WMH-DualTasker/evaluate_study.py at main · hzlab/WMH-DualTasker · 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
# generate a report for the given study, I want:
# for each model from one split:
# 1. scatter plots (both validation and test)
# 2. r2 score (correlation)
# 3. MSE score

from mypath import experiment_path
import os,re,datetime, torch, pickle
import numpy as np
import matplotlib.pyplot as plt
from sklearn.metrics import r2_score, mean_squared_error
from medcam import medcam
import nibabel as nib
import glob
import torch.nn.functional as F
import skimage.transform as skTrans

def get_model(model_type:str, task='reg'):

    if model_type == 'SFCN':
        from model.sfcn import SFCN
        model = SFCN(task=task)
    if model_type == 'SFCN4':
        from model.sfcn4 import SFCN_4
        model = SFCN_4(task=task)
    if model_type == 'SFCN3':
        from model.sfcn3 import SFCN_3
        model = SFCN_3(task=task)

    elif model_type == '3DResNet':
        from model.resnet import resnet50
        model = resnet50()
    elif model_type == 'ViT':
        if task == 'cls':
            from monai.networks.nets import ViT
            model = ViT(in_channels=1, img_size=(256, 256, 64), patch_size=16, classification=True, num_classes=30)
        else:
            from model.ViT_reg import ViT_reg
            model = ViT_reg(in_channels=1, img_size=(256, 256, 64), patch_size=16, classification=False)
    elif model_type == 'UNet':
        if task == 'reg':
            from model.unet import UNet_reg
            model = UNet_reg()

    return model

def get_preds(model, data, device, task='reg'):
    # return pred, given model (with weight) and data
    preds = []
    num_subjects = data.shape[0]
    model = model.to(device)
    model.eval()

    with torch.no_grad():
        for i in range(num_subjects):
            temp_input = torch.from_numpy(np.expand_dims(data[i], [0, 1]))
            temp_input = temp_input.to(device)

            if task == 'reg':
                temp_output = model(temp_input.float()).view(-1).detach().cpu().numpy()[0]
                preds.append(temp_output)
            else:
                temp_output = model(temp_input.float()).view(-1).detach().cpu().numpy()
                centers = np.arange(0.5, 30.5)
                preds.append(sum(temp_output*centers))

    return np.array(preds)

def evaluate_study_regression(study_name:str):
    # given study name,

    # select device
    device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
    # device = torch.device("cpu")
    # find the model type from config.txt
    if os.path.exists(experiment_path + '/{}/config.txt'.format(study_name)):
        with open(experiment_path + '/{}/config.txt'.format(study_name), 'r') as f:
            model_type = re.search(r'model=\'\w+\'', f.read()).group(0)[7:-1]
    else:
        raise("config file not found error")

    with open(experiment_path + '/{}/report.txt'.format(study_name), 'a') as f:
        f.write('log time: {}\n'.format(datetime.datetime.now()))
        f.write('exp name: {}\n'.format(study_name))
        f.write('model name: {}\n'.format(model_type))

        for (dirpath, dirnames, filenames) in os.walk(experiment_path + '/{}'.format(study_name)):
            for split in dirnames:
                print(os.path.join(dirpath, split))
                f.write('\n{}\n'.format(split))

                if os.path.exists(experiment_path + '/{}/{}/best_model.pth'.format(study_name, split)):
                    model = get_model(model_type)
                    model.load_state_dict(torch.load(experiment_path + '/{}/{}/best_model.pth'.format(study_name, split), map_location=device))
                    model.eval()

                    # load validation dataset
                    flair_val = np.load('/mnt/isilon/CSC4/HelenZhouLab/HZLHD1/Data4/Members/yileiwu/arwmc_reg/preprocessed_data/{}_val.npy'.format(split))
                    mask_val = pickle.load(open('/mnt/isilon/CSC4/HelenZhouLab/HZLHD1/Data4/Members/yileiwu/arwmc_reg/preprocessed_data/{}_val_target.p'.format(split), 'rb'))

