-
Notifications
You must be signed in to change notification settings - Fork 12
/
acr_spatial_resolution.py
376 lines (278 loc) · 14.8 KB
/
acr_spatial_resolution.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
"""
ACR Spatial Resolution (MTF)
https://www.acraccreditation.org/-/media/acraccreditation/documents/mri/largephantomguidance.pdf
Calculates the effective resolution (MTF50) for slice 1 for the ACR phantom. This is done in accordance with the
methodology described in Section 3 of the following paper:
https://opg.optica.org/oe/fulltext.cfm?uri=oe-22-5-6040&id=281325
WARNING: The phantom must be slanted for valid results to be produced. This test is not within the scope of ACR
guidance.
This script first identifies the rotation angle of the ACR phantom using slice 1. It provides a warning if the
slanted angle is less than 3 degrees.
The location of the ramps within the slice thickness are identified and a square ROI is selected around the anterior
edge of the slice thickness insert.
A rudimentary edge response function is generated based on the edge within the ROI to provide initialisation values for
the 2D normal cumulative distribution fit of the ROI.
The edge is then super-sampled in the direction of the bright-dark transition of the edge and binned at right angles
based on the edge slope determined from the 2D Normal CDF fit of the ROI to obtain the edge response function.
This super-sampled ERF is then fitted using a weighted sigmoid function. The raw data and this fit are then used to
determine the LSF and the subsequent MTF. The MTF50 for both raw and fitted data are reported.
The results are also visualised.
Created by Yassine Azma
22/02/2023
"""
import sys
import traceback
import scipy
import cv2
import os
import numpy as np
import skimage.morphology
import skimage.measure
from hazenlib.HazenTask import HazenTask
def find_n_peaks(data, n, height=1):
peaks = scipy.signal.find_peaks(data, height)
pk_heights = peaks[1]['peak_heights']
pk_ind = peaks[0]
peak_heights = pk_heights[(-pk_heights).argsort()[:n]]
peak_locs = pk_ind[(-pk_heights).argsort()[:n]] # find n highest peaks
return np.sort(peak_locs), np.sort(peak_heights)
class ACRSpatialResolution(HazenTask):
def __init__(self, **kwargs):
super().__init__(**kwargs)
def run(self) -> dict:
mtf_results = {}
z = []
for dcm in self.data:
z.append(dcm.ImagePositionPatient[2])
idx_sort = np.argsort(z)
for dcm in self.data:
if dcm.ImagePositionPatient[2] == z[idx_sort[0]]:
try:
raw_res, fitted_res = self.get_mtf50(dcm)
mtf_results[f"raw_mtf50_{self.key(self.data[0])}"] = raw_res
mtf_results[f"fitted_mtf50_{self.key(self.data[0])}"] = fitted_res
except Exception as e:
print(f"Could not calculate the spatial resolution for {self.key(dcm)} because of : {e}")
traceback.print_exc(file=sys.stdout)
continue
results = {self.key(self.data[0]): mtf_results, 'reports': {'images': self.report_files}}
return results
def centroid_com(self, dcm):
# Calculate centroid of object using a centre-of-mass calculation
thresh_img = dcm > 0.25 * np.max(dcm)
open_img = skimage.morphology.area_opening(thresh_img, area_threshold=500)
bhull = skimage.morphology.convex_hull_image(open_img)
coords = np.nonzero(bhull) # row major - first array is columns
sum_x = np.sum(coords[1])
sum_y = np.sum(coords[0])
cx, cy = sum_x / coords[0].shape[0], sum_y / coords[1].shape[0]
cxy = (round(cx), round(cy))
return bhull, cxy
