Incept-SADNet/visualization.py at main · P47Parzival/Incept-SADNet · 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
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
"merge cam.py psd.py raw_data_visual.py with one plot"
from importlib import import_module

import torch
import numpy as np
import matplotlib.pyplot as plt

import mne
from matplotlib import gridspec
from mne.time_frequency import psd_array_multitaper
from scipy.integrate import simps
from scipy.signal import resample
from matplotlib.collections import LineCollection

from grad_cam import GradCAM
from run import get_all_subjects
from cam import get_predict

state = ['normal', 'poor', 'optimal']
dataset = 'data'


def save_visualization(id, model, subject, x_data, y_data, mode, ll):
    # get label and predict
    input_tensor = torch.Tensor(x_data[id]).unsqueeze(0)
    label = y_data[id]
    out = model(input_tensor)

    # make info
    channelnames = ['Fp1', 'Fp2', 'F7', 'F3', 'Fz', 'F4', 'F8', 'FT7', 'FC3', 'FCz', 'FC4', 'FT8', 'T3', 'C3', 'Cz',
                    'C4', 'T4', 'TP7', 'CP3', 'CPz', 'CP4', 'TP8', 'T5', 'P3', 'Pz', 'P4', 'T6', 'O1', 'Oz', 'O2']
    montage = 'standard_1020'
    info = mne.create_info(ch_names=channelnames, sfreq=500., ch_types='eeg')
    info.set_montage(montage)

    # get camHeatMap
    target_layers = [model.transformer[-1]]
    cam = GradCAM(model, target_layers=target_layers, use_cuda=False)
    grayscale_cam = cam(input_tensor=input_tensor)
    grayscale_cam = grayscale_cam[0, :]

    # psd
    rawSignal = x_data[id]
    sampleLength = 1001
    sampleChannel = 30
    maxvalue = np.max(np.abs(rawSignal))
    rawsignal = rawSignal.reshape(30, 1001)
    deltapower = np.zeros(sampleChannel)
    thetapower = np.zeros(sampleChannel)
    alphapower = np.zeros(sampleChannel)
    betapower = np.zeros(sampleChannel)
    for kk in range(sampleChannel):
        psd, freqs = psd_array_multitaper(rawsignal[kk, :], 500, adaptive=True, normalization='full', verbose=0)
        freq_res = freqs[1] - freqs[0]
        totalpower = simps(psd, dx=freq_res)
        if totalpower < 0.00000001:
            deltapower[kk] = 0
            thetapower[kk] = 0
            alphapower[kk] = 0
            betapower[kk] = 0
        else:
            idx_band = np.logical_and(freqs >= 1, freqs <= 4)
            deltapower[kk] = simps(psd[idx_band], dx=freq_res) / totalpower
            idx_band = np.logical_and(freqs >= 4, freqs <= 8)
            thetapower[kk] = simps(psd[idx_band], dx=freq_res) / totalpower
            idx_band = np.logical_and(freqs >= 8, freqs <= 12)
            alphapower[kk] = simps(psd[idx_band], dx=freq_res) / totalpower
            idx_band = np.logical_and(freqs >= 12, freqs <= 30)
            betapower[kk] = simps(psd[idx_band], dx=freq_res) / totalpower
    mixpower = np.zeros((4, sampleChannel))
    mixpower[0, :] = deltapower
    mixpower[1, :] = thetapower
    mixpower[2, :] = alphapower
    mixpower[3, :] = betapower
    vmax = np.percentile(mixpower, 95)

    # plot
    fig = plt.figure(figsize=(18, 6))
    gridlayout = gridspec.GridSpec(ncols=6, nrows=2, figure=fig, wspace=0.05, hspace=0.005)
    ax0 = fig.add_subplot(gridlayout[0:2, 0:2])
    ax11 = fig.add_subplot(gridlayout[0:1, 2:4])
    ax12 = fig.add_subplot(gridlayout[1:2, 2:4])
    ax21 = fig.add_subplot(gridlayout[0, 4])
    ax22 = fig.add_subplot(gridlayout[0, 5])
    ax23 = fig.add_subplot(gridlayout[1, 4])
    ax24 = fig.add_subplot(gridlayout[1, 5])

    # raw_data_visual including resample
    resampleLength = 202  # 128Hz
    heatmap = grayscale_cam
    sampleInput = x_data[id]
    thespan = np.percentile(sampleInput, 98)
    xx = np.arange(1, resampleLength + 1)
    sampleInput = resample(sampleInput, resampleLength).reshape(30, -1)
    heatmap = resample(heatmap, resampleLength).reshape(30, -1)
    for i in range(0, sampleChannel):
        y = sampleInput[i, :] + thespan * (sampleChannel - 1 - i)
        # print(y.shape)
        dydx = heatmap[i, :]
        points = np.array([xx, y]).T.reshape(-1, 1, 2)
        # print(points.shape)
        segments = np.concatenate([points[:-1], points[1:]], axis=1)
        # print(segments.shape)
        norm = plt.Normalize(-1, 1)
        lc = LineCollection(segments, norm=norm, cmap='RdBu_r')
        lc.set_linewidth(2)
        lc.set_array(dydx)
        ax0.add_collection(lc)

