insertion_network/inference_args.py at main · MASILab/insertion_network · 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
161
162
163
164
165
166
167
168
169
170
171
172
173
174
from torch import nn
import torch
import numpy as np
from dataset.body_ct import DualStackSampler
from model.insertnet_v1 import InsertionNet
import torchvision
import os
import argparse
import json
import nibabel as nib
import albumentations as A
from nifti2npy import Nifti2Npy
from scipy.ndimage import gaussian_filter
from chart.chart import Chart, ChartGroup
import matplotlib.pyplot as plt

SLICE_NUM = 256

def prepare_network_input(im1, im2):

    def sample_slices(num_slices, num_sampled, voxel_size):

        all_indices = np.array(list(range(num_slices)))
        if num_slices < num_sampled:
            return all_indices, voxel_size

        # gap sampling
        k = int(np.ceil(num_slices / num_sampled))
        indices = all_indices[::k]
        voxel_size[2] = voxel_size[2] * k
        return indices, voxel_size

    data1 = np.array(im1.dataobj)
    data2 = np.array(im2.dataobj)

    vox1 = np.abs([im1.affine[0, 0], im1.affine[1, 1], im1.affine[2, 2]])
    vox2 = np.abs([im2.affine[0, 0], im2.affine[1, 1], im2.affine[2, 2]])

    # sample slices
    indices1, vox1 = sample_slices(data1.shape[2], SLICE_NUM-2, vox1)
    indices2, vox2 = sample_slices(data2.shape[2], SLICE_NUM, vox2)

    stack1 = np.zeros((256, 256, SLICE_NUM))
    stack2 = np.zeros((256, 256, SLICE_NUM))
    mask1 = np.zeros(SLICE_NUM)
    mask2 = np.zeros(SLICE_NUM)

    # starting dumb slice
    stack1[:, :, 0] = 1
    # ending dumb slice
    stack1[:, :, len(indices1) + 1] = -1

    stack1[:, :, 1: len(indices1) + 1] = data1[:, :, indices1]
    stack2[:, :, : len(indices2)] = data2[:, :, indices2]
    mask1[: len(indices1) + 2] = 1
    mask2[: len(indices2)] = 1

    res = {
        'stack1': stack1.astype(np.float32),
        'stack2': stack2.astype(np.float32),
        'mask1': mask1.astype(np.bool_),
        'mask2': mask2.astype(np.bool_),
        'vox1': vox1,
        'vox2': vox2,
    }

    return res

def preprocessing(nii_path):

    img = nib.load(nii_path)
    affine = img.affine

    data = img.get_fdata()

    p = Nifti2Npy()
    data = p.rescale_xy(data)
    resize = A.Resize(256, 256)

    new_affine = affine.copy()
    new_affine[0, 0] = affine[0, 0] * (img.shape[0] / 256)
    new_affine[1, 1] = affine[1, 1] * (img.shape[1] / 256)


    blurred = gaussian_filter(data, (0.8, 0.8, 0), truncate=3)
    resized = resize(image=blurred)['image']
    resized = resized.astype(np.float32)

    return nib.Nifti1Image(resized, new_affine)

parser = argparse.ArgumentParser(description="Process input arguments.")
parser.add_argument("--test_nii", type=str, help="Test file path")
parser.add_argument("--template_nii", type=str, help="Template file path")
parser.add_argument("--device", type=str, help="Device to run", default='cuda')

args = parser.parse_args()

device = args.device

im_tokenizer = torchvision.models.resnet18()
im_tokenizer.conv1 = torch.nn.Conv2d(1, 64, kernel_size=(7, 7), stride=(2, 2), padding=(3, 3), bias=False)
im_tokenizer.fc = torch.nn.Identity()

model = InsertionNet(
        im_tokenizer=im_tokenizer,
        d_model=512,
        max_seq_len=SLICE_NUM,
        n_head=8,
        n_enc_layers=2,
        n_dec_layers=2
    )

model.load_state_dict(torch.load('exp/Exp_body_ct_semi_supervised_1.0_kl_sinPE_1/checkpoints/best.pth', map_location='cpu')['model'])

test_path = args.test_nii
template_path = args.template_nii

template_image = preprocessing(template_path)
test_image = preprocessing(test_path)

input_data_np = prepare_network_input(test_image, template_image)
input_data = {}
for k, v in input_data_np.items():
    input_data[k] = torch.from_numpy(v).unsqueeze(0).to(device)

model.to(device)
model.eval()

i = 0
stack1 = input_data['stack1'][i: i + 1].permute(3, 0, 1, 2)
stack2 = input_data['stack2'][i: i + 1].permute(3, 0, 1, 2)
mask1 = input_data['mask1'][i: i + 1]
mask2 = input_data['mask2'][i: i + 1]
vox1 = input_data['vox1'][i]
vox2 = input_data['vox2'][i]

with torch.no_grad():
    atten_2to1_before_softmax = model(stack1, stack2, mask1, mask2, vox1, vox2)

valid_attn = atten_2to1_before_softmax[0, 0, : sum(mask2[0]), : sum(mask1[0])]

template_key_slices = {'2': 204, '3': 190, '4': 159, '5': 141, '6': 129, '7': 94, '8': 25}

cg = ChartGroup(1, 4)
c1 = cg.get_chart(1, 1)
c2 = cg.get_chart(1, 2)
c3 = cg.get_chart(1, 3)
c4 = cg.get_chart(1, 4)
c3.config(18, 18)
c4.config(18, 18)

c1.axis_off().slice(input_data['stack1'][0].detach().cpu().numpy(),
                    orientation='y', voxsz=input_data['vox1'][0].detach().cpu().numpy())
c2.axis_off().slice(input_data['stack2'][0].detach().cpu().numpy(),
                    orientation='y', voxsz=input_data['vox2'][0].detach().cpu().numpy())

for i, index2 in enumerate([20, 60, 100, 150, 200]):
    # index2 = template_key_slices[str(ks)]
    index_2in1 = valid_attn.argmax(axis=1).detach().cpu().numpy()[index2]

    c1.line([0, 256], [index_2in1, index_2in1], color=Chart.color_scheme2[i], linewidth=3)
    c2.line([0, 256], [index2, index2], color=Chart.color_scheme2[i], linewidth=3)

    pred_dist = torch.softmax(atten_2to1_before_softmax[0, 0], axis=1)
    x = np.array(list(range(SLICE_NUM)))
    c3.line(x, pred_dist[index2].detach().cpu().numpy(), color=Chart.color_scheme2[i], linewidth=3)

im = c4.ax.imshow(pred_dist.detach().cpu().numpy(), cmap='Blues', origin='lower')
cbar = plt.colorbar(im, ax=c4.ax, label='Value', orientation='horizontal')
cbar.ax.tick_params(labelsize=14)

cg.show()

print(1)