CraNeXt/test.py at main · biodatlab/CraNeXt · GitHub

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
import json
import argparse

import torch
import numpy as np
import nibabel as nib
from skimage.measure import label
from skimage.morphology import binary_erosion, binary_dilation

from monai.transforms import (
    AsDiscreted,
    Compose,
    CropForegroundd,
    LoadImaged,
    Orientationd,
    Resized,
    Invertd,
    ToDeviced,
    EnsureTyped,
)
from monai.data import DataLoader, PersistentDataset, decollate_batch

from model import CraNeXt


def merge_batch(batch_pred: list, post_pred: Compose):
    post_pred_batch = [post_pred(i) for i in decollate_batch(batch_pred)]
    pred_img = post_pred_batch[0]["pred"][0, :, :, :].detach().cpu().numpy()
    return pred_img


def largest_connected_component(segmentation):
    labels = label(segmentation)
    largest_cc = labels == np.argmax(
        np.bincount(labels.flat, weights=segmentation.flat)
    )
    return largest_cc


def save_prediction_array_to_nifti(
    prediction_data: np.array, orig_defective_nii_path: str, output_path: str
):
    orig_defect_img = nib.load(orig_defective_nii_path)
    orig_defect_data = orig_defect_img.get_fdata()

    implant_data = prediction_data - orig_defect_data

    implant_data[implant_data < 0] = 0
    implant_data = binary_erosion(implant_data).astype(int)
    implant_data = binary_erosion(implant_data).astype(int)
    implant_data = largest_connected_component(implant_data)
    implant_data = binary_dilation(implant_data).astype(int)
    implant_data = binary_dilation(implant_data).astype(int)

    defect_skull_data = orig_defect_data.copy()
    defect_skull_data[implant_data > 0] = 2
    defect_skull_data = defect_skull_data.clip(0, 2)
    ni_img = nib.Nifti1Image(
        defect_skull_data.astype(np.uint8),
        affine=orig_defect_img.affine,
        header=orig_defect_img.header,
    )
    ni_img.header.set_data_dtype(np.uint8)
    nib.save(ni_img, output_path)


def parse_args():
    parser = argparse.ArgumentParser()
    parser.add_argument("--name", type=str, help="define filename")
    parser.add_argument("--model_path", type=str, help="define input model path")
    args = parser.parse_args()
    return args


def main():
    args = parse_args()
    model_path = args.model_path
    name = args.name

    device = (
        torch.device("cuda:0") if torch.cuda.is_available() else torch.device("cpu")
    )

    model = CraNeXt()
    state_dict = torch.load(model_path)
    model.load_state_dict(state_dict)
    model.to(device)
    model.eval()

    model_path = f"model/{name}.pth"
    test_dataset = "test_dataset.json"
    # e.g. [{"image": "skullbreak/001_frontoorbital_defective.nii.gz"}, ...]

    with open(test_dataset) as f:
        test_files = json.load(f)

    for skull in test_files:
        output_path = (
            skull["image"]
            .replace("/data/", "/pred/")
            .replace("_defective.nii.gz", f"_{name}.nii.gz")
        )
        skull["defective_path"] = skull["image"]
        skull["output_path"] = output_path

    test_transforms = Compose(
        [
            LoadImaged(
                keys="image",
                ensure_channel_first=True,
                image_only=True,
                dtype=torch.float,
            ),
            Orientationd(keys="image", axcodes="RAS"),
            ToDeviced(keys="image", device=device),
            CropForegroundd(
                keys="image", source_key="image", margin=25, allow_smaller=False
            ),
            Resized(
                keys="image",
                spatial_size=(176, 224, 144),
                mode="area",
            ),
            EnsureTyped(keys="image", dtype=torch.float, device=device),
        ]
    )

    post_transforms = Compose(
        [
            Invertd(
                keys="pred",
                transform=test_transforms,
                orig_keys="image",
                meta_keys="pred_meta_dict",
                orig_meta_keys="image_meta_dict",
                meta_key_postfix="meta_dict",
                nearest_interp=False,
                to_tensor=True,
            ),
            AsDiscreted(keys="pred", threshold=0.5),
        ]
    )

    test_ds = PersistentDataset(
        data=test_files, transform=test_transforms, cache_dir="cache"
    )
    test_loader = DataLoader(test_ds, batch_size=1, num_workers=0)

    for batch_data in test_loader:
        inputs = batch_data["image"].to(device)
        batch_data["pred"] = model(inputs)
        bd = [x for x in decollate_batch(batch_data)]
        image_output = merge_batch(batch_data, post_transforms)
        save_prediction_array_to_nifti(
            image_output, bd[0]["defective_path"], bd[0]["output_path"]
        )


if __name__ == "__main__":
    main()