def find_rotation(self, dcm):
thresh = dcm * (dcm > 0.2 * np.max(dcm))
kernel = cv2.getStructuringElement(cv2.MORPH_RECT, (5, 5))
dilate = cv2.morphologyEx(thresh, cv2.MORPH_DILATE, kernel)
diff = cv2.absdiff(dilate, thresh)
edges = (diff >= 200) * 1
h, theta, d = skimage.transform.hough_line(edges)
accum, angles, dists = skimage.transform.hough_line_peaks(h, theta, d)
try:
angle = np.rad2deg(scipy.stats.mode(angles)[0][0])
rot_angle = angle + 90 if angle < 0 else angle - 90
except IndexError:
rot_angle = 0
return rot_angle
def y_position_for_ramp(self, res, img, cxy):
investigate_region = int(np.ceil(5.5 / res[1]).item())
if np.mod(investigate_region, 2) == 0:
investigate_region = (investigate_region + 1)
line_profile_y = skimage.measure.profile_line(img, (cxy[1] - 2 * investigate_region, cxy[0]),
(cxy[1] + 2 * investigate_region, cxy[1]),
mode='constant').flatten()
abs_diff_y_profile = np.absolute(np.diff(line_profile_y))
y_peaks = scipy.signal.find_peaks(abs_diff_y_profile, height=1)
pk_heights = y_peaks[1]['peak_heights']
pk_ind = y_peaks[0]
highest_y_peaks = pk_ind[(-pk_heights).argsort()[:2]]
y_locs = highest_y_peaks - 1
height_pts = cxy[1] - 2 * investigate_region - 1 + y_locs
y = np.min(height_pts) + 2
return y
def crop_image(self, img, x, y, width):
crop_x, crop_y = (x - width // 2, x + width // 2), (y - width // 2, y + width // 2)
crop_img = img[crop_y[0]:crop_y[1], crop_x[0]:crop_x[1]]
return crop_img
def get_edge_type(self, crop_img):
edge_sum_rows = np.sum(crop_img, axis=1).astype(np.int_)
edge_sum_cols = np.sum(crop_img, axis=0).astype(np.int_)
_, pk_rows_height = find_n_peaks(np.abs(np.diff(edge_sum_rows)), 1)
_, pk_cols_height = find_n_peaks(np.abs(np.diff(edge_sum_cols)), 1)
edge_type = 'vertical' if pk_rows_height > pk_cols_height else 'horizontal'
thresh_roi_crop = crop_img > 0.6 * np.max(crop_img)
edge_dir = np.sum(thresh_roi_crop, axis=0) if edge_type == 'vertical' else np.sum(thresh_roi_crop, axis=1)
if edge_type == 'vertical':
direction = 'downward' if edge_dir[-1] > edge_dir[0] else 'upward'
else:
direction = 'leftward' if edge_dir[-1] > edge_dir[0] else 'rightward'
return edge_type, direction
def edge_location_for_plot(self, crop_img, edge_type):
thresh_roi_crop = crop_img > 0.6 * np.max(crop_img)
naive_lsf = np.abs(np.diff(np.sum(thresh_roi_crop, 1))) > 1 if edge_type == 'vertical' else np.abs(
np.diff(np.sum(thresh_roi_crop, 0)))
edge_test = np.diff(np.where(naive_lsf == 0))[0]
edge_begin = np.where(edge_test > 1)
edge_loc = np.array([edge_begin, edge_begin + edge_test[edge_begin] - 1]).flatten()
return edge_loc
def fit_normcdf_surface(self, crop_img, edge_type, direction):
thresh_roi_crop = crop_img > 0.6 * np.max(crop_img)
temp_x = np.linspace(1, thresh_roi_crop.shape[1], thresh_roi_crop.shape[1])
temp_y = np.linspace(1, thresh_roi_crop.shape[0], thresh_roi_crop.shape[0])
x, y = np.meshgrid(temp_x, temp_y)
bright = max(crop_img[thresh_roi_crop])
dark = 20 + np.min(crop_img[~thresh_roi_crop])
def func(x, slope, mu, bright, dark):
norm_cdf = (bright - dark) * scipy.stats.norm.cdf(x[0], mu + slope * x[1], 0.5) + dark
return norm_cdf
sign = 1 if direction in ('downward', 'leftward') else -1
x_data = np.vstack((sign * x.ravel(), y.ravel())) if edge_type == 'vertical' else np.vstack(
(sign * y.ravel(), x.ravel()))
popt, pcov = scipy.optimize.curve_fit(func, x_data, crop_img.ravel(), p0=[0, 0, bright, dark], maxfev=1000)
