rerunModels.py 11.3 KB
Newer Older
sjromuel's avatar
d  
sjromuel committed
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
import numpy as np
import tensorflow as tf
import matplotlib.pyplot as plt
import tkinter as tk
import os
import matplotlib
from tkinter import filedialog
from skimage import transform
import SimpleITK as sitk
import argparse
#import os
#import pydot
#from graphviz import Digraph
#import shutil
#from tensorflow.keras import layers, models
#from utils.dataLoader import *
from utils.other_functions import *
from nets.Unet import *

def main():
    models = ["12", "34", "56", "78", "910", "1112", "1314", "1516"]
    folder_path = "finalResults/complete_seg/ae_class_cv_seg/"
    #modeltype = "Cluster_"
Robin's avatar
d    
Robin committed
24
25
    #modeltype = "Class_"
    modeltype = "Cluster_class_"
sjromuel's avatar
d  
sjromuel committed
26
    #modeltype = "Unet_"
Robin's avatar
d    
Robin committed
27
    all3dhdds = []
sjromuel's avatar
d  
sjromuel committed
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

    ##################### U-Net #####################
    for fold in models:

        file_path = folder_path+"TPs"+fold+modeltype

        ### read out files ###
        weights = np.load(file_path+"model.npy", allow_pickle=True)
        [test_patients,
         val_patients,
         number_patients,
         img_path,
         shrink_data,
         newSize,
         lr,
         batch_size,
         num_epochs,
         e,
         augment,
         save_path,
         gt_type,
         filter_multiplier] = np.load(file_path+"params.npy", allow_pickle=True)
        #  autoencoder_model__e100_switchclass1024_nohiddenclusternet needs to comment out gt_type and val_patients
        print('Training Parameters:')
        print('-----------------')
        print('Number of Patients: ', number_patients)
        print('Number of epochs: ', num_epochs)
        print('Test Patient number: ', test_patients)
        print('Image Size: ', newSize)
        print('Filter Multiplier: ', filter_multiplier)
        print('Data Augmentation: ', augment)
        print('Learning rate: ', lr)
        print('Image Path: ', img_path)
        print('Save Path: ', save_path)
        print('GT Type:', gt_type)

        ### Load test patient
        if gt_type == "thresh" or gt_type == "ctthresh_gt":
            img_path = "data/npy_thresh/"
        else:
            img_path = "data/npy/"
        full_list = os.listdir(img_path)
        seg_list = os.listdir("data/npy/")

        X_img_list = []
        GT_img_list = []
        ytrue_img_list = []
        # thresh_img_list = []
        if "mr" in save_path:
            for elem in full_list:
                if elem.endswith("T1.gipl.npy") and (elem.startswith('P' + str(test_patients[0]).zfill(2)) or elem.startswith('P' + str(test_patients[1]).zfill(2))):
                    X_img_list.append(elem)
                    if gt_type == "ctthresh_gt":
                        X_img_list.append(elem)
                        X_img_list.append(elem)

                elif elem.endswith(gt_type + ".gipl.npy") and (elem.startswith('P' + str(test_patients[0]).zfill(2)) or elem.startswith('P' + str(test_patients[1]).zfill(2))):
                    GT_img_list.append(elem)
            for elem in seg_list:
                if elem.endswith("segmr.gipl.npy") and (elem.startswith('P' + str(test_patients[0]).zfill(2)) or elem.startswith('P' + str(test_patients[1]).zfill(2))):
                    ytrue_img_list.append(elem)
                    if gt_type == "ctthresh_gt":
                        ytrue_img_list.append(elem)
                        ytrue_img_list.append(elem)

        else:
            for elem in full_list:
                if elem.endswith("ct.gipl.npy") and (elem.startswith('P' + str(test_patients[0]).zfill(2)) or elem.startswith('P' + str(test_patients[1]).zfill(2))):
                    X_img_list.append(elem)
                    if gt_type == "thresh":
                        X_img_list.append(elem)
                        X_img_list.append(elem)

