-
Notifications
You must be signed in to change notification settings - Fork 5
/
Evaluation_FROC.py
250 lines (200 loc) · 9.89 KB
/
Evaluation_FROC.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
# -*- coding: utf-8 -*-
"""
Created on Wed Jan 20 14:09:32 2016
@author: Babak Ehteshami Bejnordi
Evaluation code for the Camelyon16 challenge on cancer metastases detecion
"""
import openslide
import numpy as np
import matplotlib.pyplot as plt
from scipy import ndimage as nd
from skimage import measure
import os
import sys
def computeEvaluationMask(maskDIR, resolution, level):
"""Computes the evaluation mask.
Args:
maskDIR: the directory of the ground truth mask
resolution: Pixel resolution of the image at level 0
level: The level at which the evaluation mask is made
Returns:
evaluation_mask
"""
slide = openslide.open_slide(maskDIR)
dims = slide.level_dimensions[level]
pixelarray = np.zeros(dims[0]*dims[1], dtype='uint')
pixelarray = np.array(slide.read_region((0,0), level, dims))
distance = nd.distance_transform_edt(255 - pixelarray[:,:,0])
Threshold = 75/(resolution * pow(2, level) * 2) # 75µm is the equivalent size of 5 tumor cells
binary = distance < Threshold
filled_image = nd.morphology.binary_fill_holes(binary)
evaluation_mask = measure.label(filled_image, connectivity = 2)
return evaluation_mask
def computeITCList(evaluation_mask, resolution, level):
"""Compute the list of labels containing Isolated Tumor Cells (ITC)
Description:
A region is considered ITC if its longest diameter is below 200µm.
As we expanded the annotations by 75µm, the major axis of the object
should be less than 275µm to be considered as ITC (Each pixel is
0.243µm*0.243µm in level 0). Therefore the major axis of the object
in level 5 should be less than 275/(2^5*0.243) = 35.36 pixels.
Args:
evaluation_mask: The evaluation mask
resolution: Pixel resolution of the image at level 0
level: The level at which the evaluation mask was made
Returns:
Isolated_Tumor_Cells: list of labels containing Isolated Tumor Cells
"""
max_label = np.amax(evaluation_mask)
properties = measure.regionprops(evaluation_mask)
Isolated_Tumor_Cells = []
threshold = 275/(resolution * pow(2, level))
for i in range(0, max_label):
if properties[i].major_axis_length < threshold:
Isolated_Tumor_Cells.append(i+1)
return Isolated_Tumor_Cells
def readCSVContent(csvDIR):
"""Reads the data inside CSV file
Args:
csvDIR: The directory including all the .csv files containing the results.
Note that the CSV files should have the same name as the original image
Returns:
Probs: list of the Probabilities of the detected lesions
Xcorr: list of X-coordinates of the lesions
Ycorr: list of Y-coordinates of the lesions
"""
Xcorr, Ycorr, Probs = ([] for i in range(3))
csv_lines = open(csvDIR,"r").readlines()
for i in range(len(csv_lines)):
line = csv_lines[i]
elems = line.rstrip().split(',')
Probs.append(float(elems[0]))
Xcorr.append(int(elems[1]))
Ycorr.append(int(elems[2]))
return Probs, Xcorr, Ycorr
def compute_FP_TP_Probs(Ycorr, Xcorr, Probs, is_tumor, evaluation_mask, Isolated_Tumor_Cells, level):
"""Generates true positive and false positive stats for the analyzed image
Args:
Probs: list of the Probabilities of the detected lesions
Xcorr: list of X-coordinates of the lesions
Ycorr: list of Y-coordinates of the lesions
is_tumor: A boolean variable which is one when the case cotains tumor
evaluation_mask: The evaluation mask
Isolated_Tumor_Cells: list of labels containing Isolated Tumor Cells
level: The level at which the evaluation mask was made
Returns:
FP_probs: A list containing the probabilities of the false positive detections
TP_probs: A list containing the probabilities of the True positive detections
NumberOfTumors: Number of Tumors in the image (excluding Isolate Tumor Cells)
detection_summary: A python dictionary object with keys that are the labels
of the lesions that should be detected (non-ITC tumors) and values
that contain detection details [confidence score, X-coordinate, Y-coordinate].
Lesions that are missed by the algorithm have an empty value.
FP_summary: A python dictionary object with keys that represent the
false positive finding number and values that contain detection
details [confidence score, X-coordinate, Y-coordinate].
