-
Notifications
You must be signed in to change notification settings - Fork 11
/
visualize_results.py
241 lines (192 loc) · 11.7 KB
/
visualize_results.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
import seaborn as sns
import matplotlib.style as style
import matplotlib.pyplot as plt
style.use('ggplot')
# style.use('seaborn-poster') #sets the size of the charts
from tqdm import tqdm
import numpy as np
import pandas as pd
import sys
import os
from sklearn.metrics import f1_score, accuracy_score, balanced_accuracy_score, \
precision_score, recall_score, auc, roc_auc_score, \
confusion_matrix, roc_curve
from utils.print_pretty_confusion_matrix import plot_confusion_matrix_from_data
sys.path.insert(0, 'datasets/')
# from binary_data_loader import LAD_MPR_Loader
from estimate_model import *
from os.path import join
import re
from os import listdir
import cv2
def calculate_metrics(col_section, col_ids, col_preds, col_labels):
"""
Calculate final auc and f1 metrics on three levels: per patient, per section and per artery
:return: {dict} each metric as a key and its calculated metric as a value
"""
assert len(col_section) == len(col_ids) == len(col_preds) == len(col_labels)
metrics = {'ACC_section': 0, 'ACC_patient': 0, 'ACC_artery': 0, 'F1_section': 0, 'F1_patient': 0, 'F1_artery': 0}
dict_artery = {'LAD': ['D-1', 'D-2', 'LAD', 'D-3', '2D-2', 'D-1Original', 'LADOriginal', 'D-4'],
'LCX': ['LCX', 'OM-2', 'OM-1', 'OM-3', 'OM', 'LCX-PLB', 'LCX-PDA', 'PLV_LCX', 'PDA_LCX'],
'RCA': ['RCA', 'RCA-PLB', 'RCA-PDA', 'PLV_RCA']}
df = pd.concat([col_ids, col_section, col_preds, col_labels], axis=1)
df = df.rename(columns={col_section.name: 'section', col_ids.name: 'patient', col_preds.name:
'preds', col_labels.name: 'labels'})
df['artery'] = df['section'].apply(lambda x: [k for k in dict_artery.keys() if x in dict_artery[k]][0])
# SECTION
section_labels = df[['preds', 'labels', 'section', 'artery', 'patient']].groupby(['patient', 'section']).agg(
lambda x: max(x))
preds_section = df[['preds', 'labels', 'section', 'artery', 'patient']].groupby(['patient', 'section']).agg(
lambda x: x.value_counts().index[0])
acc = accuracy_score(preds_section['preds'], section_labels['labels'])
f1 = f1_score(preds_section['preds'], section_labels['labels'], average='weighted')
metrics['ACC_section'], metrics['F1_section'] = acc, f1
# ARTERY
sect = section_labels.reset_index()
artery_labels = sect.groupby(['patient', 'artery']).agg(lambda x: max(x))['labels']
preds_artery = preds_section.reset_index().groupby(['patient', 'artery']).agg(lambda x: max(x))[
'preds'] # x.value_counts().index[0])['preds']
acc = accuracy_score(preds_artery, artery_labels)
f1 = f1_score(preds_artery, artery_labels, average='weighted')
metrics['ACC_artery'], metrics['F1_artery'] = acc, f1
# PATIENT
art = artery_labels.reset_index()
patient_labels = art.groupby(['patient']).agg(lambda x: max(x))['labels']
# print(preds_artery.reset_index())
preds_patient = preds_artery.reset_index().groupby(['patient']).agg(lambda x: max(x))[
'preds'] # x.value_counts().index[0])['preds']
acc = accuracy_score(preds_patient, patient_labels)
f1 = f1_score(preds_patient, patient_labels, average='weighted')
metrics['ACC_patient'], metrics['F1_patient'] = acc, f1
return metrics
def label_predictions_to_images(partition, path_model, df, type_pred, p_data):
"""Draw predictions and labels on images and saves to trained model's folder"""
folder_imag = path_model + '/images'
folder_imag_correct = path_model + '/images/correct'
folder_imag_mistakes = path_model + '/images/mistakes'
if not os.path.exists(folder_imag):
os.mkdir(folder_imag)
if not os.path.exists(folder_imag_correct):
