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
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
| import os
import pandas as pd
from torch.utils.data import Dataset
from torch.utils.data import DataLoader
from PIL import Image
from torchvision import transforms
from sklearn.model_selection import train_test_split
import torch.nn as nn
import torch.optim as optim
import torchvision.models as models
from collections import Counter
from tqdm import tqdm
import torch
import random
import numpy as np
import hashlib
class KneeXRayDataset(Dataset):
def __init__(self, dataframe, transform=None, cache_size=2000):
self.df = dataframe
self.transform = transform
self.cache={}
self.cache_size=cache_size
def __len__(self):
return len(self.df)
def __getitem__(self, idx):
if idx in self.cache:
return self.cache[idx]
row = self.df.iloc[idx]
image = Image.open(row['img_path']).convert('RGB')
label = int(row['KL'][0])
image = self.transform(image)
result=(image, label)
if len(self.cache) < self.cache_size:
self.cache[idx] = result
return result
if __name__ == '__main__':
device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
print(f'device:{device}')
seed=42
random.seed(seed); np.random.seed(seed); torch.manual_seed(seed)
if torch.cuda.is_available():
torch.cuda.manual_seed_all(seed)
exps={'MedicalExpert-I','MedicalExpert-II'}
KL_levels={'0Normal','1Doubtful','2Mild','3Moderate','4Severe'}
img_exp_KLs=[]
for root, dirs, files in os.walk(f'.{os.sep}Digital Knee X-ray Images'):
for name in files:
if name.endswith('.png'):
path_parts=set(root.split(os.sep))
img_path=os.path.join(root, name)
img_exp_KL=(img_path,(path_parts&exps).pop(),(path_parts&KL_levels).pop())
img_exp_KLs.append(img_exp_KL)
dataset_df = pd.DataFrame(img_exp_KLs,columns=['img_path','exp','KL'])
dataset_df.sort_values(by='img_path',inplace=True)
dataset_df['filename'] = dataset_df['img_path'].str.split(os.sep).str[-1]
dataset_df_unique = dataset_df.drop_duplicates(subset=['KL','filename']).reset_index(drop=True)
data_transform_train = transforms.Compose([
transforms.Resize((224, 224)),
transforms.RandomHorizontalFlip(),
transforms.RandomRotation(15),
transforms.ColorJitter(brightness=0.2, contrast=0.2),
transforms.ToTensor(),
transforms.Normalize(mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225])
])
data_transform_val = transforms.Compose([
transforms.Resize((224, 224)),
transforms.ToTensor(),
transforms.Normalize(mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225])
])
X = dataset_df_unique
y = dataset_df_unique['KL']
X_train_val, X_test, y_train_val, y_test = train_test_split(
X, y, test_size=0.2, random_state=seed, stratify=y)
X_train, X_val, y_train, y_val = train_test_split(
X_train_val, y_train_val, test_size=0.25, random_state=seed, stratify=y_train_val)
class_counts=Counter(y_train)
all_counts=len(y_train)
class_weights={cls:all_counts/cnt for cls,cnt in class_counts.items()}
class_weights=torch.tensor(list(class_weights.values()),dtype=torch.float32).to(device)
train_dataset = KneeXRayDataset(dataframe=X_train, transform=data_transform_train)
val_dataset = KneeXRayDataset(dataframe=X_val, transform=data_transform_val)
test_dataset = KneeXRayDataset(dataframe=X_test, transform=data_transform_val)
train_loader = DataLoader(train_dataset, batch_size=64, shuffle=True,
num_workers=4,pin_memory=True,persistent_workers=True)
val_loader = DataLoader(val_dataset, batch_size=64, shuffle=False,