"""
max_label = np.amax(evaluation_mask)
FP_probs = []
TP_probs = np.zeros((max_label,), dtype=np.float32)
detection_summary = {}
FP_summary = {}
for i in range(1,max_label+1):
if i not in Isolated_Tumor_Cells:
label = 'Label ' + str(i)
detection_summary[label] = []
FP_counter = 0
if (is_tumor):
for i in range(0,len(Xcorr)):
HittedLabel = evaluation_mask[Ycorr[i]/pow(2, level), Xcorr[i]/pow(2, level)]
if HittedLabel == 0:
FP_probs.append(Probs[i])
key = 'FP ' + str(FP_counter)
FP_summary[key] = [Probs[i], Xcorr[i], Ycorr[i]]
FP_counter+=1
elif HittedLabel not in Isolated_Tumor_Cells:
if (Probs[i]>TP_probs[HittedLabel-1]):
label = 'Label ' + str(HittedLabel)
detection_summary[label] = [Probs[i], Xcorr[i], Ycorr[i]]
TP_probs[HittedLabel-1] = Probs[i]
else:
for i in range(0,len(Xcorr)):
FP_probs.append(Probs[i])
key = 'FP ' + str(FP_counter)
FP_summary[key] = [Probs[i], Xcorr[i], Ycorr[i]]
FP_counter+=1
num_of_tumors = max_label - len(Isolated_Tumor_Cells);
return FP_probs, TP_probs, num_of_tumors, detection_summary, FP_summary
def computeFROC(FROC_data):
"""Generates the data required for plotting the FROC curve
Args:
FROC_data: Contains the list of TPs, FPs, number of tumors in each image
Returns:
total_FPs: A list containing the average number of false positives
per image for different thresholds
total_sensitivity: A list containig overall sensitivity of the system
for different thresholds
"""
unlisted_FPs = [item for sublist in FROC_data[1] for item in sublist]
unlisted_TPs = [item for sublist in FROC_data[2] for item in sublist]
total_FPs, total_TPs = [], []
all_probs = sorted(set(unlisted_FPs + unlisted_TPs))
for Thresh in all_probs[1:]:
total_FPs.append((np.asarray(unlisted_FPs) >= Thresh).sum())
total_TPs.append((np.asarray(unlisted_TPs) >= Thresh).sum())
total_FPs.append(0)
total_TPs.append(0)
total_FPs = np.asarray(total_FPs)/float(len(FROC_data[0]))
total_sensitivity = np.asarray(total_TPs)/float(sum(FROC_data[3]))
return total_FPs, total_sensitivity
def plotFROC(total_FPs, total_sensitivity):
"""Plots the FROC curve
Args:
total_FPs: A list containing the average number of false positives
per image for different thresholds
total_sensitivity: A list containig overall sensitivity of the system
for different thresholds
Returns:
-
"""
fig = plt.figure()
plt.xlabel('Average Number of False Positives', fontsize=12)
plt.ylabel('Metastasis detection sensitivity', fontsize=12)
fig.suptitle('Free response receiver operating characteristic curve', fontsize=12)
plt.plot(total_FPs, total_sensitivity, '-', color='#000000')
plt.show()
if __name__ == "__main__":
# mask_folder = "...\\Camelyon16\\Ground_Truth\\Masks"
# result_folder = "...\\Camelyon16\\Results"
mask_folder = "/home/ar/data/Camelyon16_Challenge/check_evaluation/Ground_Truth/Mask"
result_folder = "/home/ar/data/Camelyon16_Challenge/check_evaluation/Results"
result_file_list = []
result_file_list += [each for each in os.listdir(result_folder) if each.endswith('.csv')]
EVALUATION_MASK_LEVEL = 5 # Image level at which the evaluation is done
L0_RESOLUTION = 0.243 # pixel resolution at level 0
FROC_data = np.zeros((4, len(result_file_list)), dtype=np.object)
FP_summary = np.zeros((2, len(result_file_list)), dtype=np.object)
detection_summary = np.zeros((2, len(result_file_list)), dtype=np.object)
caseNum = 0
for case in result_file_list:
print 'Evaluating Performance on image:', case[0:-4]
sys.stdout.flush()
csvDIR = os.path.join(result_folder, case)
Probs, Xcorr, Ycorr = readCSVContent(csvDIR)
is_tumor = case[0:5] == 'Tumor'
if (is_tumor):
maskDIR = os.path.join(mask_folder, case[0:-4]) + '_Mask.tif'
evaluation_mask = computeEvaluationMask(maskDIR, L0_RESOLUTION, EVALUATION_MASK_LEVEL)
ITC_labels = computeITCList(evaluation_mask, L0_RESOLUTION, EVALUATION_MASK_LEVEL)
else:
evaluation_mask = 0
ITC_labels = []
FROC_data[0][caseNum] = case
FP_summary[0][caseNum] = case
detection_summary[0][caseNum] = case
FROC_data[1][caseNum], FROC_data[2][caseNum], FROC_data[3][caseNum], detection_summary[1][caseNum], FP_summary[1][caseNum] = compute_FP_TP_Probs(Ycorr, Xcorr, Probs, is_tumor, evaluation_mask, ITC_labels, EVALUATION_MASK_LEVEL)
caseNum += 1
# Compute FROC curve
total_FPs, total_sensitivity = computeFROC(FROC_data)
# plot FROC curve
plotFROC(total_FPs, total_sensitivity)