os.mkdir(folder_imag_correct)
if not os.path.exists(folder_imag_mistakes):
os.mkdir(folder_imag_mistakes)
for index,a in df.iterrows():
im_name = [x for x in listdir(join(join(p_data, a['PATIENT_NAME']), a['SECTION'])) if a['IMG_NAME'] in x][0]
img = cv2.imread(join(p_data,a['PATIENT_NAME'],a['SECTION'],im_name),0)
label_text = 'Label: ' + str(a['LABELS']) + ' stenosis: ' + a['STENOSIS'] + '%'
pred_text = ' | Prediction: ' + str(a['PREDS']) + ' prob: ' + str(round(a['PROBABILITY']*100, 3))+ '%'
font = cv2.FONT_HERSHEY_COMPLEX
cv2.putText(img,label_text + pred_text,(10,500), font, 0.4,(255,255,255),1,cv2.LINE_AA)
if type_pred == 'correct':
cv2.imwrite( join(folder_imag_correct, im_name), img )
if type_pred == 'mistake':
cv2.imwrite( join(folder_imag_mistakes, im_name), img )
return
def categorize_stenosis_score(stenosis_scores):
temp = [' '.join(re.sub('[>,%<]', ' ', el).split('___')).replace('NORMAL', '0').split() for el in stenosis_scores]
d = {'0':1, '25':2, '25-50':3, '50':4,'50-70':5, '70':6}
temp = [max([d[i] for i in t ]) for t in temp]
stenosis_scores = [list(d.keys())[list(d.values()).index(el)] for el in temp]
return stenosis_scores
def show_bars(list_values, list_names, size_sections, name_metric, color_palette="muted", path_to_save=None, save_fig=False, plot_mode=None):
y_pos = np.arange(len(list_names))
plt.figure(figsize=(12,6))
plt.bar(y_pos, list_values, color=sns.color_palette(color_palette))
plt.xticks(y_pos, list_names, rotation=0, fontsize=20)
plt.ylim(0,1)
plt.yticks(list(plt.yticks()[0]) + [0.5, 0.7, 0.9, 0.1, 0.3], fontsize=20)
for i in range(len(y_pos)) :
plt.text(i - 0.22, list_values[i]+0.03, str(round(list_values[i], 3)) +' (' + str(size_sections[i])+ ')', color='#696969', fontweight='bold', fontsize=14)
plt.ylabel(name_metric, fontsize=30)
if not save_fig:
plt.show()
else:
plt.savefig(os.path.join(path_to_save, '{}_by_{}.png'.format(name_metric, plot_mode)))
def plot_accuracy_bystage(preds_df, path_to_save_figures):
list_accuracies, size_stages = [], []
list_stages = ['0', '25', '25-50', '50', '50-70', '70']
for stenosis_stage in list_stages:
stage_df = preds_df[preds_df['STENOSIS']==stenosis_stage]
labels_stage = stage_df['LABELS'].values
preds_stage = stage_df['PREDS'].values
probas_stage = stage_df['PREDS_PROBAS'].values
new_probas_stage = [[float(el) for el in arr[1:-1].split(' ') if len(el)>2] for arr in probas_stage]
accuracy = accuracy_score(labels_stage, preds_stage)
list_accuracies.append(accuracy), size_stages.append(len(stage_df))
show_bars(list_accuracies, list_stages, size_stages, 'Accuracy',path_to_save=path_to_save_figures, save_fig=True,plot_mode='stage')
def plot_accuracy_bysection(preds_df, path_to_save_figures):
list_accuracies,list_f1,list_recall, list_precision, size_sec = [], [], [], [], []
list_sections = preds_df['IMG_NAME'].apply(lambda x: x.split('_')[0]).unique()
for section in list_sections:
sec_df = preds_df[preds_df['IMG_NAME'].str.contains(section)]
labels_sec = sec_df['LABELS'].values
preds_sec = sec_df['PREDS'].values
probas_sec = sec_df['PREDS_PROBAS'].values
new_probas_sec = [[float(el) for el in arr[1:-1].split(' ') if len(el)>2] for arr in probas_sec]
accuracy, f1 = accuracy_score(labels_sec, preds_sec), f1_score(labels_sec, preds_sec)
recall, precision = recall_score(labels_sec, preds_sec), precision_score(labels_sec, preds_sec)
list_accuracies.append(accuracy), size_sec.append(len(sec_df)), list_f1.append(f1), list_recall.append(recall), list_precision.append(precision)
for metric in ['Accuracy', 'F1 score', 'Recall', 'Precision']:
show_bars(list_accuracies, list_sections, size_sec, metric, path_to_save=path_to_save_figures, save_fig=True, plot_mode='section')
