-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathmain_test.py
205 lines (175 loc) · 9.64 KB
/
main_test.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
import os
import torch
import numpy as np
import SimpleITK as sitk
from config import MicroadenomaConfig
from utils.logger import Logger
from model.model import AEPformer
from utils.data_utils import get_metrics, train_validate_split, NormalizationV1
from matplotlib import pyplot as plt
import pickle as pkl
os.environ['CUDA_VISIBLE_DEVICES'] = '4'
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
def get_all_data_mean(dice_per_class_list):
dice_per_class=[]
for i in range(len(dice_per_class_list[0])):
dice_per_class.append(np.mean(dice_per_class_list[:, i]))
dice_mean = np.mean(dice_per_class[1:])
return dice_per_class, dice_mean
def soft_vote(predicted_probas):
sv_predicted_prob = torch.mean(predicted_probas, dim=0)
sv_predictions = torch.argmax(sv_predicted_prob, dim=0)
return sv_predictions
def hard_vote(predicted_probas):
hv_predictions = torch.argmax(predicted_probas, dim=1)
hv_predictions = torch.mode(hv_predictions, dim=0).values
return hv_predictions
if __name__ == "__main__":
mylogger = Logger("runs", write=True, save_freq=4)
os.mkdir(mylogger.plt_dir)
config = MicroadenomaConfig()
mylogger.log(config.get_str_config())
'''======================== load model ========================'''
model_list = []
for ckpt_path in config.checkpoint_path:
model_hyper_param = mylogger.load_from_pkl(os.path.dirname(ckpt_path))
model = AEPformer(frame_num=model_hyper_param["frame_num"],
img_shape=model_hyper_param["patch_size"],
output_channel=model_hyper_param["n_class"],
resnet_depth=model_hyper_param["resnet_depth"],
resnet_out_channels=model_hyper_param["resnet_out_channels"],
dropout=model_hyper_param["dropout"]).to(device)
mylogger.log("start load checkpoint: {}".format(ckpt_path))
checkpoint = torch.load(ckpt_path)
model.load_state_dict(checkpoint)
model_list.append(model)
mylogger.log("load checkpoints succeed!")
'''===================== load testing data ====================='''
train_data_path_list, val_data_path_list = train_validate_split(config.dataset_path, train_ratio=config.train_ratio, seed=1)
test_data_list = val_data_path_list
dice_per_class_list = []
rvd_per_class_list = []
ji_per_class_list = []
asd_per_class_list = []
hd95_per_class_list = []
pre_per_class_list = []
rec_per_class_list = []
for file_itm, file_path in enumerate(test_data_list):
patient_num = file_path["image"].split('/')[-2]
image_path = file_path["image"]
label_path = file_path["label"]
orig_image_arr = sitk.GetArrayFromImage(sitk.ReadImage(image_path)) # (frm, slice, H, W)
orig_label_arr = sitk.GetArrayFromImage(sitk.ReadImage(label_path)) # (slice, H, W)
'''========================= crop image =========================='''
slice_num = orig_image_arr.shape[1]
if slice_num == 6 or slice_num == 5:
s_start, s_end = 0, 4
elif slice_num == 7:
s_start, s_end = 1, 5
elif slice_num == 8:
s_start, s_end = 1, 5
elif slice_num ==9:
s_start, s_end = 2, 6
else:
s_start, s_end = None, None
assert s_start is not None
assert s_end is not None
x_start = int(orig_image_arr.shape[2] * config.crop_ratio[0]) - 1
x_end = int(orig_image_arr.shape[2] * config.crop_ratio[1]) + 1
y_start = int(orig_image_arr.shape[3] * config.crop_ratio[2]) - 1
y_end = int(orig_image_arr.shape[3] * config.crop_ratio[3]) + 1
assert config.crop_size[0] >= (x_end - x_start + 1)
assert config.crop_size[1] >= (y_end - y_start + 1)
x_prior = (config.crop_size[0] - (x_end - x_start + 1)) // 2
x_post = config.crop_size[0] - (x_end - x_start + 1) - x_prior
x_start -= x_prior
x_end += x_post
y_prior = (config.crop_size[1] - (y_end - y_start + 1)) // 2
y_post = config.crop_size[1] - (y_end - y_start + 1) - y_prior
y_start -= y_prior
y_end += y_post
assert config.crop_size[0] == (x_end - x_start + 1)
assert config.crop_size[1] == (y_end - y_start + 1)
crop_image_arr = orig_image_arr[:, s_start:s_end + 1, x_start:x_end + 1, y_start:y_end + 1]
'''===================== uniform frame num to frame_num ====================='''
frm_gap = config.frame_num - crop_image_arr.shape[0]
flag = bool(np.random.randint(0, 2))
