-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathcreate_heatmaps.py
487 lines (399 loc) · 34.4 KB
/
create_heatmaps.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
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
from __future__ import print_function
import numpy as np
import argparse
import torch
import torch.nn as nn
import pdb
import os
import pandas as pd
from utils.utils import *
from math import floor
from utils.eval_utils import initiate_model as initiate_model
from models.model_clam import CLAM_MB, CLAM_SB
from models.resnet_custom import resnet18_baseline,resnet50_baseline
import h5py
import yaml
from wsi_core.batch_process_utils import initialize_df
from vis_utils.heatmap_utils import initialize_wsi, drawHeatmap, compute_from_patches
from wsi_core.wsi_utils import sample_rois
from utils.file_utils import save_hdf5
from HIPT_4K.hipt_4k import HIPT_4K
from torchvision import transforms
from datasets.dataset_h5 import eval_transforms
import timm
parser = argparse.ArgumentParser(description='Heatmap inference script')
parser.add_argument('--model', type=str, default='resnet50',choices=['resnet18','resnet18histo','resnet50','uni','vitl','hipt'])
parser.add_argument('--save_exp_code', type=str, default=None,
help='experiment code')
parser.add_argument('--overlap', type=float, default=None)
parser.add_argument('--config_file', type=str, default="heatmap_config_template.yaml")
args = parser.parse_args()
def infer_single_slide(model, features, label, reverse_label_dict, k=1):
features = features.to(device)
with torch.no_grad():
if isinstance(model, (CLAM_SB, CLAM_MB)):
model_results_dict = model(features)
logits, Y_prob, Y_hat, A, _ = model(features)
Y_hat = Y_hat.item()
if isinstance(model, (CLAM_MB,)):
A = A[Y_hat]
A = A.view(-1, 1).cpu().numpy()
else:
raise NotImplementedError
print('Y_hat: {}, Y: {}, Y_prob: {}'.format(reverse_label_dict[Y_hat], label, ["{:.4f}".format(p) for p in Y_prob.cpu().flatten()]))
probs, ids = torch.topk(Y_prob, k)
probs = probs[-1].cpu().numpy()
ids = ids[-1].cpu().numpy()
preds_str = np.array([reverse_label_dict[idx] for idx in ids])
return ids, preds_str, probs, A
def load_params(df_entry, params):
for key in params.keys():
if key in df_entry.index:
dtype = type(params[key])
val = df_entry[key]
val = dtype(val)
if isinstance(val, str):
if len(val) > 0:
params[key] = val
elif not np.isnan(val):
params[key] = val
else:
pdb.set_trace()
return params
def parse_config_dict(args, config_dict):
if args.save_exp_code is not None:
config_dict['exp_arguments']['save_exp_code'] = args.save_exp_code
if args.overlap is not None:
config_dict['patching_arguments']['overlap'] = args.overlap
return config_dict
if __name__ == '__main__':
config_path = os.path.join('heatmaps/configs', args.config_file)
config_dict = yaml.safe_load(open(config_path, 'r'))
config_dict = parse_config_dict(args, config_dict)
for key, value in config_dict.items():
if isinstance(value, dict):
print('\n'+key)
for value_key, value_value in value.items():
print (value_key + " : " + str(value_value))
else:
print ('\n'+key + " : " + str(value))
#decision = input('Continue? Y/N ')
#if decision in ['Y', 'y', 'Yes', 'yes']:
# pass
#elif decision in ['N', 'n', 'No', 'NO']:
# exit()
#else:
# raise NotImplementedError
model_type = args.model
args = config_dict
patch_args = argparse.Namespace(**args['patching_arguments'])
data_args = argparse.Namespace(**args['data_arguments'])
model_args = args['model_arguments']
model_args.update({'n_classes': args['exp_arguments']['n_classes']})
model_args = argparse.Namespace(**model_args)
exp_args = argparse.Namespace(**args['exp_arguments'])
heatmap_args = argparse.Namespace(**args['heatmap_arguments'])
sample_args = argparse.Namespace(**args['sample_arguments'])
patch_size = tuple([int(patch_args.patch_size/patch_args.custom_downsample) for i in range(2)])
step_size = tuple((np.array(patch_size) * (1-patch_args.overlap)).astype(int))