def calc_auc_bypatient(preds_df):
preds_df['PATIENT_ID'] = preds_df['IMG_NAME'].apply(lambda x: x.split('_')[2])
preds_df['SECTION'] = preds_df['IMG_NAME'].apply(lambda x: x.split('_')[0])
patients_ids = preds_df['PATIENT_ID'].unique()
sections_ids = preds_df['SECTION'].unique()
patients_auc = {}
patients_df = preds_df[['LABELS', 'PREDS', 'PATIENT_ID', 'SECTION']].groupby(['PATIENT_ID', 'SECTION'], as_index=False).count()[['PATIENT_ID', 'SECTION']]
patients_df['AUC'] = None
for p in patients_ids:
sections_dict = {}
for s in sections_ids:
temp_df = preds_df[preds_df['PATIENT_ID'] == p][preds_df['SECTION'] == s]
temp_pat = patients_df[patients_df['PATIENT_ID']==p][patients_df['SECTION']==s]
if len(temp_df) > 0:
acc_patient_section = accuracy_score(temp_df['LABELS'], temp_df['PREDS'])
sections_dict[s] = acc_patient_section
patients_df.xs(temp_pat.index[0])['AUC'] = float(acc_patient_section)
patients_auc[p] = sections_dict
patients_df['AUC'] = patients_df['AUC'].astype(float)
auc_bypatient = preds_df[['LABELS', 'PREDS', 'PATIENT_ID', 'SECTION']].groupby(['PATIENT_ID', 'SECTION'], as_index=False).sum()
return patients_df, auc_bypatient
if __name__ == '__main__':
DATA_SPLIT = 'val'
# DATA_SPLIT = 'test'
MODEL_NAME = 'retrained_resnet18_balanced_data_without_25_text_removed_l2_regularization'
SAVE_LABELED_IMAGES = True
p = 'prediction_results/{}/{}/'.format(MODEL_NAME, DATA_SPLIT)
p_data = 'data/binary_classification_only_lad/'
data_df = pd.read_csv(os.path.join(p_data, 'test_without_25_text_removed.csv'))
preds_df = pd.read_csv(os.path.join(p, '{}.csv'.format(DATA_SPLIT)))
path_to_save_figures = join('/home/petryshak/CoronaryArteryPlaqueIdentification/prediction_results/',MODEL_NAME, DATA_SPLIT, 'visualization')
if not os.path.exists(path_to_save_figures):
os.mkdir(path_to_save_figures)
labels = preds_df['LABELS'].values
preds = preds_df['PREDS'].values
probas = preds_df['PREDS_PROBAS'].values
preds_df['PATIENT_ID'] = preds_df['IMG_NAME'].apply(lambda x: x.split('_')[2])
preds_df['SECTION'] = preds_df['IMG_NAME'].apply(lambda x: x.split('_')[0])
preds_df['STENOSIS'] = categorize_stenosis_score(preds_df['STENOSIS_SCORE'].values)
all_probas = [[float(el) for el in arr[1:-1].split(' ') if len(el)>2] for arr in probas]
preds_df['PROBABILITY'] = [x[1] for x in all_probas]
patients_ids = preds_df['PATIENT_ID'].unique()
sections_ids = preds_df['SECTION'].unique()
# Draw graphs
calculated_metrics_dict = estimate_model(labels, preds, [x[1] for x in all_probas])
show_estimate_model(calculated_metrics_dict, path_to_save_figures)
plot_confusion_matrix_from_data(labels, preds, columns=['Normal', 'Stenosis' ], path_to_save=path_to_save_figures, annot=True, cmap='binary', fz=20, figsize=[9,9], show_null_values=0)
fpr, tpr, thresholds = roc_curve(labels, [x[1] for x in all_probas ])
plot_roc_curve(fpr, tpr, path_to_save_figures)
plot_accuracy_bystage(preds_df, path_to_save_figures)
plot_accuracy_bysection(preds_df, path_to_save_figures)
patients_df, auc_bypatient = calc_auc_bypatient(preds_df)
auc_bypatient.to_csv(join(path_to_save_figures, 'auc_by_patient_section.csv'), index=False)
if SAVE_LABELED_IMAGES:
# Save images by section
corrects = preds_df[preds_df['LABELS'] == preds_df['PREDS']]
corrects = corrects.groupby(['SECTION', 'LABELS']).head(10) # select 10 examples from each group
label_predictions_to_images(DATA_SPLIT, path_to_save_figures, corrects, 'correct')
# Save all images where mistakes have been made
mistakes = preds_df[preds_df['LABELS'] != preds_df['PREDS']]
label_predictions_to_images(DATA_SPLIT, path_to_save_figures, mistakes, 'mistake')
print("'\n'Completed!'\n'")