File size: 4,593 Bytes
5a93c05
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
import numpy as np
import os

import random
import torch
from scipy.ndimage.interpolation import zoom
from torch.utils.data import Dataset
from scipy import ndimage
from PIL import Image
from scipy.ndimage import map_coordinates, gaussian_filter
import cv2
    
def random_rot_flip(image, label):
    k = np.random.randint(0, 4)
    image = np.rot90(image, k, axes=(0,1))
    label = np.rot90(label, k, axes=(0,1))
    axis = np.random.randint(0, 2)
    image = np.flip(image, axis=axis)
    label = np.flip(label, axis=axis)
    return image.copy(), label.copy()

def random_rotate(image, label):
    angle = np.random.uniform(-25, 25)
    rotated_channels = []
    for c in range(image.shape[-1]):
        rotated_c = ndimage.rotate(
            image[..., c], angle, order=1, 
            reshape=False, mode='nearest'
        )
        rotated_channels.append(rotated_c)
    image_rot = np.stack(rotated_channels, axis=-1)
    label_rot = ndimage.rotate(
        label, angle, order=0, 
        reshape=False, mode='nearest'
    )
    return image_rot, label_rot

def elastic_transform(image, label, alpha=1000, sigma=30):
    shape = image.shape[:2] 
    random_state = np.random.RandomState(None)
    dx = gaussian_filter(
        (random_state.rand(*shape) * 2 - 1), 
        sigma, mode="constant"
    ) * alpha
    dy = gaussian_filter(
        (random_state.rand(*shape) * 2 - 1), 
        sigma, mode="constant"
    ) * alpha

    x, y = np.meshgrid(np.arange(shape[0]), np.arange(shape[1]), indexing='ij')
    indices = np.reshape(x + dx, (-1, 1)), np.reshape(y + dy, (-1, 1))
    dist_image = []
    for c in range(image.shape[-1]):
        channel = map_coordinates(
            image[..., c], indices, 
            order=3, mode='reflect'
        ).reshape(shape)
        dist_image.append(channel)
    dist_image = np.stack(dist_image, axis=-1)
    dist_label = map_coordinates(
        label, indices, 
        order=0, mode='reflect'
    ).reshape(shape)
    
    return dist_image, dist_label


class AdvancedMedicalAug(torch.nn.Module):
    def __init__(self, aug_prob=0.8):
        super().__init__()
        self.aug_prob = aug_prob 
        
    def forward(self, image, label):
        if isinstance(image, torch.Tensor):
            image = image.numpy()
        if isinstance(label, torch.Tensor):
            label = label.numpy()
        
        if random.random() < self.aug_prob:
            if random.random() > 0.5:
                image, label = random_rot_flip(image, label)
            else:
                image, label = random_rotate(image, label)
 
        if random.random() < 0.3:
            image, label = elastic_transform(image, label)
        
        image = self.intensity_augment(image)
        return image, label
    
    def intensity_augment(self, image):

        for c in range(image.shape[-1]):
            image[..., c] = np.clip(
                image[..., c] * random.uniform(0.7, 1.3), 
                0, 1
            )
            if random.random() < 0.2:
                noise = np.random.normal(0, 0.05, image[..., c].shape)
                image[..., c] = np.clip(image[..., c] + noise, 0, 1)
        return image


class Synapse_dataset(Dataset):
    def __init__(self, base_dir, list_dir, split, img_size, transform=None):
        self.transform = transform 
        self.split = split
        self.sample_list = open(os.path.join(list_dir, self.split+'.txt')).readlines()
        self.data_dir = base_dir
        self.img_size = img_size

    def __len__(self):
        return len(self.sample_list)

    def __getitem__(self, idx):
        slice_name = self.sample_list[idx].strip('\n')
        data_path = os.path.join(self.data_dir, slice_name)
        data = np.load(data_path)
        image = data['image'].astype(np.float32)
        if len(image.shape) == 2:
            image = np.expand_dims(image, axis=-1)
        if 'label' in data.files:
            label = data['label'].astype(np.int32)
        else:
            label = np.zeros_like(image, dtype=np.int32)

        if image.shape != self.img_size:
            x, y, z = image.shape 

            image = zoom(image, (self.img_size[0] / x, self.img_size[1] / y, z), order=3)  
            label = zoom(label, (self.img_size[0] / x, self.img_size[1] / y), order=0)

        if self.transform:
            image, label = self.transform(image, label)
        if self.split == 'train':
            image, label = torch.from_numpy(image), torch.from_numpy(label)
        sample = {'image': image, 'label': label, 'case_name': slice_name}
        return sample