print('patch_size: {} x {}, with {:.2f} overlap, step size is {} x {}'.format(patch_size[0], patch_size[1], patch_args.overlap, step_size[0], step_size[1]))
preset = data_args.preset
def_seg_params = {'seg_level': -1, 'sthresh': 15, 'mthresh': 11, 'close': 2, 'use_otsu': False,
'keep_ids': 'none', 'exclude_ids':'none'}
def_filter_params = {'a_t':50.0, 'a_h': 8.0, 'max_n_holes':10}
def_vis_params = {'vis_level': -1, 'line_thickness': 250}
def_patch_params = {'use_padding': True, 'contour_fn': 'four_pt'}
if preset is not None:
preset_df = pd.read_csv(preset)
for key in def_seg_params.keys():
def_seg_params[key] = preset_df.loc[0, key]
for key in def_filter_params.keys():
def_filter_params[key] = preset_df.loc[0, key]
for key in def_vis_params.keys():
def_vis_params[key] = preset_df.loc[0, key]
for key in def_patch_params.keys():
def_patch_params[key] = preset_df.loc[0, key]
if data_args.process_list is None:
if isinstance(data_args.data_dir, list):
slides = []
for data_dir in data_args.data_dir:
slides.extend(os.listdir(data_dir))
else:
slides = sorted(os.listdir(data_args.data_dir))
slides = [slide for slide in slides if data_args.slide_ext in slide]
df = initialize_df(slides, def_seg_params, def_filter_params, def_vis_params, def_patch_params, use_heatmap_args=False)
else:
df = pd.read_csv(os.path.join('heatmaps/process_lists', data_args.process_list))
df = initialize_df(df, def_seg_params, def_filter_params, def_vis_params, def_patch_params, use_heatmap_args=False)
mask = df['process'] == 1
process_stack = df[mask].reset_index(drop=True)
total = len(process_stack)
print('\nlist of slides to process: ')
print(process_stack.head(len(process_stack)))
print('\ninitializing model from checkpoint')
ckpt_path = model_args.ckpt_path
print('\nckpt path: {}'.format(ckpt_path))
if model_args.initiate_fn == 'initiate_model':
model = initiate_model(model_args, ckpt_path)
else:
raise NotImplementedError
if model_type=='resnet50':
print("CREATING HEATMAP USING RESNET50")
feature_extractor = resnet50_baseline(pretrained=True)
transforms = eval_transforms(pretrained=True)
elif model_type=='resnet18':
print("CREATING HEATMAP USING RESNET18")
feature_extractor = resnet18_baseline(pretrained=True)
transforms = eval_transforms(pretrained=True)
elif model_type=='resnet18histo':
print("CREATING HEATMAP USING RESNET18HISTO")
feature_extractor = resnet18_baseline(pretrained=True, dataset = 'Histo')
transforms = transforms.Compose(
[transforms.Resize(224),
transforms.ToTensor(),
transforms.Normalize(mean=(0.5, 0.5, 0.5), std=(0.5, 0.5, 0.5))])
elif model_type=='uni':
print("CREATING HEATMAP USING UNI")
feature_extractor = timm.create_model("vit_large_patch16_224", img_size=224, patch_size=16, init_values=1e-5, num_classes=0, dynamic_img_size=True)
local_dir = "/mnt/results/vit_large_patch16_224.dinov2.uni_mass100k/"
feature_extractor.load_state_dict(torch.load(os.path.join(local_dir, "pytorch_model.bin"), map_location="cpu"), strict=True)
transforms = transforms.Compose(
[transforms.Resize(224),
transforms.ToTensor(),
transforms.Normalize(mean=(0.485, 0.456, 0.406), std=(0.229, 0.224, 0.225))])
elif model_type=='vitl':
print("CREATING HEATMAP USING VITL")
feature_extractor = timm.create_model("vit_large_patch16_224", num_classes=0, pretrained = True)
transforms = transforms.Compose(
[transforms.Resize(224),
transforms.ToTensor(),
transforms.Normalize(mean=(0.485, 0.456, 0.406), std=(0.229, 0.224, 0.225))])
elif model_type=='hipt':
print("USING HIPT")
feature_extractor = HIPT_4K(model256_path="/CLAM/HIPT_4K/ckpts/vit256_small_dino.pth",model4k_path="/CLAM/HIPT_4K/ckpts/vit4k_xs_dino.pth",device256=torch.device('cuda:0'),device4k=torch.device('cuda:0'))
assert 1==2, "need to set the transforms for this"
else:
raise NotImplementedError
feature_extractor.eval()
device=torch.device("cuda" if torch.cuda.is_available() else "cpu")
print('Done!')