while frm_gap:
if flag:
crop_image_arr = np.insert(crop_image_arr, 0, crop_image_arr[0], axis=0) if frm_gap > 0 else np.delete(crop_image_arr, 0, axis=0)
else:
crop_image_arr = np.insert(crop_image_arr, crop_image_arr.shape[0], crop_image_arr[-1], axis=0) if frm_gap > 0 else np.delete(crop_image_arr, -1, axis=0)
flag = not flag
frm_gap -= 1 if frm_gap > 0 else -1
'''============================== normalization =============================='''
img_arr = crop_image_arr[np.newaxis, :].astype(np.float64)
img_arr = NormalizationV1(img_arr)
img_arr = torch.from_numpy(img_arr).float().to(device)
label_arr = orig_label_arr.astype(np.int16)
'''============================== model predict =============================='''
output_prob_list = []
with torch.no_grad():
for model in model_list:
model.eval()
output = model(img_arr)
output = torch.softmax(output.squeeze(dim=0).data.cpu(), dim=0)
output_prob_list.append(output)
output_prob_list = torch.stack(output_prob_list)
'''================================== vote =================================='''
# output_vot = hard_vote(output_prob_list)
output_vot = soft_vote(output_prob_list)
predict = np.zeros(orig_image_arr.shape[1:])
predict[s_start:s_end + 1, x_start:x_end + 1, y_start:y_end + 1] = output_vot.numpy()
predict = np.array(predict, dtype=np.int16) # (slice, H, W)
target = np.copy(label_arr) # (slice, H, W)
"================================== save =================================="
with open(os.path.join(mylogger.plt_dir, "{}.pkl".format(patient_num )), 'wb') as f:
pkl.dump(predict, f)
'''============================== calculate metrics =============================='''
assert predict.dtype == target.dtype
metric_dict = get_metrics(predict, target)
dsc_test = metric_dict["dsc"][1:]
dice_per_class_list.append(np.array(dsc_test))
rvd_test = metric_dict["rvd"]
rvd_per_class_list.append(np.array(rvd_test))
ji_test = metric_dict["ji"]
ji_per_class_list.append(np.array(ji_test))
asd_test = metric_dict["asd"]
asd_per_class_list.append(np.array(asd_test))
hd95_test = metric_dict["hd95"]
hd95_per_class_list.append(np.array(hd95_test))
pre_test = metric_dict["pre"]
pre_per_class_list.append(np.array(pre_test))
rec_test = metric_dict["rec"]
rec_per_class_list.append(np.array(rec_test))
mylogger.log(f"{file_itm}th/{len(test_data_list)} {patient_num} volume data:" +
" DSC: {dsc1:.5f}, {dsc2:.5f} || RVD: {rvd1:.5f}, {rvd2:.5f} || Jaccard: {ji1:.5f}, {ji2:.5f} || ASD: {asd1:.5f}, {asd2:.5f} || HD95: {hd1:.5f}, {hd2:.5f} || PRE: {pre1:.5f}, {pre2:.5f} || REC: {rec1:.5f}, {rec2:.5f}"
.format(dsc1=dsc_test[0], dsc2=dsc_test[1],
rvd1=rvd_test[0], rvd2=rvd_test[1],
ji1=ji_test[0], ji2=ji_test[1],
asd1=asd_test[0], asd2=asd_test[1],
hd1=hd95_test[0], hd2=hd95_test[1],
pre1=pre_test[0], pre2=pre_test[1],
rec1=rec_test[0], rec2=rec_test[1]))
dice_per_class_list = np.array(dice_per_class_list)
dice_per_class, dice_mean = get_all_data_mean(dice_per_class_list)
rvd_per_class_list = np.array(rvd_per_class_list)
rvd_per_class, rvd_mean = get_all_data_mean(rvd_per_class_list)
ji_per_class_list = np.array(ji_per_class_list)
ji_per_class, ji_mean = get_all_data_mean(ji_per_class_list)
asd_per_class_list = np.array(asd_per_class_list)
asd_per_calss, asd_mean = get_all_data_mean(asd_per_class_list)
hd95_per_class_list = np.array(hd95_per_class_list)
hd95_per_class, hd95_mean = get_all_data_mean(hd95_per_class_list)
pre_per_class_list = np.array(pre_per_class_list)
pre_pre_class, pre_mean = get_all_data_mean(pre_per_class_list)
rec_per_class_list = np.array(rec_per_class_list)
rec_pre_class, rec_mean = get_all_data_mean(rec_per_class_list)
mylogger.log("All test data: \tDSC: {dsc1:.5f}, {dsc2:.5f} || RVD: {rvd1:.5f}, {rvd2:.5f} || Jaccard: {ji1:.5f}, {ji2:.5f} || ASD: {asd1:.5f}, {asd2:.5f} || HD95: {hd1:.5f}, {hd2:.5f} || PRE: {pre1:.5f}, {pre2:.5f} || REC: {rec1:.5f}, {rec2:.5f}"
.format(dsc1=dice_per_class[0], dsc2=dice_per_class[1],
rvd1=rvd_per_class[0], rvd2=rvd_per_class[1],
ji1=ji_per_class[0], ji2=ji_per_class[1],
asd1=asd_per_calss[0], asd2=asd_per_calss[1],
hd1=hd95_per_class[0], hd2=hd95_per_class[1],
pre1=pre_pre_class[0], pre2=pre_pre_class[1],
rec1=rec_pre_class[0], rec2=rec_pre_class[1]))