                elif elem.endswith(gt_type+".gipl.npy") and (elem.startswith('P' + str(test_patients[0]).zfill(2)) or elem.startswith('P' + str(test_patients[1]).zfill(2))):
                    GT_img_list.append(elem)
            for elem in seg_list:
                if elem.endswith("seg.gipl.npy") and (elem.startswith('P' + str(test_patients[0]).zfill(2)) or elem.startswith('P' + str(test_patients[1]).zfill(2))):
                    ytrue_img_list.append(elem)
                    if gt_type == "thresh":
                        ytrue_img_list.append(elem)
                        ytrue_img_list.append(elem)
        list.sort(X_img_list)
        list.sort(GT_img_list)
        list.sort(ytrue_img_list)
        print("Input Image List", X_img_list)
        print("GT Image List", GT_img_list)
        print("True Segmentation Image List", ytrue_img_list)


        for j in range(2):
            if gt_type == "thresh" or gt_type == "ctthresh_gt":
                X_img_npys = np.load(img_path + X_img_list[j*3])
                GT_img_npys = np.load(img_path + GT_img_list[j*3])
                ytrue_img_npys = np.load(img_path + ytrue_img_list[j*3])
                print(GT_img_list[j*3])
                print(GT_img_list[j * 3+1])
                print(GT_img_list[j * 3+2])

                X_img_npys = np.append(X_img_npys, np.load(img_path + X_img_list[j * 3+1]), axis=0)
                GT_img_npys = np.append(GT_img_npys, np.load(img_path + GT_img_list[j * 3+1]), axis=0)
                ytrue_img_npys = np.append(ytrue_img_npys, np.load(img_path + ytrue_img_list[j * 3+1]), axis=0)

                X_img_npys = np.append(X_img_npys, np.load(img_path + X_img_list[j * 3+2]), axis=0)
                GT_img_npys = np.append(GT_img_npys, np.load(img_path + GT_img_list[j * 3+2]), axis=0)
                ytrue_img_npys = np.append(ytrue_img_npys, np.load(img_path + ytrue_img_list[j * 3+2]), axis=0)

                print("Input shape: ", np.shape(X_img_npys))
                print("GT shape: ", np.shape(GT_img_npys))
                print("True Segm shape: ", np.shape(ytrue_img_npys))

            else:
                X_img_npys = np.load(img_path + X_img_list[j])
                GT_img_npys = np.load(img_path + GT_img_list[j])
                ytrue_img_npys = np.load(img_path + ytrue_img_list[j])

                print("Input shape: ", np.shape(X_img_npys))
                print("GT shape: ", np.shape(GT_img_npys))
                print("True Segm shape: ", np.shape(ytrue_img_npys))



            X_img_npys = transform.resize(X_img_npys, (X_img_npys.shape[0], newSize[0], newSize[1]), order=0,
                                          preserve_range=True, mode='constant', anti_aliasing=False,
                                          anti_aliasing_sigma=None)
            GT_img_npys = transform.resize(GT_img_npys, (GT_img_npys.shape[0], newSize[0], newSize[1]), order=0,
                                           preserve_range=True, mode='constant', anti_aliasing=False,
                                           anti_aliasing_sigma=None)
            ytrue_img_npys = transform.resize(ytrue_img_npys, (ytrue_img_npys.shape[0], newSize[0], newSize[1]),
                                              order=0,
                                              preserve_range=True, mode='constant', anti_aliasing=False,
                                              anti_aliasing_sigma=None)
            X_test = np.reshape(X_img_npys, (X_img_npys.shape[0], X_img_npys.shape[1], X_img_npys.shape[2], 1))
            GT_test = np.reshape(GT_img_npys, (GT_img_npys.shape[0], GT_img_npys.shape[1], GT_img_npys.shape[2], 1))
            ytrue = np.reshape(ytrue_img_npys,
                               (ytrue_img_npys.shape[0], ytrue_img_npys.shape[1], ytrue_img_npys.shape[2], 1))


            test_dataset = tf.data.Dataset.from_tensor_slices((X_test, ytrue))
            test_dataset = test_dataset.batch(batch_size=1)
            print(test_patients)
            TP_num = test_patients[j]

            ###################################################################################

            detailed_images = False
            npys3d          = True

            ###################################################################################



            test_loss = []
            test_loss_hdd = []
            test_loss_hdd2 = []
            y_pred3d = []
Robin's avatar
d    
Robin committed
183
            y_true3d = []
sjromuel's avatar
d  
sjromuel committed
184
185
186
187
            for features in test_dataset:
                image, y_true = features
                y_true = onehotencode(y_true)
                y_pred = Unet(image, weights, filter_multiplier, training=False)
Robin's avatar
d    
Robin committed
188
                