label_dict = data_args.label_dict
class_labels = list(label_dict.keys())
class_encodings = list(label_dict.values())
reverse_label_dict = {class_encodings[i]: class_labels[i] for i in range(len(class_labels))}
if torch.cuda.device_count() > 1:
device_ids = list(range(torch.cuda.device_count()))
feature_extractor = nn.DataParallel(feature_extractor, device_ids=device_ids).to('cuda:0')
else:
feature_extractor = feature_extractor.to(device)
os.makedirs(exp_args.production_save_dir, exist_ok=True)
os.makedirs(exp_args.raw_save_dir, exist_ok=True)
blocky_wsi_kwargs = {'top_left': None, 'bot_right': None, 'patch_size': patch_size, 'step_size': patch_size, 'custom_downsample':patch_args.custom_downsample, 'level': patch_args.patch_level, 'use_center_shift': heatmap_args.use_center_shift, 't': transforms}
unavailable_patch_files=0
total = len(process_stack)
for i in range(total):
print('\nprogress: {}/{}'.format(i, total))
print('skipped unavailable slides: {}'.format(unavailable_patch_files))
slide_name = str(process_stack.loc[i, 'slide_id'])
if data_args.slide_ext not in slide_name:
slide_name+=data_args.slide_ext
if isinstance(data_args.data_dir, str):
slide_path = os.path.join(data_args.data_dir, slide_name)
elif isinstance(data_args.data_dir, dict):
data_dir_key = process_stack.loc[i, data_args.data_dir_key]
slide_path = os.path.join(data_args.data_dir[data_dir_key], slide_name)
else:
raise NotImplementedError
if not os.path.isfile(slide_path):
print('svs file unavailable for {}'.format(slide_name))
unavailable_patch_files = unavailable_patch_files+1
continue
print('\nprocessing: ', slide_name)
try:
label = process_stack.loc[i, 'label']
except KeyError:
label = 'Unspecified'
slide_id = slide_name.replace(data_args.slide_ext, '')
if not isinstance(label, str):
grouping = reverse_label_dict[label]
else:
grouping = label
p_slide_save_dir = os.path.join(exp_args.production_save_dir, exp_args.save_exp_code, str(grouping))
os.makedirs(p_slide_save_dir, exist_ok=True)
r_slide_save_dir = os.path.join(exp_args.raw_save_dir, exp_args.save_exp_code, str(grouping), slide_id)
os.makedirs(r_slide_save_dir, exist_ok=True)
if heatmap_args.use_roi:
x1, x2 = process_stack.loc[i, 'x1'], process_stack.loc[i, 'x2']
y1, y2 = process_stack.loc[i, 'y1'], process_stack.loc[i, 'y2']
top_left = (int(x1), int(y1))
bot_right = (int(x2), int(y2))
else:
top_left = None
bot_right = None
mask_file = os.path.join(r_slide_save_dir, slide_id+'_mask.pkl')
# Load segmentation and filter parameters
seg_params = def_seg_params.copy()
filter_params = def_filter_params.copy()
vis_params = def_vis_params.copy()
seg_params = load_params(process_stack.loc[i], seg_params)
filter_params = load_params(process_stack.loc[i], filter_params)
vis_params = load_params(process_stack.loc[i], vis_params)
keep_ids = str(seg_params['keep_ids'])
if len(keep_ids) > 0 and keep_ids != 'none':
seg_params['keep_ids'] = np.array(keep_ids.split(',')).astype(int)
else:
seg_params['keep_ids'] = []
exclude_ids = str(seg_params['exclude_ids'])
if len(exclude_ids) > 0 and exclude_ids != 'none':
seg_params['exclude_ids'] = np.array(exclude_ids.split(',')).astype(int)
else:
seg_params['exclude_ids'] = []
#for key, val in seg_params.items():
# print('{}: {}'.format(key, val))
#for key, val in filter_params.items():
# print('{}: {}'.format(key, val))
#for key, val in vis_params.items():
# print('{}: {}'.format(key, val))
wsi_object = initialize_wsi(slide_path, seg_mask_path=mask_file, seg_params=seg_params, filter_params=filter_params)
wsi_ref_downsample = wsi_object.level_downsamples[patch_args.patch_level]