surface = func(x_data, popt[0], popt[1], popt[2], popt[3]).reshape(crop_img.shape)
slope = 1 / popt[0] if direction in ('leftward', 'upward') else -1 / popt[0]
return slope, surface
def sample_erf(self, crop_img, slope, edge_type):
resamp_factor = 8
if edge_type == 'horizontal':
resample_crop_img = cv2.resize(crop_img, (crop_img.shape[0] * resamp_factor, crop_img.shape[1]))
else:
resample_crop_img = cv2.resize(crop_img, (crop_img.shape[0], crop_img.shape[1] * resamp_factor))
mid_loc = [i / 2 for i in resample_crop_img.shape]
temp_x = np.linspace(1, resample_crop_img.shape[1], resample_crop_img.shape[1])
temp_y = np.linspace(1, resample_crop_img.shape[0], resample_crop_img.shape[0])
x_resample, y_resample = np.meshgrid(temp_x, temp_y)
erf = []
n_inside_roi = []
if edge_type == 'horizontal':
diffY = (y_resample - 1) - mid_loc[0]
x_prime = x_resample + resamp_factor * diffY * slope
x_min, x_max = np.min(x_prime).astype(int), np.max(x_prime).astype(int)
for k in range(x_min, x_max):
erf_val = np.mean(resample_crop_img[(x_prime >= k) & (x_prime < k + 1)])
erf.append(erf_val)
number_nonzero = np.count_nonzero(resample_crop_img[(x_prime >= k) & (x_prime < k + 1)])
n_inside_roi.append(number_nonzero)
else:
diffX = (x_resample.shape[0] - 1) - x_resample - mid_loc[1]
y_prime = np.flipud(y_resample) + resamp_factor * diffX * slope
y_min, y_max = np.min(y_prime).astype(int), np.max(y_prime).astype(int)
for k in range(y_min, y_max):
erf_val = np.mean(resample_crop_img[(y_prime >= k) & (y_prime < k + 1)])
erf.append(erf_val)
number_nonzero = np.count_nonzero(resample_crop_img[(y_prime >= k) & (y_prime < k + 1)])
n_inside_roi.append(number_nonzero)
erf = np.array(erf)
n_inside_roi = np.array(n_inside_roi)
erf = erf[n_inside_roi == np.max(n_inside_roi)]
return erf
def fit_erf(self, erf):
true_erf = np.diff(erf) > 0.2 * np.max(np.diff(erf))
turning_points = np.where(true_erf)[0][0], np.where(true_erf)[0][-1]
weights = 0.5 * np.ones((len(true_erf) + 1))
weights[turning_points[0]:turning_points[1]] = 1
def func(x, a, b, c, d, e):
sigmoid = a + b / (1 + np.exp(c * (x - d))) ** e
return sigmoid
popt, pcov = scipy.optimize.curve_fit(func, np.arange(1, len(erf) + 1), erf, sigma=np.diag(1 / weights),
p0=[np.min(erf), np.max(erf), 0, sum(turning_points) / 2, 1], maxfev=5000)
erf_fit = func(np.arange(1, len(erf) + 1), popt[0], popt[1], popt[2], popt[3], popt[4])
return erf_fit
def calculate_MTF(self, erf, res):
lsf = np.diff(erf)
N = len(lsf)
n = np.arange(-N / 2, N / 2) if N % 2 == 0 else np.arange(-(N - 1) / 2, (N + 1) / 2)
resamp_factor = 8
Fs = 1 / (np.sqrt(np.mean(np.square(res))) * (1 / resamp_factor))
freq = n * Fs / N
MTF = np.abs(np.fft.fftshift(np.fft.fft(lsf)))
MTF = MTF / np.max(MTF)
zero_freq = np.where(freq == 0)[0][0]
freq = freq[zero_freq:]
MTF = MTF[zero_freq:]
return freq, lsf, MTF
def identify_MTF50(self, freq, MTF):
freq_interp = np.arange(0, 1.005, 0.005)
MTF_interp = np.interp(freq_interp, freq, MTF, left=None, right=None, period=None)
equivalent_linepairs = freq_interp[np.argmin(np.abs(MTF_interp - 0.5))]
eff_res = 1 / (equivalent_linepairs * 2)
return eff_res
def get_mtf50(self, dcm):
img = dcm.pixel_array
res = dcm.PixelSpacing
_, cxy = self.centroid_com(img)
rot_ang = self.find_rotation(img)
if np.round(np.abs(rot_ang), 2) < 3:
print(f'Rotation angle of the ACR phantom is {np.round(rot_ang, 3)}, which has an absolute is less than 3 '
f'degree. Results will be unreliable!')
