapely.affinity.translate(rotated, xoff=x_shift, yoff=y_shift)
disjoint_area = (
translated.difference(insert).area + insert.difference(translated).area
)
return disjoint_area / 400
x0 = np.append(np.squeeze(insert.centroid.coords), np.pi / 4)
niter = 10
T = insert.area / 40000
stepsize = 3
niter_success = 2
output = scipy.optimize.basinhopping(
minimising_function,
x0,
niter=niter,
T=T,
stepsize=stepsize,
niter_success=niter_success,
callback=callback,
)
x_shift, y_shift, rotation_angle = output.x
return x_shift, y_shift, rotation_angle
def parameterise_insert_with_visual_alignment(
x,
y,
circle_callback=None,
visual_ellipse_callback=None,
complete_parameterisation_callback=None,
):
"""Return an equivalent ellipse with visual alignment parameters."""
width, length, circle_centre = parameterise_insert(x, y, callback=circle_callback)
if complete_parameterisation_callback is not None:
complete_parameterisation_callback(width, length, circle_centre)
x_shift, y_shift, rotation_angle = visual_alignment_of_equivalent_ellipse(
x, y, width, length, callback=visual_ellipse_callback
)
return width, length, circle_centre, x_shift, y_shift, rotation_angle
def convert2_ratio_perim_area(width, length):
"""Convert width and length data into ratio of perimeter to area."""
perimeter = (
np.pi
/ 2
* (3 * (width + length) - np.sqrt((3 * width + length) * (3 * length + width)))
)
area = np.pi / 4 * width * length
return perimeter / area
def create_transformed_mesh(width_data, length_data, factor_data):
"""Return factor data meshgrid."""
x = np.arange(
np.floor(np.min(width_data)) - 1, np.ceil(np.max(width_data)) + 1, 0.1
)
y = np.arange(
np.floor(np.min(length_data)) - 1, np.ceil(np.max(length_data)) + 1, 0.1
)
xx, yy = np.meshgrid(x, y)
zz = spline_model_with_deformability(
xx,
convert2_ratio_perim_area(xx, yy),
width_data,
convert2_ratio_perim_area(width_data, length_data),
factor_data,
)
zz[xx > yy] = np.nan
no_data_x = np.all(np.isnan(zz), axis=0)
no_data_y = np.all(np.isnan(zz), axis=1)
x = x[np.invert(no_data_x)]
y = y[np.invert(no_data_y)]
zz = zz[np.invert(no_data_y), :]
zz = zz[:, np.invert(no_data_x)]
return x, y, zz