# the actual patch size for heatmap visualization should be the patch size * downsample factor * custom downsample factor
vis_patch_size = tuple((np.array(patch_size) * np.array(wsi_ref_downsample) * patch_args.custom_downsample).astype(int))
block_map_save_path = os.path.join(r_slide_save_dir, '{}_blockmap.h5'.format(slide_id))
mask_path = os.path.join(r_slide_save_dir, '{}_mask.jpg'.format(slide_id))
try:
max_size = heatmap_args.max_size
except:
max_size = 10000
vis_params['max_size'] = max_size
if vis_params['vis_level'] < 0:
best_level = wsi_object.wsi.get_best_level_for_downsample(32)
vis_params['vis_level'] = best_level
mask = wsi_object.visWSI(**vis_params, number_contours=True)
mask.save(mask_path)
features_path = os.path.join(r_slide_save_dir, slide_id+'.pt')
h5_path = os.path.join(r_slide_save_dir, slide_id+'.h5')
##### check if h5_features_file exists ######
if not os.path.isfile(h5_path) :
_, _, wsi_object = compute_from_patches(wsi_object=wsi_object,
model=model,
feature_extractor=feature_extractor,
batch_size=exp_args.batch_size, **blocky_wsi_kwargs,
attn_save_path=None, feat_save_path=h5_path,
ref_scores=None)
##### check if pt_features_file exists ######
if not os.path.isfile(features_path):
file = h5py.File(h5_path, "r")
features = torch.tensor(file['features'][:])
torch.save(features, features_path)
file.close()
# load features
features = torch.load(features_path)
process_stack.loc[i, 'bag_size'] = len(features)
wsi_object.saveSegmentation(mask_file)
Y_hats, Y_hats_str, Y_probs, A = infer_single_slide(model, features, label, reverse_label_dict, exp_args.n_classes)
del features
if not os.path.isfile(block_map_save_path):
file = h5py.File(h5_path, "r")
coords = file['coords'][:]
file.close()
asset_dict = {'attention_scores': A, 'coords': coords}
block_map_save_path = save_hdf5(block_map_save_path, asset_dict, mode='w')
# save top 3 predictions
for c in range(exp_args.n_classes):
process_stack.loc[i, 'Pred_{}'.format(c)] = Y_hats_str[c]
process_stack.loc[i, 'p_{}'.format(c)] = Y_probs[c]
os.makedirs('heatmaps/results/', exist_ok=True)
if data_args.process_list is not None:
process_stack.to_csv('heatmaps/results/{}.csv'.format(data_args.process_list.replace('.csv', '')), index=False)
else:
process_stack.to_csv('heatmaps/results/{}.csv'.format(exp_args.save_exp_code), index=False)
file = h5py.File(block_map_save_path, 'r')
dset = file['attention_scores']
coord_dset = file['coords']
scores = dset[:]
coords = coord_dset[:]
file.close()
samples = sample_args.samples
for sample in samples:
if sample['sample']:
tag = "label_{}_pred_{}".format(label, Y_hats[0])
sample_save_dir = os.path.join(exp_args.production_save_dir, exp_args.save_exp_code, 'sampled_patches', str(tag), sample['name'])
os.makedirs(sample_save_dir, exist_ok=True)
print('sampling {}'.format(sample['name']))
sample_results = sample_rois(scores, coords, k=sample['k'], mode=sample['mode'], seed=sample['seed'],
score_start=sample.get('score_start', 0), score_end=sample.get('score_end', 1))
for idx, (s_coord, s_score) in enumerate(zip(sample_results['sampled_coords'], sample_results['sampled_scores'])):
print('coord: {} score: {:.3f}'.format(s_coord, s_score))
patch = wsi_object.wsi.read_region(tuple(s_coord), patch_args.patch_level, (patch_args.patch_size, patch_args.patch_size)).convert('RGB')
## making it smaller to save space
patch = patch.resize((int(patch_args.patch_size/2), int(patch_args.patch_size/2)))
patch.save(os.path.join(sample_save_dir, '{}_{}_x_{}_y_{}_a_{:.3f}.png'.format(idx, slide_id, s_coord[0], s_coord[1], s_score)))