Robin's avatar
d    
Robin committed
189
                '''if y_pred3d == []:
Robin's avatar
d    
Robin committed
190
191
                        y_pred3d = y_pred[:,:,:,0].numpy()
                else:
Robin's avatar
d    
Robin committed
192
                    y_pred3d =  np.append(y_pred[:,:,:,0].numpy(), y_pred3d, axis=0)'''
sjromuel's avatar
d  
sjromuel committed
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209

                loss = dice_loss(y_pred, y_true)
                loss = tf.make_ndarray(tf.make_tensor_proto(loss))
                test_loss.append(loss)
                #print(test_loss)
                try:
                    y_true_np = np.squeeze(y_true[0, :, :, 0].numpy() > 0.5)
                    y_true_np = y_true_np.astype(np.float_)
                    pred_np = np.squeeze(y_pred[0, :, :, 0].numpy() > 0.5)
                    pred_np = pred_np.astype(np.float_)
                    hausdorff_distance_filter = sitk.HausdorffDistanceImageFilter()
                    hausdorff_distance_filter.Execute(sitk.GetImageFromArray(y_true_np), sitk.GetImageFromArray(pred_np))
                    test_loss_hdd.append(hausdorff_distance_filter.GetHausdorffDistance())

                    hausdorff_distance_filter2 = sitk.HausdorffDistanceImageFilter()
                    hausdorff_distance_filter2.Execute(sitk.GetImageFromArray(pred_np), sitk.GetImageFromArray(y_true_np))
                    test_loss_hdd2.append(hausdorff_distance_filter2.GetHausdorffDistance())
Robin's avatar
d    
Robin committed
210
211
212
213
214
215
216
217
218
                    
                    if y_pred3d ==[]:
                        y_pred3d = pred_np
                        y_true3d = y_true_np
                    else:
                        y_pred3d = np.append(pred_np, y_pred3d, axis=0)
                        y_true3d = np.append(y_true_np, y_true3d, axis=0)
                        
                    
sjromuel's avatar
d  
sjromuel committed
219
220
221
222
223
224
225
226
227
228
                except:
                    pass

            #plt.show()
            #print(test_loss)
            print("TestLoss Mean for P", test_patients[j], ": ", np.mean(test_loss))

            #print(test_loss_hdd)
            print("Hausdorff-Distance for P", test_patients[j],":", np.mean(test_loss_hdd))
            print("Hausdorff-Distance2 for P", test_patients[j], ":", np.mean(test_loss_hdd2))
Robin's avatar
d    
Robin committed
229
230
231
232
233
234
235
236
            
            #### Hausdorff 3D: 
            
            '''ytrue = tf.convert_to_tensor(ytrue)
            y_pred3d = tf.convert_to_tensor(y_pred3d)

            print(tf.shape(ytrue))
            y_true3d = np.squeeze(ytrue[:,:,:,0].numpy() > 0.5)
Robin's avatar
d    
Robin committed
237
            y_true3d = np.float_(y_true)
Robin's avatar
d    
Robin committed
238
239
240
241
            print(np.shape(y_true3d))
            print(type(y_true3d))
            print(type(y_pred3d))
            pred3d = np.squeeze(y_pred3d[:,:,:].numpy() > 0.5)
Robin's avatar
d    
Robin committed
242
            pred3d = np.float(pred3d)
Robin's avatar
d    
Robin committed
243
            print(np.shape(pred3d))
Robin's avatar
d    
Robin committed
244
            print(type(pred3d))'''
Robin's avatar
d    
Robin committed
245
            hausdorff_distance_filter = sitk.HausdorffDistanceImageFilter()
Robin's avatar
d    
Robin committed
246
            hausdorff_distance_filter.Execute(sitk.GetImageFromArray(y_true3d), sitk.GetImageFromArray(y_pred3d))
Robin's avatar
d    
Robin committed
247
            hdd3d = hausdorff_distance_filter.GetHausdorffDistance()
Robin's avatar
d    
Robin committed
248
249
250
251
252
253
            print("3D HDD for patient", test_patients[j], ":", hdd3d)
            all3dhdds.append(hdd3d)
    print("All 3d-HDDs: ", all3dhdds)
    print(folder_path)
    print(modeltype)
           
sjromuel's avatar
d  
sjromuel committed
254
255
256
      #####

if __name__ == "__main__":
Robin's avatar
d    
Robin committed
257
    main()