    yttics = np.zeros(sampleChannel)
    for gi in range(sampleChannel):
        yttics[gi] = gi * thespan

    ax0.set_ylim([-thespan, thespan * sampleChannel])
    ax0.set_xlim([0, resampleLength + 1])
    # ax0.set_xticks([1, 200, 400, 600, 800, 1001])
    # ax0.set_xticks([1, 100, 200, 300, 384])
    ax0.set_xticks([1, 60, 120, 180, 202])
    # ax0.set_xticks([1, 30, 60, 90, 101])
    ax0.set_title('EEG signal map', y=-0.1)

    inversechannelnames = []
    for i in range(sampleChannel):
        inversechannelnames.append(channelnames[sampleChannel - 1 - i])
    plt.sca(ax0)
    plt.yticks(yttics, inversechannelnames)

    reverse_x_data = x_data[id].reshape(30, -1) ** 2  # 电压的平方表示当前区域的能量
    reverse_cam_data = grayscale_cam.reshape(30, -1)
    topox = np.mean(reverse_x_data, axis=1)
    topoHeatmap = np.mean(reverse_cam_data, axis=1)

    im, cn = mne.viz.plot_topomap(topox, pos=info, show=False, axes=ax11, res=1200, names=channelnames, outlines='head',
                                  cmap='RdBu_r',
                                  )
    im, cn = mne.viz.plot_topomap(topoHeatmap, pos=info, show=False, axes=ax12, res=1200, names=channelnames,
                                  outlines='head',
                                  cmap='RdBu_r',
                                  )
    ax11.set_title('PSD', y=-0.09)
    ax12.set_title('Grad-cam', y=-0.1)
    # fig.colorbar(im, ax=[ax11, ax12])
    fig.suptitle(
        'Subject:' + subject + '    ' + 'Label:' + state[(int(label))] + '    ' + 'Likelihood:' + str(
            out.detach().numpy().reshape(3)))

    ax21.set_title('Delta', y=-0.1)
    ax22.set_title('Theta', y=-0.1)
    ax23.set_title('Alpha', y=-0.1)
    ax24.set_title('Beta', y=-0.1)
    im, cn = mne.viz.plot_topomap(data=deltapower, pos=info, axes=ax21, names=channelnames,
                                  outlines='head', cmap='RdBu_r', show=False)
    im, cn = mne.viz.plot_topomap(data=thetapower, pos=info, axes=ax22, names=channelnames,
                                  outlines='head', cmap='RdBu_r', show=False)
    im, cn = mne.viz.plot_topomap(data=alphapower, pos=info, axes=ax23, names=channelnames,
                                  outlines='head', cmap='RdBu_r', show=False)
    im, cn = mne.viz.plot_topomap(data=betapower, pos=info, axes=ax24, names=channelnames,
                                  outlines='head', cmap='RdBu_r', show=False)
    # fig.colorbar(im, ax=[ax21, ax22, ax23, ax24])
    plt.savefig('save_fig/' + subject + '/' + str(ll) + '/' + str(mode) + '_' + str(id) + '.jpg')
    print(str(id) + ' finished.')


if __name__ == '__main__':
    # load data and model
    model_name = 'InceptSADNet'
    subject = 's53'
    data_path = 'data/raw/'
    mode = "cross"
    # mode = "inter"
    x = import_module('models.' + model_name)
    config = x.Config(dataset)
    config.batch_size = 1
    model = x.Model(config)
    config.f1_save_path += config.model_name + '/f1_' + subject + '_' + mode + '.ckpt'
    config.auc_save_path += config.model_name + '/auc_' + subject + '_' + mode + '.ckpt'
    # checkpoint = torch.load(config.auc_save_path)
    checkpoint = torch.load(config.f1_save_path)
    model.load_state_dict(checkpoint)
    model.eval()
    x_data = np.load(data_path + subject + '/' + subject + '_x.npy')
    y_data = np.load(data_path + subject + '/' + subject + '_y.npy')
    ids = [0, 1, 2, 3, 6, 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, 121, 122, 123, 124, 125, 128, 129, 152, 170, 171, 173, 174, 175, 176, 177, 215, 216, 217, 227, 234, 235, 236, 237, 238, 239, 240, 241, 242, 243, 244, 245, 246, ]
    # statistics
    # normal, poor, optimal, error, correct = get_predict(model, x_data, y_data)
    # print("error:")
    # print(error)
    # print(len(error))
    # print("optimal:")
    # print(optimal)
    # print(len(optimal))
    # print("poor:")
    # print(poor)
    # print(len(poor))
    # print("normal:")
    # print(normal)
    # print(len(normal))
    # print("correct:")
    # print(correct)
    # print(len(correct))

    for id in ids:
        save_visualization(id, model, subject, x_data, y_data, mode, "optimal")