wsi_kwargs = {'top_left': top_left, 'bot_right': bot_right, 'patch_size': patch_size, 'step_size': step_size, 'custom_downsample':patch_args.custom_downsample, 'level': patch_args.patch_level, 'use_center_shift': heatmap_args.use_center_shift, 't': transforms}
heatmap_save_name = '{}_blockmap.tiff'.format(slide_id)
if os.path.isfile(os.path.join(r_slide_save_dir, heatmap_save_name)):
pass
else:
heatmap = drawHeatmap(scores, coords, slide_path, wsi_object=wsi_object, cmap=heatmap_args.cmap, alpha=heatmap_args.alpha, use_holes=True, binarize=False, vis_level=-1, blank_canvas=False,
thresh=-1, patch_size = vis_patch_size, convert_to_percentiles=True, max_size = max_size)
heatmap.save(os.path.join(r_slide_save_dir, '{}_blockmap.png'.format(slide_id)))
del heatmap
save_path = os.path.join(r_slide_save_dir, '{}_{}_roi_{}.h5'.format(slide_id, patch_args.overlap, heatmap_args.use_roi))
if heatmap_args.use_ref_scores:
ref_scores = scores
else:
ref_scores = None
if heatmap_args.calc_heatmap:
compute_from_patches(wsi_object=wsi_object, clam_pred=Y_hats[0], model=model, feature_extractor=feature_extractor, batch_size=exp_args.batch_size, **wsi_kwargs,
attn_save_path=save_path, ref_scores=ref_scores)
if not os.path.isfile(save_path):
print('heatmap {} not found'.format(save_path))
if heatmap_args.use_roi:
save_path_full = os.path.join(r_slide_save_dir, '{}_{}_roi_False.h5'.format(slide_id, patch_args.overlap))
print('found heatmap for whole slide')
save_path = save_path_full
else:
continue
file = h5py.File(save_path, 'r')
dset = file['attention_scores']
coord_dset = file['coords']
scores = dset[:]
coords = coord_dset[:]
file.close()
heatmap_vis_args = {'convert_to_percentiles': True, 'vis_level': heatmap_args.vis_level, 'blur': heatmap_args.blur, 'custom_downsample': heatmap_args.custom_downsample, 'max_size': max_size}
if heatmap_args.use_ref_scores:
heatmap_vis_args['convert_to_percentiles'] = False
heatmap_save_name = '{}_{}_roi_{}_blur_{}_rs_{}_bc_{}_a_{}_l_{}_bi_{}_{}.{}'.format(slide_id, float(patch_args.overlap), int(heatmap_args.use_roi),
int(heatmap_args.blur),
int(heatmap_args.use_ref_scores), int(heatmap_args.blank_canvas),
float(heatmap_args.alpha), int(heatmap_args.vis_level),
int(heatmap_args.binarize), float(heatmap_args.binary_thresh), heatmap_args.save_ext)
if os.path.isfile(os.path.join(p_slide_save_dir, heatmap_save_name)):
pass
else:
heatmap = drawHeatmap(scores, coords, slide_path, wsi_object=wsi_object,
cmap=heatmap_args.cmap, alpha=heatmap_args.alpha, **heatmap_vis_args,
binarize=heatmap_args.binarize,
blank_canvas=heatmap_args.blank_canvas,
thresh=heatmap_args.binary_thresh, patch_size = vis_patch_size,
overlap=patch_args.overlap,
top_left=top_left, bot_right = bot_right)
if heatmap_args.save_ext == 'jpg':
heatmap.save(os.path.join(p_slide_save_dir, heatmap_save_name), quality=100)
else:
heatmap.save(os.path.join(p_slide_save_dir, heatmap_save_name))
if heatmap_args.save_orig:
if heatmap_args.vis_level >= 0:
vis_level = heatmap_args.vis_level
else:
vis_level = vis_params['vis_level']
heatmap_save_name = '{}_orig_{}.{}'.format(slide_id,int(vis_level), heatmap_args.save_ext)
if os.path.isfile(os.path.join(p_slide_save_dir, heatmap_save_name)):
pass
else:
heatmap = wsi_object.visWSI(vis_level=vis_level, view_slide_only=True, custom_downsample=heatmap_args.custom_downsample, max_size=max_size)
if heatmap_args.save_ext == 'jpg':
heatmap.save(os.path.join(p_slide_save_dir, heatmap_save_name), quality=100)
else:
heatmap.save(os.path.join(p_slide_save_dir, heatmap_save_name))
with open(os.path.join(exp_args.raw_save_dir, exp_args.save_exp_code, 'config.yaml'), 'w') as outfile:
yaml.dump(config_dict, outfile, default_flow_style=False)