                    # load test dataset
                    flair_test = np.load("/mnt/isilon/CSC4/HelenZhouLab/HZLHD1/Data4/Members/yileiwu/arwmc_reg/preprocessed_data/test.npy")
                    mask_test = pickle.load(open('/mnt/isilon/CSC4/HelenZhouLab/HZLHD1/Data4/Members/yileiwu/arwmc_reg/preprocessed_data/test_target.p', 'rb'))
                    # get prediction
                    preds_val = get_preds(model, flair_val, device)
                    preds_test = get_preds(model, flair_test, device)
                    # calcuate MSE and r2
                    mse_score_val = mean_squared_error(mask_val, preds_val)
                    mse_score_test = mean_squared_error(mask_test, preds_test)
                    r2_score_val = r2_score(mask_val, preds_val)
                    r2_score_test = r2_score(mask_test, preds_test)
                    f.write("Mean square Error for Validation Set is : {}\n".format(mse_score_val))
                    f.write("Mean square Error for Test Set is : {}\n".format(mse_score_test))
                    f.write("R2 score for Validation Set is : {}\n".format(r2_score_val))
                    f.write("R2 score for the test Set is: {}\n".format(r2_score_test))
                    f.write("GT for Val: {}\n".format(mask_val))
                    f.write("Preds for Val: {}\n".format(preds_val))
                    f.write("GT for Test: {}\n".format(mask_test))
                    f.write("Preds for Test: {}\n".format(preds_test))
                    # plot figure
                    plt.scatter(mask_val.astype(float), preds_val, marker="o", label="mse={:.3f}, r2={:.3f}".format(mse_score_val, r2_score_val))
                    plt.scatter(mask_test.astype(float), preds_test, marker="o", label="mse={:.3f}, r2={:.3f}".format(mse_score_test, r2_score_test))
                    plt.legend()
                    plt.savefig(experiment_path + '/{}/{}.png'.format(study_name, split))
                    plt.clf()

                else:
                    f.write("model not found !!! \n")

    pass


# WIELD ISSUE: CANNOT WORK WITH GPU
def generate_CAM(study_name:str, fold:str):
    # generate CAM for all test subjects given study

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

    # load test dataset
    flair_test = np.load("/mnt/isilon/CSC4/HelenZhouLab/HZLHD1/Data4/Members/yileiwu/arwmc_reg/preprocessed_data/test.npy")
    mask_test = pickle.load(open('/mnt/isilon/CSC4/HelenZhouLab/HZLHD1/Data4/Members/yileiwu/arwmc_reg/preprocessed_data/test_target.p', 'rb'))

    # find the model type from config.txt
    if os.path.exists(experiment_path + '/{}/config.txt'.format(study_name)):
        with open(experiment_path + '/{}/config.txt'.format(study_name), 'r') as f:
            model_type = re.search(r'model=\'\w+\'', f.read()).group(0)[7:-1]
    else:
        raise("config file not found error")

    layer='feature_extractor'
    model_weight_path = experiment_path + '/{}/fold{}/best_model.pth'.format(study_name, fold)
    if os.path.exists(model_weight_path):
        model = get_model(model_type)
        model.load_state_dict(torch.load(model_weight_path, map_location=device))
        model = medcam.inject(model, output_dir=experiment_path+'/{}/fold{}/cam'.format(study_name, fold),layer=layer,  save_maps=True)
    model.eval()
    model = model.to(device)

    # get prediction
    get_preds(model, flair_test, device, task='reg')

    if not os.path.exists(experiment_path+'/{}/fold{}/cam/resized_cam'.format(study_name, fold)):
        os.mkdir(experiment_path+'/{}/fold{}/cam/resized_cam'.format(study_name, fold))

    if not os.path.exists(experiment_path+'/{}/fold{}/cam/flair_test'.format(study_name, fold)):
        os.mkdir(experiment_path+'/{}/fold{}/cam/flair_test'.format(study_name, fold))

    for each in glob.glob(experiment_path + '/{}/fold{}/cam/{}/*'.format(study_name, fold, layer)):
        temp_array = nib.load(each).get_fdata()
        resized_array = skTrans.resize(np.transpose(temp_array, (2, 0, 1)), (256, 256, 64))

        print(resized_array.shape)
        temp_image = nib.Nifti1Image(resized_array, affine=np.eye(4))
        temp_image.to_filename(experiment_path+'/{}/fold{}/cam/resized_cam/{}'.format(study_name, fold, each.split('/')[-1]))

    for i in range(flair_test.shape[0]):
        temp_array = flair_test[i]
        temp_image = nib.Nifti1Image(temp_array, affine=np.eye(4))
        temp_image.to_filename(experiment_path+'/{}/fold{}/cam/flair_test/{}.nii.gz'.format(study_name, fold, i))

# evaluate_study_regression('sfcn_mse_1e-3')
# evaluate_study_regression('sfcn4_mse_1e-3')
# evaluate_study_regression('sfcn3_mse_1e-3')

generate_CAM(study_name='sfcn3_mse_1e-3', fold='0')