else:
print(f'Rotation angle of the ACR phantom is {np.round(rot_ang, 3)}')
ramp_x, ramp_y = int(cxy[0]), self.y_position_for_ramp(res, img, cxy)
width = int(13 * img.shape[0] / 256)
crop_img = self.crop_image(img, ramp_x, ramp_y, width)
edge_type, direction = self.get_edge_type(crop_img)
slope, surface = self.fit_normcdf_surface(crop_img, edge_type, direction)
erf = self.sample_erf(crop_img, slope, edge_type)
erf_fit = self.fit_erf(erf)
freq, lsf_raw, MTF_raw = self.calculate_MTF(erf, res)
_, lsf_fit, MTF_fit = self.calculate_MTF(erf_fit, res)
eff_raw_res = round(self.identify_MTF50(freq, MTF_raw), 2)
eff_fit_res = round(self.identify_MTF50(freq, MTF_fit), 2)
if self.report:
edge_loc = self.edge_location_for_plot(crop_img, edge_type)
import matplotlib.pyplot as plt
import matplotlib.patches as patches
fig = plt.figure(figsize=(8, 8))
gs = fig.add_gridspec(3, 2)
ax0 = fig.add_subplot(gs[:2, 0])
ax2 = fig.add_subplot(gs[2:, 0])
ax3 = fig.add_subplot(gs[0, 1])
ax4 = fig.add_subplot(gs[1, 1])
ax5 = fig.add_subplot(gs[2, 1])
ax0.imshow(img, interpolation='none')
rect = patches.Rectangle((ramp_x - width // 2 - 1, ramp_y - width // 2 - 1), width, width, linewidth=1,
edgecolor='w', facecolor='none')
ax0.add_patch(rect)
ax0.axis('off')
ax2.imshow(crop_img)
if edge_type == 'vertical':
ax2.plot(np.arange(0, width - 1), np.mean(edge_loc) - slope * np.arange(0, width - 1), color='r')
else:
ax2.plot(np.mean(edge_loc) + slope * np.arange(0, width - 1), np.arange(0, width - 1), color='r')
ax2.axis('off')
ax2.set_title('Cropped Edge', fontsize=14)
ax3.plot(erf, 'rx', ms=5)
ax3.plot(erf_fit, 'k', lw=3)
ax3.set_ylabel('Signal Intensity')
ax3.set_xlabel('Pixel')
ax3.grid()
ax3.set_title('ERF', fontsize=14)
ax4.plot(lsf_raw, 'rx', ms=5)
ax4.plot(lsf_fit, 'k', lw=3)
ax4.set_ylabel(r'$\Delta$' + ' Signal Intensity')
ax4.set_xlabel('Pixel')
ax4.grid()
ax4.set_title('LSF', fontsize=14)
ax5.plot(freq, MTF_raw, 'rx', ms=8, label=f'Raw Data - {round(eff_raw_res, 2)}mm @ 50%')
ax5.plot(freq, MTF_fit, 'k', lw=3, label=f'Weighted Sigmoid Fit of ERF - {round(eff_fit_res, 2)}mm @ 50%')
ax5.set_xlabel('Spatial Frequency (lp/mm)')
ax5.set_ylabel('Modulation Transfer Ratio')
ax5.set_xlim([-0.05, 1])
ax5.set_ylim([0, 1.05])
ax5.grid()
ax5.legend(fancybox='true', bbox_to_anchor=[1.05, -0.25, 0, 0])
ax5.set_title('MTF', fontsize=14)
plt.tight_layout()
img_path = os.path.realpath(os.path.join(self.report_path, f'{self.key(dcm)}.png'))
fig.savefig(img_path)
self.report_files.append(img_path)
return eff_raw_res, eff_fit_res