Low_Field_MRI_Recon/unrolling_fns.py at main · Image-X-Institute/Low_Field_MRI_Recon · 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
import torch
import torch.nn as nn
import numpy as np
import scipy.io as scio

# UnrollNet and DisCorNet (runs_20260714 onwards) share the same architecture and state-dict
# keys, so this import works for both. The class was renamed; the weights are compatible.
from trained_models.Unrolling.UnrollNet_DC import *
import time
from collections import OrderedDict
import sigpy as sp

from ulf_recon_fns import coil_combine

def zero_filling(x, factor = 16):
    H = x.size(0)
    W = x.size(1)
    D = x.size(2)

    newH = torch.ceil(torch.tensor(H / factor)) * factor
    newW = torch.ceil(torch.tensor(W / factor)) * factor
    newD = torch.ceil(torch.tensor(D / factor)) * factor

    tmp = torch.zeros(newH.int(), newW.int(),newD.int())

    pos = torch.zeros(2, 3)

    a = torch.ceil((newH - H) / 2)
    b = torch.ceil((newW - W) / 2)
    e = torch.ceil((newD - D) / 2)

    c = (a + H)
    d = (b + W)
    f = (e + D)

    a = a.int()
    b = b.int()
    c = c.int()
    d = d.int()
    e = e.int()
    f = f.int()

    tmp[a:c, b:d, e:f] = x

    pos[0, 0] = a
    pos[0, 1] = b
    pos[1, 0] = c
    pos[1, 1] = d
    pos[0, 2] = e
    pos[1, 2] = f

    return tmp, pos


def zero_removing(x, pos):

    a = pos[0, 0]
    b = pos[0, 1]
    c = pos[1, 0]
    d = pos[1, 1]
    e = pos[0, 2]
    f = pos[1, 2]

    a = a.int()
    b = b.int()
    c = c.int()
    d = d.int()
    e = e.int()
    f = f.int()

    x = x[a:c, b:d, e:f]
    return x

def unrollingRecon(inputKspace,mask,mps,model_pth):
    with torch.no_grad():


            ## load trained network
            state_dict = torch.load(model_pth, map_location=lambda storage, loc: storage)
            # create new OrderedDict that does not contain `module.`

            new_state_dict = OrderedDict()
            for k, v in state_dict.items():
                if k[0:6]=='module':
                    name = k[7:] # remove `module.`
                else:
                    name=k  #DEJW - not sure why module is gone
                #
                new_state_dict[name] = v
            # load params

            #print(len(state_dict))
            #print(state_dict.keys())
            mx_size = inputKspace.shape
            #print(mx_size)

            ## the state_dict contains parameters, 14 items per unroll layer
            Unrolling_chi = UnrollNet(int(len(state_dict)/14), (mx_size[1], mx_size[3]), ini_flag = False)

            Unrolling_chi.load_state_dict(new_state_dict)

            Unrolling_chi.eval()

            #mask_name = "trained_models/Unrolling/testing_R4_rand_patt_all/ksp_R4_NA256/test" + str(idx+1) + '.mat'
            #matImage = scio.loadmat(mask_name)
            #Ahb = matImage['AHb']

            zf_vol = sp.ifft(inputKspace,axes=[1,2,3])

            recon_vol_mc = np.zeros(inputKspace.shape,dtype='complex64')


            mask = torch.from_numpy(abs(mask))
            mask = torch.unsqueeze(mask, 0)
            mask = torch.unsqueeze(mask, 0)
            mask = mask.byte()

            for j in range(inputKspace.shape[0]):
                for k in range(inputKspace.shape[2]):

                    Ahb = np.array(np.squeeze(zf_vol[j,:,k,:]))

                    ksp = np.fft.ifftshift(Ahb, axes=(0 ,1))
                    ksp = np.fft.fftn(ksp, axes=(0 ,1))
                    ksp = np.fft.fftshift(ksp, axes=(0 ,1))

                    Ahb_in = Ahb

                    Ahb_r = np.real(Ahb)
                    Ahb_r = torch.from_numpy(Ahb_r).float()
                    Ahb_r = torch.unsqueeze(Ahb_r, 0)

                    Ahb_i = np.imag(Ahb)
                    Ahb_i = torch.from_numpy(Ahb_i).float()
                    Ahb_i = torch.unsqueeze(Ahb_i, 0)

                    Ahb = torch.cat([Ahb_r, Ahb_i], dim = 0).unsqueeze(0)

                    ksp_r = np.real(ksp)
                    ksp_r = torch.from_numpy(ksp_r).float()
                    ksp_r = torch.unsqueeze(ksp_r, 0)
                    ksp_i = np.imag(ksp)
                    ksp_i = torch.from_numpy(ksp_i).float()
                    ksp_i = torch.unsqueeze(ksp_i, 0)
                    ksp = torch.cat([ksp_r, ksp_i], dim = 0).unsqueeze(0)


#                    R_cal_OP = R_cal((mx_size[1], mx_size[3]), Ahb, mask, 'cpu')  # old no DC version
                    R_cal_OP = R_cal((mx_size[1], mx_size[3]), Ahb, mask, ksp, 'cpu')  #DC version
                    #R_cal_OP = nn.DataParallel(R_cal_OP)


                    pred_chi, _, _ = Unrolling_chi(Ahb, R_cal_OP, 'cpu')
                    pred_chi = R2C(pred_chi)
                    pred_chi = torch.squeeze(pred_chi, 0)
                    pred_chi = torch.squeeze(pred_chi, 0)
                    pred_chi = pred_chi.to('cpu')
                    pred_chi = pred_chi.numpy()

                    recon_vol_mc[j,:,k,:] = pred_chi

            recon_vol_abs = np.sum(np.abs(recon_vol_mc)**2, axis=0)**0.5


            recon_vol_cplx = coil_combine(recon_vol_mc,mps)


            return recon_vol_abs, recon_vol_cplx