-
Notifications
You must be signed in to change notification settings - Fork 1
/
brain_segmenter.py
132 lines (105 loc) · 5.34 KB
/
brain_segmenter.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
from pathlib import Path
import numpy as np
from pytorch_lightning import LightningDataModule, Trainer, LightningModule
from tqdm import tqdm
from dataset.patch_dataset import BrainPatchesDataModule
from models.UNetModule import get_model, UNet3
from dataset.roi_extraction import slice_image, reconstruct_patches
from utils import z_score_norm
import SimpleITK as sitk
import torch
from models.EM import ExpectationMaximization
import cv2
import matplotlib.pyplot as plt
import yaml
import time
from scipy.stats import mode
class BrainSegmenter:
def __init__(self, model_checkpoint_path: Path,
device: str = 'cuda:2'):
"""Constructor for BrainSegmenter class
Args:
model_checkpoint_path (Path): Path to lightning model checkpoint.
config_dump.yaml file should be in the
/outputs/experiment_name folder
"""
super().__init__()
self.device = device
with open(model_checkpoint_path.parent.parent.parent/'config_dump.yml', 'r') as f:
self.cfg = list(yaml.load_all(f, yaml.SafeLoader))[0]
self.model = UNet3(**self.cfg['model'])
self.model = self.model.load_from_checkpoint(model_checkpoint_path,
**self.cfg['model']).to(self.device)
# disable randomness, dropout, etc...
self.model.eval()
print('Model loaded')
def segment(self, image: np.ndarray,
progress: bool = False,
ssegm_image=None) -> np.ndarray:
segm_reconstructed = np.zeros_like(image)
if progress:
range_iter = tqdm(range(image.shape[0]))
else:
range_iter = range(image.shape[0])
for slice in range_iter:
image_slices = slice_image(image[slice, :, :],
self.cfg['dataset']['patches']['window_size'],
self.cfg['dataset']['patches']['stride'])
image_slices = [z_score_norm(slice, non_zero_region=True) for slice in image_slices]
image_slices = np.expand_dims(np.asarray(image_slices, dtype=np.float32), axis=1)
image_slices = torch.tensor(image_slices, requires_grad=False).to(self.device)
if self.cfg['model']['in_channels'] == 2 and ssegm_image is not None:
ssegm_slices = slice_image(ssegm_image[slice, :, :],
self.cfg['dataset']['patches']['window_size'],
self.cfg['dataset']['patches']['stride'])
ssegm_slices = torch.tensor(ssegm_slices, dtype=torch.float).to(self.device)
ssegm_slices = ssegm_slices.unsqueeze(1)
image_slices = torch.cat((image_slices, ssegm_slices), dim=1)
# predict with the model
y_hat = self.model(image_slices).detach().cpu().numpy()
y_hat = np.argmax(y_hat, axis=1)
segm_reconstructed[slice, :, :] = reconstruct_patches(y_hat,
image[slice, :, :],
self.cfg['dataset']['patches']['window_size'],
self.cfg['dataset']['patches']['stride'])
segm_reconstructed[image == 0] = 0
return segm_reconstructed
def segment_and_compare(self,
image: np.ndarray,
mask: np.ndarray,
ssegm_image: np.ndarray|None = None,
ensemble: bool=False):
if ensemble:
mask_pred = self.segment_ensemble(image, progress=False, ssegm_image_xyz=ssegm_image)
else:
mask_pred = self.segment(image, progress=False, ssegm_image=ssegm_image)
res = ExpectationMaximization.compute_dice(mask, mask_pred,
map_dict={1:1, 2:2, 3:3})
res['avg_dice'] = np.mean(list(res.values()))
return mask_pred, res
def segment_ensemble(self, img_xyz: np.ndarray,
progress: bool = False,
ssegm_image_xyz=None) -> np.ndarray:
# rearrange axes with numpy
img_yzx = np.transpose(img_xyz, (1, 2, 0))
img_zxy = np.transpose(img_xyz, (2, 0, 1))
if ssegm_image_xyz is not None:
ssegm_yzx = np.transpose(ssegm_image_xyz, (1, 2, 0))
ssegm_zxy = np.transpose(ssegm_image_xyz, (2, 0, 1))
else:
ssegm_yzx = None
ssegm_zxy = None
pred_xyz = self.segment(img_xyz,
progress=progress,
ssegm_image=ssegm_image_xyz)
pred_yzx = self.segment(img_yzx,
progress=progress,
ssegm_image=ssegm_yzx)
pred_zxy = self.segment(img_zxy,
progress=progress,
ssegm_image=ssegm_zxy)
pred_xyz2 = np.transpose(pred_yzx, (2, 0, 1))
pred_xyz3 = np.transpose(pred_zxy, (1, 2, 0))
ens_pred = np.stack((pred_xyz, pred_xyz2, pred_xyz3), axis=0)
final_pred, _ = mode(ens_pred, axis=0, keepdims=False)
return final_pred