num_workers=4,pin_memory=True,persistent_workers=True)
test_loader = DataLoader(test_dataset, batch_size=64, shuffle=False)
train_dataset_hash=hashlib.md5(''.join(X_train['filename'].to_list()).encode('utf-8')).hexdigest()
val_dataset_hash=hashlib.md5(''.join(X_val['filename'].to_list()).encode('utf-8')).hexdigest()
print(f'train_dataset_hash:{train_dataset_hash},val_dataset_hash:{val_dataset_hash}')
model=models.resnet50(pretrained=True)
num_ftrs = model.fc.in_features
model.fc = nn.Linear(num_ftrs, 5)
save_epoch=100
latest_model_path=f'.{os.sep}models{os.sep}model_{train_dataset_hash}_epoch{save_epoch}.pth'
while(os.path.exists(f'.{os.sep}models{os.sep}model_{train_dataset_hash}_epoch{save_epoch}.pth')):
latest_model_path=f'.{os.sep}models{os.sep}model_{train_dataset_hash}_epoch{save_epoch}.pth'
save_epoch+=100
if os.path.exists(latest_model_path):
model.load_state_dict(torch.load(latest_model_path,map_location=device))
save_epoch-=100
print(f'Loaded latest model from {latest_model_path}')
else:
save_epoch=1
model.to(device)
criterion = nn.CrossEntropyLoss(weight=class_weights)
optimizer = optim.Adam(model.parameters(), lr=1e-5, weight_decay=1e-4)
scheduler = optim.lr_scheduler.ReduceLROnPlateau(optimizer, 'min', patience=10, factor=0.5, min_lr=1e-7)
num_epochs = 2000
print(f'Training for {num_epochs} epochs')
best_val_loss=float('inf')
early_stopping_patience = 50
patience_counter = 0
for epoch in range(save_epoch,num_epochs+save_epoch+1):
model.train()
running_loss = 0.0
for inputs, labels in tqdm(train_loader, desc=f'Epoch {epoch}/{num_epochs}'):
inputs, labels = inputs.to(device), labels.to(device)
optimizer.zero_grad()
outputs = model(inputs)
loss = criterion(outputs, labels)
loss.backward()
optimizer.step()
running_loss += loss.item()
model.eval()
val_loss=0.0
correct = 0
total = 0
with torch.no_grad():
for inputs, labels in tqdm(val_loader, desc=f'Epoch {epoch}/{num_epochs}'):
inputs, labels = inputs.to(device), labels.to(device)
outputs = model(inputs)
val_loss += criterion(outputs, labels).item()
_, predicted = torch.max(outputs.data, 1)
total += labels.size(0)
correct += (predicted == labels).sum().item()
epoch_val_loss = val_loss/len(val_loader)
scheduler.step(epoch_val_loss)
print(f"Epoch {epoch}/{num_epochs}, Loss: {running_loss/len(train_loader)}, Val Loss: {epoch_val_loss}, Val Accuracy: {100*correct/total}%, lr: {optimizer.param_groups[0]['lr']}")
with open(f'.{os.sep}models{os.sep}best_model_{train_dataset_hash}_on_{val_dataset_hash}.log','a') as f:
f.write(f"epoch:{epoch},train_loss:{running_loss/len(train_loader)},val_loss:{epoch_val_loss},val_acc:{100*correct/total}%,lr:{optimizer.param_groups[0]['lr']}\n")
if (epoch) % 100 == 0:
torch.save(model.state_dict(), f'.{os.sep}models{os.sep}model_{train_dataset_hash}_epoch{epoch}.pth')
if epoch_val_loss < best_val_loss:
best_val_loss = epoch_val_loss
torch.save(model.state_dict(), f'.{os.sep}models{os.sep}best_model_{train_dataset_hash}_on_{val_dataset_hash}.pth')
with open(f'.{os.sep}models{os.sep}best_model_{train_dataset_hash}_on_{val_dataset_hash}.log','a') as f:
f.write(f'epoch:{epoch},saved best model\n')
patience_counter=0
else:
patience_counter+=1
if patience_counter >= early_stopping_patience:
print(f'Early stopping triggered at epoch {epoch}')
with open(f'.{os.sep}models{os.sep}best_model_{train_dataset_hash}_on_{val_dataset_hash}.log','a') as f:
f.write(f'epoch:{epoch},early stopping\n')
break
|
.png)
